264 Commits

Author	SHA1	Message	Date
Logan Johnson	a2d28eb6c4	fix(relay): apply channel setting changes atomically and lock every membership writer (#8516) ... 🤖 ## Summary This PR fixes two ways the relay could report a channel change that did not fully happen. **1. Channel settings could be saved only in part, while the app was told that they all worked.** When an owner or admin changed a channel's settings (name, description, topic, public or private, message lifetime, archived or not, who can join), the relay saved the request first and then applied each setting with its own database write. If one write failed, the relay only logged the error. It skipped the remaining settings and still reported success. Sending the request again did nothing, because the relay saw it as already saved. Example: a switch from public to private could fail while the app showed success, and the channel stayed public. Now saving the request, applying every setting and saving the "settings changed" messages in the channel happen in one database transaction. Either all of it is saved or none of it is: - If anything fails, nothing is saved and the client gets an error. It can send the same request again. - A repeat of a request that already succeeded is reported as a duplicate and changes nothing. This is also true when the repeat arrives while the first copy is still being saved. - A request to restore a channel that is not archived, or to change an archived channel without restoring it, is refused and nothing is saved. Before, the relay saved the request and a "settings changed" message, and then the restore failed. - Live updates go out only after the save commits. These are: removing access for people who lose it, and pushing the new messages and channel info to clients. **2. A published member list could include someone who had just been removed.** For some channels the relay publishes a list of the channel's members. While it builds that list, it holds a per-channel membership lock. Kicks by a relay admin and the automatic kick after a deploy did not take that lock. A kick could commit after the list was read, so the published list still showed the removed member. Channel deletion also did not wait for the lock. Kicks, channel deletion and the automatic add when someone joins a Huddle (voice call) now all take the lock first. The published list then includes the change, or the change waits until the list is saved. ## Details - **One lock order for all channel settings writes.** Before, each writer took its own locks. Review kept finding writers that took them in a different order or forgot one. `lock_channel_for_settings_write` (`crates/buzz-db/src/store/channel.rs`) is now the only place that takes them. The order is: 1. the membership lock, only when the public/private setting or the "who can join" setting (`join_role`) changes. Without it, a member add could read the old rule while the change commits. 2. the message-lifetime (TTL) lock: exclusive for a lifetime change, shared otherwise. 3. the channel row (`FOR NO KEY UPDATE`). This does not block the key lock that a member insert takes. The row lock also orders two copies of the same request, so the second copy sees the first as already stored. - The function returns a `ChannelSettingsWriter`. The writer borrows the transaction that took the locks, so the compiler stops it from being used with another transaction or after commit. The settings writes (`update_channel_on`, `set_topic_on`, `set_purpose_on`, `set_archived_on`) require the writer. They refuse a change its locks do not cover. - `Db::apply_channel_metadata_event` stores the channel-edit event (kind:9002) and its mentions, applies the settings and stores the "settings changed" messages (kind:40099) in one transaction. The older single-purpose writes (`update_channel`, `set_topic`, `set_purpose`, `archive_channel`, `unarchive_channel`) use the same helpers and behave as before. - `ingest.rs` and `validate_admin_event`: channel edits no longer check archived/active state before the write. The transaction checks it after its duplicate check, under the row lock. A refusal there goes back to the client as `invalid:`. Channel edits have their own save path and skip the general post-save step. `handle_side_effects` refuses them, so no caller can go around the transaction. - `side_effects.rs`: `emit_system_message` is split into signing and sending. The messages are signed before the transaction and sent after it. All "settings changed" messages from one edit share one timestamp and come in a fixed order (topic, purpose, visibility, lifetime, archive). - A change to "who can join" (`join_role`) is applied in the same transaction and announced with its own `join_role_changed` message. - **Membership lock outside settings** (`relay_admin_actions.rs`, `channel_members.rs`, `audio/handler.rs`): both kick paths (`execute_kick_with_marker`, `deploy_kick_member`) take the membership lock before they remove the member. The marker kick still undoes the removal if its lease check fails after waiting. The Huddle join takes its locks through a new `ChannelMembershipLocks` token, and the auto-add is a method on that token, so it cannot run without the lock. `soft_delete_channel` takes the full settings lock order. - **Limit of the membership lock:** it covers member changes, public/private and "who can join" changes, and channel deletion. It does not cover archiving: archive writes and the cleanup of expired temporary channels do not take it. The cleanup keeps its single UPDATE. A comment in `channel.rs` explains why that is safe and why it must not be split into a read and a write. ### Related issue Follow-up to the review of #8143 (which added the "who can join" setting). That review found that channel edits could be applied in part. No separate issue. ### Testing New PostgreSQL tests in `crates/buzz-relay/src/handlers/edit_metadata_postgres_tests.rs`: - `failed_edit_stores_nothing`: a test trigger makes one write fail partway through the transaction. Nothing is stored or changed, on the first send or a resend. Without the trigger, the same signed request applies in full. - `resent_edit_is_a_duplicate`, `resent_archive_and_unarchive_are_duplicates`: a resend after success is a duplicate and adds nothing. - `resend_validated_after_the_first_commit_is_a_duplicate`: a resend validated after its first copy committed is a duplicate, with no second change and no second message. - `concurrent_copies_of_a_restore_apply_once`: two copies of one restore, held at the channel lock until both are waiting, apply once. The other copy is a duplicate. - `unarchive_of_an_active_channel_is_refused`, `join_role_edit_is_applied_and_announced`. Lock tests in buzz-db: each one holds a member list in progress and checks that the writer waits for it. They cover a "who can join" change (`channel_members.rs`), the deploy kick and the marker kick (`relay_admin_actions.rs`), and channel deletion (`channel.rs`). The kick and delete tests fail when the new locks are removed. These tests and the existing channel-edit tests in `group_state_postgres_tests` pass locally against a migrated PostgreSQL database. --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-22 10:32:43 -04:00
Luke Tornquist	2f2f6e2774	fix(db): bound every statement and the #p mention scan (#8461) ... ## Problem An agent `#p` + `#h` + `since` REQ (the `require_mention` subscription in `buzz-acp`) ran for over 45 minutes on a production writer and held `AccessShare` on every `events` partition. There were two causes: 1. **The mention side of the `#p` join was unbounded.** `build_query_events_sql` joined `event_mentions` on `(community_id, event_id)` only. The planner picked a nested loop whose outer side was every mention of the pubkey, ever, and whose inner side range-scanned the channel index from `since` on each partition. Cost ≈ total mentions × channel rows since `since`. 2. **Nothing cancelled it.** The writer `statement_timeout` defaulted to `0`. Only `#e` reads had a deadline (the 20s cap from #7854), and read-replica sessions had none at all. `pg_stat_statements` shows this shape has a multi-thousand-second max going back months, so it is chronic and was not introduced by a recent change. ## Change **`#p` mention bound** (`store/event.rs`, `push_mention_join`, shared by the page SELECT and COUNT) - Join on the full event key, `AND e.created_at = m.event_created_at`. Mention rows copy the event's `created_at` at insert (migration 0007 already relies on this equality), so the join can prune `events` partitions by time. - Repeat `since`/`until` on `m.event_created_at`, so the outer scan is a range on the existing `idx_event_mentions_pubkey_created`. `until` is a superset bound for the keyset cursor; the exact `(created_at, id)` cursor test stays on `e`. - Repeat `kinds` on `m.event_kind` when the filter has kinds. Every production mention writer binds the event kind (`insert_mentions` in `store/runtime/mod.rs`, and the replaceable-event path in `store/replaceable.rs`, back to the first Postgres schema), and no migration inserts mention rows, so this drops no rows written by the relay. Rows imported out of band with a NULL `event_kind` would no longer match a kinds-filtered `#p` read; a read-only production count on 2026-10-20 found 0 such rows on both production reader databases. The pushdown makes `idx_event_mentions_pubkey_kind_created` available to the planner for the `require_mention` kind-9 shape; its use has not been observed (see below). An E2E fixture with a hand-set NULL `event_kind` row confirmed the caveat: base returns 8 events, head 7. - The feed's mentions and needs-action queries (`store/feed.rs`, `build_mentions_query` / `build_needs_action_query`) join `event_mentions` the same way, so each joined row probes one `events` partition instead of all of them. Their `ORDER BY m.event_created_at ... LIMIT` already kept them from running away; this removes the per-row cross-partition probe. - Observed plan (E2E at `9cb6638b1`, before the `kinds` pushdown; 60,000 historical mentions + 2,500 recent decoys, stock planner, `EXPLAIN (ANALYZE, BUFFERS)` of the captured production SQL): the mention side is a bounded range on `idx_event_mentions_pubkey_created` (13 rows), then a hash join with a scan of the recent `events` partition. Base scanned 2,509 recent events and probed `event_mentions_pkey` 2,509 times. Head: **1.1 ms / 319 shared hits**; base: **8.5 ms / 10,049 shared hits**. Not measured against the production replica. A fresh `EXPLAIN (ANALYZE, BUFFERS)` replay at `a7065c367`, after the pushdown, ran on a smaller fixture where the planner chose a sequential mention scan, so it neither confirms nor replaces these numbers. **Reads: 20s deadline on every generic event read** - `query_events_on` and `count_events_on` always run under the transaction-local 20s deadline that `#e` reads used (`fetch_with_read_deadline`, formerly `fetch_with_e_tag_deadline`). It only tightens: shorter operator or caller caps are kept. - Read-replica sessions get a 20s session default through `after_connect`, using the same tighten-only SQL. - Cost: every generic REQ history read and COUNT now pays BEGIN + `set_config` + ROLLBACK (three extra round trips; before, only `#e` reads did). In the local E2E the end-to-end history read was still faster at head (median 4.1 ms vs 10.2 ms at base, 9 samples each, warm cache), so the round trips are dominated by the bounded plan; this does not isolate them or establish production p99. The round trips also lengthen how long each such read holds its writer connection; that hold time has not been measured against production pool sizing. - A replica read cancelled at its deadline (57014) is returned, not re-run on the writer. Usually the cancel means expensive work the writer would only repeat. It can also mean the replica waited on a lock only it holds (a replayed `ACCESS EXCLUSIVE` from primary DDL), when the writer would have answered quickly; that miss is the accepted price of keeping runaway reads off the primary. The rationale now lives once on `DbError::is_statement_cancelled` instead of in a dozen copied comments. The generic routed reads and `ReadSession` already did this; the eight routed reads with their own SQL (feed mentions, needs-action and activity; thread replies; channel window; relay membership; thread window and its auxiliary scan) now do too. A cancelled auxiliary scan also keeps its replica session instead of degrading it to the writer for the rest of the request. Other replica failures (e.g. hot-standby recovery conflicts) still fall back to the writer. - Paths with their own tighter `SET LOCAL` (artifact 2–5s, thread window 4s, personal reads, usage 5–15s, serving lease 1.5s) are unchanged. **Writes: every writer session is bounded** - The default `statement_timeout` changes from `0` to 120s (`DEFAULT_STATEMENT_TIMEOUT_MS`). - `BUZZ_DB_STATEMENT_TIMEOUT_MS` still overrides it. `0` or any value above 300s installs 300s (`MAX_STATEMENT_TIMEOUT_MS`), so a session can no longer disable it, and a clamped value logs a `warn!` when the pool is built. Only the operator-approved, fenced destruction paths below lift it, per transaction. **Long-running writer paths, audited** | Path | Behavior | |---|---| | Migrations / schema destruction (`with_exclusive_schema_destruction_lock`) | Unchanged: these lift lock and statement timeouts on their own detached connection | | Partition maintenance | Unchanged: own 5s `statement_timeout` | | Partition audit (`buzz-admin`) | Unchanged: 30s | | Community deletion inventory, purge and logical verification (`inventory_schema` whole-tenant counts, `purge_postgres` whole-community `DELETE` per table in one tx, `verify_postgres_logically_deleted` whole-tenant `EXISTS` per table right after the purge) | Lifted (`statement_timeout = 0`) transaction-locally by `lift_statement_timeout`, like migrations; lock and idle timeouts still bound them. Inventory counts each table in its own lifted transaction, so each table's lock and snapshot are released as soon as its count finishes, as the autocommit counts did. A cancelled purge would retry from scratch and eventually block, so a cap here would turn large-tenant deletion into an operator override | | Audit appends (`buzz-audit`, audit pool inherits the writer cap) | Waits on the audit-chain advisory lock. With `BUZZ_DB_LOCK_TIMEOUT_MS=0` (or at least the statement cap), `statement_timeout` (57014) cancels the wait instead of `lock_timeout` (55P03). The relay's audit worker now retries both with the same backoff instead of dropping the entry on 57014; admission refusal stays terminal | On HTTP routes a statement timeout already maps to the distinct retryable 503 (`api/mod.rs`, `statement_timeout_maps_to_distinct_503`). A cancelled REQ fails its history read instead of hanging. ## Tests - `p_tag_query_bounds_mention_side_by_time` (unit): page and COUNT SQL carry the `created_at` join and the `since`/`until`/`kinds` mention bounds, and the bounds are absent when the filter has none. - `p_tag_read_with_mention_bounds_keeps_results` (Postgres): through the production mention insert, unbounded, `since`, window and keyset-cursor reads return exactly the mentioned rows, and COUNT matches. Every bound sits exactly on a mentioned row's second and the cursor page includes a same-second sibling after the cursor id, so an off-by-one on either mention-side bound fails it. - `plain_read_is_cancelled_at_deadline_when_timeout_disabled` / `count_events_cancels_at_read_deadline_when_timeout_disabled` (Postgres): the production `#p` + `#h` + `since` shape and a plain COUNT are cancelled at 20s with the session timeout disabled. Previously these covered `#e` only. - `writer_statement_timeout_is_always_bounded` (unit) and `session_statement_timeouts_bound_writer_and_reader_sessions` (Postgres): through `Db::new`, writer sessions install `2min` / `5min` (for `0` and for 600s) / `45s`, and replica sessions install `20s`. - `deletion_whole_tenant_stages_outlive_the_writer_statement_timeout` (Postgres): on a scratch desired-state database with a 500ms writer cap, the real `inventory_schema`, `purge_postgres` and `verify_postgres_logically_deleted` each wait behind a held `ACCESS EXCLUSIVE` lock on `events` for over 1s and still complete, a plain statement is cancelled (57014), and pooled sessions still read `500ms` afterwards. A lifted transaction reads `statement_timeout = 0`, so a bounded lift (e.g. `'300000'`) fails it, as does removing any of the three `lift_statement_timeout` calls. - `mentions_query_is_tenant_scoped_and_joins_mentions_by_composite_key` / `needs_action_query_is_tenant_scoped_and_joins_mentions_by_composite_key` (unit): both feed queries carry the `e.created_at = m.event_created_at` join. - `replica_statement_cancel_propagates_without_writer_rerun` (Postgres): with the replica locked past its cap, each of the six routed reads above returns 57014 instead of the writer's copy. Removing the cancel arm in any of them fails it. - `thread_window_statement_cancel_propagates_without_writer_rerun` (Postgres): a thread-window read parked behind a replica lock and cancelled (57014; the window's own 1s `lock_timeout` would otherwise fire first) returns the cancel instead of the unlocked writer's window, and an auxiliary scan timed out on a replica session returns 57014 and leaves the session on the replica. Removing either guard fails it. - `audit_worker_retries_statement_timeout_until_original_entry_is_appended_once` (Postgres, `buzz-relay`): the production audit worker, with `lock_timeout` disabled and a 100ms statement cap, keeps the entry through repeated 57014 cancels while the chain lock is held and appends it exactly once after release. Treating only 55P03 as retryable fails it. - The existing writer and audit-pool GUC tests now expect `5min` for a disabled statement timeout. ## After deploy, watch - `buzz_db_route_decision{decision="replica",reason="statement_cancelled"}`: replica reads that now fail instead of falling back to the writer, including the replica-only lock case above. - `buzz_db_pool_acquire_wait_seconds` and `buzz_db_pool_waiters` on the writer pool: every generic REQ history read and COUNT now holds its connection for three extra round trips. ## Database checklist (`docs/database.md` § Review checklist) This PR changes schema-dependent `buzz-db` queries (the `#p` join and its `m.event_kind` pushdown). It does not change the schema. - **Phase:** none. No schema change, so this is not an expand, backfill or contract step. Nothing before or after it is required. - **Triggers, functions, procedures, rules, foreign keys:** none added. - **Row mutations / backfill:** no migration, and no row is mutated or backfilled. The one data-shape dependency is mention rows with a NULL `event_kind`: a kinds-filtered `#p` read no longer matches them (see the `kinds` bullet under Change). Relay writers always bind `event_kind`, so only rows imported out of band are affected. A read-only production count on 2026-10-20 found 0 such rows on both production reader databases. - **Contract migration:** N/A. - **Rule 4 locations:** none touched. No table, index or desired-state file changes, so no parity test applies. The new bounds use the existing `idx_event_mentions_pubkey_created` and `idx_event_mentions_pubkey_kind_created`. - **Deployed relay vs. new schema:** there is no new schema. The currently deployed relay and this relay run against the same tables. - **Lock and statement timeouts:** this PR is about them. Generic reads get a 20s deadline, replica sessions a 20s default, and writer sessions a 120s default capped at 300s. Community-deletion transactions lift `statement_timeout` per transaction; migrations and schema destruction are unchanged (see "Long-running writer paths, audited"). No index builds and no operator prebuild step. ## What got simpler - One read-deadline path instead of an `#e`-only branch in `query_events_on` and `count_events_on`. - The `#p` join SQL that was duplicated between SELECT and COUNT now lives in one `push_mention_join`. - The deletion tests share one scratch desired-state database helper instead of an inline copy. - "Is this query bounded?" now has one answer: yes, unless it runs in a migration, schema destruction, or community-deletion transaction. ## Identified debt (not in this PR) - `buzz-acp` re-sends the same filter after a timeout `CLOSED` (no backoff or narrowing). With the `#p` bound the `require_mention` read should finish well inside 20s, so this is a follow-up, not a blocker. --------- Signed-off-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz> Co-authored-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz>	2026-10-21 19:35:03 -04:00
Logan Johnson	d97f6b4c6e	feat(relay): add announce channels with a join_role setting (#8143) ... 🤖 ## Summary Channels are Buzz's main way to control access, but today they cannot separate reading from writing. In an open channel, anyone who can read can also write. So there is no way to say "many people read, a few people write". Announcements, release feeds, plugin release records and shared configuration all need that. This PR adds **announce channels**: anyone can follow and read them, but only members can post. It does this with one new channel setting and without changing what any role means. - **Roles mean the same thing in every channel.** A *guest* can read and leave, but cannot write. A *member* (or *bot*) can write. *Admins* and *owners* also manage the channel. The relay already had the guest role, but no client used it until now. - **A new setting, `join_role` (`member` | `guest`), picks the role a new person gets** when nobody chooses one: when they join by themselves, or when someone adds them without a role. The default, `member`, keeps today's behavior for every existing channel. - **An open channel with `join_role = guest` is an announce channel.** Anyone can find, read and join it. People who join are guests, so they cannot write until someone makes them a member. The relay enforces these rules, so a client cannot get around them. What each role can do (the same in every channel): | Action | Guest | Member or bot | Admin / owner | | --- | --- | --- | --- | | Read, leave | ✅ | ✅ | ✅ | | Post, reply, edit, delete own, react, artifacts, canvas, forum, pins, topic, typing, workflow messages, save or run workflows, huddles | ❌ | ✅ | ✅ | | Add people | ❌ | ✅ as member, guest or bot | ✅ any role | | Make a guest a member (person or agent) | ❌ | ✅ | ✅ | | Any other role change | ❌ | ❌ | ✅ | | Change name, about, visibility, TTL, `join_role` | ❌ | ❌ | ✅ | Other rules: - **Joining by yourself.** A person who joins (kind:9021), or adds themselves (kind:9000 with their own key), gets the channel's `join_role`. They may ask for `join_role` or a lower role, never a higher one. So in an announce channel nobody can join as a member by themselves. - **Non-members.** A non-member may write only where joining would let them write: an open channel with `join_role = member`. That is today's behavior for open channels. - **Agents.** Agents follow the same rules as people. Any member who can write may make a guest agent a member. Limits that only an agent's owner may lift belong to the later IFC work (its "restricted" agent level), not to this PR. That work only refuses more actions, so it can be added later without breaking anything here. - **Leaving always works.** A member may leave an archived channel too (kind:9022, or kind:9001 naming themselves). The last owner still cannot leave. - **Agent owners keep edit and delete.** Today the human owner of an agent can edit (kind:40003) and delete (kind:9005) that agent's messages without being a member. This PR keeps that existing behavior in announce channels too, and documents and tests it. It does not let the owner post. - **Changing the setting changes no roles.** Existing members can still write after a change to `guest`. Existing guests stay read-only after a change back to `member`. A refused write gets `restricted: only members can post in this channel`. ## Details - **One write rule.** The relay's channel write check decides from the actor's role and the channel's `join_role`. Some event types skip that check and run their own: message edit, deleting your own message (kind:9005), topic and purpose (kind:9002), add-member (kind:9000), workflow messages and joining a huddle. Each of these now uses the same rule. Membership changes (add, remove, leave) do not need write access, so a guest can always leave. If the relay finds a stored `join_role` value it does not know, it treats it as `guest`, so an unknown value can never give more access. - **Admission is decided in the database transaction.** The membership write (`put_member`) makes the whole admission decision inside its transaction, after it takes the channel's membership lock: `join_role` for a missing role tag, the self-join ceiling, guests cannot add people, who may grant admin or owner, and role changes (`MemberRole::may_change_role`). The relay's request check uses the same rules, but the database is the authority, so a write that skips the request check cannot get around them. A change to `visibility` or `join_role` takes the same lock, so an add and a setting change cannot interleave. - **No creator exception.** `put_member` used to give owner to a channel's creator who added themselves to a private channel without an active membership. A removed creator could use that: their join (kind:9021) passes the open-channel check, the channel becomes private before the join takes the membership lock, and the join restored them as owner. Channel creation already makes the creator owner in the same transaction, so the exception is removed. A removed creator is now admitted like anyone else. - **Huddles.** Joining a huddle needs write access in the huddle channel **and** in its parent channel. The relay checks both before the join. When it commits the join, it applies the same rule again while it holds both channels' membership locks: a member needs a role that can post, and a non-member needs an open channel whose `join_role` is `member`. A guest in either channel cannot join. A join is also refused when the joiner loses membership or the channel's `join_role` changes before the commit. - **One lock order.** Each writer that takes more than one of a channel's locks takes them in this order: membership lock, TTL lock, channel row. Settings changes, roster publication, TTL refresh and huddle joins all use this order. Before this change, a `join_role` change on an ephemeral channel at the same time as a roster publication could deadlock. - **Workflows.** Saving a workflow (kind:30620) skips the generic write check, so it now checks the sender's role itself and needs a writing role before anything is stored. Every path that starts a workflow run, including the manual trigger (kind:46020), needs the workflow's owner to hold a writing role now (and an elevated role for an elevated workflow). So a workflow saved while its owner was a member stops running after the owner is made a guest. - **Channel names are checked before storage.** kind:9002 and kind:9007 refuse a name longer than the `channels.name` column (255 characters) before the event is stored. Before this, the relay accepted the event, then the name update failed, and the rest of the update (for example `join_role`) was silently skipped. This PR only adds this check. Storing the event and applying the update in one transaction is a larger change, and it is not in this PR. - **Changing the setting.** Owners and admins send a channel-metadata update (kind:9002) with `["join_role", "guest"]` or `["join_role", "member"]`. Buzz applies these updates as partial updates; a missing tag means "no change". The relay posts a `join_role_changed` system message in the channel. - **What clients see.** The relay signs a channel description event (kind:39000) that clients read. One shared builder now writes it for both the relay and the `buzz-admin` backfill, which had a copy that had drifted from the relay's. It always publishes an explicit `["join_role", …]` tag, plus the standard NIP-29 group access flags: | `visibility` | `join_role` | `public` / `private` | `restricted` | `closed` | | --- | --- | --- | --- | --- | | open | `member` | public | | | | open | `guest` | public | ✓ | | | private | either | private | ✓ | ✓ | This also fixes an old bug: `closed` ("join requests are ignored") was published for every channel, even open channels that accept joins. - **CLI and SDK.** `buzz channels update --join-role guest|member`. `buzz_sdk::build_update_channel` takes a new `join_role` argument and always writes it as an explicit tag. - **Docs.** `NOSTR.md` has an "Announce channels" section with the role table and the flag table. - **Tests.** Relay end-to-end tests, run in the CI relay E2E step: - `test_nip29_announce_channel_wire`: who can set `join_role`, the kind:39000 flags, joining as a guest, refused guest and non-member posts, reactions and adds, member posts, a member making a guest a member who then posts, changing the setting back without changing roles, and leaving. - `test_nip29_member_raises_guest_agent`: a member who is not the agent's owner can make a guest agent a member. - `test_nip29_announce_channel_wire_workflow_send_needs_write_role`: a guest's workflow message is refused; a member's is accepted. - `test_nip29_departure_wire_archived_channel_allows_self_leave`: guests and members can leave an archived channel. - `test_private_channel_any_member_can_invite` now expects a private-channel guest to be refused when it adds someone (it used to expect success). - `test_nip29_announce_channel_wire_workflow_commands_need_write_role`: saving and triggering a workflow, by a member and by a guest, in announce and normal channels; a guest cannot run a workflow saved while it was a member. - `test_nip29_announce_channel_wire_overlong_name_is_refused_before_storage`: a kind:9002 with an overlong name and `join_role` is refused, and kind:39000 still shows the old settings. - `test_nip29_announce_channel_wire_agent_owner_exception`: an agent's owner, as a guest and as a non-member, can edit and delete the agent's messages, while ordinary guests are still refused. - `test_nip29_guest_is_read_only_in_normal_channel`: a guest in an ordinary channel cannot write, and joining an ordinary channel still gives `member`. Unit tables cover the self-join rule for each `join_role`, and every actor, current role and requested role for the role-change rule. Database tests check the admission rules in `put_member` (`put_member_enforces_announce_admission`), that an add which waits on an uncommitted change to `guest` is decided on the new setting (`put_member_waits_for_a_concurrent_join_role_change`), and that a member can make a guest a member but not an admin. Unit tests cover the name limit and the workflow owner rule (`owner_authority_denies_roles_that_cannot_write`). A relay test covers huddle joins with different roles in the huddle and its parent, and a commit after a demotion, a `join_role` change or a lost membership (`huddle_join_needs_a_write_role_in_the_huddle_and_its_parent`). Lock-order tests: `join_role_change_waits_for_a_roster_publication_on_an_ephemeral_channel` reproduced a real deadlock (`40P01`) against the old order. `f2d_join_lock_order_serializes_membership_and_settings_writers` checks that membership and settings writers wait for a join and then succeed. `removed_creator_cannot_rejoin_a_private_channel` checks that `put_member` refuses a removed creator on a private channel, and `removed_creator_join_racing_a_private_switch_is_refused` holds the membership lock so a removed creator's kind:9021 passes the open-channel check, then makes the channel private before the join continues. Each new test was run against the old code and failed. ## Rollout 1. **Before the deploy, count guest rows** (`SELECT count(*) FROM channel_members WHERE role = 'guest' AND removed_at IS NULL`). We expect none. Any guest that exists loses the ability to post, change the topic and add people when this ships. If some exist, decide first whether to make them members. 2. Deploy. Every channel stays `join_role = member`, so nothing else changes. 3. **Do not set any channel to `join_role = guest` until no old relay is running.** Old relays ignore the column, so they would let guests and non-members write in that channel. 4. Rerun the `buzz-admin` channel reconcile so open channels drop the wrong `closed` flag. ## Migration and rollback - Migration `0067` adds `channels.join_role TEXT NOT NULL DEFAULT 'member'` with a CHECK constraint. Postgres stores a constant default in table metadata, so existing rows are not rewritten and there is no backfill. The relay runs the migration at startup. - Existing channels keep the old `closed` flag until their next change, or until the reconcile in Rollout step 4. - **Before rolling back the relay**, set every announce channel back to `join_role = member`. An older relay ignores the column and does not stop guests from writing, so those channels would accept posts from their guests and give `member` to new joiners again. ## Not in this PR Client UI: a Follow button, hiding the composer for guests, and an admin toggle. Clients that ignore the new tag keep working. A guest who tries to post sees the `restricted` error. ## Stack 1. #8143 (this PR): announce channels 2. #8494: store the bot role only for keys with a recorded agent owner --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-21 16:01:11 -04:00
