A

Slice 43 - Scheduled Sync Retry Policy

Scope#

Scheduled sync should have an explicit unattended policy without adding a local queue, hosted worker, or new daemon.

Local automation still launches foreground codealmanac sync. The scheduled command now supplies:

  • stable claim owner: codealmanac.automation.sync
  • pending timeout: 24h
  • failed-attempt budget: 3

Shape#

AutomationService
  -> ScheduledJob(program_arguments=(..., "sync", "--claim-owner", ...))
  -> CLI dispatch
  -> RunSyncRequest(claim_owner=..., max_failed_attempts=...)
  -> SyncWorkflow
  -> SyncLedgerEntry.failed_attempts

SyncLedgerEntry.failed_attempts is durable. It increments when an Ingest run fails or when sync reconciles a linked failed pending run. Successful absorb resets it to zero.

When a failed entry reaches max_failed_attempts, sync returns sync-retry-budget-exhausted as needs-attention instead of retrying forever.

Tests#

  • automation install includes stable sync owner, pending timeout, and failed-attempt budget in scheduled argv
  • failed sync ingest records failed_attempts = 1
  • exhausted failed entries report sync-retry-budget-exhausted
  • focused sync, automation, and CLI tests cover the command path

Cosmic Python Note#

Chapter 11's async messaging section shaped this slice: scheduled sync is an external command source. The adapter should translate that source into explicit command data before workflow code runs. The workflow still owns the durable state transition.