alepha@docs:~/docs/reference/providers$
cat JobProvider.md | pretty
2 min read
Last commit:

#JobProvider

#Import

typescript 
1import { JobProvider } from "alepha/api/jobs";

#Overview

Coordinates cron and push jobs with a durable outbox table and a single reconciliation sweep. The actual delivery channel (queue / direct) is abstracted behind JobDispatcher, substituted by DI:

  • DirectJobDispatcher (default, registered by AlephaApiJobs) - runs the handler in-process right after push() returns.
  • QueueJobDispatcher (registered by AlephaApiJobsQueue): sends the executionId through AlephaQueue so a pool of workers can pick it up.

Push flow: push() → INSERT row (pending) → dispatcher.dispatch(jobName, id) worker → claim → UPDATE running → handler → UPDATE ok (or DELETE when successes are not recorded) → UPDATE error (or DELETE) / scheduled (retry) on failure

Cron flow: scheduler tick → claim the instant → acquire lock → executeInline (no retry) → enqueue + dispatch (retry declared)

Sweep responsibilities (every sweepCron), one per status, declared in sweepTable():

  • scheduled with scheduledAt <= now → pending + dispatch
  • pending untouched for staleThreshold → re-dispatch
  • running past its lease → failed, then the retry policy

Trim runs on its own cron (trimCron, default hourly):

  • rows of a job name nothing registers are purged first, every status (a renamed job loses its rows, see purgeUnregistered)
  • then each job's retention: per status, rows past the newest last or completed more than days ago (see JobRetentionProvider)
  • decoupled from sweep because trim cost scales with job count, not retry latency - running it every sweep is wasted work for most apps.