Logan Johnson	c37d6b8eb6	test(relay): cover system channels in the workflows API and push matcher (#8509) ... 🤖 ## Summary Test-only. This adds end-to-end coverage for the last two system channel read paths that had none. It changes no behavior. A system channel is a channel type the relay leaves out of ordinary reads. A client sees one only when it names the channel or opts in. The system channels spec lists every read path that must follow this rule. The Postgres end-to-end tests already covered subscriptions, counts, search, the HTTP bridge and membership notices, but not these two: - **Workflows API.** A workflow names its channel, so reading a workflow in a system channel follows ordinary channel access. Members can read it. Outsiders can read it only when the channel is open. For a private one, an outsider gets 404, the same answer as for a missing workflow. - **Push notifications.** This is database-backed matcher coverage: the test stores real channel rows, memberships and push leases, loads the matcher's context from the database, and runs the production matcher. It builds the match batch directly, so it does not cover the outbox or gateway delivery. A member's push filter that names no channel gets a wake for a message in an ordinary channel, but not for a message in a system channel. A member's filter that names the system channel by `#h` gets a wake for that channel only (the rule since #8517). A non-member's filter gets neither. ### Related issue Part of the system channels spec (risks section: end-to-end cases for the workflows API and push). Rebased on #8507 and #8517, which made push follow the read rule: a system channel pushes only to a member's filter that names it by `#h`. No duplicate PR found. ### Testing - Both tests pass against local Postgres and Redis on this branch. - Each test fails when the behavior it covers is removed: without the named-channel check in the workflows API, a member gets 404 for its own system channel's workflow; if the push matcher lets a filter that does not name a system channel match it, the member's broad lease also wakes; if the matcher never pushes system channels, the named lease does not wake. - clippy (`--tests -D warnings`) and fmt pass. --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-21 13:09:45 -04:00
Logan Johnson	cd8a968179	fix(relay): push system channel events to a lease that names them (#8517) ... 🤖 ## Summary Mobile clients can now get push notifications for a system channel, but only when they ask for that channel by its ID. Before this change, the relay never sent push for system channels, even when a client asked for one directly. System channels are channels the app uses for its own purposes (for example, agent attention). They are hidden from clients that do not know about them. #8507 made push use the same visibility rule as normal reads: a push subscription sees a system channel, or any channel type added later, only if its filter names that channel ID in `#h`. With that rule, an older client can never get system channel pushes by accident. The extra "never push system channels" rule in NIP-CL now only blocks clients that ask on purpose, such as a future Live Activities subscription. This PR removes that special case. ### Details - `push_runtime.rs`: system channels now follow the same rule as other named-only channel types. The relay still checks membership before it sends a wake. - `NamedOnlyChannels::is_system` had no other caller, so it is removed. - NIP-CL and NOSTR.md now describe the shared rule instead of the ban. - No behavior change for current clients. They do not name system channels in their push subscriptions. ### Related issue Follow-up to #8507 (see https://github.com/block/buzz/pull/8507#issuecomment-6316180106). ### Testing - `named_only_channel_pushes_only_to_a_filter_that_names_it` now expects a wake for a system channel named by `#h`. It also checks that a broad filter still gets no wake, and that a non-member still gets no wake. - `cargo test -p buzz-relay --lib push_runtime`, `cargo clippy -p buzz-relay --all-targets -D warnings`, and `cargo fmt --check` pass. Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-21 10:38:57 -04:00
Logan Johnson	4dea51f644	fix(relay): scope push matches with the read path's channel rule (#8507) ... 🤖 ## Summary This is a consistency and forward-compatibility fix. It does not change what users of current clients see. Mobile push notifications now decide which channels can notify a device with the same rule that reads use, with one exception that the protocol requires: system channels never push. Before this change, push and reads used two separate rules: - **Reads** (subscriptions, counts, search and the bridge) show only the original channel types (stream, forum, DM, workflow) unless the filter names the channel. - **Push** skipped only `system` channels. Today `system` is the only non-original type, so the two rules give the same result. But a channel type added later would be hidden from reads and would still send push notifications to every member. This PR removes that gap: push now calls the same function that reads use, so a future type gets the correct push behavior with no extra work. ## What changes - A channel of an original type pushes as before. Membership is still required. - A system channel never pushes, even when a push filter names it. NIP-CL says: "The relay MUST NOT send push notifications for system channels." This is the same as before. - A channel of a type added later pushes only to a push filter that names that channel with `#h` (the channel tag), as reads do. - A push filter that names one channel does not get pushes from other named-only channels. - Events that are not in a channel (for example gift wraps) match as before. Push filters cannot carry the `#t` or `#P` opt-in that reads accept, so for push, naming the channel means `#h` only. ## Impact on current clients: none - System channels still never push, and no current channel type is affected by the later-type rule. - The mobile app names channels by `#h` only for joined, unarchived DM channels (`mobile/lib/shared/push/push_subscription_provider.dart`). DMs are an original type. - Naming a channel cannot get around membership. A push filter that names a channel still gets nothing unless its owner is a member of that channel. - Database cost is unchanged: one per-batch query replaces another, on the same channel list, with no extra round trip. - The push docs (NIP-PL and the push suppression doc) do not mention channel types, so they need no update. ## Details - The push matcher loads the named-only channels in each batch and checks each lease filter with `filter_channel_scope`, the function REQ, COUNT, search and the bridge already use. Before that check, it rejects system channels outright. - This removes `system_channel_ids_among`, which push was the last caller of. ### Related issue Part of the system channels spec (§2, rule 5). Builds on #8181 and #8469. No duplicate PR found. ### Testing - A `match_job` unit test covers: unnamed filters on an original-type channel, on a channel-less event, on a system channel and on a later-type channel; a later-type channel named by `#h` (1 wake); a system channel named by `#h` (0 wakes); an unrelated `#h` (no widening); and non-members. Removing either the shared scope check or the system exception fails it. - Before this PR, no test covered the push channel filter, not even for `system`. --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-21 08:42:30 -04:00
Ian Oberst	ca3be55018	fix(relay): advance the partition catch-all by default (#8464) ... ## Summary The migrations leave `events` and `delivery_log` with a catch-all partition that already covers every month from 2026-07. Only catch-all advancement replaces it with monthly partitions. Advancement was off by default, so any deployment that left out `BUZZ_PARTITION_MANAGER_ADVANCE_ENABLED` routed every new row into the catch-all. An operator then had to repair it by hand: drain the catch-all, swap it for monthly partitions, and re-attach. This has already happened in one staging environment. - **Advancement on by default.** `BUZZ_PARTITION_MANAGER_ADVANCE_ENABLED` now defaults to `true`, and an explicit `false` still disables it. `BUZZ_PARTITION_MANAGER_CREATE_ENABLED=false` remains the kill switch for all automatic partition DDL. - **Prompt retry while the current month routes to the catch-all.** If a startup or periodic pass leaves the current month routed to an empty catch-all that the maintenance planner would replace, the audit loop retries on its existing backoff. That is 5 s, doubling to a 60 s cap, for at most 8 retries (about 5 minutes). After that it returns to `BUZZ_PARTITION_AUDIT_INTERVAL_SECS`. Before this change, a failed first pass, such as a lock timeout during a rollout, waited the full 15 minutes. - The routine monthly advance, where only the last lookahead month falls into the catch-all, stays on the normal interval. Each attempt queues an `ACCESS EXCLUSIVE` request on the parent, so it shouldn't be retried without urgency. - The budget resets once the catch-all no longer awaits advancement. - A failed attempt keeps the last known state. - Layouts the planner refuses, such as a catch-all that holds rows, need an operator, so they are never retried early. Neither is a policy with create or advance disabled. - The check is `PartitionTableAudit::awaits_catch_all_advance(policy)`, plus a `PartitionAudit` helper that applies it to every table. It is derived from `plan_maintenance`, so it can't drift from the planner's rules. - **Fresh-install test.** A new `#[ignore = "requires Postgres"]` test runs every migration in a dedicated database. It checks that both tables start out waiting on the catch-all and that create-only maintenance does not fix them. It then checks that the first pass with advancement produces canonical monthlies, an empty `{table}_p_future`, trigger parity and a non-degraded audit, and that a second pass is a no-op. ### Related issue None found. ### Testing **Automated:** - `cargo test -p buzz-db --lib partition`, including `--ignored` against local Postgres: 21 unit tests and 55 Postgres tests pass. - `cargo test -p buzz-relay` for the config tests and the partition schedule tests passes. - `cargo clippy -p buzz-db -p buzz-relay --all-targets -- -D warnings` and `cargo fmt --check` are clean. - Mutation checks on `awaits_catch_all_advance`. Each fails the new unit test: - matching any window month instead of the current month; - replacing the planner check with the policy flags alone; - short-circuiting the current-month gate. **Local relay** (`BUZZ_AUTO_MIGRATE=true`, no advance env var): - **Fresh database:** startup committed `advanced` for both tables, creating `p2026_07` through `p2027_04` plus `_p_future`. - **Failed first pass:** I held `LOCK TABLE ONLY communities IN SHARE ROW EXCLUSIVE MODE` for 4 s across startup. The startup pass failed with `lock_timeout` at 17:07:20.8, and the first periodic retry committed `advanced` for both tables at 17:07:29. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Ian Oberst <ioberst@block.xyz> Co-authored-by: Ian Oberst <ioberst@block.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-20 16:20:18 +00:00
Luke Tornquist	039e1c8d59	feat(relay): write the NIP-43 member list in the membership transaction (#8482) ... ## Summary The relay-signed NIP-43 member list (kind:13534) was published after each membership change had already committed. If that publish failed, or if a change never republished at all (startup `bootstrap_owner`), the list stayed wrong until the background reconcile sweep found and fixed it. #8467 moved that sweep off the startup path. This PR removes the reason the sweep exists. Now every `buzz-db` function that changes `relay_members` writes the new list **in the same transaction**, as its last statement before commit: - It takes the per-community list lock last, reads the members, signs the list, retires the old one and inserts the new one. - If the list write fails, the membership change rolls back and the caller gets an error. - A call that changes nothing (already a member, role mismatch, host exists, limit reached) writes no list. The functions return `MembershipWrite<T> { outcome, snapshot }`. Callers push `snapshot` to live subscribers after commit. That push is the only thing left outside the transaction; if it fails, live clients see the change on their next fetch, and the stored list is already correct. Write paths covered (RPR #2, every write path): `add_relay_member`, `claim_relay_membership`, `claim_relay_invite`, `remove_relay_member`, `remove_relay_member_if_role`, `update_relay_member_role`, `bootstrap_owner` / `provision_owner`, `transfer_ownership`, `backfill_from_allowlist`, and the owner insert in `create_community_with_owner[_and_nip_fi_issuers]`. Community deletion removes the members together with the community, so it writes no list. The relay keypair is passed in explicitly instead of being stored on `Db`, so the compiler finds every caller. ### Behavior changes - **Open relays get an accurate list too.** Provisioning and ownership transfer used to publish only when `require_relay_membership` was on. The write is now unconditional. - **Same-second changes can't reuse an id.** The list's `created_at` is one second past the newest earlier list for that community, including retired ones, or the current time if that is later. Before, adding and then removing a member in the same second could produce an event with the same id as an earlier list. The insert then failed, and the stale list stayed live. A burst of changes can push `created_at` ahead of the clock by up to one second per change; it returns to the clock once the burst ends. - **`buzz-admin add-member` / `remove-member`** no longer build their own list (`publish_membership_list_with_bump` is deleted). They push the list the database wrote. Concurrent CLI runs now serialize on the database lock rather than depending on `run.sh` running one at a time. - **Owner mutations lock the community row `FOR NO KEY UPDATE`.** `transfer_ownership` and `bootstrap_owner` used `FOR UPDATE`. The list insert takes `FOR KEY SHARE` on that row through `events.community_id`'s foreign key, so a repair holding the list lock, or a removal or role update holding a member row, deadlocked with a concurrent owner change. `FOR NO KEY UPDATE` still conflicts with archive, unarchive and deletion admission, so the lifecycle ordering is unchanged. - **Startup delivers its list to other replicas.** When backfill or `bootstrap_owner` changes members at startup, the relay fans the newest list out through pub/sub once its state is ready, so clients connected to other replicas during a rolling deploy see an owner rotation without a refetch. - **New `buzz-admin republish-members`.** Rewrites the community's list from the stored members in one community-write transaction and pushes it live. This replaces the sweep as the repair for a list that is missing or stale: a `relay_members` change made directly in SQL, a restore, or a deployment that turns on `BUZZ_REQUIRE_RELAY_MEMBERSHIP` after running without it. ## Rollout Nothing rewrites a list automatically once the sweep is gone. A community whose list drifted before this deploys stays wrong until its next membership change or until an operator runs `buzz-admin republish-members`. Either deploy #8467 and let its sweep complete one pass first, or run `republish-members` for each community after this deploys. The command repairs only the community that `RELAY_URL` maps to, so on a relay hosting many communities, run it once per community host. ## What got simpler Deleted: - `reconcile_nip43_membership_snapshots`, `spawn_nip43_membership_reconciler`, its leader lock key and #8467's standby retry - the three `nip43_membership_snapshot_needs_reconciliation*` drift reads (one already deprecated) - `publish_nip43_membership_list` / `publish_nip43_membership_locked`, the separate after-commit republish at every call site, and `buzz-admin`'s duplicate publisher - the reconcile metrics `buzz_nip43_membership_reconciliations_total` and `buzz_nip43_membership_reconciliation_failures_total` - the after-commit publish metrics `buzz_nip43_membership_publications_total` and `buzz_nip43_membership_publication_seconds`, and the `publish_nip43_membership_locked` transaction metric label. A failed list write is now a failed membership write and shows up in the caller's existing error handling and the transaction metrics of its write path. - the `BUZZ_NIP43_RECONCILE_INTERVAL_SECS` setting There is now one way the list gets written, and it can't disagree with `relay_members` after commit. **Unused deploy config (outside this repo):** hosted deployments that set `BUZZ_NIP43_RECONCILE_INTERVAL_SECS` can drop it. It is now ignored. Dashboards or alerts on the four deleted metrics will stop receiving data. ## Testing New Postgres tests (`buzz-db`): - `relay_members::every_membership_write_commits_its_list`: for each write path, the returned list is the single live kind:13534 and names exactly the committed members. No-op calls return no list and leave the live list alone. - `relay_members::a_failed_list_write_rolls_back_the_membership_change`: holds the list lock, waits for the join to block on it, cancels that statement, and checks that the member was not added and the previous list is still live. - `relay_members::concurrent_joins_all_land_in_the_list`: 8 concurrent joins; the final list names all 8. - `relay_members::same_second_return_to_an_earlier_membership_writes_a_new_list`: add, remove, add, remove within a second; every list is newer than the last and no id repeats. - `relay_members::owner_transfer_and_list_repair_both_commit` and `relay_members::owner_transfer_and_member_removal_both_commit`: hold a repair (list lock) or removal (member row) open, wait until a transfer blocks behind it, then finish it. Both transactions commit; with `FOR UPDATE` back in place, one of them fails. - `community::create_community_with_owner_is_atomic_and_create_only` and `relay_invite::bounded_claim_exhausts_and_existing_member_retry_does_not_consume` now also check the list. Each new test was checked to fail against a mutation of the new code that brings back the old behavior (details in the thread). Stacked on #8467. Merge order: #8467, then #8468, then this. --------- Signed-off-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz> Co-authored-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz>	2026-10-20 11:39:13 -04:00
Luke Tornquist	c99872f749	fix(relay): delete the startup NIP-33 d_tag backfill (#8468) ... ## Summary Delete the NIP-33 `d_tag` backfill that ran on every relay boot. On every boot, before the listener opened, the relay ran: ```sql SELECT DISTINCT community_id FROM events WHERE kind BETWEEN 30000 AND 39999 AND d_tag IS NULL ``` No index covers `d_tag IS NULL`. The only `d_tag` index is the partial `idx_events_parameterized`, which is `WHERE d_tag IS NOT NULL`. So this query scanned every partition of `events`. In production it averaged about 11s per boot, max 19s (`buzz_startup_phase_seconds{phase="d_tag_backfill"}`), and found nothing. These rows shouldn't exist anymore: - The backfill was the one-time migration for #246, the change that added `d_tag`. - In `crates/buzz-db/src`, every production write path sets the column for kinds 30000–39999 through `extract_d_tag`: the main insert, `replaceable.rs`, `channel_members.rs`, `relay_members.rs`, and `push.rs`. A new source test, `every_production_event_insert_names_d_tag`, now enforces that. - The `events` inserts that omit `d_tag` are all test fixtures. `scripts/backfill-d-tag.sql` is still there as the manual fallback for any database that predates #246 and somehow never booted a relay since. ### What got simpler - Removed `Db::backfill_d_tags` and its admission test. - Removed the source-shape assertion in `observability_source.rs`. - Removed the `d_tag_backfill` startup phase. - `StartupStep::ALL` is now a slice, so removing a step no longer means editing a hand-kept array length. ### Notes - The `buzz_startup_phase_seconds{phase="d_tag_backfill"}` series stops reporting. - **Rebased onto main after #8467 merged.** Both remove a startup phase. With the fixed-length `ALL` array the two merged cleanly in git into a 17-entry `[Self; 18]`, which doesn't compile. This PR makes `ALL` a slice. - **Before merging**, run this read-only query against production and confirm it returns `0`: `SELECT count(*) FROM events WHERE kind BETWEEN 30000 AND 39999 AND d_tag IS NULL`. It scans every partition, so use a read replica. If it isn't `0`, run `scripts/backfill-d-tag.sql` first. The cutover script copies legacy `d_tag` values as-is, so only this query proves historical data is clean. - **Self-hosted installs** should run the same check before upgrading. ### Related issue None found. Companion to #8467, which also targets relay startup time. ### Testing - New `every_production_event_insert_names_d_tag` (`crates/buzz-db/tests/observability_source.rs`): every production `INSERT INTO events` must name `d_tag` in its column list. Dropping `d_tag` from `push.rs` makes it fail and name the function. - Runtime E2E against a real relay: boot has no `d_tag_backfill` phase; kind:30023 replacement via WebSocket and HTTP keeps only the newer event, an unrelated `d` survives, and an older replay can't win (confirmed in SQL). - Stacked on #8467 at `501dfd83`: a fresh two-replica boot runs all enabled startup phases, with neither `nip43_reconcile` nor `d_tag_backfill` present. Both replicas pass readiness and return the same signed message over authenticated HTTP and WebSocket, and outsiders are denied. - Blox, at `501dfd83`: `just fmt-check clippy security-review-check file-size-check` passes. Every `test-unit` step and `just test` pass except the `buzz-acp` `git_bootstrap` native-git tests, which fail the same way at base `b6aa5539` on Blox, and one `buzz-agent` catalog timing test that fails only under full-suite load and passes when run alone. - `e27139de` adds only a doc comment on `every_production_event_insert_names_d_tag`, saying which SQL shapes the text match covers. No code changed, so the results above still apply. --------- Signed-off-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz> Co-authored-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz>	2026-10-20 11:22:09 -04:00
Logan Johnson	25bcb791b4	fix(relay): show only legacy channel types to reads that name no channel (#8469) ... 🤖 ## Summary Every Buzz channel has a **type**, such as an ordinary chat channel (`stream`), a forum, a DM or a workflow. A client app can only show the types it was built to understand. This PR makes the relay (the Buzz server) follow one rule everywhere: **a client sees a channel only if it knows that channel's type.** #8406 applied this rule to membership notices. This PR applies it to the reads that list channels without naming one: the channel list, "channels I belong to", the sidebar's unread list and artifact lists. Before, these reads left out one type by name: system channels. Now they return only the four types that every client has always known: `stream`, `forum`, `dm` and `workflow`. Today the result is the same for every user, because `system` is the only other type. ### Backward compatibility: old clients stay safe - A client that was released before a new channel type exists will never receive channels of that type in its lists. It cannot show such a channel correctly, so it does not get one. - Requests that old clients send today return exactly what they returned before. Nothing changes for them now or when a type is added later. ### Forward compatibility: new types can be added safely - Before, the rule was a deny list: "everything except `system`". Each new type would appear in every deployed client by default, unless someone remembered to update five separate checks in the relay. A missed check would show the new type to clients that cannot handle it. - Now the rule is an allow list that is frozen: the four original types. A new type is hidden by default, with no relay change needed. Clients that understand it can request it on purpose, the same way current clients request system channels today. Each new type gets its own opt-in, so adding a type cannot break clients that are already in use. ## Details - One frozen list, `buzz_core::channel::LEGACY_CHANNEL_TYPES`. Membership notices (#8406) now use the same constant, so channel reads and notices cannot drift apart. The duplicate `LEGACY_NOTICE_CHANNEL_TYPES` is removed. - One SQL helper, `legacy_channel_type_sql`, builds `channel_type IN ('stream', 'forum', 'dm', 'workflow')` from that list. It replaces the five `<> 'system'` checks: - `get_accessible_channel_ids` (both halves): the implicit channel scope of any filter without `#h`, for example kind 39000 channel metadata, REQ, COUNT and search; - `get_accessible_channels`: the channel list; - the personal-read sidebar roster; - artifact queries. - A filter that names a channel (`#h`, or `#d` on kinds 39000–39003) reads it under the ordinary rules, whatever its type: the reader is a member, or the channel is open. The uncached named-channel lookup, `get_readable_named_only_channels`, now returns every readable non-legacy channel with its type, not only system channels. Before, a non-member could not read an open later-type channel, and `#d` read nothing. Workflow run reads and thread windows use the same lookup. - The relay keeps each named channel in the scope of the filter that names it only. This also holds when it re-checks membership for a named channel that the cache lacks. - The opt-in to system channels (`#t:["system"]` or `#P` on kinds 39000–39003) adds readable **system** channels only. It never adds a channel of another type, even one that a sibling filter in the same request names. The relay scope type `NamedOnlyChannels` keeps the type with each channel, so the opt-in can tell them apart. - NIP-CL and NOSTR.md state the rule. ### Related issue Follows #8406 (the same rule for membership notices). ### Testing New Postgres tests use a scratch database with a `future_type` value added to `channel_type`, as a later migration would add one: - `reads_that_name_no_channel_leave_out_a_type_added_later`: the channel scope, the channel list and the sidebar roster leave out the later type and keep the legacy types. - `artifacts_of_a_later_channel_type_are_read_only_by_name`: artifact current and history queries and counts without `#h` leave out the later type. With `#h` naming the channel, they return it. - `channels_of_a_later_type_are_read_only_by_name`: production read paths (WS REQ, WS COUNT, HTTP `/query` and `/count`). A member reads a private later-type channel and anyone reads an open one, by `#h` and by `#d`; an outsider cannot read the private one; a read that names no channel returns neither. In one request, a filter that names the channel plus a `#P` or `#t:["system", <label>]` sibling: the sibling does not get it. - `a_named_channel_of_a_later_type_stays_in_its_named_filter`: the membership re-check for a channel joined mid-request, with an unnamed sibling and a `#P` sibling. Only the naming filter gets the channel. Each test fails when its predicate goes back to `<> 'system'`, when the named lookup loads only `system`, or when the opt-in adds every named-only channel. The #8406 notice tests and the system channel Postgres tests pass on this branch. --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-20 11:04:29 -04:00
