Metered billing outbox reconciler.

Scans accrue_meter_events for rows stuck in stripe_status = "pending" more than 60 seconds and retries the Stripe call. Closes the "row committed but the process died before the synchronous Stripe call returned" gap in Accrue.Billing.MeterEventActions.report_usage/3.

Wiring

Accrue does not start its own Oban — the host app schedules this worker in its Oban cron config:

config :my_app, Oban,
  plugins: [
    {Oban.Plugins.Cron,
     crontab: [
       {"* * * * *", Accrue.Jobs.MeterEventsReconciler}
     ]}
  ]

The worker runs on queue :accrue_meters (host must declare the queue in its Oban config). Each tick processes up to 1 000 rows.

Failure handling

On Stripe error the row flips to failed — the reconciler does NOT keep retrying the same row in the same tick, avoiding the "stuck row infinite retry" footgun. Failures emit [:accrue, :ops, :meter_reporting_failed] with source: :reconciler so ops can distinguish inline vs deferred failures.