ci: bound Hermit fetches and let relay builds run without sccache (#8739)

## Summary

Two CI-workflow fixes from the 23–30 Oct flake audit. This is one of six
PRs split from the reviewed `flakey-tests` branch (siblings below). Each
PR applies on its own to `main`.

### Hermit tool downloads (5 runs, 1 on main)

- **Runs:** 40635610550 and 40941660945, in Detect Changed Paths
(`node`). The selection step printed nothing for 9–10 min, the job hit
its cap, and every gate failed as cancelled. Three more, 40373404267 on
main (PostgreSQL Tests and Desktop E2E Integration 1/3, `pgschema`),
40373003333 (Mobile Android, `cmake`) and 40373467575 (Desktop Build
macOS, `just`), got 504s during GitHub's release-asset outage on 10-26,
~17:54–18:00 UTC.
- **Cause:** jobs start with an empty Hermit cache, so a tool's first
call downloads it, with no deadline. Hermit v0.54.1 tries three times, a
second apart, then gives up.
- **Change:** a new `scripts/ci-hermit-retry.sh` runs a no-op tool
command with a 90 s deadline per attempt, three attempts and 15 s × n
backoff. The deadline is plain bash: job control gives each attempt its
own process group, which gets TERM and then KILL. macOS runners have no
`timeout(1)` and may run bash 3.2. A "Fetch Hermit tools" step calls it
for:
- `node` and `just` in Detect Changed Paths, where one stall cancels
every gate. One `sh -c` invocation covers both, so the worst case is 345
s, inside the 10-minute cap. Two invocations would take 690 s.
- `pgschema` in PostgreSQL Tests, Desktop E2E Integration and Backend
Integration (via `--help`, since it has no `--version`).
  - `just` and `cmake` in the Android lane.
  - `just` in Desktop Build (macOS).

The test, `scripts/ci-hermit-retry.test.mjs`, joins Detect Changed
Paths' `node --test` list, and the script joins the rust and mobile path
filters. One stalled attempt costs ~105 s, which fits every job's cap.
- A killed attempt doesn't break the retry. Hermit downloads to a temp
file and unpacks to a temp dir, and renames each into place only once
it's complete. Its lock is a kernel flock.

### sccache storage failures (2 runs)

- **Runs:** 40355289475 (Desktop E2E Relay postgres-tests, 10-26,
"Egress is over the account limit") and 40714034945 (relay-binaries,
10-28, "send http request").
- **Cause:** the sccache server checks its GHA cache storage at startup
and won't start if that fails. Cargo's first `sccache rustc -vV` spawns
it, so the build fails. `SCCACHE_IGNORE_SERVER_IO_ERROR` doesn't help
here: in v0.16.0 it is read only when a running server's compile
response fails.
- **Change:** a "Start sccache" step runs `sccache --start-server`. It
outputs `wrapper=sccache`, or an empty wrapper and a warning if the
server won't start. The three build steps take `RUSTC_WRAPPER` from it,
and cargo treats an empty `RUSTC_WRAPPER` as no wrapper.
- The fallback doesn't fail the job later, in sccache-action's post
step. At the pinned fc920bf, that step runs `sccache --show-stats` and
the JSON form, and fails the job if either exits non-zero. With no
server running, v0.16.0's `--show-stats` doesn't start one. It returns
zero stats from the client side and exits 0. So the step keeps its stats
annotation, and `disable_annotations` stays unset.

### Related issue

None found. **Overlaps #8213** ("ci: fix empty root-workspace Rust
caches and merge test-unit nextest runs"). #8213 already conflicts with
`main`, and this PR adds a conflict in `.github/workflows/ci.yml` (path
filters).

### Testing

Verified locally:
- With a fresh Hermit state dir, `node` was SIGKILLed mid-download, then
mid-unpack while holding the lock, then stopped with TERM by the
script's own 1 s deadline. Each retry produced a working v24.15.0, at
clean-unpack speed. The `sh -c` form fetched both tools, and a 1 s
deadline on `sh -c 'sleep 37 && ...'` stopped `sh` and its child,
leaving no orphan. Tested under macOS bash 3.2 and Debian bash 5.2.
- sccache v0.16.0 (aarch64-apple-darwin) ran against a GHA cache whose
`ACTIONS_RESULTS_URL` refused connections, in READ_ONLY and READ_WRITE
modes. `--start-server` exited 2 ("Server startup failed: cache storage
failed to read"). Both `--show-stats` forms then exited 0, with zero
stats and valid JSON. `--stop-server` exited 2, which shows no server
had started.