Luke Tornquist	e6c56ee5ee	fix(relay): take NIP-43 snapshot reconcile off the startup path (#8467) ... ## Summary Before the relay opened its listener, it reconciled the NIP-43 membership snapshot for every active community. Each community costs two writer round trips, one after the other. On a production deployment with about 80k communities, this one step averaged 174s of a roughly 180s boot (max 324s). Pods in an availability zone farther from the writer took 3–4× longer than pods in the writer's zone. The step doesn't need to block startup. The snapshots live in the shared database, and the replicas that are already running serve the same snapshots. A new replica has nothing of its own to repair before it serves. The sweep before bind did the same work as the background maintenance sweep; only the metric labels differed. This PR: - Removes the NIP-43 sweep that ran before bind. - Moves the background loop into `side_effects::spawn_nip43_membership_reconciler`. Its first sweep now runs as soon as the task starts. Before, the loop skipped its first tick, so the first maintenance sweep waited a full `BUZZ_NIP43_RECONCILE_INTERVAL_SECS`. - Runs that loop on one replica only. The replica holding the `NIP43_RECONCILE_LOCK_KEY` session advisory lock sweeps; the others retry the lock every `NIP43_STANDBY_RETRY` (5s, capped at the interval). This reuses the usage-metrics leader lease (a detached Postgres session), renamed from `UsageMetricsLeader`/`try_lock_usage_metrics` to the generic `SessionLeader`/`try_lock_session_leader`. - Waits a full interval after each sweep (sleep after sweep), so a sweep longer than the interval never runs straight into the next one. - `reconcile_nip43_membership_snapshots` is now the only entry point and keeps its original signature (`-> Result<usize>`). It's no longer deprecated. ### What got simpler Deleted: - `Nip43ReconciliationPurpose` (the `Bootstrap`/`Maintenance` purposes) and `reconcile_nip43_membership_snapshots_with_purpose` - `Db::bootstrap_community_hosts` and `Db::nip43_membership_snapshot_needs_reconciliation_for_bootstrap` - The startup progress log (`NIP-43 startup reconciliation progress`, `nip43_progress_due`) and its test - The `nip43_reconcile` startup phase: `StartupStep::ALL` goes from 19 to 18 steps The leader lock reuses the existing usage-metrics lease rather than adding a second leader primitive. `StorageAccountingLeader` is still a near-duplicate of it; folding that in is left as identified debt. ### Trade-offs - **Repair before serving.** A newly started replica no longer guarantees it has repaired stale snapshots before serving. The running fleet doesn't guarantee that today either: the background sweep is what repairs drift, and the leader starts it as soon as it takes the lock. - **Rollout load.** Before, every replica swept before binding, and every replica also ran the background loop. Now there is one sweeper in steady state, so writer load from this job drops from one sweep per replica to one per deployment, and rollouts no longer stack boot sweeps. Standbys add one lock attempt every 5s each. - **Leader handoff.** When the leader's process exits (including SIGKILL), its connection closes, Postgres releases the session lock, and a standby retries the lock at least every `NIP43_STANDBY_RETRY` (5s) and sweeps as soon as it gets it. Each attempt also waits for a writer connection, so under database pressure takeover can take longer than 5s. Before this, a standby waited a full interval (3600s in production), so after a rollout replaced the leader nothing swept for up to an hour. If the leader is lost with its host or network instead, the lock is held until the database server's TCP keepalive drops the session, and nobody sweeps meanwhile. The usage-metrics leader already has the same exposure. The lease is checked only between sweeps, so a leader that loses its session mid-sweep finishes that sweep while the new leader starts one; the per-community publication lock keeps that overlap safe. - **Snapshot window.** A missing or stale kind:13534 snapshot is served until the leader's sweep reaches that community; before, a booting replica blocked until it was repaired. Admission reads canonical `relay_members`, so this window doesn't change who's admitted. It does affect clients that read the snapshot: desktop onboarding shows the membership-denied page to a member missing from it, the desktop/mobile member lists and join alerts lag, and desktop mesh discovery keeps its existing allowlist when the snapshot is missing. In production the inline publications that keep it current have not failed since 2026-10-07, and the sweep repaired no snapshots in the last 30 days. - **Dashboards and alerts.** The `buzz_startup_phase_seconds{phase="nip43_reconcile"}` series stops reporting. Anything charting or alerting on it should move to the sweep's existing counters (`buzz_nip43_membership_reconciliations_total`, `buzz_nip43_membership_reconciliation_failures_total`). ### Related issue None found. ### Merge order #8468 is stacked on this PR. Both remove a startup phase, and with the old fixed-length `StartupStep::ALL` array, merging both would have built a 17-entry `[Self; 18]` that doesn't compile. Merge this PR first; #8468 makes `ALL` a slice. ### Testing - `background_reconciler_sweeps_immediately` (private scratch database): spawns the production reconciler with a 24h interval and 24h standby retry, and requires the first sweep to publish a missing snapshot. Fails if the skipped first tick is restored. - `background_reconciler_standby_takes_over_on_short_retry` (private scratch database): with a 24h interval and 100ms standby retry, nothing publishes while another session holds the leader lock; once it's released the standby must publish within 10s. Fails if the standby sleeps the full interval again (checked by reverting that line). - `test_usage_metrics_lock_has_single_owner_and_releases_on_drop` covers the renamed lease. - `maintenance_sweep_skips_archived_and_deleted_communities` now calls the single entry point. - Runtime E2E at `574b28b3`: two real relays sharing one database, 3600s sweep interval, production 5s standby retry. While the leader was alive, the standby made only lock attempts (no sweep or snapshot-comparison reads). SIGKILL of the actual lock holder, 3/3 trials: the survivor took the lock in 4.81–4.86s and repaired the snapshot within 10ms after that. Both replicas served authenticated reads while the initial publication was stalled. - Blox, at `574b28b3`: `just fmt-check clippy security-review-check file-size-check` passes. Every `test-unit` step and `just test` (118 suites) passes except the `buzz-acp` `git_bootstrap` native-git tests, which fail the same way at base `b6aa5539` on Blox (environment, not this change). --------- Signed-off-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz> Co-authored-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz>	2026-10-20 10:19:46 -04:00
Wes	b6aa5539e3	feat(mobile): account onboarding, key recovery and hosted reconnect (#8375) ... ## Behavior Adds optional mobile Builderlab sign-in, identity backup/recovery, community discovery/creation, and account-aware reconnect. **Builderlab features are disabled by default:** when `BUZZ_AUTH_DOMAINS` is unset or empty, account login, creation, backup/recovery, reconnect prompts, and background account restoration are hidden/disabled. QR/code pairing, links, and invites retain Nostr-only entry. Saved account state is retained for later opt-in. `--dart-define=BUZZ_AUTH_DOMAINS=communities.buzz.xyz,communities.public.builderlabstaging.xyz` opts those DNS domains and their subdomains into required account setup. The same policy gates pairing, deep links, durable community admission, and reconnect. A nonempty list enables Builderlab features; `BUZZ_BUILDERLAB_URL` alone does not. Enabled flows still require their backend services. Existing Nostr relay authentication remains mandatory. `BUZZ_HOSTED_RELAY_AUTH=true` enables scoped relay assertions and reconnect prompts only for configured authentication domains. Configure assertion issuance and each community's issuer binding before enabling assertions; enable relay enforcement last. Account credentials stay at Builderlab. Wrong-account recovery preserves keys; failed writes are not replayed; authentication is bounded by request deadlines. Hosted flows preserve keys across cancellation/restart, reconcile interrupted creation/invitation acceptance, and support manual nsec file import/export. Account identities are scoped to backend and authenticated subject; reusing an existing local key for an unbound account requires an explicit selection. Recovery reconciles the advertised and recovered inventories before automatic selection. Successful self-hosted admission retires its pending key, and iOS startup removes interrupted export files. Adds authenticated relay `GET /operator/communities/joined` for the companion Builderlab membership endpoint. ## Validation - Follow-up review repair at `66a12ac92`: self-hosted entry presents readable choices and import despite incomplete enumeration, while generation still requires a complete empty inventory; explicit selection uses exact-key lookup. Consuming a value-equal duplicate now advances dispatch. Reconnect requires an existing account binding or a subject-scoped backup receipt before signing an assertion, and pins the subject inside the serialized account request. An empty different account is refused without sending key proof. Focused checks: configured domains 58 passed; default 49 passed, 9 hosted-only skips. Separate isolated mutations fail the incomplete-inventory sheet, exact-key reuse, duplicate dispatch and reconnect journeys (including an empty-account-only bypass); restored controls pass. Evidence uses synthetic storage/network boundaries, not physical Android or live backend association. All mandatory commit/push hooks passed; fresh GitHub CI is pending. - Bounded review repair at `d3cc0003f`: incomplete secure-storage enumeration now blocks automatic identity generation while explicit known-key reads retain their fallback; declined/failed onboarding links release the queue; HTTP 429/5xx keeps Retry regardless of response encoding. Default app-gate coverage exercises a bind link followed by ordinary community entry. Focused checks: configured domains 47 passed; default 38 passed, 9 hosted-only skips. Removing the three production fixes in an isolated source/Git/build copy fails the identity-resolution, declined/failed queue, and Retry journey assertions; clean and restored controls pass. Storage and HTTP failures are synthetic; physical Android and live outage behavior were not exercised. Published after all mandatory commit/push hooks passed (mobile analysis/tests, branch-skew and file-size gates); fresh GitHub CI remains pending. - Existing whole-PR source review cleared the reconnect defects at `fb15d5b82`. - Domain-policy increment: focused environment, entry, app-gate, pairing and onboarding journeys pass with empty and configured domains. Existing pairing provider/page tests pass. Changed Dart files analyze cleanly. - Isolated source/Git/build mutation: forcing paired credentials through hosted login fails the default-build journey; clean and restored controls pass. CI now runs configured-domain journeys in addition to the default-build suite. - Integrated main through `5b433d929` at `0067c069c`, including its pinned Flutter upgrade. Preserved the PR welcome UI with the updated transition API and added Material surfaces around onboarding selection/consent rows. Focused post-merge runs pass in both configurations; all mandatory local push gates passed (mobile analysis/tests, Rust tests, desktop-native checks, branch-skew and file-size checks). - Merge-review repair at `a315bce11`: focused identity/storage/admission and widget journeys pass, including account switching with an empty or already-bound second account, explicit reuse, and self-hosted removal/rejoin with another key. Default-domain journeys pass too (hosted-only cases intentionally skip there). - Isolated source/Git/build mutation checks: bypassing account scoping, reuse consent, pending-key retirement, or recovery inventory reconciliation fails the relevant journey/invariant assertions; clean and restored controls pass. - Native `RunnerTests/IdentityFilePluginTests`: both tests pass on the iPhone 18 Pro iOS 27 simulator with an isolated app identifier. At `24a0dcaac`, the picker journey seeds residue in Library/Caches, matching path_provider_foundation, and verifies plugin startup removes it while preserving an unrelated file. The earlier tmp-only fixture did not exercise the actual export location. This is synthetic native cleanup/picker evidence, not a real terminated-share or live account round trip. - Follow-up: Caches and tmp cleanup; fallback to individually known identity slots when Android bulk enumeration fails; explicit replacement of a refused self-hosted key preserves its original before swapping the retry slot. Configured-domain journeys/invariants: 34 passed; default: 31 passed, 3 hosted-only skips. Isolated enumeration/retry mutations fail with passing controls. Bulk-read fallback preserves but cannot enumerate arbitrary other-account slots until bulk reads recover. - Configured-domain CI exposed a stale pairing-status assertion after account scoping. The repaired test checks the admission mismatch, preserved account/transfer keys, and no admitted community. The configured-domain CI selection passes locally (56 tests); removing the production mismatch guard fails the updated test, with clean/restored controls passing. This is synthetic coordinator evidence, not physical QR acceptance. - Final review repair (`40e1e6101`): native iOS export creation uses atomic complete file protection; normal and restart cleanup remain. Account/policy/claim HTTP uses one 30-second total deadline with request abort and body cancellation. Removed the unrelated NIP-FA spec edit. - Focused account/admission/export tests: 25 pass; configured-domain account/admission/onboarding journeys: 33 pass. Isolated deadline and native-dispatch mutations fail, with passing clean/restored controls. Native export/picker tests: 2 pass on iOS 27 simulator. The real file-protection attribute assertion is physical-device-only because simulator provides no Data Protection attribute; locked-device protection and interrupted real sharing have not been observed. - Opt-in UI: configured-domain CI selection passes (57 tests); default selection passes (7 tests, 10 hosted-only skips); URL-only plus whitespace/empty domains passes (6 tests, 9 hosted-only skips). Changed Dart files analyze cleanly. Independent mutations that unconditionally initialize the coordinator in the app gate, welcome entry, or link form fail the default journey; clean/restored controls pass. - Native welcome screens exercised on iPhone 18 Pro / iOS 27 simulator in default and configured builds: default exposes community link / QR / pairing code only; configured build additionally exposes Builderlab connection and community creation. No live hosted login was exercised in this check. - Delivery at `e48fe73a9`: all mandatory push hooks passed (full mobile analysis/tests, Rust tests, desktop-native checks, branch and file-size checks). GitHub checks had not yet populated at the immediate post-push snapshot. ## Review disposition | Feedback | Disposition | | --- | --- | | Explicit self-hosted entry under unreadable inventory, duplicate invite dispatch, reconnect cross-account binding ([entry](https://github.com/block/buzz/pull/8375#discussion_r4318421622), [queue](https://github.com/block/buzz/pull/8375#discussion_r4318421640), [reconnect](https://github.com/block/buzz/pull/8375#discussion_r4318455512)) | Repaired at `66a12ac92`; evidence above. Reconnect refuses unproven ownership rather than silently establishing a binding. | | Incomplete Android identity inventory, stranded deep-link queue, transient HTTP retry classification ([review](https://github.com/block/buzz/pull/8375#pullrequestreview-5576864143)) | All three repaired in `d3cc0003f`; focused journeys and isolated mutation evidence above. | | Account cross-link, incomplete recovery inventory, committed invite checkpoint, pending local key retirement/replacement | Repaired in the preceding commits; journey and mutation evidence above. | | iOS interrupted key export ([review](https://github.com/block/buzz/pull/8375#pullrequestreview-5575253252)) | Native cleanup plus complete protection at creation implemented. Physical-device lock and real interrupted sharing acceptance remains pending. | | Trickle responses bypass idle timeout; unrelated NIP-FA edit ([review](https://github.com/block/buzz/pull/8375#pullrequestreview-5575764556)) | Total request deadline/abort implemented; spec edit removed. | | Production domains rejected | Ordinary entry now needs no Builderlab configuration. Opt-in account-required domains must still use supported community mappings. | | Lost Android Keystore recovery | Backup exclusions prevent the reviewed stale-restore scenario; genuine key loss still has no in-app recovery/reset and remains deferred. | | Native secp256k1 dependency question | Obsolete against the current base: upstream C integration landed in #8398; no package diff remains here. | | Operator membership endpoint contract | Documentation now states the operator chooses the pubkey; no end-user session binding is implied. Additional handler-level coverage is optional follow-up. | | Optional Android partial document cleanup/activity retention; local work after backup cancellation | Follow-up hardening, not claimed implemented. Cancellation prevents remote backup upload. | | Semgrep crypto inventory notices | Informational inventory, not demonstrated defects. | ## Remaining draft gates Native QR/browser-return acceptance remains unverified for this revision. Optional Builderlab onboarding still needs backend backup-status and joined-community routes, recovery policy/configuration, and real backup/recovery/reconnect acceptance. The empty-domain build removes these services as prerequisites for ordinary pairing; it does not make the optional hosted flows backend-ready. Human acceptance and current CI are still required before merge. Review dispositions retained: production auth-domain configuration must use supported community mappings; enabling `builderlab.xyz` does not add support for `buzz.block.builderlab.xyz`. Recovery after genuinely losing an Android Keystore key remains a follow-up, not a newly implemented capability. Human acceptance steps: build once without `BUZZ_AUTH_DOMAINS` (including with a previously saved account) and confirm no Builderlab controls/restoration while ordinary link and QR/code entry work. Build with supported domains and confirm connection/creation appear; exercise browser return and live backup/recovery/reconnect. On physical iOS, exercise export and interrupted sharing/locked-device protection. Record results before marking review complete. --------- Signed-off-by: Wes <wesbillman@users.noreply.github.com> Co-authored-by: Brain <1a02c72794dcd0f07058a353bc3a81f4028b8c77c92c87fce6d5c8b85970a20b@buzz.block.builderlab.xyz> Co-authored-by: Brain <1a02c72794dcd0f07058a353bc3a81f4028b8c77c92c87fce6d5c8b85970a20b@wes.communities.public.builderlabstaging.xyz>	2026-10-19 15:46:58 -06:00
Logan Johnson	cc6336d985	feat(relay): tag every membership notice with its channel type (#8406) ... 🤖 ## Summary - When someone is added to or removed from a system channel, the relay now sends them the usual "you were added / removed" notice. Before, it sent nothing, so a client could learn about a new system channel only by rereading every channel. - **A client sees a notice only if it knows the channel's type.** Every membership notice now carries its channel type. Clients that do not send `#t` (every client today) see only the types that existed before system channels, so they see no change. A type that Buzz adds later is hidden from them by default. - The notices are stored like every other membership notice, so a client that was offline can catch up with `since` instead of rereading all its channels. ## Details - Every kind:44100 (added) and kind:44101 (removed) notice now carries `["t", <channel_type>]`, for example `["t","stream"]` or `["t","system"]`. - **One place reads the channel type:** `emit_membership_notification` in `side_effects.rs`, where the relay signs the notice. A channel's type is set when it is created and no code path changes it, so the tag stays correct. If the channel row is missing, the relay sends no notice rather than an untagged one. - **The rule (NIP-CL):** - a filter with `#t` matches a notice only when the notice's type is one of its `#t` values; - a filter without `#t` matches only the **legacy types**: `stream`, `forum`, `dm`, `workflow`. This list is frozen (`LEGACY_NOTICE_CHANNEL_TYPES`), so a new type never joins it; - a notice without a `t` tag was stored before this change. It is always a legacy type, so the relay treats it as one, and replay across the upgrade stays complete. - **Fails closed.** The rule lives in two shared places, and both default to the legacy types. A read path that is not updated behaves as it does today: - `filters_match` / `notice_visible` (buzz-core). Live fan-out, the REQ history post-filter, COUNT, the HTTP bridge and push all match through it. - `EventQuery::notice_channel_types` (buzz-db). Its SQL keeps only visible notices before `LIMIT`, so page limits and COUNT stay exact. `None` means legacy types. Only `filter_to_query_params` sets it, from the filter's `#t`. - A client that knows every current type sends one filter: `{"kinds":[44100,44101],"#p":[<me>],"#t":["stream","forum","dm","workflow","system"]}`. NIP-CL and NOSTR.md describe this. - **Only the target receives a notice.** A notice names the target, the channel and who made the change, and the relay signs it, so knowing its id is not permission to read it. 44100/44101 are now in `RESULT_GATED_KINDS`, the one list used by the `ids` exemption check, the per-result reader check (WS and HTTP history, push), live fan-out (same pod and Redis) and COUNT. This also closes the same id lookup for ordinary channels' notices. - No migration: 44100/44101 are already excluded from search indexing. ## Testing - `buzz-core`: a table test covers untagged, `stream`, `system` and a future type against filters without `#t`, with `#t` for one type, and with `#t` for every type. Other kinds keep the plain `#t` match. - Postgres tests (`system_channel_postgres_tests.rs`): live fan-out, REQ history, COUNT and the HTTP bridge, for a filter without `#t`, a `#t:["system"]` filter and an all-types filter; an untagged legacy notice; a non-target with an `ids` filter on each path; and a raw `EventQuery` that does not set the field. - I broke each part on purpose and the tests failed: no type tag on the notice, no rule in `filters_match`, no SQL rule, and no untagged clause in the SQL. - End to end: I built a relay container from f24f8e7ba and ran buzz-app's `container-relay.e2e.test.ts` (block/buzz-app#1290) against it. The live system channel test passed 5 of 5 runs. Paired client change: block/buzz-app#1290. It works unchanged against this relay. --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-19 16:29:32 -04:00
Will Pfleger	91e7b469b8	fix(relay): wait for the peer's Close before dropping any WebSocket (#8426) ... 🤖 A NIP-42 login refused for a NIP-FI key mismatch could reach the client as an abnormal close (`1006`, no reason) instead of the `NOTICE "restricted: authorization denied"` followed by close `1008` that every other authorization denial produces. The relay did write the NOTICE and the `1008` Close, but the connection task dropped TCP as soon as the writer finished, before the peer finished the close handshake. The client could then lose frames it hadn't read yet. It depends on timing, so it showed up intermittently and is not specific to key pairing. ## Relay route - `recv_loop` now hands back the read half even while a frame handler is still running. Frames are handled one at a time while the connection is live. Once it is cancelled, an in-flight handler (for example a Nostr `CLOSE` waiting on the subscription lock behind a REQ stuck in Redis) is spawned to finish instead of holding the reader. It is never dropped partway, so subscription retirement stays atomic, and cleanup joins it before the expiry barrier and application cleanup. - As soon as the writer finishes, `await_client_close` keeps the transport open until the client's Close arrives or the stream ends. It runs concurrently with joining the heartbeat, auth-timeout, expiry and auth-cancel tasks and the connection's application cleanup (subscriptions released, registry entry removed, presence cleared). Cleanup never waits on the peer, so a nonresponsive client cannot delay registry counts, presence or `has_admitted_connection`. The socket never waits on cleanup, so a stalled Redis (the pool has no wait timeout) or an expiry join blocked on its quiescence barrier cannot hold the socket and its file descriptor open past the bounded wait. The active-connection gauge and the connection permit are released after both finish. - The connection permit (`BUZZ_MAX_CONNECTIONS`) is now taken at the start of `handle_connection`, before the NIP-FI gate and the community-active bootstrap check, and held through the active handler or the refusal drain and its bounded wait. Previously a socket in bootstrap, or refused there, held no permit, so a burst of refusals could open sockets beyond the limit. Over the limit, the socket is still dropped with no frames. `buzz_ws_connections_active` still counts only connections that reached the active handler. ## Huddle-audio route - The relay and audio bootstrap drains share one bounded helper. A socket refused before admission waits after its community registry entry is released. - Every exit of the audio handler gets the bounded peer wait. Exits whose cleanup follows their terminal frames (`room_full`, `room_ended`, version mismatch, archived channel, owner rejection or unreachable, failed commit confirmation) and the normal teardown hand the socket to a background task that waits for the client's Close, so lease releases, `48102` writes and room archiving on Postgres and Redis never hold the socket open. The remaining exits send their frames last, and `handle_active_audio_connection` waits once they return. The wrapper joins the background task before returning, so the audio connection permit is held until the socket ends, as on the relay route. - Audio error exits now follow their existing JSON error with a Close frame: | Exit | Close code | | --- | --- | | `huddle_relay_draining` | `1012` service restart | | `join_rejected` (owner resolution failed), `huddle_owner_unreachable`, failed channel lookup | `1011` internal error | | `huddle_audio_unavailable`, `unsupported_version`, `upgrade_required`, `room_full`, `room_ended`, archived channel, owner rejection, NIP-FI authorization denial | `1008` policy | Audio login failures and authorization refusals outside NIP-FI `authorization_denied` still send only their existing error frame, then get the bounded wait. - The audio read half is an enum, so the hand-off has no `expect()` that can panic. ## Timing - The peer wait is bounded by `WS_TERMINAL_FLUSH_TIMEOUT` (1s), and drains to end-of-stream so a client-initiated Close gets its reply. The terminal writer flush has its own 1s budget, so the worst case for terminal transport work is about 2s per refused or closed socket. - Terminal frame writes (`flush_terminal_frames`, `send_exit_frames_bounded`) still give up on the first failure, but a failed or timed-out send is now logged at `debug` instead of being discarded. - The graceful-shutdown comment in `main.rs` now describes per-stage budgets (restart jitter, restart-Close acknowledgement, terminal flush, peer wait) rather than a per-socket total. The 30s hard drain bounds every socket, and the 35s process shutdown budget fits within the chart's 60s `terminationGracePeriodSeconds`. ## Tests Real-socket tests in the infra-free relay lane of `just test-unit` and `scripts/run-tests.sh` unless marked PostgreSQL: - The key-mismatch login and a session-expiry denial (as a control) both assert NOTICE then `1008` / `authorization denied` on the wire. - A raw-TCP client that never answers Close: the relay holds the socket for at least the bounded wait, has deregistered the connection by the time the socket ends, and ends it on its own under a 10s watchdog. - A raw-TCP client that withholds its Close until the connection has deregistered: the relay ends the socket in under the 1s cap, which fails if cleanup waits on the peer. - A raw-TCP client that never answers while final cleanup is stalled at a test pause point: the socket still ends after the bounded wait. - A raw-TCP client that never answers Close on the relay bootstrap drain, the audio bootstrap drain and an early audio-handler exit: the relay holds each socket for the bounded wait. - On the relay route with a one-slot budget through the real `handle_connection`: a refused bootstrap holds the slot until its socket ends, and a bootstrap stalled on the community check already holds the slot before the AUTH challenge is sent. - An established relay session whose reader is blocked in a Nostr `CLOSE` handler on a held subscription lock when the session expires: the socket ends within the bounded wait while the lock is held, and cleanup completes with the subscription retired once it is released. - An established relay session whose expiry task is stalled behind a held session effect permit: the socket still ends within the bounded wait. - PostgreSQL: an established audio session whose teardown is stalled after its Close, and a stalled archived-channel exit, both end within the bounded wait from the hand-off. - PostgreSQL: a fast refusal (unsupported protocol version) through the real `handle_audio_connection` with a one-slot permit budget keeps the permit taken until that socket ends. - PostgreSQL: `huddle_audio_unavailable` sends its error and then Close `1008` as the very next frame. --------- Signed-off-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz> Co-authored-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>	2026-10-19 16:13:27 -04:00
Will Pfleger	5b433d9292	feat(relay): resolve NIP-FI issuer bindings from the database (#8377) ... Stack: #8372 (merged) → this PR → #8399 → #8401 Makes relay login checks read the per-community NIP-FI issuer bindings that #8372 stores. Without this, the relay can't keep up with bb-public, where new communities are created at every signup and a static `BUZZ_NIP_FI_COMMUNITIES` list won't do. **Source switch.** `BUZZ_NIP_FI_COMMUNITY_SOURCE=database` resolves each community's trusted issuers from `community_nip_fi_issuers`, filtered by the issuer registry, instead of the static list. Each community gets its own policy ID: the base ID no longer hashes every community audience, and the final ID hashes the base plus this community's audience, so policy IDs change once (see `docs/operator-nip-fi-bindings.md`). `Config::from_env` refuses to start, in every mode, on an unknown source, on `database` together with `BUZZ_NIP_FI_COMMUNITIES`, and on `database` without `BUZZ_NIP_FI_PROVISION_BINDING`. `BUZZ_NIP_FI_RESOLVER_*` and `BUZZ_NIP_FI_SNAPSHOT_*` are validated in every mode, so a bad value fails the deploy that sets it. In `off` mode no resolver is built and no snapshot is loaded. **Resolver (`nip_fi_resolver.rs`).** - **Snapshot:** each pod loads every binding into memory, capped at `BUZZ_NIP_FI_SNAPSHOT_MAX_ROWS` (going over is a failed load). It refreshes every 30 s, start to start, with jitter that only shortens the delay and backoff after failures. Each attempt is limited by `BUZZ_NIP_FI_SNAPSHOT_REFRESH_TIMEOUT_MS`, which must be shorter than the interval. A refresh whose deadline passes before it publishes publishes nothing. The old snapshot is dropped off the runtime. - **Freshness:** a host in a snapshot loaded under two minutes ago is answered from memory. Any other host, and every host once the snapshot is stale, goes to a live lookup. Every answer carries a deadline two minutes after the read behind it started, and WS, HTTP and audio recheck it just before admitting. - **Live lookups:** bounded and single-flight. In-flight lookups are capped by `BUZZ_NIP_FI_RESOLVER_MAX_INFLIGHT` (default 8, clamped below the database pool size), and the pending hosts and their waiters have fixed caps. Positive answers are used for that admission only. A "not found" is cached for 5 s, up to `BUZZ_NIP_FI_RESOLVER_NEGATIVE_CACHE_MAX` hosts, and only when no snapshot was published and no binding was created while it ran. A lookup that panics still leaves the pending map. - **Removal:** a removed binding stops admitting new connections within 30 s plus the refresh timeout (40 s by default) while refreshes succeed, and within two minutes of the last good load while they fail. Open WebSocket sessions are not rechecked when a binding is removed; they end at their own session deadline. **Shadow mode** resolves through the same path as enforce, including the database lookup and the freshness rules. It keeps the resolved binding's deadline and checks it at the same point as enforce; a binding that goes stale during the tenant lookup is recorded as `stage=deadline, outcome=unavailable`, without changing the legacy response or closing the socket. With `source=database` its `community` metric label is only `bound` or `unmapped`; with `source=env` it keeps the existing per-community audience label, so current shadow dashboards are unchanged. Each request resolves its Host once: the router guard's resolution, failures included, is reused by shadow recording and admission. **Provisioning.** Both operator provisioning paths drop this pod's cached "not found" for the new host after a commit that bound issuers, so the creator's first login looks it up. **Startup and readiness.** With `source=database` and a mode other than `off`, the first snapshot loads before any listener binds, under its own startup phase `nip_fi_snapshot_load`. A failed load exits non-zero, so a pod is never ready without a snapshot. **Operational consequence:** boot already fails when the database is unreachable; what is new is that a snapshot over `BUZZ_NIP_FI_SNAPSHOT_MAX_ROWS` or a load past its timeout also crash-loops new pods. After two minutes the old pods' snapshots go stale and most logins fall through to the 8-slot live lookup, which sheds the rest with 503. The way out is raising the cap or timeout, or `mode=off`. **Metrics.** `buzz_nip_fi_binding_snapshot_*` covers loads, failures, rows and freshness. `buzz_nip_fi_binding_snapshot_loaded_timestamp_seconds` is written on every refresh attempt, so alert on `time() - x > 120` together with `absent(x)`. `buzz_nip_fi_binding_snapshot_rows` against `buzz_nip_fi_binding_snapshot_max_rows` gives the headroom to the row cap. `buzz_nip_fi_resolver_*` covers live lookups and the negative cache. The env vars and alerts are documented in `.env.example`, the chart README and `docs/operator-nip-fi-bindings.md`, which also has the staged off → shadow → enforce rollout for `source=database` with go/no-go signals at each step, and the over-cap and load-timeout failure modes with their ways out. **Regression coverage.** Router-level tests run the database resolver through the real `build_router` for root WS, audio and protected HTTP, in enforce and shadow mode. Over PostgreSQL and Redis, two pods using the env community source (two exact bindings) replay assertions retained from before a disconnect through a real root NIP-42 session, the audio deny-set check and signed NIP-98 requests: they are denied on both pods, and a root session upgraded before the disconnect is refused by the post-registration deny-set check. Postgres tests in `buzz-db` check that archived and quiescing communities, and issuers missing from the registry, never reach the snapshot or the live lookup. Process-level `boot_lifecycle` tests spawn the real relay with `source=database`: an over-cap or timed-out startup load exits non-zero with no health listener bound, and a loadable snapshot lets readiness turn ready. --------- Signed-off-by: Will Pfleger <pfleger.will@gmail.com> Co-authored-by: Hayt <211b96e6a2b7f45fd4047988976c7bbbeeda0c15f3ae7b32eec20834b5a55118@buzz.block.builderlab.xyz> Co-authored-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>	2026-10-19 10:32:18 -04:00
