A claim that failed with anything but sql.ErrNoRows parked the worker
on the wake channel, so a job queued across a restart (which has no
wake token) stayed queued until another enqueue or restart. The worker
now retries such a claim with backoff from 50ms doubling to 5s, reset
on any successful or empty claim, and stops promptly on shutdown. The
claim error is logged by category only.
The claim query is a Runner field so a test can inject one failing
claim; TestClaimErrorIsRetried drops the wake token and asserts the
queued job still runs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Boot recovery now writes the required fixed text to the job error column,
and the recovery test asserts that literal text instead of the constant.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server now registers the framework's built-in noop kind through
registerJobKinds before the runner starts, so production can run a job; a
noop job succeeds with {"ok":true}. Each feature's kinds are registered in
the same function.
Handlers report failure only with a *jobs.Failure code their kind declared
at registration. The runner stores and returns only that code, or
internal_error, interrupted or unknown_kind, and logs only the job id, kind,
code and a fixed reason: handler error text is never stored, returned or
logged. Results implement jobs.Public, whose PublicJSON names exactly the
data that is stored and returned.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
internal/jobs persists jobs in the jobs table, which is also the queue: a
bounded pool of workers claims queued jobs oldest first. Handlers are
registered per kind before Start with typed payloads and results, and can
publish partial results to the row while they run. Only typed results and
*jobs.Failure messages are stored; any other handler error is logged and
stored as "internal error", so dependency error text cannot surface a
secret. Start fails jobs a previous process left running with
"interrupted by restart" before any worker runs.
GET /api/v1/jobs/{id} returns {id, kind, status, result, error} inside the
protected /api/v1 group; an unknown id is a 404 not_found.
Migration 0003_jobs_status rebuilds jobs with status queued, running,
succeeded, failed, mapping legacy pending and done rows.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Feature 03 made the OIDC variables required, so the Feature 04 boot tests'
loadConfig helper now sets them alongside the secret key file.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dev_agentandClaude Opus 5.5