Only hosted CI can show:
- the script under the runners' real shells, and that healthy runs only
pay a few seconds;
- whether ~3 min of spaced attempts rides out the next outage. The 10-26
outage lasted ~6 min, longer than this covers;
- that the explicit `--start-server` behaves like cargo's implicit start
against the real GHA cache. The fallback branch needs a storage outage
to run in CI.

Not covered here: first fetches of other tools (rustup, uv, pnpm) still
have no deadline. A per-job Hermit package cache is the follow-up.

Pushed with `git push --no-verify`, because the pre-push suites fail
when the laptop is under load.

### Readiness

An agent review of the combined `flakey-tests` branch (b432453af) found
no blockers. `buzz-review-completed` is not added, because the human
test step (AGENTS.md "Before opening a PR", step 3) hasn't happened yet.

### Siblings

- #8740 test(relay): thread_window aux-page hold and `api::media`
Postgres tests
- #8741 test(mobile): RunnerTests title fade and alert wait
- #8742 fix(desktop): deep-linked message hold (product change)
- #8743 test(desktop): four specs racing scroll settles and focus
- #8744 test(desktop): five settle classes and remote-owned-mentions

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Signed-off-by: Matt Toohey <contact@matttoohey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

.github/workflows/_ci-clients.yml

@@ -177,6 +177,13 @@ jobs:
             ~/.gradle/wrapper
           key: gradle-${{ runner.os }}-${{ hashFiles('mobile/android/**/*.gradle*', 'mobile/android/gradle/wrapper/gradle-wrapper.properties', 'mobile/pubspec.lock') }}
           restore-keys: gradle-${{ runner.os }}-
+      # The Hermit package cache above holds only Flutter, so the build's first
+      # just and cmake calls download them.
+      - name: Fetch Hermit tools
+        if: matrix.lane == 'android'
+        run: |
+          scripts/ci-hermit-retry.sh ./bin/just --version
+          scripts/ci-hermit-retry.sh ./bin/cmake --version
       # SDK component and plugin downloads intermittently fail on hosted
       # runners; retry once only for those, so real build failures still fail.
       - name: Build Android debug APK

.github/workflows/_ci-desktop-macos.yml

@@ -31,6 +31,8 @@ jobs:
     steps:
       - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
       - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1
+      - name: Fetch Hermit tools
+        run: scripts/ci-hermit-retry.sh ./bin/just --version
       - name: Verify macOS icon representations
         run: scripts/macos-icon.sh --check
       - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32

.github/workflows/_ci-relay.yml

@@ -88,6 +88,21 @@ jobs:
         uses: Mozilla-Actions/sccache-action@fc920bf0ec8de6ee65d409111f7ec508035751ba # v0.0.11 # zizmor: ignore[cache-poisoning] Bounded trial: only PR 5224 writes to its isolated merge-ref scope; trusted pushes retain production writes.
         with:
           version: v0.16.0
+      # The sccache server checks its GHA cache storage at startup and refuses to
+      # start when that fails (egress over the account limit, a failed read),
+      # which used to fail cargo's first `sccache rustc -vV` and the build.
+      # SCCACHE_IGNORE_SERVER_IO_ERROR doesn't cover startup in v0.16.0, so start
+      # it here and build with plain rustc if it won't start.
+      - name: Start sccache
+        id: sccache
+        if: steps.relay-artifacts-cache.outputs.cache-hit != 'true'
+        run: |
+          if sccache --start-server; then
+            echo "wrapper=sccache" >> "$GITHUB_OUTPUT"
+          else
+            echo "::warning::sccache did not start; building without it"
+            echo "wrapper=" >> "$GITHUB_OUTPUT"
+          fi
       - name: Install cargo-nextest
         if: steps.relay-artifacts-cache.outputs.cache-hit != 'true' && matrix.artifact != 'relay-binaries'
         uses: taiki-e/install-action@0fd46367812ee04360509b4169d9f659d6892bb2 # v2.79.15
@@ -96,12 +111,12 @@ jobs:
       - name: Build relay binaries
         if: steps.relay-artifacts-cache.outputs.cache-hit != 'true' && matrix.artifact == 'relay-binaries'
         env:
-          RUSTC_WRAPPER: sccache
+          RUSTC_WRAPPER: ${{ steps.sccache.outputs.wrapper }}
         run: cargo build --profile ci -p buzz-relay -p git-credential-nostr
       - name: Build backend integration test archive
         if: steps.relay-artifacts-cache.outputs.cache-hit != 'true' && matrix.artifact == 'backend-integration-tests'
         env:
-          RUSTC_WRAPPER: sccache
+          RUSTC_WRAPPER: ${{ steps.sccache.outputs.wrapper }}
         run: |
           cargo nextest archive \
             --cargo-profile ci \
@@ -116,7 +131,7 @@ jobs:
       - name: Build PostgreSQL test archive
         if: steps.relay-artifacts-cache.outputs.cache-hit != 'true' && matrix.artifact == 'postgres-tests'
         env:
-          RUSTC_WRAPPER: sccache
+          RUSTC_WRAPPER: ${{ steps.sccache.outputs.wrapper }}
         run: |
           postgres_package_args=()
           while IFS= read -r package; do
@@ -191,6 +206,9 @@ jobs:
     steps:
       - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
       - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1
+      # pgschema has no --version; --help exits 0 once Hermit has fetched it.
+      - name: Fetch Hermit tools
+        run: scripts/ci-hermit-retry.sh ./bin/pgschema --help
       - name: Tune disposable PostgreSQL
         env:
           PGPASSWORD: ${{ env.BUZZ_TEST_POSTGRES_PASSWORD }}
@@ -243,6 +261,9 @@ jobs:
     steps:
       - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
       - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1
+      # pgschema has no --version; --help exits 0 once Hermit has fetched it.
+      - name: Fetch Hermit tools
+        run: scripts/ci-hermit-retry.sh ./bin/pgschema --help
       - name: Start integration services
         run: |
           for attempt in 1 2 3; do
@@ -433,6 +454,9 @@ jobs:
     steps:
       - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
       - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1
+      # pgschema has no --version; --help exits 0 once Hermit has fetched it.
+      - name: Fetch Hermit tools
+        run: scripts/ci-hermit-retry.sh ./bin/pgschema --help
       - name: Install cargo-nextest
         uses: taiki-e/install-action@0fd46367812ee04360509b4169d9f659d6892bb2 # v2.79.15
         with:

.github/workflows/ci.yml

@@ -34,6 +34,12 @@ jobs:
         with:
           fetch-depth: 2
       - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1
+      # The first call of each Hermit tool downloads it. A stalled Node download
+      # hung the selection step until this job's cap cancelled every gate. One
+      # invocation fetches both tools, so its worst case (under 6 minutes)
+      # still leaves the rest of the job inside the 10-minute cap.
+      - name: Fetch Hermit tools
+        run: scripts/ci-hermit-retry.sh sh -c './bin/node --version && ./bin/just --version'
       - uses: dorny/paths-filter@ceb8a2b8f2d89434be7ff52d3de7ec3738c5cc9d # v4.0.3
         id: filter
         with:
@@ -69,6 +75,7 @@ jobs:
               - 'scripts/model-capabilities.json'
               - 'scripts/normative-corpus.json'
               - 'scripts/ci-apt-retry.sh'
+              - 'scripts/ci-hermit-retry.sh'
               - 'Justfile'
             desktop:
               - '.github/workflows/ci.yml'
@@ -104,6 +111,7 @@ jobs:
               - 'scripts/test-mobile-release-contract.sh'
               - 'scripts/test-mobile-release-candidate-publisher.sh'
               - 'scripts/test-mobile-worktree-overrides.sh'
+              - 'scripts/ci-hermit-retry.sh'
               - '.github/workflows/mobile-release-candidate.yml'
               - '.github/workflows/ci.yml'
               - '.github/workflows/_ci-*.yml'
@@ -144,7 +152,7 @@ jobs:
       - name: CI required-context isolation contract
         run: scripts/test-ci-required-context-isolation.sh
       - name: CI path selection regression tests
-        run: node --test scripts/ci-selection.test.mjs scripts/ci-required-gates.test.mjs scripts/ci-apt-retry.test.mjs
+        run: node --test scripts/ci-selection.test.mjs scripts/ci-required-gates.test.mjs scripts/ci-apt-retry.test.mjs scripts/ci-hermit-retry.test.mjs
       - name: File size policy
         run: just file-size-check
 

lefthook.yml

@@ -14,8 +14,9 @@
 # they diverge and CI remains the authoritative gate for that ref.
 # Deliberate deviations:
 # - The `.github/workflows/ci.yml` path CI adds to its `rust`/`desktop`/`mobile`
-#   filters, and `scripts/ci-apt-retry.sh` in `rust`/`desktop`, are omitted;
-#   a CI-plumbing-only edit doesn't need a local test run.
+#   filters, `scripts/ci-apt-retry.sh` in `rust`/`desktop`, and
+#   `scripts/ci-hermit-retry.sh` in `rust`/`mobile`, are omitted; a
+#   CI-plumbing-only edit doesn't need a local test run.
 # - `desktop-test` doesn't trigger on `crates/**` changes, though CI's Desktop
 #   JS job does: a few node:test cases read relay/DB crate sources to pin shared
 #   constants. Running the full desktop suite on every crate push would be

