Recurring archive and delete
Run PostgreSQL, MySQL or MongoDB cycles to S3, Azure Blob or GCS, with verified deletion gating each new cycle.
Enroll a policy
Create and test a policy using PostgreSQL, MySQL or MongoDB with S3, Azure Blob or GCS. Recurring enrollment requires explicit source-deletion authorization and an interval from 60 seconds to one year. Use immutable-rows for PostgreSQL only when eligible values and membership stay stable through the cycle; MongoDB requires quiescent-copy. Stability must hold during downtime and retries too.
coldctl agent scheduler recurring create pg-policy --every-seconds 86400 --source-stability immutable-rows --confirm-original-source --delete-after-archive
coldctl agent scheduler recurring create mongo-policy --every-seconds 86400 --source-stability quiescent-copy --confirm-original-source --delete-after-archive
coldctl agent scheduler runCycle boundaries and late arrivals
A cycle freezes its retention cutoff and configuration, exports, verifies and deletes verified unchanged records. A new cycle is admitted only after durable evidence of completed deletion. The interval starts after completion is observed, not on a fixed calendar schedule. Each fresh cycle rescans eligible data, including lower-key and backdated arrivals between cycles. This is not CDC or an incremental watermark. Missed intervals coalesce rather than creating a catch-up flood.
Inspect and recover
The worker runs in the foreground; use an OS service manager to keep it running. Ctrl+C stops the worker safely and restart resumes the same saved cycle. A failed cycle holds the schedule; after fixing the cause, recurring retry retries that cycle rather than exporting a new one. MySQL sources are not supported for recurring enrollment.
coldctl agent scheduler recurring list
coldctl agent scheduler recurring show RECURRING_ID
coldctl agent scheduler recurring retry RECURRING_ID --confirm-original-sourceStop and policy safety
recurring stop is terminal, cancels pending work and requests cancellation of an incomplete job. An in-flight batch may commit; already committed deletions remain. A stopped schedule reserves its policy history, so re-enrollment needs a separately reviewed policy. Manual exports for a reserved policy and overlapping policies are not a supported bypass. Use one worker per local state directory; distributed scheduling is not implemented. Run agent sync --watch separately for background Cloud metadata synchronization.
coldctl agent scheduler recurring stop RECURRING_ID