Bradley Axen	3ffbfa693c	feat(relay): start a channel with its members in one create (#8421) ... ## What A kind:9007 can now name the channel's first members with `p` tags. Those members are added in the same database transaction that creates the channel. Resending a create whose OK was lost is now accepted instead of refused. ## Why Starting a new Solo chat today takes about 6–8 sequential relay round trips before the first message goes out: create, read, add each agent (kind:9000), read the roster, then send. With members in the create, a client can start a chat with its agents as two writes: the create, then the message as soon as the create is OK. The resend fix makes the create safe for an outbox to retry. Today an identical resend after a lost OK is refused with `duplicate: channel already exists`, so the client can't tell success from a real UUID collision. ## How - **buzz-db:** `Db::create_channel_with_event` writes the channel, its owner, its members (role `member`, `invited_by` = creator) and the 9007 event itself in one transaction. A channel row therefore never exists without its create event. `create_channel_with_id` is unchanged for its other callers. - **Validation** (`side_effects::initial_members` and the `validate_admin_event` 9007 branch): - Each `p` must be a 32-byte hex pubkey. - Duplicates and the creator are dropped. - At most 32 members. - An `h` tag is required when there are members. - Each member goes through the same `decide_channel_add_policy` check a kind:9000 uses, so an agent that refuses third-party adds keeps the whole create out. - **Side effects:** each initial member gets the same `member_joined` system message and member-added notification a kind:9000 would post. The fallback in `handle_create_group` creates a bare channel when the pre-created row is missing; it announces no members, because it added none. - **Resend:** when the UUID is already taken, the same transaction checks whether this exact event is the stored create. - If it is, the create is OK'd as `duplicate:` (it is the same signed create). - Anything else, including a changed create from the same author, is refused and writes nothing, so it can never re-run side effects or announce members. - A resend that arrives while the first create is still uncommitted waits on its channel insert. It is confirmed once the first commits, or creates the channel itself if the first rolled back. Before this, a resend in that window was refused, and the app then withdrew a chat that existed. - **No compensation:** ingest no longer soft-deletes a pre-created channel when the event insert fails. The rollback covers it. - **Announcements:** a failed `member_joined` notice for one initial member is logged at `error` and the loop continues, so the members after it still get theirs. A resend writes nothing, so nothing would retry them. ## Compatibility A 9007 with malformed `p` tags, or more than 32 of them, used to be accepted with the tags ignored. It is now rejected with `invalid:`. The relay's own SDK and CLI builders send no `p` tags on 9007. ## Out of scope - The kind:9000 OK still doesn't prove the membership change committed: side-effect failures are logged and OK'd, and a duplicate replay skips side effects. The new path doesn't depend on kind:9000. - Events in separate HTTP requests are still unordered. Clients send the first message after the create's OK. ## Tests `store::channel::postgres_tests` (buzz-db): - `a_resend_beside_an_open_create_confirms_it_once_committed`: holds a create open, checks that the resend waits, commits, and expects `Exists { resent: true }` with one owner and one member. - `a_resend_beside_a_failed_create_creates_the_channel`: the same, but the first create rolls back. - `a_create_whose_event_fails_leaves_no_channel`: an event the insert refuses leaves no channel row, and a retry creates the channel. - `another_create_for_a_taken_channel_writes_nothing`. Two mutations, each caught: committing the channel before its event fails the "leaves no channel" test, and looking up the event before the channel insert waits fails the "confirmed once committed" test. `handlers/create_with_members_postgres_tests.rs`, all through the real ingest path: - `create_adds_its_p_tags_as_members` checks: - the roles and the 39002 roster - exactly one `member_joined` notice - that the agent can post with no separate add - `create_respects_the_members_add_policy`: an agent with policy `nobody` gets an `invalid:` rejection, and no channel is created. - `resending_a_create_confirms_it`: - an identical resend is accepted - a changed create (from the creator or from someone else) is refused - the roster is unchanged and nobody new is announced - Unit tests for `initial_members`: dedupe, creator excluded, bad hex, the 32-member cap, and the missing `h` tag. Each new test was mutation-checked against the change it covers: no members, no policy check, refuse every resend, and let a changed creator create through. Each mutation turns its test red. Agent exercise on a local relay, over `POST /events` the way the app sends: - The 9007 with `p` = agent was accepted in 54–97ms (debug build). - The kind 9 was accepted 12–15ms after the create's OK. - The agent was in the 39002 roster, and the operator listener delivered the mention. - An identical resend came back `duplicate:` (accepted), and a changed create was refused. Local suite runs at 10adcf6dd (the transaction fix), on PG14: - `postgres-test-run.sh -p buzz-relay -p buzz-db`: 28 failures. That is the same 27 as below, plus `personal_read_frontier_is_monotonic…`, which hit a statement timeout under load and passes when run alone. - `cargo test -p buzz-db -p buzz-relay`: the same 6 `api::media` failures and the flaky `mesh_demo` test. Earlier local suite runs: - New tests pass and clippy is clean. - `postgres-test-run.sh -p buzz-relay -p buzz-db`: 27 failures (partition maintenance, a huddle links test, a redis boot test), and the same 27 fail on a96ace4e5. - `cargo test -p buzz-relay`: 6 `api::media` failures (they need a DB at the default URL) also fail on base. - One `api::mesh_demo` test is flaky: it passed 2 of 3 branch runs and failed once on a timeout. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Bradley Axen <baxen@squareup.com> Co-authored-by: mbot <3754f8729004d95654c46dbab3129e4ab9ef05cc2534e2a3fbfc155983bd637b@buzz.block.builderlab.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: claude00 <fb8b76a01257932b8ca35060862c29bd09b2137638573731c2d5074fc02c5385@buzz.block.builderlab.xyz> Co-authored-by: ART <ea39c1c8acf57180161ef4ba4ed68abf8589f5291956b909a5aa991fc66664ca@buzz.block.builderlab.xyz>	2026-10-18 11:32:09 -07:00
Will Pfleger	6286a54321	feat(relay): store per-community NIP-FI issuer bindings (#8372) ... Stack: this PR → #8377 → #8399 → #8401 bb-public creates communities all the time, so a hand-edited `BUZZ_NIP_FI_COMMUNITIES` map can't list them. This PR stores an explicit NIP-FI issuer allowlist for each community in the database and writes it when the community is created. Login checks don't read the table yet; that comes in #8377 together with the database-backed resolver. Operator steps are in [`docs/operator-nip-fi-bindings.md`](docs/operator-nip-fi-bindings.md). **Storage.** Migration `0066_community_nip_fi_issuers` adds `community_nip_fi_issuers (community_id, issuer, created_at)`, with primary key `(community_id, issuer)` and the community write fence attached. The assertion audience is `https://<communities.host>`, so the table has no `aud` column. Zero rows means no binding. The table is included in whole-community deletion and purge. **One bindable rule.** `BUZZ_NIP_FI_PROVISION_BINDING` (`{"host_suffix": ..., "authorized_issuers": [...]}`) is parsed in `buzz-auth` with `BUZZ_NIP_FI_ISSUERS`, whatever `BUZZ_NIP_FI_MODE` is set to, and both the relay and `buzz-admin` use the same parser. Each authorized issuer must be a configured issuer (1 to 2048 bytes), and the suffix must not cover any host declared in `BUZZ_NIP_FI_COMMUNITIES`; an unreadable declaration refuses startup even in mode `off`. `ProvisionBindingPolicy::is_bindable(host)` is true for a host exactly one label under the suffix that is not reserved. It decides creation-time binding, explicit `bind`, and what the lookup route reports. **Binding on creation.** Rows are written in the same transaction as the community insert on all three creation paths (`POST /operator/communities` `create_only`, its legacy convergence branch, and startup bootstrap), and only when the call actually created the community, so existing rows are never rewritten. **Reserved host.** A host whose `https://<host>` equals a configured `command_audiences` value never holds a binding. Operator provisioning of it is rejected with a 400 and the `reserved_host` error code, bootstrap creates it unbound, and the availability route reports it unavailable with an error code. `main.rs` calls `verify_reserved_hosts_unbound` at startup; if a reserved host has rows the relay refuses to start and names the fix (`buzz-admin nip-fi unbind --host <h> --issuer <i>`, or revert the command audience). A database failure during that check is a separate `Check` error. **Operator commands.** - `buzz-admin nip-fi backfill` pages unbound communities by host (1000 per page) and binds each page with one `INSERT ... SELECT` limited to active communities. It skips communities being deleted, always prints `{rows_bound, unbound, unbound_hosts, truncated}`, and exits non-zero while any bindable community is still unbound. Archived communities are bound on purpose. Both bind paths take the shared deletion lock and then lock the community row (`FOR NO KEY UPDATE`, in id order for a batch), the same order deletion uses, and the backfill insert also requires the community to still have no binding, so an operator `bind` that lands after the backfill listed a community wins and the default issuer is not added beside it. - `buzz-admin nip-fi bind --host --issuer` refuses a reserved host, a host outside the suffix, or an unconfigured issuer. - `buzz-admin nip-fi unbind --host --issuer` refuses to remove a community's last issuer (archive or delete the community instead), with one exception: when a valid provisioning policy shows that backfill can't bind the host (it's reserved, or outside the suffix), the last issuer can be removed. That is the reserved-host recovery path in the runbook. A missing or malformed policy doesn't block `unbind`; it just means the exception never applies. **Lookup route.** `GET /operator/communities/nip-fi-binding?host=` (NIP-98, `RELAY_OPERATOR_PUBKEYS`) reads the database and returns `{host, active, issuers[]}`. `host` is the stored host. It returns 404 for an unknown host, 200 with `active:false` for an archived or quiescing community, and `issuers:[]` when no provisioning policy is set. It lists only issuers that are configured and on a bindable host. kgoose depends on this contract; it is also documented in the runbook. **NIP-FA.** The `relay_url` bullet now also says: "An adapter MAY obtain its exact relay list from a configured, authenticated source; it MUST still compare `relay_url` exactly." **Deployment dependency.** The kgoose config side is already handled by [cash-server#137671](https://github.com/squareup/cash-server/pull/137671). In public-community mode it requires the command audience not to be a community audience. --------- Signed-off-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz> Co-authored-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>	2026-10-17 13:13:49 -04:00
Brad Seiler	2d0f78c6e1	feat(auth): support explicit mesh JWKS transport (#8382) ... ## Summary Allow an issuer to opt into mesh HTTP for its JSON Web Key Set (JWKS). Public HTTPS remains the default. This supports deployments whose service mesh authenticates and protects the key endpoint. The sealed source contract selects transport and contributes to policy identity. Mesh mode requires a DNS hostname and checks every resolved address against an explicit allow-list before pinning one address. Both modes disable proxies, reject redirects, cap responses, and enforce a whole-fetch deadline. The relay cannot verify mesh security: operators must enforce mutual TLS, endpoint authorization, and egress restrictions. ### Related issue No duplicate mesh-JWKS issue or PR found. Existing NIP-FI relay enforcement: #7224. No deployment change is included. ### Testing - At `eccb8cb19420abda44cf6e5e62b1361e2843bd1d`: full auth package and auth/relay all-target clippy passed. Full relay package passed: 1,457 library tests, 37 binary tests, and 9 boot tests; 446 tests were ignored. - At prior head `113939838cd81d5e8d0b0f8783c42ffaae44a4d3`: the full relay library suite failed the demo echo test (1,455 passed, 1 failed, 440 ignored). Independent checks reproduced that timeout at base `d61fe2071d2b3d8e56dd931947f16bfe0d6173fe`; the receive-cancellation defect is outside this change. A later pass does not establish that defect is fixed. - Three falsifying mutations at `eccb8cb194` were caught: mixed DNS answers, proxy bypass, and whole-fetch deadline. The full locked auth package passed after restoring the exact source. - At `eccb8cb194`, `cargo build --locked --release -p buzz-relay -p buzz-cli -p buzz-admin` passed with command-local `CARGO_PROFILE_RELEASE_STRIP=none`. This avoids a macOS loader error without a source, lockfile, or project-profile change. - Earlier hooks failed the ACP signal test and Tauri login-shell PATH test. The ACP failure also reproduced on unmodified main `93ffc37c`. Main merged without conflict. The full Tauri package then passed with `RUST_TEST_THREADS=1`: 3,422 library tests and 3 integration tests, 19 ignored. All pre-push hooks passed with that same command-local setting; no hook or test was skipped. - At `eccb8cb194`, isolated `just test` passed all 107 runner steps. The monolithic `just ci` run timed out after 1,120 seconds; all recipes passed across that run and separate `just desktop-tauri-test`, `just web-build`, and `just mobile-test` invocations. No monolithic pass is claimed. - Exact-head release live-local checks passed: loopback HTTP mesh JWKS warm; valid assertion 101; expired/wrong issuer/wrong audience 403; valid signed disconnect 200; wrong command audience 403. Redirect-only JWKS returned assertion 503 without a request to its target. Both omitted/default and explicit public transport rejected HTTP at startup. - A separate NIP-FI-off release CLI smoke test created an open channel, sent an accepted message, read it back, and fetched its thread. The assertion probe closes after upgrade; it does not prove a complete NIP-42 session. Local loopback tests do not prove deployment mesh authentication, endpoint authorization, egress controls, or real `240/4` routing. Generated with Codex ## Human Testing valid assertion / configured community audience: HTTP 101 (expected 101) expired assertion: HTTP 403 (expected 403) wrong issuer: HTTP 403 (expected 403) wrong community audience: HTTP 403 (expected 403) valid command JWT / configured command audience: HTTP 200, body=b'{"disconnected": true}' (expected 200) valid command signature / wrong audience: HTTP 403 (expected 403) --------- Signed-off-by: seiler <seiler@squareup.com> Signed-off-by: Local Builder <1c476cb1127ea270cb900bdc3891060140bdd1173f91498030ee9b4db01c2f20@buzz.block.builderlab.xyz> Co-authored-by: Codex <noreply@openai.com> Co-authored-by: Local Builder <1c476cb1127ea270cb900bdc3891060140bdd1173f91498030ee9b4db01c2f20@buzz.block.builderlab.xyz>	2026-10-16 14:26:08 -04:00
Bradley Axen	d039d9a2d5	fix(relay): confirm requested channels in one membership query (#8378) ... ## What A request that names channels with `#h` used to check each channel missing from the cached access set with its own `is_member` query. These checks ran one after another, so the database cost of a request grew with the number of channels it named. That cost is why the relay capped explicit `#h` values at 128, and why mobile splits large communities into many requests ([#8342](https://github.com/block/buzz/pull/8342)). This PR makes the check one query, and then raises the cap. - **One membership query.** WS REQ, WS COUNT, and HTTP `/query` and `/count` now share `confirm_requested_channel_access`. It checks every unconfirmed channel with one `membership_pairs` statement (`channel_id = ANY($2)`), so one request costs one query, whatever number of channels it names. `membership_pairs` uses the same SQL rules as `is_member`: deleted channels and removed members don't count, and it runs on the writer pool. - **Hash-set lookups.** Comparing the requested channels with the accessible set no longer costs requested × accessible comparisons. This covers `apply_channel_scope_to_query`, the system-channel scope helpers, the COUNT and bridge loops, and `#h` dedupe. - **`MAX_EXPLICIT_CHANNEL_VALUES` goes from 128 to 2048.** Huddle liveness used the same constant to limit session IDs. It now has its own constant, `MAX_HUDDLE_LIVENESS_SESSIONS`, which stays at 128, because each linked session still costs a separate mesh-directory lookup. Authorization rules are unchanged: - A scoped token still sets the upper limit. A channel outside the token is never checked. - Channels already cached and readable system channels are skipped. - Repaired channels are added in request order, each once, so `keep_repaired_system_channels_named` still sees the same repaired tail. - A database error still closes the request (or returns 500 over HTTP). One difference: the old loop could leave a partial repair if a later query failed, and that can no longer happen. ## Not in this PR - **Client chunk sizes.** `desktop/src/shared/api/relayClientShared.ts:39` and `mobile/lib/shared/relay/nostr_filters.dart:8` stay at 128. They can go up only after every relay they talk to runs this change. - **Per-connection limit on live channel scope.** Each authorized channel in a live WS subscription still gets a fan-out index entry for each kind, plus a Redis topic retain. Real membership limits this, not the cap. Still, an account in at least 2048 channels could now register up to 1024 × 2048 channel scopes on one connection; before, the most was 1024 × 128. If we want to keep the old worst case, the fix is a limit on the total channel scope per connection, not per REQ. ## Testing - `cargo clippy -p buzz-relay --tests` and `cargo fmt` are clean. - `cargo test -p buzz-relay --lib`: 1449 passed, 6 failed. All 6 are in `api::media::tests` and fail because the local Postgres on :5432 has no `buzz` role. Nothing in this change touches them. - I ran the Postgres lane (`scripts/postgres-test-run.sh`) for the bridge, req, count, group-state, and system-channel tests on local PG14: 55 passed. That includes the new `named_channels_missing_from_the_cache_are_confirmed_together`, which checks that the reader's channels missing from the cache are added in request order, each once, and that a stranger's channel stays out. - Not done yet: a live run against a local relay sending a REQ with more than 128 `#h` values. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Signed-off-by: Bradley Axen <baxen@squareup.com> Co-authored-by: murderbot <3754f8729004d95654c46dbab3129e4ab9ef05cc2534e2a3fbfc155983bd637b@buzz.block.builderlab.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-15 17:22:37 -07:00
Logan Johnson	f148c0f22c	feat(relay): add profile search over name and display_name only (#8301) ... 🤖 > **Review as a pair with block/buzz-app#1084** (the app change that uses this endpoint). Either can merge first: the app falls back to the old search on relays without this endpoint. ## Summary Searching for a person by name over the relay's general search returns profiles that only *mention* the text somewhere else: in an avatar image embedded as text, in the bio, or in the verification address. A search for "wes" could fill every result page with profiles whose avatar data happens to contain "Wes", and never return the person actually named Wes. This adds a relay endpoint, `POST /profiles/search`, that looks only at a profile's **name** and **display name**. Every other profile field is ignored. **Matching and paging are the same as the relay's existing full-text profile search** (NIP-50 search over kind:0 profiles with `search_mode: "prefix"`). A client can switch to this endpoint without a visible change in which people a query finds: - Each query word must match a whole word of the name; the last word can be a prefix. Words can be in any order, so "johnson wes" finds "Wes Johnson". Case is ignored. - Pages work the same way: up to 100 results per page, up to page 1000. - The results are exactly the old results minus profiles that match only outside their names. The order is better for people search: exact name first, then names that start with the query, then names with a later word that starts with it, then other matches. Ties are ordered by `ts_rank_cd` over the two name fields only (not the whole profile, as in NIP-50), then newest first, then event id. The response is a JSON array of the original signed profile events (Nostr kind:0), so clients can verify them as they do today. The buzz-app client change is block/buzz-app#1084. ## Details - Request: `{"query": "wes", "limit": 20, "page": 1}`. `query` is 1–100 characters after trimming; `limit` is 1–100 (default 20); `page` is 1–1000 (default 1); unknown fields are rejected with 400. - The search reuses the full-text search's own query builder (`buzz-search`), so the word rule cannot drift from it. The existing full-text index on events stays the access path; the name check runs only on rows the index already matched. There is no new index. One small migration (`0061`) adds the function `profile_name_fields()`; it reads no table data, so it runs instantly. - Only current profiles count: a replaced profile stops matching. Profiles that are not a JSON object, or whose name fields are not strings, never match. Content that PostgreSQL cannot read as `jsonb` (for example a `\^@` escape or very deep nesting) also never matches, and it cannot make the search fail for everyone else. - The handler loads the matched events through the normal routed event read and applies the same reader visibility check as other reads, in search order. - Access uses the same checks as the relay's GIF search endpoint: a signed HTTP request (NIP-98), identity admission (NIP-FI), rate limit, replay protection, and relay membership. That code moved out of `gifs.rs` into one shared helper in `bridge.rs`; GIF behavior is unchanged. - The route is added to the existing protected-route and NIP-FI shadow/enforce route tables. Docs: `docs/profile-search.md`. - Two test-only lines get `sadscan:disable` comments, as other fixtures already do: an existing public ES256 test key in `router.rs`, and the local Postgres test URL in the search integration test. ### Related issue None found. ### Testing - Postgres test for the search query (`profile_names_match_name_words_only`): only name fields match (avatar data URL, `about`, `nip05`, `banner`, non-string names and non-object content are excluded); words in any order and across both name fields; whole words only for completed words; ranking order; pages split without overlap or gaps; replaced profile; other community; `%` literal. - Postgres end-to-end test through the router: an admitted member gets only the name-matching signed event, and the returned event still verifies. - Request validation unit tests; existing GIF admission tests and router route-table tests pass. - Local end-to-end run with buzz-app#1084 against a relay built from this head, with about 195 seeded profiles. For every query tried, the endpoint's results were a subset of the full-text search's results; the only profiles missing were avatar-only matches. --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-15 17:35:33 -04:00
Kalvin C	cdfc38ecbd	feat(search): add newest-first sort to bridge search (#8371) ... ## Summary Adds an opt-in newest-first order to bridge search, so clients can offer "Most recent" alongside relevance. - `buzz-search`: `SearchQuery` gains `sort: SearchSort`. `Relevance` keeps the existing `ts_rank_cd` ordering. `Newest` orders hits by `created_at DESC, id`, and `page` steps through that order. Newest skips the exact-profile-lexeme prioritization, which only applies to relevance ranking. - Bridge `POST /query`: search filters accept `search_sort: "newest"`. When the field is missing or has another value, results stay in relevance order. Community scope, channel intersection and per-hit authorization are unchanged. - NIP-11 lists `buzz-search-sort` in `supported_extensions`. Clients only send the field to relays that advertise it, so they never get relevance results labelled as newest. - WebSocket NIP-50 search stays relevance-only. `NOSTR.md` documents the bridge extension. ### Related issue N/A ### Testing - New Postgres integration test `newest_sort_pages_by_time_then_id_within_scope`. It pages a channel in newest order, including an equal-timestamp tie split across a page boundary and a newer hit outside the channel scope. It checks that the pages join up in `(created_at desc, id asc)` order and that relevance still ranks the strongest match first. - Unit tests: - `bridge_search_sort_extension_selects_newest_only_when_named` - `advertises_nip_cl_and_search_sort_extensions` - Ran the buzz-search Postgres integration suite, the relay `nip11` and `bridge_search` unit tests, `cargo fmt --check`, and `cargo clippy -p buzz-search -p buzz-relay --all-targets -D warnings` on Linux. - Local relay exercised through signed `POST /query`: - NIP-11 lists `buzz-search-sort`. - Relevance and newest return different orders. - Newest pages 1–3 return 25 hits with no gaps or duplicates, including same-second ties across page boundaries. - A non-member gets no hits from a private channel; a member gets all of them. - Client integration is in a companion block/buzz-app change. It was exercised end to end against this relay through the production broker: switching sort, paging, scope changes and capability fallback. Deployment note: clients only offer newest sorting after a relay with this change advertises `buzz-search-sort`. --------- Signed-off-by: Kalvin Chau <kalvin@block.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-15 13:27:49 -07:00
Logan Johnson	d28f6b2f70	feat(relay): let channel-state filters opt in to system channels (#8182) ... 🤖 ## Summary After part 2, a client can read a system channel only if it already knows the ID. This PR lets a client find system channels on purpose, with an ordinary Nostr filter and no new filter key. A filter on the channel-state kinds (39000–39003) now also returns the system channels the reader can read when it has either: - `#t:["system"]`, the channel type tag that every channel-state event already carries; or - any `#P`, the creator tag from #8174. Asking for the channels of a named creator returns a small result, so it does not fill generic channel lists. The filter then matches as usual. Existing clients send neither on these kinds, so they still never see a system channel. | Reader wants | Filter | |---|---| | My system channels | `{kinds:[39002], #p:[me], #t:["system"]}`, then their kind:39000 by `#d` | | Open system channels | `{kinds:[39000], #t:["system"]}` | | One creator's labeled channel, system or not | `{kinds:[39000], #P:[creator], #t:[label]}` | Private system channels stay hidden from non-members, as for any private channel. Values in one tag filter combine with OR, so `#t:["system", label]` means `system` **or** the label; use `#P` plus the label instead. Part 3 of 3, stacked on part 2. ### Related issue None found. ### Details - The opt-in is decided per filter in `handlers/system_channels.rs` and applies on REQ, COUNT and the HTTP bridge. It applies only when every kind in the filter is 39000–39003. - A label alone is not an opt-in. - `NOSTR.md` documents the system channel query rules. - **Spec.** NIP-CL says how a filter asks for system channels. - The relay information document (NIP-11) now lists `"nip-cl"` in `supported_extensions`. Clients use this to tell if a relay supports labels and system channels, and refuse those calls on a relay that does not. NIP-CL documents this. ### Testing - End-to-end relay test: `#t:["system"]` finds both system channels for a member and only the open one for a non-member; 39002 `#p:[me]` with `#t:["system"]` lists the member's system channels; `#P` includes a creator's system channels; creator plus label returns exactly one channel with `limit: 1` and excludes another creator's channel with the same label; a label alone does not return the system channel; COUNT and HTTP `/query` and `/count` agree. - The test fails when the opt-in is disabled. - Unit tests for which filters opt in. ### Stack These are two stacks. Merge them in this order. Client support (creating, labeling and finding these channels) will be in block/buzz-app, not the buzz SDK and CLI. **Stack 1: channel labels and creator lookup (#8174–#8176)** 1. #8174 channel type and creator tags, and the channel creation time 2. #8175 channel labels 3. #8176 database-side label and creator match, and the rule for choosing one channel **Stack 2: system channels (#8180–#8182)**, stacked on stack 1 because it uses labels and the creator tag 4. #8180 the system channel type 5. #8181 leave system channels out of reads that do not name them 6. #8182 channel-state filters opt in to system channels ← this PR --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-15 08:40:16 -04:00