scripts/ci-hermit-retry.sh

@@ -0,0 +1,86 @@
+#!/usr/bin/env bash
+# =============================================================================
+# ci-hermit-retry.sh — fetch a Hermit tool with a per-attempt deadline
+# =============================================================================
+# CI jobs start with an empty Hermit package cache, so the first run of a bin/
+# stub downloads its package. That download has no deadline: a stalled Node
+# download hung Detect Changed Paths until the job's cap cancelled every gate.
+# Hermit tries a download three times, a second apart, so a short run of
+# GitHub release-asset 504s fails the job outright.
+#
+# Run a cheap command for each tool through this before the job's first real
+# use (`scripts/ci-hermit-retry.sh ./bin/node --version`). Each attempt runs in
+# its own process group, which is stopped at the deadline (TERM, then KILL)
+# and retried after a growing backoff.
+#
+# A stopped attempt leaves nothing Hermit reuses. Hermit downloads to a
+# temporary file and unpacks to a temporary directory, renaming each into place
+# only once complete, and its state lock is an flock that the kernel drops with
+# the process. The leftover temporaries only cost disk.
+#
+# With the defaults one invocation is bounded at under 6 minutes. Size job
+# timeouts to fit that plus the job's real work. A job with a tight cap can
+# fetch several tools in one invocation through `sh -c`, as Detect Changed
+# Paths does.
+#
+# Runs under macOS's /bin/bash 3.2 and does not use timeout(1), which macOS
+# runners lack.
+#
+# Usage:
+#   scripts/ci-hermit-retry.sh <command> [args...]
+#
+# Env:
+#   CI_HERMIT_ATTEMPTS            Attempts before failing (default: 3).
+#   CI_HERMIT_ATTEMPT_SECONDS     Deadline per attempt (default: 90). Healthy
+#                                 fetches finish well inside a minute.
+#   CI_HERMIT_KILL_AFTER_SECONDS  Wait after TERM before KILL (default: 10).
+#   CI_HERMIT_BACKOFF_SECONDS     Sleep after failed attempt n is n times this
+#                                 (default: 15).
+# =============================================================================
+set -euo pipefail
+
+attempts="${CI_HERMIT_ATTEMPTS:-3}"
+deadline="${CI_HERMIT_ATTEMPT_SECONDS:-90}"
+kill_after="${CI_HERMIT_KILL_AFTER_SECONDS:-10}"
+backoff="${CI_HERMIT_BACKOFF_SECONDS:-15}"
+
+if [ "$#" -eq 0 ]; then
+  echo "usage: $0 <command> [args...]" >&2
+  exit 2
+fi
+
+for attempt in $(seq 1 "$attempts"); do
+  # Job control puts each background job in its own process group, so the
+  # watchdog's signals reach everything the attempt started.
+  set -m
+  "$@" &
+  pid=$!
+  (
+    sleep "$deadline"
+    kill -TERM -- "-$pid" 2>/dev/null || exit 0
+    sleep "$kill_after"
+    kill -KILL -- "-$pid" 2>/dev/null || true
+  ) >/dev/null 2>&1 &
+  watchdog=$!
+  set +m
+  started=$SECONDS
+  rc=0
+  # Bash reports a job killed by a signal on wait's stderr; the warning below
+  # says the same thing.
+  wait "$pid" 2>/dev/null || rc=$?
+  kill -- "-$watchdog" 2>/dev/null || true
+  wait "$watchdog" 2>/dev/null || true
+  [ "$rc" -eq 0 ] && exit 0
+  # The leader can exit on TERM while a child that ignores it lives on.
+  kill -KILL -- "-$pid" 2>/dev/null || true
+  if [ $((SECONDS - started)) -ge "$deadline" ]; then
+    echo "::warning::attempt $attempt/$attempts timed out after ${deadline}s: $*"
+  else
+    echo "::warning::attempt $attempt/$attempts failed (exit $rc): $*"
+  fi
+  if [ "$attempt" -lt "$attempts" ]; then
+    sleep $((attempt * backoff))
+  fi
+done
+echo "::error::all $attempts attempts failed: $*"
+exit 1

