santifer/career-ops · warning

concurrent reservation test flaked

Error message

concurrent reservation test flaked (${e.message}). Retrying once...

What it means

test-all.mjs stress-tests reserve-report-num.mjs by spawning concurrent reservations and asserting the returned ranges never overlap. Because concurrent file-based allocation can transiently collide, the test wraps the assertion in a retry loop: the first failure logs this warning and retries once; only a second consecutive failure is recorded as a real test failure.

Solutions

1. If it flakes once and passes on retry, ignore it — that is the designed behavior for timing-sensitive concurrency.
2. If it fails consistently, debug reserve-report-num.mjs's atomic claim path (sentinels, locks, GC of stale sentinels) with the overlapping ranges from the error message.
3. Reduce external load or run the concurrency tests on a local filesystem rather than a network mount.
4. Increase retries locally while bisecting, but do not paper over a deterministic overlap.

Example fix

// before (diagnosis step)
node reserve-report-num.mjs --count 4 & node reserve-report-num.mjs --count 4 & wait
// after (compare the two printed ranges; overlap => bug in claim logic, not a flake)

// pre-flight: ensure the allocator works serially before stressing it
if (run('node', ['reserve-report-num.mjs', '--count', '1']) === null) fail('allocator broken; skip concurrency stress');

try { assertNoOverlap(rangeX, rangeY); } catch (e) { if (retries-- > 0) warn(`flaked (${e.message}), retrying`); else fail(e.message); }

When it happens

Trigger: Two concurrent reservation workers return ranges X and Y sharing an overlap, or another assertion inside the try throws — on the first attempt while reserveRetries > 0, producing this warning; a second failure calls fail().

Common situations: Heavily loaded CI runners slowing atomic claim/lock timing, filesystems with weak locking semantics (some network mounts), or a genuine regression in reserve-report-num.mjs's claim-and-release logic (would fail both attempts).

Related errors

• merge-tracker concurrent write test flaked
• Skipped # : status changed from to during review
• archive render skipped — no Playwright browser in env
• Cannot import generate-pdf.mjs
• Cannot verify tracker lock ownership at

AI-assisted analysis of santifer/career-ops@e7abd431fc (2026-09-16). Data as JSON: /api/errors/79d54cf5d3f2a7d7. Report an issue: GitHub.

Appendix: source

Thrown at test-all.mjs:10152

        let stdout = '';
        child.stdout.on('data', chunk => { stdout += chunk; });
        child.on('close', () => resolve(stdout.trim()));
      });
      const [rangeX, rangeY] = await Promise.all([spawnReserve(), spawnReserve()]);
      const toNums = r => {
        const [s, e] = r.split('-').map(Number);
        return Array.from({ length: e - s + 1 }, (_, i) => s + i);
      };
      const overlap = toNums(rangeX).filter(n => toNums(rangeY).includes(n));
      if (rangeX && rangeY && overlap.length === 0) {
        pass(`concurrent --count 4 reservations are disjoint (${rangeX} vs ${rangeY})`);
      } else {
        throw new Error(`concurrent ranges overlap: ${rangeX} vs ${rangeY} share [${overlap}]`);
      }
      break;
    } catch (e) {
      if (reserveRetries > 0) {
        warn(`concurrent reservation test flaked (${e.message}). Retrying once...`);
        reserveRetries -= 1;
      } else {
        fail(`concurrent reservation test failed: ${e.message}`);
        break;
      }
    } finally {
      rmSync(concTmp, { recursive: true, force: true });
    }
  }

  // --release with a range deletes every sentinel in it.
  const reserveRunFail = (args, dir) => {
    try {
      execFileSync(NODE, [RESERVE, ...args], {
        encoding: 'utf-8',
        stdio: ['pipe', 'pipe', 'pipe'],
        env: { ...process.env, CAREER_OPS_REPORTS_DIR: dir, CAREER_OPS_TRACKER: join(dir, 'applications.md') },
      });

View on GitHub (pinned to e7abd431fc)