buzz/scripts/channel-labels-e2e

Native channel-label command E2E

This runs the candidate CLI against two separate candidate relay processes sharing isolated PostgreSQL and Redis. It uses the production HTTP bridge and NIP-98 admission. A loopback proxy discards one successful response after commit, records request hashes, then forwards the exact-event retry. This is not a mocked receipt server.

Requirements and execution

• An empty, disposable PostgreSQL database and disposable Redis service. Do not point this harness at Desktop, staging, production, or another test's database.
• Built buzz, buzz-relay, and buzz-admin binaries from the same checkout.
• Python 3.10+ (Hermit uv can use an existing managed Python). No Python packages.
• Local loopback ports; seven free ports are selected. A bind race fails setup.

. ./bin/activate-hermit
bin/cargo build -p buzz-cli -p buzz-relay -p buzz-admin
# Create a fresh DB using your isolated PostgreSQL administration connection.
# The harness migrates it; it never drops a database or changes DB credentials.
export LABEL_DATABASE_URL='postgres://user:password@127.0.0.1:5432/labels_unique_run'
export LABEL_REDIS_URL='redis://127.0.0.1:6379'
bin/uv run --no-project --offline python scripts/channel-labels-e2e/run.py   --evidence /tmp/buzz-e2e/labels-unique-run   --confirm-isolated-empty-database

Use --bin-dir for another Cargo target directory. The evidence directory must not exist. It is mode 0700 and contains generated fixture private keys: do not publish it wholesale. Share sanitized logs and public artifact hashes only. The harness records source revision, dirty status and all three binary SHA-256s in SOURCE.json. FLOWS.stdout lists assertions; PROXY.jsonl binds the lost ACK and exact retry to identical command bodies and different transport proofs. TEARDOWN.json records stopped processes; the database remains for inspection. Remove the run-owned database and evidence manually when no longer needed.

Evidence boundaries

Measured here: labeled creation, verified reads, committed-but-lost ACK, opposing mutation, exact retry without reapplication, no-op snapshot reuse, ordinary topic/purpose preservation (including empty labels), concurrent CLI additions, generic label discovery, unauthorized rejection, both backend processes exercised, and final readiness of both processes.

The concurrent additions do not prove lock overlap. Deterministic lock ordering, rollback, archive/deletion/removal races, current ACLs, cross-community UUID/event collisions, stale discovery caches, mixed/id reads, activation, and pagination are covered by the PostgreSQL/router suites instead. Run complete packages with scripts/postgres-test-run.sh -p buzz-db -p buzz-relay -p buzz-admin; see crates/buzz-db/TESTING.md for isolated runner configuration. Core and CLI unit suites cover malformed grammar, snapshot trust and torn/exclusive journals.

Test-only BUZZ_GIT_CONFORMANCE_PROBE=false avoids requiring S3 for this non-Git workflow. Git/media admission, Docker relay images, Kubernetes, forced live cross-process overlap, multi-tenant bystander progress, process-kill recovery, and deployment credential fencing are not established by this harness. The offline cutover declaration enables these isolated test processes; it does not exercise a real fleet upgrade. See docs/channel-labels-rollout.md.

The runner's readiness polling is bounded to 60 seconds; every command has a bounded timeout and all run-owned processes are terminated on completion/failure. Generated credentials never leave loopback test infrastructure.