scripts/ci-hermit-retry.test.mjs

@@ -0,0 +1,103 @@
+import assert from "node:assert/strict";
+import { spawnSync } from "node:child_process";
+import {
+  chmodSync,
+  existsSync,
+  mkdtempSync,
+  readFileSync,
+  writeFileSync,
+} from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { test } from "node:test";
+import { setTimeout as sleep } from "node:timers/promises";
+import { fileURLToPath } from "node:url";
+
+const script = fileURLToPath(new URL("./ci-hermit-retry.sh", import.meta.url));
+
+// Run the wrapper on a fake tool that counts its calls, then runs `body` with
+// $n set to the call number and $dir to a scratch directory.
+function run(body, env = {}) {
+  const dir = mkdtempSync(join(tmpdir(), "ci-hermit-retry-"));
+  const calls = join(dir, "calls");
+  writeFileSync(calls, "");
+  const tool = join(dir, "tool");
+  writeFileSync(
+    tool,
+    `#!/bin/bash
+dir="${dir}"
+echo call >> "$dir/calls"
+n=$(( $(wc -l < "$dir/calls") ))
+${body}
+`,
+  );
+  chmodSync(tool, 0o755);
+  const started = Date.now();
+  const result = spawnSync("bash", [script, tool], {
+    encoding: "utf8",
+    timeout: 20_000,
+    env: {
+      ...process.env,
+      CI_HERMIT_ATTEMPT_SECONDS: "1",
+      CI_HERMIT_KILL_AFTER_SECONDS: "1",
+      CI_HERMIT_BACKOFF_SECONDS: "0",
+      ...env,
+    },
+  });
+  return {
+    dir,
+    status: result.status,
+    seconds: (Date.now() - started) / 1000,
+    calls: readFileSync(calls, "utf8").split("\n").filter(Boolean).length,
+    stdout: result.stdout,
+  };
+}
+
+test("a first-attempt success runs the tool once and keeps its output", () => {
+  const r = run("echo v1.2.3");
+  assert.equal(r.status, 0);
+  assert.equal(r.calls, 1);
+  assert.equal(r.stdout, "v1.2.3\n");
+});
+
+// A stalled download: the first call hangs on a child that would write a
+// marker at 2s if it outlived the attempt's 1s deadline.
+test("a stalled attempt is stopped with its children and retried", async () => {
+  const r = run(`
+if [ "$n" -eq 1 ]; then
+  ( sleep 2; touch "$dir/survived" ) &
+  wait
+fi
+echo fetched`);
+  assert.equal(r.status, 0, r.stdout);
+  assert.equal(r.calls, 2);
+  assert.match(r.stdout, /::warning::attempt 1\/3 timed out after 1s/);
+  assert.match(r.stdout, /fetched\n$/);
+  assert.ok(r.seconds < 10, `took ${r.seconds}s`);
+  await sleep(Math.max(0, 3000 - r.seconds * 1000));
+  assert.ok(
+    !existsSync(join(r.dir, "survived")),
+    "the stalled attempt's child outlived it",
+  );
+});
+
+test("an attempt that ignores TERM is killed", () => {
+  const r = run("trap '' TERM; sleep 30", { CI_HERMIT_ATTEMPTS: "1" });
+  assert.equal(r.status, 1, r.stdout);
+  assert.match(r.stdout, /attempt 1\/1 timed out after 1s/);
+  assert.ok(r.seconds < 10, `took ${r.seconds}s`);
+});
+
+test("persistent failure backs off between attempts and stops at the budget", () => {
+  const r = run("exit 7", { CI_HERMIT_BACKOFF_SECONDS: "1" });
+  assert.equal(r.status, 1);
+  assert.equal(r.calls, 3);
+  assert.match(r.stdout, /attempt 3\/3 failed \(exit 7\)/);
+  assert.match(r.stdout, /::error::all 3 attempts failed/);
+  // Backoff sleeps 1s after attempt 1 and 2s after attempt 2.
+  assert.ok(r.seconds >= 3, `took ${r.seconds}s`);
+});
+
+test("no command is a usage error", () => {
+  assert.equal(spawnSync("bash", [script]).status, 2);
+});

scripts/ci-selection.test.mjs

@@ -244,6 +244,7 @@ const scenarios = [
   ],
   ["Justfile", ["Justfile"], ["desktop", "mobile", "rust"]],
   ["CI apt retry", ["scripts/ci-apt-retry.sh"], ["desktop", "rust"]],
+  ["CI Hermit retry", ["scripts/ci-hermit-retry.sh"], ["mobile", "rust"]],
   ["integration services", ["docker-compose.yml"], ["rust"]],
   ["CI integration images", ["docker-compose.ci.yml"], ["rust"]],
   [