Logan Johnson	cdf03f618a	feat(relay): leave system channels out of reads that do not name them (#8181) ... 🤖 ## Summary After part 1, a system channel is still returned by every query that returns ordinary channels, so existing clients would show it as a chat. This PR makes the relay leave system channels out of every read that does not name them. Clients do not need to filter, so clients that know nothing about system channels stay correct. | Query | Before | After | |---|---|---| | A query that does not name a channel (no `#h`), COUNT, full-text search, the HTTP `/query` and `/count` routes, artifact queries (`{"artifact": ...}`), the sidebar | Includes system channels | Leaves them out | | "Channels I belong to" (kind:39002 `#p:[me]`) | Includes them | Leaves them out | | A filter that names the channel with `#h`, or with `#d` on kinds 39000–39003 | Ordinary read rules | Unchanged: members can read it, and anyone can read it if it is open | Push notifications and "you were added to a channel" notifications (kinds 44100/44101) skip system channels, because a client that received one would show the channel. A client that is given a system channel's ID, for example in a pasted link, can still open it. Naming the ID is the opt-in. This keeps system channels ordinary NIP-29 groups. Part 2 of 3, stacked on part 1. ### Related issue None found. ### Details - The rule lives in one place: `get_accessible_channel_ids`, the set of channels a query may read when it does not name one. It is the union of the reader's memberships and every open channel; both halves now exclude `channel_type = 'system'`. Because the filter is in SQL before `LIMIT`, pages stay full. - A filter that names a system channel adds it to that filter's scope only (`handlers/system_channels.rs`). A sibling filter in the same request keeps the implicit set, so naming one system channel cannot widen another filter. - Thread windows and the workflows API name their channel, so they follow the same named-channel rule. - Artifact queries have their own read check in SQL. It reads a system channel only when the query's `#h` names it. - When a request re-checks membership for a named channel that its first lookup missed, a confirmed system channel joins only the filters that name it. Before, a reader who joined a system channel during the request could have it added to a sibling filter's scope. - The system-channel lookup runs only when a filter names a channel outside the implicit set, so ordinary queries do not pay for it. - **Spec.** NIP-CL lists the reads that leave system channels out, and says when a filter names one. `AGENTS.md` notes this for `get_channels`. ### Testing - End-to-end relay test (Postgres and Redis) with a private system channel, an open system channel and an ordinary channel, as a member and as a non-member: REQ without `#h`, NIP-50 search, COUNT, HTTP `/query`, HTTP search, HTTP `/count`, the channel list, and 39002 `#p:[me]` return only the ordinary channel. Named reads by `#h` and `#d` follow member-or-open on REQ, COUNT, search and HTTP. - Relay test: adding a member to a system channel sends no kind:44100; an ordinary channel still does. - Both tests fail when the SQL exclusion or the named-channel scope is removed. - Database test: an artifact in an open system channel is missing from `{"artifact":"current"}` and present with `#h:[<system channel>]`, for a member and a non-member. It fails without the channel-type condition. - Relay test: the membership re-check confirms a system channel that the system-channel lookup missed. The channel goes to the filter that names it, and a sibling filter's scope does not include it. It fails without the fix. - Unit tests for the per-filter scope. They now run in `just test-unit` and `scripts/run-tests.sh`. ### Stack These are two stacks. Merge them in this order. Client support (creating, labeling and finding these channels) will be in block/buzz-app, not the buzz SDK and CLI. **Stack 1: channel labels and creator lookup (#8174–#8176)** 1. #8174 channel type and creator tags, and the channel creation time 2. #8175 channel labels 3. #8176 database-side label and creator match, and the rule for choosing one channel **Stack 2: system channels (#8180–#8182)**, stacked on stack 1 because it uses labels and the creator tag 4. #8180 the system channel type 5. #8181 leave system channels out of reads that do not name them ← this PR 6. #8182 channel-state filters opt in to system channels --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-15 08:40:15 -04:00
Bradley Axen	8b60dcf8e3	feat(relay): deliver full signed events to operator listeners (#8309) ... ## Summary Operator-listener mention notifications now carry the full signed event, so a listener such as kgoose can wake an agent without first fetching the event (one signed round trip less on time-to-first-reply). Because the notification now carries message content, two mitigations come with it, as agreed in #buzz-security: 1. **Each target signs its own registration.** `POST /operator/listener/pubkeys` takes `{"registrations": [proof, …]}`. Each proof is a kind `24245` event signed by the target pubkey, with exactly one `p` tag (the listener pubkey) and one `u` tag (`{RELAY_OPERATOR_API_ORIGIN}/operator/listener/pubkeys`), and a `created_at` within ±60s. The old unsigned `{"pubkeys": […]}` POST body is rejected with 400. `DELETE` is unchanged. 2. **Only notify a target that can read the event.** Before each delivery the relay re-checks, on the writer, that: - the registration still exists; - the event isn't deleted; - neither the target nor its recorded NIP-OA owner is banned; - the target passes the relay-membership gate (with the owner path when `allow_nip_oa_auth` is on); - the channel is open, or the target is an active member. A delivery that fails the check retries with backoff instead of being dropped, because a mention can commit just before the target is added to the channel. It fails permanently, with no HTTP request, once attempts are exhausted. The payload is now `v: 2`: the v1 fields plus `event`, the full signed NIP-01 event. There's no migration. Existing registrations stay in place and age out under the 30-day retention; only a configured listener could have created them. Listeners must sign proofs to register new targets, so receivers have to accept `v: 2`. Companion kgoose change: squareup/cash-server#136794 signs registrations with the agent key. The kgoose receiver in squareup/cash-server#136004 currently accepts only `v: 1` and bodies up to 4 KB, so it needs to accept v2 and a larger body. ## Testing - Unit tests: - proof verification (forged signer, wrong kind, stale or future, other or ambiguous listener, other or missing endpoint); - signed-only request bodies; - delivery payload carries a verifiable event; - unreadable or failed loads retry, then fail, without an HTTP request. - Postgres tests: - every access predicate (open, member and non-member private, channel-less, closed community with target or owner membership, owner ban, removed member, deleted channel or event, removed registration); - the route test (signed registration accepted; a proof for another listener and the unsigned body both rejected). - Mutation runs confirmed that the membership, owner, ban and reader-visibility guards each fail a test when removed. - Local relay exercise with a fake listener: - signed registration returned 200; unsigned, stale and wrong-listener proofs returned 400; - open-channel mentions delivered v2 with an event whose id and signature verify independently; - a private-channel mention of a non-member retried without delivering, then delivered once the target was added. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Bradley Axen <baxen@squareup.com> Co-authored-by: murderbot <3754f8729004d95654c46dbab3129e4ab9ef05cc2534e2a3fbfc155983bd637b@buzz.block.builderlab.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-14 17:53:25 -07:00
Will Pfleger	e645918c63	fix(relay): match pushed-down tags by position, not containment (#8335) ... 🤖 The relay pushes several tag filters into SQL as JSONB containment (`tags @> [[name, value]]`). Containment ignores element position, so a tag like `["x","e",<id>]` or `["meta","t",v]` satisfies `#e:[<id>]` or `#t:[v]`, and `#t:["t"]` matches `["t","other"]`. Because the predicate runs before `LIMIT`, a newer event that only matches by containment can take a `limit: 1` slot; the exact NIP-01 post-filter then drops it and the real match is never returned. Where no post-filter follows, the decoy is returned or counted outright. `push_exact_tag_match` in `buzz-db/src/store/event.rs` keeps containment as the GIN-indexed prefilter (one array-bound `tags @> ANY(...)`, so the e-tag planner shape is unchanged) and adds a positional `EXISTS (... tag->>0 = name AND tag->>1 = ANY(values))`. It now backs every `EventQuery` tag pushdown: | Pushdown | Before | Effect of over-match | |---|---|---| | `#e` (`e_tags`, query and COUNT) | containment | REQ `limit` slot theft; COUNT has no post-filter and counted decoys; thread aux closure returned decoy reactions/replies | | artifact `#d` (`d_tag_values`) | containment | `limit` slot theft on NIP-AR identity lookups | | `custom_tag` (bridge `#buzz-channel`) | containment | no post-filter: a decoy project/repository was returned for its channel | | shared-gated visibility (`shared_gated_reader`) | containment of `["shared","true"]` | a private event tagged `["x","shared","true"]` took a page slot before `event_visible_to_reader` dropped it, so the page came back short. Never a visibility leak: the post-filter stays in place | **Bridge `/count` honors `#buzz-channel`.** `POST /query` scoped project/repository reads by `#buzz-channel`, but `POST /count` ignored it, so a count for one channel counted matching projects in every channel the reader can see. `/count` now sets the same `custom_tag` on both its channel-scoped and access-scoped paths, and the COUNT SQL builder pushes `custom_tag` down exactly like the query builder, so `/count` and `/query` select the same events on both the fast count and the post-filtered fallback. Left unchanged: the retention-free purge predicate for mesh status (`tags @> [["k","buzz-mesh-status"]]`). It mirrors the migration 0019 trigger, has no `LIMIT`, and only affects how the author's own deleted event is purged. REQ `#t` outside group-state kinds is still post-filtered only. --------- Signed-off-by: Will Pfleger <pfleger.will@gmail.com> Co-authored-by: Hayt <211b96e6a2b7f45fd4047988976c7bbbeeda0c15f3ae7b32eec20834b5a55118@buzz.block.builderlab.xyz>	2026-10-15 00:37:11 +00:00
Logan Johnson	71d7aa0436	feat(relay): add the system channel type (#8180) ... 🤖 ## Summary Buzz uses channel membership to decide who can read what, and every channel a person belongs to shows up in their client as a chat. So there is no way to make a group that exists only for access control, for example "only this agent and its owner can read the agent's config". Such a channel would appear in both people's sidebars, search results and unread counts. This stack adds a **system channel**: an ordinary channel that queries leave out unless they ask for it. This first PR adds the type itself. - `kind:9007` (create channel) with `["channel_type", "system"]` creates a system channel. - It is private unless the creator sends `["visibility", "open"]`. - While it is private, its channel metadata event (kind:39000) carries the NIP-29 `hidden` marker, which means "hide this group's metadata from non-members". Until now the relay set `hidden` only for DMs. On its own this PR does not yet hide system channels from queries. Part 2 does that. Part 1 of 3, stacked on #8176 (the channel labels and creator lookup stack, which this feature uses). ### Related issue None found. ### Details - Migration 0062 adds `system` to the `channel_type` Postgres enum. No new columns. - `buzz_core::channel::ChannelType::System`, and `system` is now a reserved label name (labels cannot equal a channel type name). - `buzz-admin` reconcile writes `hidden` with the same rule as the relay. Usage metrics count system channels. - **Rollback.** A relay older than this change cannot read `channel_type = 'system'`. Do not roll back past it while system channels exist. - **Spec.** NIP-CL gets a "System channels" section. `ARCHITECTURE.md` lists the `System` type. ### Testing - Relay test: a system channel is private and `hidden` by default; opening it removes `hidden`; its first `t` tag is `system`. - Migration embed and usage-metrics tests. ### Stack These are two stacks. Merge them in this order. Client support (creating, labeling and finding these channels) will be in block/buzz-app, not the buzz SDK and CLI. **Stack 1: channel labels and creator lookup (#8174–#8176)** 1. #8174 channel type and creator tags, and the channel creation time 2. #8175 channel labels 3. #8176 database-side label and creator match, and the rule for choosing one channel **Stack 2: system channels (#8180–#8182)**, stacked on stack 1 because it uses labels and the creator tag 4. #8180 the system channel type ← this PR 5. #8181 leave system channels out of reads that do not name them 6. #8182 channel-state filters opt in to system channels --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-14 19:03:34 -04:00
Logan Johnson	a3909acc67	feat(relay): match label and creator filters before the page limit (#8176) ... 🤖 ## Summary After parts 1 and 2, a client can ask for "channels with label X" or "channels created by key K". But the relay applied these tag filters after it read one page of the newest channel events. So a request such as `{kinds:[39000], #t:["workspace"], limit:10}` could return fewer than 10 channels, or none, while matching channels existed. This PR moves the label and creator match into the database query, before the page limit. Part 3 of 3, stacked on part 2. ### Related issue None found. ### Details - Applies only when every kind in the filter is a relay-signed channel-state kind (39000–39003). Other kinds keep today's behavior. - The database match uses JSONB containment on `tags`, which is GIN-indexed. Containment can match a little too much (for example, a tag that contains both strings in another position), never too little. So the relay still checks each returned event with the normal Nostr filter rules, and `COUNT` requests keep that check too. - The same match is applied to the query path and the count path. - **Spec.** NIP-CL now says that `#t` and `#P` match before `limit`. - **Choosing one channel.** The relay does not make labels unique, so two creates for one label can both succeed, for example when two processes of one agent start at the same time. NIP-CL gets a "Choosing one channel" section: a client that needs one channel per creator and label looks it up with `#P` + `#t`, creates it only if none exists, and otherwise uses the oldest one (the lowest `created_at` tag, then the lowest channel ID). Every reader that follows this rule picks the same channel, even after an edit. ### Testing - Relay test: many channels, one match, small `limit`. The request returns the match for both `#t` and `#P`. This test fails without the change. - Unit test: `#t` and `#P` reach the database query only for channel-state kinds. - Relay test: one creator sends two creates for the same label at once. Both channels exist, a `#P` + `#t` lookup finds both, and the oldest-channel rule picks the same one in either order, even after the older channel's metadata is re-published by an edit. ### Stack These are two stacks. Merge them in this order. Client support (creating, labeling and finding these channels) will be in block/buzz-app, not the buzz SDK and CLI. **Stack 1: channel labels and creator lookup (#8174–#8176)** 1. #8174 channel type and creator tags, and the channel creation time 2. #8175 channel labels 3. #8176 database-side label and creator match, and the rule for choosing one channel ← this PR **Stack 2: system channels (#8180–#8182)**, stacked on stack 1 because it uses labels and the creator tag 4. #8180 the system channel type 5. #8181 leave system channels out of reads that do not name them 6. #8182 channel-state filters opt in to system channels --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-14 19:03:34 -04:00
Logan Johnson	2f79374f48	feat(relay): channel labels (#8175) ... 🤖 ## Summary A plugin or agent often needs to tag its own channels and then find them again, for example "my attention-config channel" or "all channels of this workspace". Today a channel has no labels: the relay ignores any `t` tags in create or edit requests. This PR adds channel labels. - A create request can carry up to 8 labels as `["t", "<label>"]` tags. - An edit request with `t` tags replaces the label set. `["t", ""]` alone clears it, the same convention as `["ttl", ""]`. Only owners and admins can change labels, as for the name. - The relay publishes each label as a `t` tag on the channel's state events, after the channel type tag. Part 2 of 3, stacked on part 1. ### Related issue None found. ### Details - **Rules.** A label uses 1–64 characters from `a-z`, `0-9`, `.`, `:` and `-`. The relay refuses a label equal to a channel type name (`stream`, `forum`, `dm`, `workflow`, `system`), so a label cannot make a channel look like another type. - **No trust from labels.** Any admin can set a label. Only the relay-written creator tag (part 1) says who made a channel. - **Storage.** New migration `0061_channel_labels.sql` adds `channels.labels TEXT[] NOT NULL DEFAULT '{}'`. Channels without labels look and behave as before. - **Readers.** The CLI and `buzz-acp` read the channel type from the first `t` tag. Before this PR they used the last one, which would now be a label. - **Spec.** NIP-CL gets a "Channel labels" section with these rules. ### Testing - Relay tests: labels at creation appear after the type tag, edit replaces and clears them, invalid labels are refused, and members who are not admins cannot change them. - Unit tests for the label rules: allowed characters, length, reserved type names, the 8-label cap and the clear marker. - `buzz-acp` test: the channel type comes from the first `t` tag, not a label. ### Stack These are two stacks. Merge them in this order. Client support (creating, labeling and finding these channels) will be in block/buzz-app, not the buzz SDK and CLI. **Stack 1: channel labels and creator lookup (#8174–#8176)** 1. #8174 channel type and creator tags, and the channel creation time 2. #8175 channel labels ← this PR 3. #8176 database-side label and creator match, and the rule for choosing one channel **Stack 2: system channels (#8180–#8182)**, stacked on stack 1 because it uses labels and the creator tag 4. #8180 the system channel type 5. #8181 leave system channels out of reads that do not name them 6. #8182 channel-state filters opt in to system channels --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-14 17:25:46 -04:00
Will Pfleger	b3795236cc	fix(relay): reject unreadable write targets as missing (#8249) ... 🤖 A sender who cannot use or read a referenced event, artifact or workflow now gets exactly the response it would get for an ID that does not exist. #8197 (merged) set this rule for deletions; this applies it to the other writes that look up a target, plus the workflow read and webhook endpoints. Artifact `create` and workflow definition saves are the exceptions, described below. Access is checked before any message that depends on the target, readability goes through `event_visible_to_reader` (including #8197's private-deletion rule), and soft-deleted targets keep matching missing ones. | Path | Checks, in order | Response for any unusable target | |---|---|---| | Reaction (kind 7) | target lookup → `event_visible_to_reader` → every channel gate for the derived channel, including archived | `invalid: reaction target event not found` | | Edit (kind 40003) | target in the edit's `h` channel and readable → author / agent-owner check, including an author since removed from a private channel | `invalid: edit target not found or not editable by you` | | Forum vote (kind 45002) | target in the vote's `h` channel and readable → kind check | `invalid: vote target event not found` | | Report by event (kind 1984 `e`) | `event_visible_to_reader` → token channel scope → channel membership or open | `invalid: report target event not found` | | Reply (NIP-10 and workflow `send_message` replies) | parent in the reply's channel and readable | `reply parent not found` | | Artifact update / delete / restore / move | head exists and is homed in `h` (move: in the source the sender can write) → revision, type, deleted and root checks | `conflict: artifact head unavailable` | | Workflow trigger (kind 46020) | workflow exists and sender is its owner | `invalid: workflow not found or not triggerable by you` | | `GET /workflows/{id}/runs` and `/runs/{run_id}/approvals` | workflow exists, has a channel, and the caller can read it | `404 {"error":"workflow not found"}` | | `POST /hooks/{id}` | workflow exists and the provided secret matches its stored secret → definition and trigger-type checks | `404 {"error":"workflow not found"}` | Two of these reject writes that used to be accepted: a reaction to an event the sender cannot read (such as someone else's author-only event), and a report on an event the reporter cannot read. Reports by pubkey and by blob are unchanged. In an archived channel, writes whose channel comes from their target (reactions and deletions) get that target's not-found response, and an artifact move out of an archived source channel gets `conflict: artifact head unavailable` like any other source the sender cannot write; every other write keeps `invalid: channel is archived`. A webhook call with an absent or wrong secret, or to a workflow without a webhook secret, now gets `404` instead of `401`/`400`. Artifact `create` and workflow definition saves still report a taken ID, as NIP-AR allows for artifacts. Removed messages, none of which any client depends on from these endpoints: `edit target event not found`, `must be event author to edit`, `target event belongs to a different channel`, `target event has no channel`, `parent event belongs to a different channel`, `parent event has no channel association`, `artifact home changed`, `workflow is not accessible`, `workflow is not channel-scoped`, the webhook's `authentication failed` and `webhook secret required but not configured`, `forbidden: not authorized to trigger this workflow` for non-owners (it remains for an owner whose channel authority lapsed), and `restricted: not a channel member` for an edit by an author removed from a private channel. `e2e_existence_oracle` checks each event write over HTTP and WS, and the workflow read and webhook endpoints over HTTP, against missing, soft-deleted, other-channel, private-channel, archived, author-only and wrong-kind targets as applicable, with a positive control per path. It runs in the Backend Integration job alongside `e2e_author_only_deletion`. `removed_author_edit_does_not_reveal_target` checks that a removed author's edits of their remaining message, their deleted message and a missing ID get identical answers. In the PostgreSQL suite, `head_outside_authorized_channel_matches_missing_head` covers the artifact transaction, `workflow_reply_to_unusable_parent_matches_missing` covers workflow replies through `RelayActionSink::send_message`, `scoped_token_report_does_not_reveal_target` and `scoped_token_reaction_does_not_reveal_target` cover channel-scoped tokens, and `channelless_workflow_runs_read_as_missing` covers the runs read for a workflow with no channel. --------- Signed-off-by: Will Pfleger <pfleger.will@gmail.com> Co-authored-by: Hayt <211b96e6a2b7f45fd4047988976c7bbbeeda0c15f3ae7b32eec20834b5a55118@buzz.block.builderlab.xyz>	2026-10-14 17:18:38 -04:00
Luke Tornquist	d25384a4f3	buzz-db: admit every fenced-table writer in app code (#8219) ... ## Summary This makes every serving writer of a community-write-fenced table take community admission in application code. That's the "Dual enforcement" step of the Supported Writer Contract in `ARCHITECTURE.md`. Admission means the shared community deletion lock plus the active/not-deleted check from `deletion::guard_community_write`. It's taken before the transaction's first tenant write or row lock and held until commit. The 32 `community_write_fence_*` triggers, the six fence SQL functions and `validate_serving_catalog` are unchanged. They stay authoritative. Removing them is a separate migration that will be checked table by table. ## Changes - **New admission entry point.** `begin_community_write_transaction` (`runtime/mod.rs`) builds an `AdmittedTx` for non-event writers. It sits next to the existing event-write entry points and shares their implementation. - **Writers moved onto it:** - `api_tokens`, `archived_identities`, `channels`, `channel_members` (including DM and admin paths), `community_bans`, `git_repo_names` - `moderation_actions` and `moderation_reports`, `personal_read_accounts` and `personal_read_frontiers`, `pubkey_allowlist` - `push_leases`, `push_match_queue`, `push_wake_outbox` - `relay_invites`, `relay_members` (including the new-community owner bootstrap and the startup `pubkey_allowlist` backfill), `join_policy_acceptances` - `thread_metadata`, `users`, `workflows`, `workflow_runs`, `workflow_approvals`, `scheduled_workflow_fires` - Personal-read helpers now require `&mut AdmittedTx`. - **`buzz-audit`.** An append now passes admission first, then takes a transaction-scoped audit-chain lock (`pg_advisory_xact_lock`). This replaces the session lock and the `catch_unwind` unlock cleanup. The relay audit worker retries a `lock_timeout` on either lock and keeps the entry; an admission refusal is terminal, as the trigger's refusal was. - **Fleet-wide statements.** These keep `community_write_allowed(...)` in the same SQL statement, because there's no single community to admit up front: - scheduled-fire prune - invite expiry - push matcher claim and exhausted-job cleanup - startup `d_tag` backfill - channel TTL cleanup - **Relay-member removal.** `remove_relay_member` and `remove_relay_member_if_role` run `delete_member_with_acceptances` inside their admitted transaction. The helper no longer opens its own transaction, so the member lock, the `join_policy_acceptances` delete and the member delete all hold admission until commit. `delete_workflow` and `delete_workflow_for_owner` are gone on `main`, and the coordinate deletion path they used to back was already admitted. - **Unchanged exceptions.** The deletion-executor generation exception and the serving-lease path work as before. - **Tables with no production writer.** `delivery_log` and `subscriptions` are only written by a historical cutover script and a test fixture. Their triggers stay. - **No cross-community moves.** No production `community_id` move exists on a fenced table. The only `SET community_id = NULL` is the deletion-executor purge. So no dual-community lock ordering was needed. ## What got simpler - Per-store ad-hoc writer helpers are gone: `begin_event_write_transaction` and `acquire_event_write_connection` in `channel.rs`, and `begin_operation_transaction` in `push.rs`. - Audit append drops the session advisory unlock, the `catch_unwind`/`resume_unwind` panic handling, `log_inner` and the `futures-util` dependency. ## Notes for reviewers - **New crate dependency.** `buzz-audit` now depends on `buzz-db` only to call the free function `buzz_db::deletion::guard_community_write`. `ARCHITECTURE.md` names this as the one sanctioned subsystem edge, and its audit section is updated to match. - **Hot paths.** `ensure_user` first checks whether the user exists and returns without admission when it does (no write happens). Only a real insert takes admission. `touch_api_token` has no production caller today. - **No-op writers stay read-only.** The startup allowlist backfill returns early on plain reads when the legacy table is gone, it has no rows for the community, or the community already has members, so a quiescing deployment community still starts. It re-checks under admission before inserting. - **Owner mutations.** When admission refuses a transfer or a provisioning and a deletion request exists, both report `DeletionPending`, as before. For a transfer, the operator API keeps its coded `deletion_lifecycle_conflict` 409. For provisioning, it keeps the existing uncoded 409, `owner_conflict: community deletion is pending`. - **Per-community worker loops.** `claim_due_push_wakes` and similar now return `AccessDenied` for a community that goes quiescing between the active-host scan and the claim. The worker logs it at `debug` and the next scan skips the community. - **Kick side effects.** `membership_removal_fence` admits before the membership lock to keep the lock order of `add_member`. A community that leaves `active` between the kick and the fence skips live subscription eviction; the membership cache is already invalidated and fan-out re-checks membership. This is documented on the function. ## Testing - **New refusal tests.** 19 tests: the 18 in `store::admission_postgres_tests`, plus `audit_append_requires_application_admission_with_triggers_bypassed`. Each runs with ordinary triggers disabled (`session_replication_role = replica`), checks that a raw fenced insert would succeed, and asserts that the writer is refused for both `quiescing` and `fenced` communities. Active writes still succeed. - **The tests catch a missing check.** Measured in review at `6fc073c3f`: with the guard in `AdmittedTx::admit` bypassed, 25 tests fail. That includes 16 of the 18 database admission tests. The audit test stays green there, because audit calls `guard_community_write` directly. - **Affected PG suites, with triggers on:** 183/183 passed across 16 store suites, push 18/18 run isolated, audit 7/7. The existing artifact, reaction and replaceable admission tests pass. - **Review fixes, each checked to fail without its fix:** - `allowlist_backfill_requires_application_admission` and `owner_mutations_report_pending_deletion_when_admission_refuses` fail with the old `relay_members.rs`. - `empty_allowlist_backfill_skips_admission` fails (`AccessDenied … write-fenced (quiescing)`) without the allowlist-rows early return. - `audit_worker_retries_admission_lock_timeout_until_original_entry_is_appended_once` holds the community deletion lock exclusive and fails without the admission retry arm. - The existing audit-chain retry test now looks for the transaction-scoped lock and passes. - The two owner-revoke mini-schema tests (`bridge.rs`, `ingest.rs`) now include `communities` and `community_deletion_lock_key`, which admission reads; both failed without that. - **Static checks:** `cargo fmt --check` and workspace `cargo clippy --all-targets -D warnings` pass. - **Not run locally:** the `mobile-check` step of `just ci` (the build machine couldn't reach pub.dev, and clean base fails the same way) and the full `just test` lane. CI on this PR covers both. `mesh_demo::demo_join_forwarded_arm_round_trips_echo` fails on the build machine at base `394d4a434` as well; it uses no database. Refs BUZZ-314. --------- Signed-off-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz> Co-authored-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz>	2026-10-14 10:58:12 -04:00
Logan Johnson	8bb49a1b6a	feat(relay): publish channel type and creator on every channel-state event (#8174) ... 🤖 ## Summary Clients cannot tell who created a channel, and only the channel-metadata event says what type of channel it is. This PR makes every relay-signed channel-state event carry both facts, so a client can find "the channel that this key created" with one ordinary Nostr filter. - Channel metadata, admin list and member list events (kinds 39000, 39001 and 39002) now all carry the channel type as their first `t` tag, then the creator as a `P` tag. - The creator comes from the relay's own record of who signed the channel's create request. No client can set it, and it does not change when ownership moves. - Channel metadata (kind:39000) also carries `["created_at", <unix seconds>]`, the time the channel was created. The event's own `created_at` changes on every edit, so before this a client could not tell which of two channels is older. This is part 1 of a 3-part stack (channel labels and creator lookup). The system channels stack (#8180–#8182) builds on it. ### Related issue None found. ### Details - **Why `P`.** Several Nostr specs (NIP-22, NIP-34, NIP-72) use uppercase `P` for the author of the root object, and lowercase `p` for people involved. Lowercase `p` is already used on these events for members and DM participants, and a filter matches tag values, not their meaning, so `#p` would mix them. - **Trust.** A client that trusts a channel because of its creator must also check that the relay key signed the event. - **Order.** The type `t` tag comes first, so clients that read the first `t` tag still get the type after part 2 adds labels as more `t` tags. - **Creation time.** The relay reads it from `channels.created_at`, so an edit does not change it. #8176 uses it to pick one channel when two creates race. - `buzz-admin reconcile` writes the same tags when it rebuilds these events. `NOSTR.md` documents the tags. - **Spec.** New `docs/nips/NIP-CL.md` defines these tags, the trust rule and lookup by creator. Later PRs in both stacks add their parts to it. `NOSTR.md` links to it. ### Testing - Relay tests in `group_state_postgres_tests.rs`: each of the three kinds carries the type first, then the creator; the creator stays the same after an ownership transfer; kind:39000 carries the stored creation time, and still does after an edit. - Unit test for the shared tag builder that both the relay and `buzz-admin reconcile` use. ### Stack These are two stacks. Merge them in this order. Client support (creating, labeling and finding these channels) will be in block/buzz-app, not the buzz SDK and CLI. **Stack 1: channel labels and creator lookup (#8174–#8176)** 1. #8174 channel type and creator tags, and the channel creation time ← this PR 2. #8175 channel labels 3. #8176 database-side label and creator match, and the rule for choosing one channel **Stack 2: system channels (#8180–#8182)**, stacked on stack 1 because it uses labels and the creator tag 4. #8180 the system channel type 5. #8181 leave system channels out of reads that do not name them 6. #8182 channel-state filters opt in to system channels --------- Signed-off-by: Logan Johnson <loganj@squareup.com> Co-authored-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>	2026-10-14 09:22:42 -04:00
Luke Tornquist	52d13efc08	fix(buzz-db): delete workflow and membership children explicitly (#8212) ... ## What Workflow deletion and relay-member removal relied on `ON DELETE CASCADE` to remove child rows. Both paths now delete their own children inside the same transaction, so a later migration can drop those foreign-key actions without changing behavior. **Workflow deletion** (`store/workflow/deletion.rs`, the coordinate path used by NIP-09 deletes of workflow definitions): - Locks the target workflow row with `SELECT … FOR UPDATE`, under the same owner and coordinate predicate as before. A concurrent run or fire insert has to lock that row for its foreign key, so the child deletes see every committed child. - Deletes scheduled fires first. A new fire has to lock the workflow row, and a cron attach locks the fire before the run, so deleting fires before the run lock keeps the lock order consistent and avoids a deadlock. - Then locks the workflow's runs `FOR UPDATE` in `id` order. An approval that names another workflow but references one of these runs key-share locks only the run, so the workflow lock alone doesn't exclude it. - `delete_workflow_with_children` then deletes approvals (for the workflow or any of its runs), then runs, then the workflow, and returns its `channel_id`. **Relay-member removal** (`store/relay_members.rs`): `remove_relay_member` and `remove_relay_member_if_role` share a new `delete_member_with_acceptances` helper. It locks the member row under the existing predicate (any non-owner, or an exact role), deletes `join_policy_acceptances`, then deletes the member, in one transaction. The `RemoveResult` outcomes are unchanged. **Removed:** `delete_workflow` (no callers) and `delete_workflow_for_owner` (test-only), plus their `Db` wrappers. The community-confinement test now drives the production coordinate deletion. ## Tests - `test_support::retire_foreign_key_delete_actions` rewrites every `CASCADE`, `SET NULL` and `SET DEFAULT` foreign key in the test's own per-test database as `NO ACTION`, which is the target schema of the follow-up migration. It asserts that none remain. All triggers stay live, including foreign-key checks and the community write fence. It refuses to run unless `current_database()` has the per-test wrapper's exact shape (`buzz_nt_` plus 24 hex digits), which excludes the run's `_desired` template. It checks the name only. - `workflow_deletion_removes_children_without_cascades` and `member_removal_deletes_acceptances_without_cascades` run the production paths on that schema and require every child row to be gone. One approval references another workflow but this workflow's run, which pins the `run_id` branch. The owner, role-mismatch and not-found outcomes still hold. - `workflow_deletion_waits_for_and_removes_concurrent_run` and `member_removal_waits_for_and_removes_concurrent_acceptance` hold a child insert open in a second transaction, wait until the removal is blocked on a lock, and then commit. The removal must succeed with no children left. - `workflow_deletion_waits_for_concurrent_fire_attach_without_deadlock` holds a fire row the way the cron attach does, then attaches a run to it while the deletion waits. Both must succeed. - `workflow_deletion_waits_for_and_removes_concurrent_cross_workflow_approval` does the same with an approval for another workflow that references this workflow's run. - The bridge test schema now includes `join_policy_acceptances`. ## What got simpler Two delete paths no longer depend on implicit constraint actions, and two unused workflow-delete functions are gone. This is the first step of retiring the `CASCADE` and `SET NULL` foreign keys. **Identified debt (later PRs):** `channel_members` and `personal_read_frontiers` (no hard-delete path found yet for their parents), the deletion `manifest_keys` cascade, the `SET NULL` actions on `executor_heartbeats` and `product_feedback`, and the `users` self-reference. That last one is a composite `(community_id, agent_owner_pubkey)` `ON DELETE SET NULL` that would try to null the `NOT NULL` `community_id`. Also, a scheduled fire for workflow B can link to a run of A (that FK is already `NO ACTION`), which would block A's run delete; the scheduler always attaches a run to its own workflow's fire, so this needs same-workflow enforcement in the cascade work. ## Verification Head `081be4c0d`, tree `ab649487`, on a Linux cloud workstation: fmt, buzz-db and workspace clippy, source policy and buzz-db lib pass; the six new tests pass; buzz-db PostgreSQL passes 490/490 and buzz-relay PostgreSQL 378/378. A sixth mutation, moving the fire delete back after the run lock, makes `workflow_deletion_waits_for_concurrent_fire_attach_without_deadlock` fail with `40P01` (deadlock detected). Previous head: Head `787c7c42d`, tree `6b9cf5d5`, on a Linux cloud workstation: fmt, buzz-db and workspace clippy, source policy and buzz-db lib pass; the new tests pass 5/5; buzz-relay PostgreSQL passes 378/378; buzz-db PostgreSQL passes 488/489, the one failure being `parent_contention_times_out_without_blocking_the_other_table` in `store/partition/`, which this PR doesn't touch and which passes on its own at the same tree. A fifth mutation, removing the run lock, makes the new cross-workflow approval test fail with the reviewer's exact 23503 on `workflow_approvals_community_id_run_id_fkey`. Previous head `00cba2f4a`, tree `2f89fed4`, base `e9269cbdf`. Results from a Linux cloud workstation: - fmt, buzz-db and workspace clippy, the source policy and buzz-db lib all pass. - The buzz-db PostgreSQL suite passes 488/488. - The buzz-relay PostgreSQL suite passes 378/378. Each of four mutations makes exactly one of the new tests fail with the foreign-key violation you'd expect (23503): - swapping the fires and runs deletes - deleting approvals by `workflow_id` only - removing the workflow row lock - removing the member row lock GitHub CI is green at this head. buzz-review-completed --------- Signed-off-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz> Co-authored-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz>	2026-10-14 08:27:26 -04:00
Jarrod Sibbison	8599efb6a3	feat(relay): allow HTTP operator listener delivery URLs (#8287) ... ## Summary Operator listener configuration currently rejects HTTP URLs, preventing delivery to sibling services over internal HTTP connections. Accept HTTP and HTTPS delivery URLs while retaining the existing pubkey, duplicate-route, and URL restrictions. Add configuration coverage for internal hostnames and custom ports, verify signed delivery through the real HTTP client, and include the listener suite in both unit-test runners. Shared NIP-98 signing and push gateway code are unchanged. ### Related issue Related: #7934 refactors operator listener delivery tests. No matching open issue found. ### Testing - `cargo test -p buzz-relay --lib operator_listener`: 13 passed, including signed HTTP delivery and redirect rejection. - `cargo fmt --all -- --check`, shell syntax, Justfile parsing, and diff whitespace checks passed. - `just test`: passed. - `just ci`: stopped at mobile analysis with an existing undefined `buzzDevPushAppProfile` in `mobile/test/features/settings/notification_toggle_test.dart:81`. Workspace and native desktop formatting/Clippy, desktop frontend checks, and web checks passed. The mobile file is unchanged by this PR. Locally verified with a spiked listener client Signed-off-by: jsibbison-square <jsibbison@squareup.com>	2026-10-14 08:30:20 +11:00
Ian Oberst	85b7331784	fix(relay): partition audit observability follow-ups (#7994) ... > [!IMPORTANT] > **The alert rules changed in c3a314e71.** If you reviewed the alert guidance earlier, please look again. > - **The two range-based rules now apply only to live pods.** A range selector such as `[30m]` keeps reading a terminated pod's samples until they leave the window. So `BuzzPartitionAuditNotRunning` could fire for about 10 minutes after normal pod turnover, and `BuzzPartitionAuditFailing` could keep firing for up to an hour after the pod was gone. Each rule now adds `and <counter>`, which requires the counter to be exported right now: > ```yaml > expr: | > sum without (outcome) (increase(buzz_partition_audit_runs_total[30m])) == 0 > and sum without (outcome) (buzz_partition_audit_runs_total) > ``` > ```yaml > expr: | > increase(buzz_partition_audit_failures_total[1h]) >= 2 > and buzz_partition_audit_failures_total > ``` > - **The audit counters start at 0.** When the relay starts, it exports every `buzz_partition_audit_runs_total{table, outcome}` series and `buzz_partition_audit_failures_total` at 0, before the first audit. `increase()` ignores a new series' first sample. Before this change, it could miss the first periodic failure, or read a table's first `degraded` run as a stopped loop. This change is also what makes the `and <counter>` gate work: the counters now exist for a relay's whole life. > > The doc's old claim that pod turnover can't leave an alert firing was true only of the instant-selector stale rule. The doc now explains the difference. ## Why These are the observability follow-ups deferred from #6515 ("fix(db): audit partition catalog before creation"): - **Timestamp-age alerts go silent** ([review comment](https://github.com/block/buzz/pull/6515#discussion_r4127212141)). Partition gauges are set only when a table audits successfully, and the exporter evicts idle gauges after 2,700 s by default. After sustained failures, `buzz_partition_audit_last_success_timestamp_seconds` disappears. A table that has never audited successfully never has it. An alert on `time() - last_success` then has no input. - **Startup fallback failures are not counted** ([review comment](https://github.com/block/buzz/pull/6515#discussion_r4127212158)). When the startup fallback audit fails, it logs an error but does not increment `buzz_partition_audit_failures_total`. Periodic failures do. - **Repaired environments are always `degraded`**. `PartitionTableAudit::degraded()` counts every legacy leaf, even one wholly in the past. The `bb-block-staging` audit of the repaired catalog reports `degraded` only because of `events_p2026_07_08_legacy` `[2026-07-01, 2026-09-01)`. With this rule, every repaired environment stays `degraded`, so nobody can alert on the outcome. - **Partition config tests can see a poisoned env** ([review comment](https://github.com/block/buzz/pull/6515#discussion_r4137416297)). Two partition config tests take only `ENV_MUTEX`. A NIP-FI test running in the same process can make them see its deliberately invalid environment. Partition health is still diagnostic only, through `/_status.partition_catalog` and metrics. Nothing here affects `/_readiness`. Handling concurrent creation (#6515 item 4) is out of scope and stays with the catch-all advancement work. ## What - **Success timestamp survives failures.** Each failed table audit refreshes `buzz_partition_audit_last_success_timestamp_seconds` with `increment(0.0)`, which leaves the value unchanged. It still resets the exporter's idle tracking, the same mechanism as `refresh_legacy_active_gauge_recency`. Sustained failures therefore keep the last success exported and its age rising. A table that has never succeeded exports `0`. A failed audit still never makes the last success look fresh. - **Alert guidance.** New `docs/partition-catalog-monitoring.md` covers the metrics, how long each series lives, and the audit outcomes. It notes that `runs_total` has no `unsafe` outcome. It gives three alert rules: - timestamp age, which also covers never-successful tables and sustained failures; - `sum without (outcome) (increase(buzz_partition_audit_runs_total[30m])) == 0` for a stopped audit loop. Counters are never idle-evicted, and this rule fires before the stopped loop's timestamp is evicted, so there is no false resolve; - an aggregate failure-counter rule. Both range-based rules also require their counter to be exported right now (`and <counter>`), so a terminated pod's leftover samples can't raise or hold an alert. - **Zero counter baselines.** `try_install` exports each `buzz_partition_audit_runs_total{table, outcome}` series (`ok`, `degraded`, and `error` for both tables) and `buzz_partition_audit_failures_total` at 0 before the startup audit, the same way the auth series already start at 0. A failure in the startup audit itself can still happen before the first scrape. The doc says so; the stale rule covers that case. - **Failure counter.** `buzz_partition_audit_failures_total` now counts failed explicit audit attempts: the startup fallback audit and each periodic audit. Both paths go through one helper, `counted_partition_audit`. A failed `ensure_future_partitions` call, such as a DDL error that the fallback then recovers, is not counted by itself. A single attempt never counts twice. - **Historical legacy leaves.** A legacy leaf is historical when its finite upper bound is at or before the current month start. It is still reported under `children` as `legacy_leaf`, but it no longer degrades the table. A legacy leaf that still covers the current month or any lookahead month stays degraded. All other degradation causes are unchanged. The verdict now lives in `PartitionAuditReport::outcome()`, which `buzz-admin partition-audit` uses. The JSON outcome names and exit codes are the same as before. - **Test hygiene.** Both partition config tests now use `env_guards()`. ## How it was tested - `cargo fmt --all -- --check` and `cargo clippy -p buzz-db -p buzz-relay -p buzz-admin --all-targets -- -D warnings`. - New unit tests: - The repaired incident layout with an injected clock: `degraded` in August 2026 and `ok` on 2026-09-01T00:00Z, when the upper bound equals the current month start, and on 2026-09-15. The legacy leaf is still in the serialized report. The `outcome` label on the run counter follows the verdict. - A table test of the legacy-leaf predicate: the upper bound before, at, and one second after the month start; lookahead and beyond-horizon leaves; historical leaves with trigger drift; anomalous and `DEFAULT` children; no audited month. - Outcome precedence (error, then unsafe, then degraded) and the `buzz-admin` exit-code mapping. - Startup: a failed fallback counts once, a successful fallback after a DDL failure counts nothing, and a successful ensure never runs the fallback. Each periodic failure counts once. Against an unreachable database, startup counts one aggregate failure while each table still records two `error` runs. - On the production exporter, never-successful audits leave run counters and a `0` success timestamp for both tables. - On the production exporter, all six run-counter series and the failure counter read 0 before any audit. After one failed audit, `error` reads 1 and `ok`/`degraded` stay at 0. Removing the initialization fails this test. - PostgreSQL lane (`scripts/postgres-test-run.sh`) with new or extended tests: - The repaired layout is not degraded in September. The same catalog is degraded in August. - With a 1 s idle timeout, a failed audit leaves the success timestamp unchanged, and sustained failures keep it exported past the idle timeout. When audits stop, the timestamp is evicted but the run counters remain. The next success exports a fresh timestamp. - Mutation check: removing the failure-path refresh fails both alert fixtures. - These fixtures check the emitted series. They do not evaluate the rules themselves, including the `and <counter>` gate. `promtool` isn't available locally or in CI, and the doc points to `promtool test rules` for that. - `scripts/test-postgres-test-discovery.sh`. - `config::` and `nip_fi_config::` tests run together in one process five times. ## Related This PR conflicts textually with #7993 (catch-all advancement) in `crates/buzz-db/src/store/partition.rs` and `crates/buzz-relay/src/main.rs`. Whichever PR merges second will be rebased onto the other. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Ian Oberst <ioberst@block.xyz> Co-authored-by: Ian Oberst <ioberst@block.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-13 13:57:54 -07:00
Will Pfleger	40c1c474e7	fix(relay): hide deletions of author-only events from other readers (#8197) ... 🤖 Opened by an agent (Hayt) for Will Pfleger. The relay keeps `AUTHOR_ONLY_KINDS` (event reminders `30300`, push leases `30350`, private managed agents `30179`) readable only by their author, but a NIP-09 deletion request (`kind:5`) for one of those events was readable by any community reader. The deletion carries no `h` tag, so ingest stores it with its target's channel (none), and every read gate classified it by its own kind, `5`. Another reader could read and count it by author, by id, live, or over HTTP, learning the deleted ids, how many, and when. A deletion is now exactly as private as the event it deletes: - **Classification by the deletion's own `k` tag.** `buzz_core::kind::is_author_only_event_kind` treats a `kind:5` that names an author-only kind in a `k` tag as author-only. Only the canonical decimal spelling counts (`"30300"`, not `"030300"` or `"+30300"`), so Rust classification, ingest and the admin SQL (which matches the tag text) always agree. Because it reads the deletion, not a target lookup, the deletion stays private after its target is soft-deleted or gone. It is generic over `AUTHOR_ONLY_KINDS`, so a kind added to the list later (for example `44300`) is covered automatically. - **Every read surface.** The shared per-event gate (`is_author_only_event`, used by WS REQ historical and lookup by id, COUNT fallbacks, HTTP `/query`, `/count`, search hydration and the thread window), live fan-out in `handlers/event.rs`, the COUNT fast-path guard (`filter_can_match_author_only_kinds` now treats any filter that can match `kind:5`, including unfiltered and id-only filters, as needing per-event filtering), the channel-window aux closure in the bridge, and admin event preview plus report list and detail (SQL suppression of `kind:5` rows with a tag whose first two elements are `k` and an author-only kind, matching the Rust classifier; soft-deleted rows included). - **Enforcement still reaches the author.** The admin report query takes the private-deletion `k` list from its caller next to the hidden kinds. HTTP-facing reads pass both; `admin_get_report_for_enforcement` passes neither, so report-directed bans, timeouts and action recovery still resolve the stored author of a reported private deletion. - **Ingest agreement.** `validate_standard_deletion_event` rejects a deletion of a stored (including soft-deleted) author-only event, by `e` or by `a` coordinate, unless it carries the matching canonical `["k","<kind>"]`. It also rejects an author-only `k` on a deletion of any other kind, so a public deletion can't hide itself from the readers who need it. A deletion whose target is itself a private deletion is rejected with `invalid: cannot delete a private deletion request`: it would otherwise be stored as a public event revealing the private deletion's id and timing, and NIP-09 gives a deletion of a deletion no effect. A deletion of a public deletion is unchanged. The `k` tag only narrows who can read the request; it grants no one new deletion authority. - **Unauthorized deletions learn nothing about the target.** On every delete-by-id path, no target-dependent error (its kind, the artifact and `k`-marker errors, the artifact-removal-marker error, or where it lives) is returned until every access check that could block the deletion has passed. A sender who fails any of them gets `invalid: deletion target not found or not deletable by you`, the same text and status as for a missing target. Errors that depend only on the request, never on the target, are unchanged: for example a `kind:9005` whose own `h` channel is archived still gets `channel is archived`, for a missing target and a real one alike. "May not delete" includes an author who has since been removed from the target's private channel, an author whose token is scoped to a different channel, and anyone deleting in an archived channel. The single-target rule (exactly one `e` or `a` tag) now runs before any target lookup, so every target faces the gates. For `kind:5` with an `e` target, ingest decides authority (effective author or the owning human of the authoring agent) first, then takes the channel from that same lookup, soft-deleted targets included, so tombstones face the same gates; an authorized re-deletion of a soft-deleted target now keeps the target's stored channel instead of being treated as global, so historical access checks, `#h` filtering and live routing all use that channel: readers outside a private channel no longer see it, and that channel's members and subscribers do. The token channel scope, scoped-token global-event, membership and archived checks then reject with the shared denial, all before `validate_standard_deletion_event` returns its artifact and `k`-marker errors. Other kinds, and `kind:5` with only `a` coordinates, keep their existing errors. For NIP-29 `kind:9005`, a target outside the `h` channel is denied like a missing one, so the relay no longer returns "different channel" or "no channel"; `may_delete_channel_event` (the author while still a member or the channel is open, a channel owner/admin, or the owning human of the author agent) is then checked in the `h` channel before the artifact-removal-marker error. No access restriction is removed and nothing becomes newly deletable. An authorized sender who names the wrong `h` channel for a `kind:9005` now gets the shared denial. Whether kind:5 content is full-text indexed depends on how the database was built. The desired-state schema (`schema/schema.sql`, which CI applies) and databases created before migration `0008` index it; an empty database initialized through `0008` gets a positive allowlist that doesn't. Where it is indexed, the per-event gate on search hydration is what keeps a private deletion out of other readers' search results. No current client is affected by the new rejection. `buzz-app`'s signer only allows `kind:5` for channel messages (with `h`) and managed-agent `30177` coordinates, and neither `buzz-app` nor this repo's desktop or CLI publishes deletions for `30300`, `30350` or `30179` today. NIP-ER already requires `k=30300` on reminder deletions. Coverage: - `crates/buzz-test-client/tests/e2e_author_only_deletion.rs` (Backend Integration job): a non-author gets nothing from WS REQ, WS COUNT, HTTP `/query` and `/count` (by author, mixed kinds, and id), search, or live delivery, while the author gets the deletion. Search runs WS and HTTP against a private deletion and a public control deletion sharing a reason word; with `BUZZ_TEST_SEARCH_INDEXES_DELETIONS=1` (set by the CI step) the author must get exactly both and another reader exactly the control; ingest rejects a missing, wrong, non-canonical, or public-target marker, and an untagged deletion of an already soft-deleted target. - `e2e_author_only_deletion.rs` also sends one non-owner's deletions over HTTP and WS for a missing, live, soft-deleted, private-channel and global target, plus the `kind:9005` equivalents including a target in another channel and one with no channel, and requires the identical rejection text and HTTP status in every case; another member's deletion of a private event never names its kind. - `e2e_author_only_deletion.rs` `removed_author_deletion_does_not_reveal_target`: an author removed from a private channel gets the missing-target rejection over HTTP and WS for `kind:5` and `kind:9005` (with `h` set to a channel they can write to) against their retained message; the channel owner can still delete the removed author's message, and the author can still delete in a channel they can write to. - `e2e_author_only_deletion.rs` `archived_target_deletion_does_not_reveal_target`: over HTTP and WS, `kind:5` against a message, the same message with a wrong `k`, and an artifact in an archived channel, and `kind:9005` against the archived-channel message with `h` set to an active channel, all match the missing-target rejection; a `kind:9005` for an artifact-removal marker in the archived channel matches a missing target with the same `h`; in an active channel the owner still gets the `k`-mismatch and artifact errors. - `e2e_author_only_deletion.rs` `deletion_of_private_deletion_rejected`: over HTTP and WS, the author's deletion of their private deletion, with no `k` and with `k=5`, gets the rejection and stores nothing; a non-author's gets the missing-target rejection; a `kind:9005` against it gets the missing-target rejection; the private deletion stays hidden from another reader in WS REQ and COUNT, HTTP `/query` and `/count`, and live delivery; a deletion of a public deletion is still accepted. - `e2e_author_only_deletion.rs` `soft_deleted_target_keeps_its_channel`: an author removed from a private channel, and an author in an archived channel, re-deleting their own soft-deleted message get the missing-target rejection; a distinct successful re-deletion is readable by a channel member and not by an outsider, by id and by `#h`, and is delivered live to the member's `#h` subscription but not to the outsider's. - `e2e_author_only_deletion.rs` `multi_target_deletion_error_ignores_targets`: deletions with two `e` tags, or an `e` and an `a` tag, get the identical single-target error whether the first target is missing, live, in another private channel, a private reminder or an artifact. - `handlers::ingest::postgres_tests::scoped_token_deletion_rejection_does_not_reveal_target`: a non-owner and the target's author, each with a token scoped to another channel, get the same rejection for a missing, live, soft-deleted and global target, and the author's `kind:9005` naming the token's channel gets it for a missing target, a live target in another channel and a global target; a token scoped to the target's channel still deletes, and its re-deletion of the soft-deleted target is stored with that target's channel. - `thread_window::postgres_tests`: stored private deletions never reach a non-signer through the thread-window or channel-window aux closures, or by id once their target is gone. - `api::admin::postgres_tests`: ban and stranded-ban recovery on a reported private deletion reach the stored author while report detail hides it; `admin_moderation` tests treat non-canonical markers as public. --------- Signed-off-by: Will Pfleger <pfleger.will@gmail.com> Co-authored-by: Hayt <211b96e6a2b7f45fd4047988976c7bbbeeda0c15f3ae7b32eec20834b5a55118@buzz.block.builderlab.xyz>	2026-10-13 14:45:59 -04:00
Bradley Axen	1d9f2c81a5	feat(relay): accept an owner's presence for its agents (kind 20003) (#8261) ... ## Summary App-hosted agents (buzz-app agents2) run inside the desktop app without a socket per agent, so they can't send their own kind 20001 presence and always read as offline — including to their owner on another device. This adds **kind 20003, an owner's presence for its agents**: - Owner-signed; `content` is the status (`online` / `away` / `offline`); each `p` tag names an agent (1–256 distinct). - WebSocket only (rejected on HTTP ingest, like 20001). - The relay keeps only the agents the signer owns (`users.agent_owner_pubkey`) — cached `true` from `observer_owner_cache`, the rest resolved with one `list_agents_for_owner` query. Only positive links are cached, so a just-linked agent isn't hidden for the cache TTL and the observer path isn't poisoned. - Sets (180s TTL) or clears each agent's presence in one Redis round trip, the same keys kind 20001 writes, so the presence snapshot reports them with no client change. - Not fanned out: live presence is matched by author, and the author here is the owner. A new kind rather than `p` tags on 20001, so a relay without this ignores the batch instead of storing it as the owner's own presence. ## Testing - `agent_presence_request_names_distinct_agents_with_a_known_status` (unit). - `presence_storage_postgres_tests::agent_presence_stores_only_owned_agents` (Postgres + Redis): only the owner's agent goes online, another owner's agent doesn't, a pre-cached `false` doesn't hide a linked agent, and `offline` clears it. All 5 presence Postgres tests pass via `scripts/postgres-test-run.sh -p buzz-relay --lib -E 'test(/presence/)'`. - `buzz-pubsub` ignored Redis test `test_presence_many_sets_with_ttl_and_clears` passes. - clippy clean. Plain `cargo test` failures locally were environmental only (buzz-media "role buzz does not exist", mesh_demo 504). - Not yet exercised end to end against a local relay + app. App side: block/buzz-app#862. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Bradley Axen <baxen@squareup.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-12 21:43:24 -04:00
Bradley Axen	326e2301cb	perf(relay): drop per-pulse ban reads and owner-link writes on HTTP (#8233) ... ## Why Agents send an HTTP typing pulse (`POST /events`, kind 20002) every 3–5 seconds while they work. #8207 already routes the pulse's write-restriction and serving checks through #8208's short-TTL caches. Two per-pulse costs were still left at the membership step: - **A fresh ban read** for the agent, plus its owner when the agent comes in through NIP-OA (`deny_banned` → `community_ban_outcome`). - **Four writer round-trips for owner-admitted agents**, on every request: `materialize_nip_oa_owner` runs 2× `ensure_user` + the `agent_owner_pubkey` upsert + an `is_agent_owner` confirm, even though the link already exists. ## What changes - **Typing's ban check reads the restriction cache.** `submit_event` already parses the event before the membership step, so typing pulses call a new `enforce_relay_membership_ephemeral`. Its ban check reads through `restriction_state_cached`, the same 30s bound the pulse's write-restriction check already has. Every other HTTP route, persistent `POST /events`, and socket auth still read fresh (`cached: false`). - **`materialize_nip_oa_owner` skips a link it already knows.** If `observer_owner_cache` already holds `(community, agent, owner) = true`, it returns `true` right away. `agent_owner_pubkey` has one production writer, which sets it only `WHERE agent_owner_pubkey IS NULL`, and nothing clears it, so the skipped writes were no-ops. This helps every caller (HTTP, socket auth, audio). The owner-link disconnect only runs on the first write, so it still runs. - **`observer_owner_cache` capacity goes from 1k to 100k**, so concurrently active agents actually hit it. That's tens of MB per pod at most, the same scale as `restriction_cache`. Per-pulse Postgres work at the membership step: | Agent | Before | After | |---|---|---| | Direct relay member | host lookup (writer) + membership + ban | host lookup (writer) + membership | | Owner-admitted (NIP-OA) | host lookup + 2 memberships + 2 bans + 4 writer trips | host lookup + 2 memberships | ## Not in this PR - **Caching host → community.** It's the tenant-binding step every HTTP request goes through, and it reads from the writer. Community deletion depends on new requests not binding once a community leaves `active`. A cache would keep binding a quiescing community for its TTL. The serving-write leases may make that harmless, but that needs a closer look than a "safe fixes" PR. - **Caching relay membership.** Removing a member has no HTTP equivalent of the socket disconnect, so a cache would delay removal by its TTL. - **The desktop client reusing one connection.** That's going in a separate buzz-app PR. ## Unrelated one-liner The `sadscan` pre-commit hook scans whole staged files. It blocked this commit on a pre-existing test-only URI in `handlers/auth.rs` (`postgres://buzz:buzz_dev@127.0.0.1:1/buzz`, a deliberately unreachable stub). I added the hook's own `// sadscan:disable np.postgres.1` marker to that line. ## Testing - New Postgres tests. I confirmed each one fails when its guard is removed: - `typing_ban_gate_uses_cached_restriction_row`: a member banned in Postgres with a cached clear row can still type. Fails if typing reads fresh. - `persistent_post_ban_gate_reads_fresh`: a stale cached ban doesn't refuse a kind-9 post that Postgres allows. Fails if every kind reads the cache. - `known_owner_link_skips_materialize_writes`: after the link is cached, deleting the agent row behind the cache's back stays deleted. Fails without the short-circuit. - `scripts/postgres-test-run.sh -p buzz-relay --lib --tests`: 397/397 passed. This was local PG14. - `cargo test -p buzz-relay`: 1430 passed and 7 failed. The same 7 fail without this commit in this environment: 6 `api::media` tests can't find a `buzz` role on :5432, and `mesh_demo` returns a 504. - `cargo clippy -p buzz-relay --all-targets -D warnings` is clean. - An independent agent review found no blockers. Its optional suggestions are test gaps: nothing pins the owner-cascade cached read, or the `false` literals at socket auth. - **Not yet done:** a run against a live local relay, and human testing. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Signed-off-by: Honey <8e307ae0076a4dab6b94b036ea3edc7e08f823a625269c1e6919e881a048b4d2@buzz.block.builderlab.xyz> Co-authored-by: Honey <8e307ae0076a4dab6b94b036ea3edc7e08f823a625269c1e6919e881a048b4d2@buzz.block.builderlab.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-09 23:15:16 +00:00
Bradley Axen	6c388bde62	feat(relay): accept typing indicators over HTTP POST /events (#8207) ... ## Summary Agents hosted inside the Buzz app sign their own events but don't hold a WebSocket, so they couldn't show "typing…" in a channel. `POST /events` rejected kind 20002 (typing indicator) with `restricted: unknown event kind`. Now, when `POST /events` receives a kind 20002 event, the bridge passes it to the same channel broadcast the WebSocket path uses. Live subscribers in the channel get the pulse. Nothing is written to Postgres. Before broadcasting, the relay checks, in order: - the event has an `h` tag naming the channel - the serialized event is at most 2 KiB (real typing events are ~600 bytes). Measuring the whole JSON, not just content and tag values, means piling on empty tags can't get around the cap - the signature and id are valid - bans and timeouts - the community write fence (over HTTP, a quiescing community's host already fails to bind with a 404, so this catches quiescing that starts after the tenant is bound) - the sender is a channel member, or the channel is open (same rule as WebSocket) - the signer is the pubkey that authenticated the request (checked first; a mismatch returns 403) The existing HTTP admission rate limit and relay-membership gate still run before any of this. Other ephemeral kinds are still rejected over HTTP. `publish_channel_ephemeral` is the old WebSocket channel-broadcast code moved into a shared helper, so both paths use the same membership check and fan-out. Unlike the WebSocket path, HTTP requires the `h` tag. Over WebSocket, a 20002 with no channel goes to every subscriber. Points left for review: - HTTP typing pulses count against the same per-pubkey HTTP quota as message posts (default 300/min). One pulse every 3 s is 20/min per active channel. - A failed Redis publish is logged and the POST still returns `accepted: true`, same as the WebSocket path. A pulse is resent every 3 s and expires after 8 s, so a retry would only resend a stale pulse. The `sadscan:disable` comment on an existing test URL in `ingest.rs` is there because the secret scanner reads every file a commit touches. It uses the same form as the other local test fixtures in the repo. ### Related issue Needed for Agents2 typing indicators in buzz-app. The client PR depends on this one and will follow as a draft. ### Testing - 9 new Postgres-backed tests in `api/typing_postgres_tests.rs` cover these cases: a member's pulse is published and not stored; a non-member gets a 400; a timed-out member gets a 403; a signer mismatch gets a 403; a missing `h` tag gets a 400; a tampered event gets a 400; kind 20001 is still rejected; oversized events get a 400 (2 KiB of content, one 2 KiB tag value, or `h` + 10k empty tags); a community that starts quiescing after the tenant is bound gets a 400 `community writes are fenced`, and nothing is published. Two mutation checks: with the fence removed, the fence test fails; with the old content-plus-tag-values cap, the empty-tags case is accepted and the test fails. - The full buzz-relay Postgres lane on 27b5a294: 386/387 pass. The one failure is `boot_lifecycle::hanging_redis_peer_times_out_before_the_health_listener_binds`: it went over its timing slack under full-suite load (8.03 s) and passed when rerun alone. - buzz-relay unit tests pass (1476). One mesh demo test timed out once under load and passed when rerun alone. - `cargo clippy -p buzz-relay -p buzz-cli -p buzz-acp --all-targets -D warnings` is clean. - Live test against a local relay built from this branch (27b5a294), run by ART at baxen's request as the sign-off test. alice subscribed over WebSocket to kind 20002 in a private channel; bob (a bot member with no WebSocket) and carol (outsider) used `POST /events` with NIP-98. 16/16 checks passed: - bob's top-level pulse → 200 `accepted: true`, and alice received it live - bob's thread pulse (`h` + root `e` + reply `e`, the buzz-acp shape) → delivered with tags intact - 2 KiB content, a 2 KiB tag value, or `h` + 10k empty tags → 400 `typing indicator too large`; alice received nothing - carol → 400 `restricted: not a channel member`; alice received nothing - carol posting an event signed by bob → 403 - no `h` tag → 400; kind 20001 → 400 `only accepted via WebSocket` - 0 kind-20002 rows in Postgres afterwards 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Bradley Axen <baxen@squareup.com> Signed-off-by: ART <ea39c1c8acf57180161ef4ba4ed68abf8589f5291956b909a5aa991fc66664ca@buzz.block.builderlab.xyz> Signed-off-by: Honey <8e307ae0076a4dab6b94b036ea3edc7e08f823a625269c1e6919e881a048b4d2@buzz.block.builderlab.xyz> Co-authored-by: Honey <8e307ae0076a4dab6b94b036ea3edc7e08f823a625269c1e6919e881a048b4d2@buzz.block.builderlab.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: ART <ea39c1c8acf57180161ef4ba4ed68abf8589f5291956b909a5aa991fc66664ca@buzz.block.builderlab.xyz>	2026-10-09 22:44:31 +00:00
Tom Brow	868e2f8f32	refactor(push): one app profile per gateway (#8206) ... Buzz push currently selects an application profile, such as `dogfood`, during client registration. We want each gateway to serve one Apple app configured by the operator instead. This removes profile selection from the client, relay, and gateway, while allowing internal and public gateways to use their own Apple identities. This PR implements that model in the Buzz source and upstream deployment chart. Discovery, enrollment, grants, and leases no longer carry an application-profile selector. Each gateway has one configured Apple application and APNs transport. The protocol changes are incompatible with the old registration format, so relays advertise `buzz-push-v1` instead of `nip-pl`. Old clients do not discover the new push capability. Updated clients use separate Keychain records and register again. The [push notes](https://github.com/block/buzz/blob/push-capability-v1/docs/buzz-push.md) describe the experimental contract. Landing this PR is the first step. Publish the matching gateway image and upstream chart, then land the [shared chart update](https://github.com/squareup/cash-helm-charts/pull/5740) and [gateway deployment update](https://github.com/squareup/builderbot-platform-core-infrastructure/pull/1085). Before cutover, stop old gateway instances and check legacy registration state. Migrations refuse to proceed if old gateway installations or active relay leases remain; they do not delete that state. Physical-device verification remains necessary before production activation. ### Validation Gateway and focused Flutter push tests pass, as do the Swift package, simulator build, and native notification/enrichment tests. A local relay advertises `buzz-push-v1` without `nip-pl` or `app_profiles`. The PostgreSQL suite passed 868 tests; its two migration-test failures were corrected and pass a focused rerun together with the gateway cutover guard. Chart rendering and release-contract checks pass. Desktop JavaScript tests, desktop and web builds, native desktop compilation, mobile analysis/tests, and desktop native tests pass. The full Rust unit lane and prescribed `just test` command pass, including workspace integration tests against task-isolated local services. The production overlay now requires an explicitly supplied APNs topic, with omission and rendered-value regression coverage. The corrected Swift checksum assertion passes the complete package suite: 78 XCTest tests and 42 Swift Testing tests. No physical-device enrollment or production cutover is claimed. Local merge validation: the embedded migration test and formatting pass after reserving profile migration 0060 behind rollback migration 0059. The local pre-push Rust lane failed ACP timing/steering tests (`idle_resets_on_stdout_activity`, `keepalive_resets_idle_past_deadline`, and the failed/missing-outcome steering tests). ACP and its dependency manifests are unchanged by this branch; all four pass in an isolated single-thread rerun. The documented local-check exception skips only the Rust pre-push lane for publication; required CI remains mandatory. Mobile and desktop pre-push lanes passed. Combined migration validation: after rollback 0059 landed, the main merge preserves profile migration 0060 and both embedded migration assertions. The focused migration test and formatting pass. The next local push again failed only the unchanged ACP idle/keepalive timing tests (1053 passed, 2 failed, 3 ignored); other lanes passed. Publication uses the same documented local Rust-lane exception; required CI remains mandatory. Follow-up review fix: removed the legacy profile column from the copied-lease fixture. All four event-follow-up contract tests pass on isolated PostgreSQL databases, covering desired and migrated schemas in dual and app-only modes. Formatting passes. Publication retains the prior passing local-lane evidence and uses the authorized pre-push exception for the known unrelated ACP timing failures; required CI remains mandatory. --------- Signed-off-by: Tom Brow <tomb@block.xyz>	2026-10-09 15:09:53 -07:00
Bradley Axen	bcc46fe835	perf(relay): cache ephemeral-path serving and restriction reads (#8208) ... ## Summary Typing pulses and agent observer frames each ran two Postgres reads on the WebSocket path in `handlers/event.rs`: whether the community still serves writes (`is_serving_active`), and the sender's ban/timeout row (`moderation_restriction_state`). Both now read through short-TTL moka caches on `AppState`, following the membership cache's pattern. - **Serving state, keyed per community, 10 s TTL.** Every typer in a community shares one entry, so a pod does about one read per community per 10 s. - **Raw restriction row, keyed per (community, pubkey), 30 s TTL.** It caches the row, not the verdict, so the verdict is still taken at "now" and a cached timeout lifts the second it ends. One exception: `banned` is computed when the row is read, so a ban with an expiry can outlast that expiry by up to the TTL. - **The TTL is the staleness bound.** On every pod, and for every way a ban or timeout changes (signed commands, the admin API, report resolution, an owner's moderation reaching their agents), the ephemeral path sees the change within 30 s (10 s for archive/delete). As a best effort, signed ban/unban/timeout/untimeout (9040–9043) also drop the target's own entry on the pod that handles the command. Bans also keep their existing cluster-wide socket disconnect, and archiving a community disconnects all of its sockets. - **Hit/miss counters** `buzz_serving_active_cache_{hits,misses}_total` and `buzz_restriction_cache_{hits,misses}_total`, matching the membership cache, so the saved reads can be confirmed in prod. - **The durable fence stays uncached.** Persistent ingest, moderation commands, auth, and NIP-43 leave (28936, ephemeral-range but ingested) still read Postgres directly. The worst staleness case is a just-timed-out user's typing dots or observer frames getting through on another pod for up to 30 s. Real messages are never affected. ### Related issue N/A. This is the ephemeral-path Postgres read cost raised in the typing-indicator load discussion. Searched for existing PRs and found none. ### Testing - New Postgres tests: - The ephemeral gate honours a cached ban and admits an expired cached timeout. - The ephemeral fence honours a cached "fenced" community. - A cache miss fills both caches. - At the `handle_event` seam, a stored write (kind 1) is accepted despite a cached ban or a cached "fenced" state, and NIP-43 leave skips the cached ban. Mutation-checked: switching either ingest check to its cached variant, or dropping the 28936 carve-out, turns one red. - Timeout and untimeout commands drop the cached row. Mutation-checked: removing the invalidation turns it red. - Live test on a local relay: 100 typing pulses made 0 restriction/serving reads; 10 real messages made 10 of each. A timeout written straight to the DB refused messages at once and typing after ~30 s; an archive written straight to the DB fenced typing after ~10 s. - `scripts/postgres-test-run.sh -p buzz-relay --lib --tests`: 384/384 passed (local PG14). - `cargo nextest run -p buzz-relay --lib --tests`: 1470 passed. The 6 `api::media` failures are environmental: those tests seed through the default `:5432` database, which locally has no `buzz` role. - `cargo clippy -p buzz-relay --all-targets -D warnings` is clean. - Review follow-up commit pushed with the full pre-push hook passing. - Also adds a `sadscan:disable` marker to an existing dev-only test URL in `ingest.rs`. The commit hook flagged it once the file was staged, and the identical URLs elsewhere in that file already carry the marker. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Bradley Axen <baxen@squareup.com> Signed-off-by: Honey <8e307ae0076a4dab6b94b036ea3edc7e08f823a625269c1e6919e881a048b4d2@buzz.block.builderlab.xyz> Signed-off-by: ART <ea39c1c8acf57180161ef4ba4ed68abf8589f5291956b909a5aa991fc66664ca@buzz.block.builderlab.xyz> Co-authored-by: Honey <8e307ae0076a4dab6b94b036ea3edc7e08f823a625269c1e6919e881a048b4d2@buzz.block.builderlab.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: ART <ea39c1c8acf57180161ef4ba4ed68abf8589f5291956b909a5aa991fc66664ca@buzz.block.builderlab.xyz>	2026-10-09 14:53:27 -07:00
Tom Brow	5b907b838a	fix(push): isolate push enqueue and stop workers on rollback (#8216) ... Push enqueueing previously ran inside the message transaction, so a push lock or database error could block or reject a message. Configured relay writers now submit bounded best-effort work only after commit. A dedicated connection and bounded queue isolate enqueue waits and failures from message-write connections. Overload, errors, process loss, or shutdown may drop a notification; the committed message remains authoritative. `BUZZ_PUSH_ENABLED=false` disables the application producer and workers. Updated writer connections always bypass any surviving enqueue trigger. The overlap migration updates an existing function without resurrecting retired objects. Apply it before relying on isolation with a surviving trigger, and replace every old enabled process and writer connection: missing settings preserve legacy behavior during overlap. SIGTERM/Ctrl-C cancels enqueue, matcher, and delivery work before listener drain. Durable queues, claims, leases, preferences, and delivery configuration remain intact. Claims expire normally; gateway requests already accepted may still deliver. Reactivation requires an explicit operator decision about retained work. There is no automatic purge, startup interlock, or backfill of messages written while disabled. Built on merged [#8135](https://github.com/block/buzz/pull/8135), including [#8178](https://github.com/block/buzz/pull/8178). Trigger retirement [#8210](https://github.com/block/buzz/pull/8210) remains open; this change is tested with the trigger present and after removal. Operational details are in [the rollback and reactivation guide](https://github.com/block/buzz/blob/relay-push-rollback/docs/push-gateway-deployment.md#relay-rollback-and-reactivation). ### Validation [9d3d710](9d3d710dc1) changes documentation and Rust comments only, with no executable or assertion changes. Formatting and diff checks passed; [12f1d1e](12f1d1e814) additionally restores the reviewer-requested caveat that active lease renewal may suppress a notification. This is also documentation-only; Owner approval was confirmed on 12f1d1e. [a4f602d](a4f602df49) subsequently merges main to resolve a migration-count conflict after main added migration 0058. Both 0058 and rollback migration 0059 are retained; the focused embedded-migrator test, formatting, and applicable desktop pre-push checks passed. Current-head [owner approval](https://github.com/block/buzz/pull/8216#pullrequestreview-5474958334) is confirmed. [CI attempt 2 passed](https://github.com/block/buzz/actions/runs/37985516452/attempts/2), with required checks passing or appropriately skipped. The first attempt failed only the unchanged replica-floor probe test with `MaskedActivity { masked: 1 }`; one failed-jobs retry passed without code changes. All four push contract tests passed. The current-head [advisory security review](https://github.com/block/buzz/actions/runs/37985514346/job/114006457769) has remained in its review step since 20:18:51 UTC without a published result for over 25 minutes. It is stalled under the advisory-review policy, not a clean security review. [Required CI checks passed or were skipped](https://github.com/block/buzz/actions/runs/37964035783) on [3164b25](3164b25cde). The advisory security review was cancelled without a published result. Codex's durable-retry suggestion was [rebutted and resolved](https://github.com/block/buzz/pull/8216#discussion_r4232674980) under the explicit best-effort notification contract. The [lease activation-order observation](https://github.com/block/buzz/pull/8216#discussion_r4233051799) is correct. Under the clarified product requirement, either notification outcome is acceptable during overlapping registration and message arrival. Exact commit ordering is not required; deliberate historical backfill remains prohibited. The bot thread is resolved against that clarification; [Owner review accepted the clarified overlap contract](https://github.com/block/buzz/pull/8216#pullrequestreview-5474720278) on 9d3d710. All 488 PostgreSQL tests passed. The final four-way producer contracts passed on desired/migrated schemas, with/without triggers. Added coverage verifies message commit under push-gate and queue-table locks, recovery after enqueue failure, in-flight enqueue cancellation, the lease timestamp filter (not production activation ordering), serving-lease writers, rollback preservation, reconnects, and reactivation without backfill. Queue saturation and all three relay worker shutdown tests passed. Rust lint and all 17 source-policy checks passed. Full local CI/repository runs encountered the unchanged ACP `keepalive_resets_idle_past_deadline` timing failure. Earlier desktop subprocess timing tests also failed under parallel load and passed as an isolated module. Under Tom's local pre-push exception, this update excluded `rust-tests` and `desktop-tauri-checks`; remaining hooks stayed enabled. Required CI validation remains necessary before merge. A focused run competing with the full database suite hit wall-clock deadlines; tests now use a short database lock timeout to distinguish unwanted push dependencies, and their final isolated run passed. A local process-replacement rehearsal on 3164b25 used the real relay binary and a mock gateway holding a delivery request open. SIGTERM closed that request in under 1 ms; the relay exited in 5.02 seconds. Restarting the same binary with push disabled produced no further gateway requests over 35 seconds, beyond the 30-second claim expiry. The outbox row, lease, and its subscriptions remained byte-for-byte unchanged. This used a seeded outbox and did not exercise APNs or a mobile device. Production writer inventory, shutdown deadline, and retained-queue disposition still require operational validation before rollout. The PR is marked ready for review. The overlap behavior is accepted by the product requirement; owner review accepted that contract. No production deployment has occurred. --------- Signed-off-by: Tom Brow <tomb@block.xyz>	2026-10-09 13:48:06 -07:00
Bradley Axen	1e80066c93	feat(relay): reshape read-state API to the sidebar wire format (#8220) ... ## Summary Changes the private read-state API (`/buzz/v1`) to the sidebar wire format agreed in #buzzin-in-secret. **The counting code is unchanged.** Requests and responses take the new shape, so the client integration can build against it now and later behavior fixes (counting rules, followed threads, another-device pings) ship without another wire change. Nothing ships against v1 yet (no buzz-app use on `main`), so this replaces version 1 directly. Supersedes #8188 (same branch content, rebased onto current `main`, plus the thread-cap and doc changes below). ``` GET /buzz/v1/me/sidebar?limit=&cursor= | ?channel_ids=a,b,… → { channels: [channel], next_cursor } channel: { channel_id, unread: bool, mentions, read_through_id|null, latest_id|null, threads: [thread] } thread: { root_id, unread: bool, mentions, read_through_id|null, latest_id } POST /buzz/v1/me/read-state { intents: [{ type: "mark_through", target: { channel_id [, root_id] }, message_id }] } → { outcomes: [applied|blocked|invalid|unknown], channels: [channel] } ``` What changed and why: - **Channel and thread positions are fully separate.** Channel `unread`/`mentions` count only top-level messages. Thread replies count only on their thread row. `mark_channel_read` and its whole-channel cut are gone. "Mark channel read" sends a channel `mark_through` at `latest_id`, plus one per thread row to clear threads too. - **Channel `latest_id` is always a top-level message.** It used to be able to point at a reply, and a channel `mark_through` rejects a reply as its anchor. It also skips a reply-marked message whose ancestry never resolved, because marking through one would fail on every retry. - **`read_through_id`:** frontiers stored only the anchor's arrival time. Migration 0058 adds `personal_read_frontiers.through_message_id`, which follows the greatest arrival, and drops the now-unused `threads_through_timestamp`. This is the only storage change. - **Counts are plain numbers.** The exact/at_least/unknown wrappers are gone. Evidence the relay can't use (unusable tags, unresolved ancestry, undecided conversation membership, past the scan cap) leaves the message out, so counts can undercount. - **POST returns the updated row** of each distinct channel it applied to, so the client doesn't refetch after every read. If those rows can't be read after the writes commit, `channels` is omitted and the outcomes still stand. The row refresh shares the 8 s intent deadline. - **Up to 25 thread rows per channel** (was 5). Five hid a mention in a sixth unread thread. 25 is the most that keeps a POST's rows for `MAX_INTENTS` (100) channels inside the 1 MiB response limit; 50 would be ~1.4 MB, a 503 that hides already-committed outcomes. A unit test builds that worst-case response and fails if the cap outgrows the limit. - **The doc separates shape from behavior.** `/buzz/v1` promises a stable shape (fields, types, status variants, what `mark_through` does). What the counts include (which messages are unread or directed, which threads are listed, the bounds) may change while v1 is pre-release, so clients replace their rows with each response instead of reconciling against local counts. - **Removed from the wire:** the explicit-contexts read (`GET /me/read-state?targets=`), `account`, channel name/type/archived/hidden (these come from Nostr metadata), `latest_message_at`/`_complete`, and the NIP-11 `max_contexts`/`max_context_messages`/`max_thread_summaries`. Deliberately deferred (behavior or additive, no shape change): - Thread counting rule (relay counts a reply if you wrote or replied to its direct parent; the client uses "posted anywhere in the thread or followed"), the workflow-owner attribution p-tag, and thread-row `mentions` counting every counted reply rather than directed ones. - `follow`/`unfollow` intents, listing followed threads that are already read, and a truncation flag for the thread list. Thread: buzz://channel/6f773b94-34d0-4004-bec9-a100b6ad17d5 (#buzz-bff, root 31020fff) ### Related issue Follows #8157. Supersedes #8188; replaces the wire parts of the closed #8155. ### Testing - `cargo test --no-fail-fast -p buzz-db -p buzz-relay` at 1f041e46: all pass except 7 relay unit tests (`api::media::tests` ×6 needing a seeded local DB, `api::mesh_demo` round trip) that fail identically on `main` (394d4a43) in the same environment. - `scripts/postgres-test-run.sh -p buzz-db -p buzz-relay` against local PG14 + Redis at 88d2f6fc: 836 passed, 26 failed. 25 also fail on `main` in the same environment (partition maintenance tests and `huddle_started_links`, which need PG16). The 26th, `boot_lifecycle::hanging_redis_peer_times_out…`, is a timing assertion that failed under full-suite load and passed 3/3 when rerun alone on this branch. - All `personal_read`, `migration`, `buzz_v1`/`accessory` and NIP-FI wire-contract tests pass, including the new response-size bound test and the 25-thread `summarize` cap test. - `cargo fmt --check` and `cargo clippy -p buzz-db -p buzz-relay --tests -D warnings` are clean. - Not done: running a standalone relay binary (the dev object store rejected writes with HTTP 403). The relay tests drive the real HTTP router with NIP-98-signed requests. A human test against a local relay is still needed. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Honey <8e307ae0076a4dab6b94b036ea3edc7e08f823a625269c1e6919e881a048b4d2@buzz.block.builderlab.xyz> Co-authored-by: Honey <8e307ae0076a4dab6b94b036ea3edc7e08f823a625269c1e6919e881a048b4d2@buzz.block.builderlab.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-09 13:07:19 -07:00
Ian Oberst	a8fe2b612c	feat(db): advance an empty partition catch-all under bounded locks (#7993) ... ## Summary Adds bounded, serialized advancement of an empty right-edge partition catch-all for `events` and `delivery_log`. The repaired catalogs currently end in an empty catch-all from 2027-01-01, and the merged manager (#6515) only audits them. Without this change, January writes land in the catch-all, after which only an operator can move them. When `BUZZ_PARTITION_MANAGER_ADVANCE_ENABLED=true`, each relay keeps six dedicated monthlies after the current month. It does this at startup and on every audit interval by replacing the empty catch-all with canonical monthlies and a new canonical `<table>_p_future` after the horizon. From today's catalogs the first run creates January–March 2027 and moves the catch-all to April 2027. ## Safety design - **Refuse before locking.** The following return `operator_required` without taking any table lock: - a populated catch-all, or one whose occupancy is unknown; - a catch-all with child-only triggers, which the replacements would not inherit; - a misaligned bound; - a `DEFAULT`, anomalous or pending-detach child; - a name collision. - **One winner.** A schema-scoped `pg_try_advisory_xact_lock` is taken before any DDL. Losers report `skipped_locked` and never issue DDL. The winner re-plans from a fresh audit before taking any table lock and checks for collisions. A runner that lost a race therefore sees the committed layout as a no-op, not as colliding names. - **Writer lock order.** First `ONLY` the parent is locked `ACCESS EXCLUSIVE` under a 2 s `lock_timeout`. Then each foreign-key counterpart (today only `communities`) is locked in OID order `SHARE ROW EXCLUSIVE NOWAIT`, then the catch-all `NOWAIT`. - The counterparts are read again once the parent is locked, and that set is the one locked. Any FK DDL on or to the parent conflicts with the parent lock, so the set can't change after that. An FK committed while maintenance waited is therefore locked `NOWAIT` too, so attaching never waits on it under the parent lock. The read before the lock remains as a cheap early refusal. - This is the order an insert takes locks (parent, then the FK check on `communities`), so maintenance cannot deadlock with ingest. The internal manual runbook used to lock `communities` first; it now uses this order too. - `SHARE ROW EXCLUSIVE` is the strongest lock the DDL takes on `communities`; measured on PG 16 and 17, the catch-all `DROP` takes none, as long as the catch-all's only foreign keys are those inherited from the parent. A catch-all with its own foreign key, or one that another table's foreign key references, is refused: once before any lock, and again after the catch-all lock, where the check is authoritative. Readers and other writers' `FOR KEY SHARE` checks proceed. A concurrent `communities` writer makes the run fail fast with `lock_timeout`. - A `40P01` deadlock victim is also reported as `lock_timeout`. - Parents that are referenced by a foreign key, or that reference an inherited table, are refused, including when such a key commits while maintenance waits on the parent. - **Re-plan under lock.** The catalog is re-audited under those locks. The transaction proceeds only if the plan is unchanged. - **Bounded.** `statement_timeout` 5 s, `idle_in_transaction_session_timeout` 5 s, and a 10 s client deadline per table. A deadline-abandoned connection is closed, never returned to the pool, and the server rolls back anything uncommitted. If the deadline fires after `COMMIT` was sent, the change may have landed; the next audit reports the truth. The orphaned backend keeps its locks until its in-flight statement ends, at most `statement_timeout`. - **Verify before commit.** The changed catalog is re-audited inside the transaction for bounds, canonical kinds, trigger parity and catch-all emptiness, and serving safety must not regress. It also checks that every new child has valid, ready clones of each parent index and validated clones of each PK/unique/FK constraint. - **Audits tolerate a concurrent advance.** Catalog rows come from the statement snapshot, but `pg_get_expr` renders from current catalog state. So an audit that overlaps another relay's `DROP` of a catch-all can see a partition that no longer exists. The read-only audit detects this (a NULL bound, or `42P01` from the emptiness probe) and retries up to twice with a fresh snapshot. - **Fail fast on audit probes.** If another relay's audit is probing the catch-all's emptiness, the catch-all `NOWAIT` lock fails and the run reports `lock_timeout`. The parent lock is then released immediately, and the next runner or interval completes the advance. Uncovered-month creation (`BUZZ_PARTITION_MANAGER_CREATE_ENABLED`, default `true`) now runs through the same locked transaction, replacing the unlocked per-month `CREATE`. That flag remains the kill switch for all automatic partition DDL: with it off, advancement is disabled too. With advancement disabled (the default) the periodic loop stays read-only and startup behaves as before, apart from the locking. A runner that planned DDL always re-audits before returning, even when a peer did the work, so it never reports the pre-maintenance layout. A serialized re-plan that finds only disabled work reports `skipped_disabled`. ## Behavior changes - The relay audit horizon and the `buzz-admin partition-audit` default are now 6 months (`PARTITION_MANAGER_MONTHS_AHEAD`), up from 3. Until advancement is enabled, months already covered by the catch-all make `degraded` true. That correctly reports a runway under six months. - New metric `buzz_partition_maintenance_runs_total{table,outcome}` with outcomes `noop | skipped_disabled | created | advanced | skipped_locked | operator_required | lock_timeout | deadline | error`. `buzz_partition_create_attempts_total` keeps its per-month `created | skipped_covered | error` labels. `error` there still means a DDL or collision failure; `lock_timeout` and `deadline` runs add nothing to it, and only `buzz_partition_maintenance_runs_total` records them. - `ensure_future_partitions` (free function and `Db` method) is removed; `maintain_partitions` is the one entry point. - New advisory-lock label `lock_type="partition_maintenance"`. ## Rollout Enable `BUZZ_PARTITION_MANAGER_ADVANCE_ENABLED` per environment in BPCI: `bb-block-staging` first, then production. The staging canary proves a real advance because the six-month horizon already extends past January. Fallback: rerun the internal manual runbook, which uses the same parent-first lock order as of 2026-10-05. It keeps `ACCESS EXCLUSIVE` on the counterparts, because its detach/attach can drop FK triggers on them. ## Testing - `cargo test -p buzz-db --lib partition`: 17 unit tests. The 10 new ones cover: - plans: repaired layout, policy gating, no-op, refusal states including child-only triggers, December→January; - the create kill switch also stopping advancement; - the month cap at exactly 120; - per-month metric outcomes for transient failures versus DDL failures; - the mid-audit-drop retry classifier. - `cargo test -p buzz-db --lib partition -- --ignored` (PostgreSQL 17): 48/48 on 9 consecutive runs, including 17 new tests: - advance to the canonical layout, with routing and idempotent rerun; - replacing a canonically named catch-all in one transaction; - populated catch-all refused in under 1 s while the parent is locked; - a `communities` writer making the run fail fast and roll back; - a `communities` reader not blocking the advance; - an in-flight insert's FK check, and its write to `communities`, behind queued maintenance completing without a deadlock; - a foreign key committed while maintenance waits on the parent, with a writer on its new counterpart, failing fast as `lock_timeout` on that table instead of waiting under the parent lock; - a foreign key referencing `events`, committed while maintenance waits on the parent, being refused under the parent lock; - a catch-all's own foreign key, or a foreign key referencing it, being refused before the parent lock with the key intact, and a catch-all key committed while maintenance waits on the parent being refused after the catch-all lock; - a `40P01` deadlock victim being classified as `lock_timeout`; - parent contention timing out without blocking the other table; - held advisory lock skipping with no DDL; - concurrent runners converging, with at most a transient `lock_timeout`; - a stale audit after a concurrent advance being a no-op, and the losing runner returning the committed layout; - a re-plan that finds only disabled work reporting `skipped_disabled`; - an audit retrying when a listed partition is dropped mid-audit; - a catch-all with a child-only trigger being refused; - client deadline abandoning the attempt and closing the session (scoped to the test's own blocked backend); - catch-all name collision; - incoming foreign key refused. - Mutation checks: - restoring the old counterpart-first `ACCESS EXCLUSIVE` order fails the reader test (`lock_timeout`) and the writer test (a ~1 s deadlock wait); - dropping `NOWAIT` from the counterpart lock, or allowing a populated catch-all, fails the matching tests; - removing the parent-locked counterpart re-read fails the committed-FK tests (the parent is held 2.06 s waiting on the new counterpart), and ignoring its refusal fails the referencing-FK test; - locking the counterparts `SHARE ROW EXCLUSIVE` before the parent fails the writer test; - removing either catch-all foreign-key check fails its test. - Full PostgreSQL lane: 716/718. Both failures are timing-sensitive tests outside partition code, and each passes 3/3 when run alone: - `hanging_redis_peer_times_out_before_the_health_listener_binds`; - `readiness_check_cancellation_balances_waiter_and_inflight_connection`. - `cargo clippy -p buzz-db -p buzz-relay -p buzz-admin --all-targets -D warnings` and `cargo fmt --check` are clean. ## Related Refs #2396. This PR conflicts textually with #7994 (partition audit observability follow-ups) in `crates/buzz-db/src/store/partition.rs` and `crates/buzz-relay/src/main.rs`. Whichever PR merges second will be rebased onto the other. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Signed-off-by: Ian Oberst <ioberst@block.xyz> Co-authored-by: Ian Oberst <ioberst@block.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-08 12:59:10 -07:00
Bradley Axen	7afdc2aac9	feat(relay): record when a read-state account starts (#8157) ... ## Goal Add `started_at` to read-state accounts from the day the table is created. This prevents users seeing ALL their channels as unread when they first install a client update that starts using these APIs --------- Signed-off-by: hMB2 <0400eb0da32afbbf16cdd1dee6ffd243587e8f3d6a2823fe7dfa95d16b002b7d@buzz.block.builderlab.xyz> Signed-off-by: ART <ea39c1c8acf57180161ef4ba4ed68abf8589f5291956b909a5aa991fc66664ca@buzz.block.builderlab.xyz> Signed-off-by: hMurderbot <219c58afff8ac2c76e4e527a329aa61c2ebbb896786b1916d4f9cca5140ecd04@buzz.block.builderlab.xyz> Co-authored-by: hMB2 <0400eb0da32afbbf16cdd1dee6ffd243587e8f3d6a2823fe7dfa95d16b002b7d@buzz.block.builderlab.xyz> Co-authored-by: hMurderbot <219c58afff8ac2c76e4e527a329aa61c2ebbb896786b1916d4f9cca5140ecd04@buzz.block.builderlab.xyz> Co-authored-by: ART <ea39c1c8acf57180161ef4ba4ed68abf8589f5291956b909a5aa991fc66664ca@buzz.block.builderlab.xyz> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-10-08 10:04:02 -07:00
Luke Tornquist	f8bf2c529c	refactor(db): type-level admitted event-write transactions (AdmittedTx) (#8130)	2026-10-08 08:47:07 -04:00
Luke Tornquist	1972b7d256	feat(admin): add community archive commands (#7107)	2026-10-07 18:31:00 -04:00
Will Pfleger	032198ce43	feat(relay): add community reads to the admin API (#8011) ... The in-app Admin Console needs to work on any community the relay serves, not only the one the desktop is connected to. This adds four read-only routes under `/api/admin/v1` that the stacked desktop PR builds on, puts every admin route under one explicit access rule, and binds every admin request's signature to its real method, which fixes signed `HEAD` on reads. ## Access rule | Mode | Moderation reads | Writes | Staffing (`/operators`) | |---|---|---|---| | `nip98` (default) | signed, any staff role | signed, role-checked | signed Operator | | `disabled` | served to anyone who can reach the relay | `403` | `403` | "Moderation reads" means every read: reports, feedback and attachments, restrictions, and the four community routes below. `nip98` behavior is unchanged, and the Operator API (`/operator/*`) is untouched. **Existing `BUZZ_ADMIN_AUTH=disabled` deployments start serving the four new reads** (community directory, member search, member lookup without staff status, and stored events by ID) with no config change. The startup `WARN`, `docs/admin/README.md`, `.env.example`, `deploy/compose/.env.example` and the `just admin` comment list exactly what disabled mode exposes, say it should be reachable only by people allowed to see all moderation data in every community, and say the whole relay port must be private: the admin API is on the relay's main listener and its hostname check uses the caller's `Host` header. `just admin` binds the relay to `127.0.0.1` unless `BUZZ_ADMIN_AUTH` is exactly `nip98`, so an unset, padded or misspelled value stays local. In code, `authorize()` returns `AdminAccess`, either `NetworkTrusted` (disabled mode, no identity) or `Staff(AdminPrincipal)` (`nip98`, any rostered Operator or Moderator). Each route makes one check: **view** (`authorize()` succeeding), **act** (`AdminAccess::act()`, must be `Staff`, then the existing role checks) or **operator** (`AdminAccess::operator()`, must be an Operator). Act and operator refuse `NetworkTrusted` with `403 this endpoint requires BUZZ_ADMIN_AUTH=nip98`; no principal is ever created for an anonymous caller. `every_route_makes_its_check_in_both_auth_modes` lists every mounted admin route with its check and asserts the result in both modes, and `every_route_is_listed` fails when `router()` mounts a route that list does not name. ## Routes All four live in `api/admin/reads.rs`; `mod.rs` only registers them. Their queries reject unknown fields, like `ReportQuery`. The three `?communityHost=` routes share `CommunityQuery`, which now rejects unknown fields too, so the existing unban, untimeout and direct-action routes that use it do the same; their desktop callers send only `communityHost`. Every admin route that takes `limit` validates it with the module's shared `limit()` helper, so an out-of-range value is `400 invalid_limit`. | Route | Returns | |---|---| | `GET /communities?q=&cursor=&limit=` | `{items: [{id, host, icon\|null}], nextCursor\|null}` | | `GET /members/search?communityHost=&q=&limit=` | `{items: [{pubkey, displayName\|null, nip05\|null, avatarUrl\|null}]}` | | `GET /members/{pubkey}?communityHost=` | `{pubkey, profile: {displayName, nip05, avatarUrl, about}\|null, role: "owner"\|"admin"\|"member"\|null, banned, mutedUntil\|null, isStaff: bool\|null}` | | `GET /events/{id}?communityHost=` | `{id, authorPubkey, kind, content, createdAt, deletedAt\|null, channelId\|null}` | - **One community per read.** Community-scoped routes bind `communityHost` through the tenant binder (`400 unknown_community_host`) and read only that community's `users`, `relay_members`, `community_bans` and `events` rows. - **Directory.** It uses the binder's liveness rule (not archived, not deleted, `deletion_state = 'active'`), so every listed host can be opened. `q` is an escaped, case-insensitive host prefix of at most 255 characters (`400 invalid_query`). Results are ordered and paged by `lower(host)`, which is unique, so the opaque base64url cursor carries only the last host; a malformed cursor is `400 invalid_cursor`. `limit` is 1 to 100 with a default of 50. - **Member search.** It reuses `search_users`, which already escapes, binds and scopes, with no second escaping layer. `q` is trimmed and must be 1 to 100 characters (`400 invalid_query`), and `limit` is 1 to 50 (default 20). It searches community profiles, so former members appear. - **Member lookup.** `isStaff` uses the same roster lookup as the direct-action staff guard and is returned to signed staff only: it is `null` in disabled mode, where the roster is never read. A roster lookup failure is a 500, never "not staff". - **Event preview.** It reads the event by `(community_id, id)` and never returns a private kind (see below). A malformed id, an id stored only in another community and a private kind are all `404 event_not_found`, the same answer as an absent event. ## Private events Admin reads never reveal events the relay keeps private. `admin_hidden_kinds()` combines the four private kind sets in `buzz_core::kind`: `AUTHOR_ONLY_KINDS`, `RESULT_GATED_KINDS`, `P_GATED_KINDS` and `SHARED_GATED_KINDS`. The whole shared-gated set is hidden, including events tagged `["shared","true"]`; the shared tag is never evaluated. Every HTTP response applies it the way it treats a missing event: event preview is `404 event_not_found`, report detail returns the report with `message: null` and `targetAuthorPubkey: null`, and the report list returns a null target author. Moderators can still act on these events, they just cannot read them. Report detail is one query with the hidden-kind list as a parameter, exposed two ways: `admin_get_report` hides private kinds and backs every HTTP response (GET detail, the resolve response, reopen, cancel and the reporter-notice worker), while `admin_get_report_for_enforcement` reads the target unfiltered and is used only by the resolve handler's enforcement load (the staff-target check and ban, timeout and kick target derivation) and by the recovery worker for interrupted actions. Its result is never serialized, so a ban from a report on a private persona bans the real author while GET still shows nothing. ## Method binding Axum serves `HEAD` through `GET` handlers, but the read handlers authorized with a literal `"GET"`, so a correctly signed `HEAD` was rejected. Admin reads now authorize through `authorize_read` and writes through `authorize_write`, both of which use the request's real method and full target, so no handler hardcodes a method string. A database error behind any admin route is now logged before it becomes a 500. ## Directory query plan `idx_communities_host` is on `lower(host)` in the default collation, so the prefix `LIKE` is a filter, not an index condition. The plan walks the index in order and stops once it has `limit` matches. A common prefix is cheap (on local Postgres 16 with 21,359 communities, a first page skipped 2,728 index rows in 3.9 ms), but a prefix with few or no matches walks every index entry, so its cost grows with the number of communities. A `COLLATE "C"` index would turn the prefix into an index range; it is left out because the admin directory is low-traffic and the cost is small at current scale. The cursor condition is `lower(host) > COALESCE($2, '')`, with no `OR`, so it remains an index condition under a generic plan: ```text -> Bitmap Index Scan on idx_communities_host Index Cond: (lower((host)::text) > COALESCE($2, ''::text)) ``` --------- Signed-off-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Will Pfleger <wpfleger@block.xyz> Co-authored-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz> Co-authored-by: Hayt <211b96e6a2b7f45fd4047988976c7bbbeeda0c15f3ae7b32eec20834b5a55118@buzz.block.builderlab.xyz>	2026-10-07 16:07:38 -04:00
Luke Tornquist	6d07a29137	perf(relay): bound fleet usage metrics collection (#7176) ... ## Why The usage poller pays memory and query costs proportional to community count even when per-community emission is disabled. Every pod loads the full host map, and the leader materializes all per-community aggregates before suppressing their labels. That makes telemetry collection itself a scaling constraint. ## What changed - Fleet-only usage metrics are now the default. `all` remains available as an explicit rollback mode. - In fleet-only mode, the poller skips the community host map and all community-sized query results. - Exact community counting is replaced with the PostgreSQL planner's table-size estimate. - Stock totals are collected as one fixed row, scanning each stock table once per hour. - 1d/7d/30d active-user totals come from one bounded 30-day scan per day. - Fleet database snapshots run only on a proved read replica, with a telemetry-specific freshness budget and server-side statement timeouts. If no such replica is available, the collection is skipped; it never falls back to the writer. - Successful fixed-row snapshots are cached and re-emitted on each poll tick, so gauges stay scrape-visible without raising query frequency. - `buzz_usage_snapshot_available{family}` reports whether the latest due refresh of that family succeeded. After a skipped or failed refresh, the last values stay visible (with `buzz_usage_snapshot_age_seconds`) but the family reads unavailable until a 60-second retry succeeds. - The `all` rollback collector publishes the same availability gauges, so dashboards gated on availability keep working during a rollback. - Each due refresh outcome is applied by `apply_stock_refresh` / `apply_activity_refresh`, the single production mapping from `Ok(Some)` / `Ok(None)` / `Err` to cache, availability, skip counter, and retry; the regression test drives that mapping directly. - `buzz_usage_snapshot_age_seconds` is emitted in the `all` rollback too, so age-based alerts keep firing through a rollback. - Every fleet telemetry attempt that completes lands in `buzz_db_route_decision`; a query error on a proved reader records `skipped/replica_error`. An attempt abandoned at the relay deadline records no route decision and appears only as `buzz_usage_query_skipped_total{reason="timeout"}`. The shared `proved_reader` warnings no longer claim "routing to writer", because the telemetry caller never does (serving callers still record their writer decision in the route counter). - The PostgreSQL production-method matrix now exercises `usage_fleet_stock_snapshot`, so the new `reader/maintenance` pool pair is proven to be emitted by a real Db method. - Each fleet collection, reader proof included, runs under a relay-side `tokio::time::timeout` (10s stock, 20s activity, both above their server-side `statement_timeout`). A reader that stops answering expires into the same failure mapping with `buzz_usage_query_skipped_total{reason="timeout"}`, so it cannot stall the leader's poller, hold the advisory lock past a liveness check, or leave a frozen snapshot reporting available. - Fleet telemetry reader checkouts (`ReaderOperation::Maintenance`) opt into SQLx `close_on_drop` before their first post-acquire await, so an abandoned collection's connection is closed within SQLx's 5s `CLOSE_ON_DROP_TIMEOUT` instead of going through the unbounded return-to-pool ping while it holds a slot in the reader pool that serving reads share. Serving reads keep pooled reuse; the cost is one reader reconnect per fleet refresh. - On leader demotion, cached snapshots and availability are cleared, and a fresh collection is forced after the lock is reacquired. - Fleet-only mode passes no host map to the worker-snapshot storage emitter. It therefore omits community attribution instead of reporting every byte as `buzz_storage_unmapped_community_bytes`. - Lifetime stored-message and active-channel totals are no longer queried in fleet-only mode. Event throughput remains available through the existing stored-event counter. - No test-selector changes: main's `config::tests::` selector in `just test-unit` and the `run-tests.sh unit` fallback already runs the new replica-budget config test, and the new fleet schedule, availability, and storage tests run through the existing bin `tests::` and `storage_sweep::tests::` selectors. - Added seeded PostgreSQL contract tests and operator docs covering rollout, rollback, availability, the no-replica behavior, and leader election. No migrations, tables, indexes, or new persistent storage are added. ### What got simpler The relay-side storage half of the original branch (totals-only S3 listing and `buzz_storage_community_breakdown_available`) was dropped. Main moved S3 listing out of the relay in #7543 and #7845, so that code was dead on arrival. The rework adds no storage code path; it only gates attribution on the presence of a host map. ## Risk Medium. The default changes which telemetry is emitted: - In fleet-only mode, `buzz_communities_estimated` replaces the exact `buzz_communities_total`. - `buzz_total_messages` and `buzz_total_active_channels` are not refreshed in fleet-only mode (also called out in the rollout steps of `docs/usage-metrics.md`). - Deployments without a verified reader (including local compose and single-node self-hosts) skip database-derived fleet snapshots instead of querying the writer, and report both families unavailable. - Multi-pod dashboards must filter database and storage series to `buzz_usage_poller_is_leader == 1`. Setting `BUZZ_USAGE_METRICS_PER_COMMUNITY=all` restores the legacy collector for temporary rollback. ## Verification At head `1879ee3aedf4227942120a3b521c8d35fc7839fe` (tree `78477e74`; merges main `da651ad6`; the PR's diff against main is byte-identical to `1613823f`'s, minus the Justfile and `run-tests.sh` selector lines that main's `config::tests::` selector now subsumes): - On Blox: `cargo fmt --all -- --check` and `cargo clippy -p buzz-db -p buzz-relay --all-targets -- -D warnings` pass. Relay bin `tests::` 34/34, `config::tests::` 55/55, `storage_sweep::tests::` 10/10 and `store::usage::tests` pass, as do the four `fleet_` and lock PostgreSQL tests that use scratch databases, including `abandoned_fleet_collection_releases_its_reader_slot`. - Against a freshly migrated PostgreSQL database: `store::usage::postgres_tests::` 14/14, `storage_sweep::postgres_tests::` and `buzz-db --test observability_source` 14/14 pass. The new reader code passes main's crate-wide typed-acquisition and serving-table policy checks. - E2E at `1613823f` is reused for this head. It ran in an isolated relay against a real PostgreSQL primary and streaming standby. In 2/2 runs a stock read went dark mid-result; the deadline abandoned it at about 10s, the reader pool's active and size counts dropped to zero about 1s later, and a fresh activity read and the scheduled stock retry succeeded on new backends while the dark session stayed open. The `all`-mode and healthy-replica controls also passed. Main's merged changes are on the writer path and in startup step timing, and touch none of the reader checkout, the `close_on_drop` seam, the fleet poller or the deadlines. At head `1613823fa8ef1a423bc1a499c57bc40d692fcd4f` (tree `12bc6d1b`; test/docs-only on top of `57749418`), on a Blox workstation: - `cargo fmt --all -- --check` and `cargo clippy -p buzz-db -p buzz-relay --all-targets -- -D warnings` pass. Relay bin `tests::` 34/34 pass. - New PostgreSQL test `abandoned_fleet_collection_releases_its_reader_slot`: a one-slot reader pool behind a loopback proxy that goes dark mid-session; the collection is cut at a deadline, then a later acquire must succeed in under 8s on a fresh connection. The dark call runs on a fresh `Db` with a cold Aurora capability cache, so the first post-acquire await is the capability probe. It passes in 3.4s. With `close_on_drop` removed, or moved below the probe await, it fails with `PoolTimedOut` after the 15s acquire timeout. The other three `fleet_` PostgreSQL tests that use scratch databases pass. - `each_fleet_family_is_abandoned_at_its_own_deadline` now has an outer timeout: replacing the stock deadline with a bare `.await` fails it (`Elapsed`) instead of hanging. At head `fd194ec79d6c6796d6764564de225ad56485cfc7` (tree `79e94727`, main merged at `932228936`), on a Blox workstation: - `cargo fmt --all -- --check` and `cargo clippy -p buzz-relay -p buzz-db --all-targets -- -D warnings` pass. - Relay bin `tests::` nextest selection: 27/27 pass (plus `store::usage::tests`), including `fleet_refresh_outcomes_drive_availability_through_production_mapping` and the new `stalled_fleet_collection_times_out_into_failed_refresh` (paused time: two never-completing collections are abandoned at exactly 10s + 20s; both families read unavailable with cached values kept, `reason="timeout"` skips, retry scheduled; the next tick runs and recovers; no route decision is recorded for an abandoned attempt) and `each_fleet_family_is_abandoned_at_its_own_deadline`. Compile-time asserts keep each deadline above its `statement_timeout`. - Falsifiability: replacing the stock deadline with a bare `.await` makes the new test fail (the tick stalls); swapping the two deadlines fails the per-family test; deleting `schedule.mark_stock_failure(now)` from the `Ok(None)` arm fails the mapping test. Trees restored and re-verified. - At `970414db`: CI PostgreSQL Tests 809/809 including the production-method matrix with `reader/maintenance`; Unit Tests ran `finish_usage_read_records_every_attempt_in_route_decisions`. - `just ci` / `just test` at `5b18914a` on Blox: not green for environmental reasons (`mobile-check` could not reach pub.dev; `_ensure-services` failed so service-backed suites did not run; `buzz-acp git_bootstrap` fails identically on `origin/main`). PostgreSQL and integration coverage comes from GitHub CI at this head. E2E at the previous head `0a01e5f22` (isolated relay, real Postgres primary plus streaming replica, worker storage snapshot, desktop frontend in relay mode): with no reader, both families report skipped/unavailable and route `skipped/disabled` (no writer queries); fleet-only storage emits totals without `buzz_community_storage_*` or unmapped bytes; `all` mode reports both availability gauges `1` with per-community storage series; with a fresh replica, both fleet paths route `replica/fresh` on `reader/maintenance` acquisitions; desktop send/receive works. The cached success → due failure → recovery transition was not reached live (hourly/daily cadence); it is covered by the seam test above. The original branch also recorded live verification at `aa3e7023`, before this rework: reduced-idle cache re-emission, two-relay leader takeover with demoted availability clearing, the no-reader skip with 60-second retry, `all` rollback restoring per-community series, and a desktop smoke test. Those runs predate the availability and storage changes and are not claimed for this head. ## Linear tracking Related work: BUZZ-167. --------- Signed-off-by: tornquist <tornquist@squareup.com> Signed-off-by: Luke Tornquist <tornquist@squareup.com> Signed-off-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz> Co-authored-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz>	2026-10-07 15:46:18 -04:00
Luke Tornquist	e097c42996	feat(relay): time post-metrics-bind startup steps (#8132) ... ## Summary Relay pods on large deployments can take over 5 minutes to open their health port, and right now nothing shows which startup step is slow. Startup timing ends at `metrics_bind`, and the slow work all happens after it. This PR times every startup step from Postgres connect through public listener bind. It does not change startup behavior. - Each step logs `Startup phase started` and `Startup phase finished`, with `phase`, `elapsed_ms`, and `status` (`succeeded` / `degraded` / `failed`, the same values `buzz_process_lifecycle` records use). - Gauge `buzz_startup_phase_current{phase}` is `1` while a step runs and `0` after it ends. A pod that is still starting shows exactly which step it is on. - Gauge `buzz_startup_phase_seconds{phase}` records each finished step's duration. - The startup NIP-43 membership reconcile logs `NIP-43 startup reconciliation progress` every 5,000 communities, with `done`, `total`, `republished`, `failed`, and `elapsed_ms`. These are gauges rather than counters on purpose. Each value is written once per boot, and a monotonic-counter scrape drops a series' first sample, so a one-shot counter never shows up. ## Contract change `deploy/charts/buzz/README.md` ("Early-startup telemetry contract") said the early lifecycle phases are log-only. That still holds: `crypto_init` through `metrics_bind` are unchanged and emit no metrics. The README now also documents the post-bind steps. The guard in `boot_lifecycle.rs` now allows exactly the two `buzz_startup_phase_*` families. It still rejects any other startup, boot, or lifecycle family, and it also rejects any exported label value that names an early phase. ## Design - `crates/buzz-relay/src/startup_steps.rs` adds a closed `StartupStep` vocabulary (18 steps) and a `StepTimer` guard. - A step ends exactly once: - `finish()` → `succeeded` - `degrade()` then `finish()` → `degraded` (a non-fatal error branch) - dropped without finishing → `failed` (covers `?` early returns and panics) - The NIP-43 and large-roster reconciles now return `ReconcileSummary { repaired, failed }`. A sweep where any item fails ends its step `degraded`, and the final log line includes the `failed` count. NIP-FI warm ends `degraded` when any issuer did not warm. The deprecated `reconcile_nip43_membership_snapshots` wrapper still returns `repaired`. - Steps that are disabled by config (migrate, audit connect, NIP-FI warm, git probe, NIP-43 reconcile) are not timed, so their series never appear. - The gauges are set once, so they age out after the existing gauge idle timeout (2700 s with default intervals; logged as `idle_timeout_secs`). A step that runs longer than that loses its `_current` series. The README says so. - Progress logging runs only for the `Bootstrap` reconcile. The 60 s maintenance sweep stays quiet. ## What got simpler Little got simpler in the code: this mostly adds instrumentation. One CI gap closed along the way: the 54 `config` tests and the 9 non-ignored `boot_lifecycle` tests, which ran in no CI lane, now run in `test-unit`. The instrumentation is scoped as small as I could make it, and it is the evidence needed for the follow-up that should delete code: moving the per-community NIP-43 reconcile off the pre-bind path, after which the 900 s startup-probe override can go back to the chart default. ## Testing - `startup_steps` unit tests (`DebuggingRecorder`): - a running step is current and has no duration - a finished step clears current and records its elapsed time - dropped and degraded steps clear current and record a duration - log capture: `finish`, `degrade`+`finish`, drop, and panic each emit exactly one `Startup phase finished` record with the right `status`. Mutating `degrade()` to a no-op and removing the ends-once guard makes these tests fail. - the vocabulary is unique and shares no names with the early lifecycle phases - Tests that capture logs or check callsite interest now share `test_support::tracing_dispatch_lock()`. These are the `startup_steps` helpers, two `telemetry` tests, and the `config` and `bridge` log-capture helpers. tracing-core caches each callsite's interest across the whole process. Without the lock, the new panic test lost its record in 60 of 150 parallel runs of `telemetry:: startup_steps:: config::`. With the lock it failed 0 of 300. Removing the lock from `gauges()` brings the failures back (53 of 100 runs). - `nip43_progress_logs_every_interval_inside_the_startup_sweep_only` checks the cadence at 79,942 communities (15 progress lines, none at the start or end) and table-tests that the maintenance sweep never logs progress. - `boot_lifecycle::successful_main_emits_complete_lifecycle_and_reports_current_startup_step` boots the real relay binary against a TCP peer that never speaks Postgres. It asserts that `/metrics` shows `buzz_startup_phase_current{phase="db_connect"} 1` and no finished-step durations, and that no early phase leaks into metrics. - **CI routing.** `just test-unit` picks relay tests by hand-maintained filter. Before this PR, CI's filters selected none of the `startup_steps`, `telemetry`, `config` or non-ignored `boot_lifecycle` tests above; the NIP-43 cadence test is picked up by the existing `handlers::side_effects::tests::` clause. Its infra-free relay stanza now selects `startup_steps::tests::`, `telemetry::tests::` and `config::tests::` (anchored so `nip_fi_config` is not selected twice), and a new `cargo nextest run -p buzz-relay --test boot_lifecycle` line runs the non-ignored `boot_lifecycle` tests (they spawn the real binary and need no Postgres or Redis; the Postgres cases stay `#[ignore]`d). `scripts/run-tests.sh` mirrors both, running `telemetry` and `config` single-threaded because they mutate process env. ### Manual check 1. `just relay` (or run `buzz-relay` against local Postgres and Redis). 2. Once the relay is up: `curl -s localhost:9102/metrics | grep buzz_startup_phase`. Every step that ran has a `_seconds` value, and every `_current` value is 0. 3. In the relay logs, `"Startup phase finished"` lines show each step's `elapsed_ms`. --------- Signed-off-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz> Co-authored-by: Elrond <28d6302a099e5225b02c4155ac4236e4912603df2ab08dbfc2f4fef08ce598c8@buzz.block.builderlab.xyz>	2026-10-07 10:15:43 -04:00
Tyler	1a2ddf9eaa	feat(relay): private read-state accessory API (#7906) ... Buzz BFF: Private read-state accessory API (`/buzz/v1`)	2026-10-06 16:12:03 -04:00