Scheduled jobs and retention

Cluster ERP runs periodic work in two places: the cron table deploy/cluster-erp.cron of the service user, and background loops that every server worker starts in src/cluster_erp/server/app.py. This page lists both, the lock that keeps a loop to one worker, and the retention passes in app.py and src/cluster_erp/server/control_retention.py. The admin-configured record retention window is explained on the records module page.

Scope and source¶

• Defined in: deploy/cluster-erp.cron, src/cluster_erp/server/app.py (lifespan, _leader, and the _*_loop functions), src/cluster_erp/server/control_retention.py (sweep_control_plane), src/cluster_erp/server/membership.py.
• Count: four cron entries (two active, two commented), counted from the file; four background loops and one one-off background task, counted from the asyncio.create_task calls in lifespan (src/cluster_erp/server/app.py:164-194).

Exact cron times of a node are deployment data; the table gives the frequency only.

Cron entries¶

install.sh installs the file as the service user's cron table on every run (deploy/install.sh:210). The cron comment calls the membership sweep a second line of defence next to the in-process loop (deploy/cluster-erp.cron:9-10).

Background loops¶

Each 24-hour loop sleeps first, so it never runs at start-up; the start-up runs its own catch-up passes instead (src/cluster_erp/server/app.py:158-193). One database that fails is logged, and the loop continues with the next database (src/cluster_erp/server/app.py:452-454, src/cluster_erp/server/app.py:538-540).

Start-up passes¶

Leader lock¶

_leader takes pg_try_advisory_lock on the database the pass serves, with a key derived by CRC-32 from cluster_erp:<name>. The worker that gets the lock runs the pass; the others skip it. The lock is released at the end of the pass, so a crashed worker never blocks the next one (src/cluster_erp/server/app.py:357-369).

Log retention¶

_daily_log_retention runs on the default database and on every tenant, suspended tenants included (src/cluster_erp/server/app.py:1267-1284). On each database it:

1. deletes expired rows of the token blacklist (src/cluster_erp/server/app.py:1168-1193);
2. deletes activity_log rows older than activity_log_retention_days and error_log rows older than error_log_retention_days, in batches of log_purge_batch_rows (src/cluster_erp/server/app.py:1196-1245);
3. writes an evidence row of kind logs, even when nothing was removed (src/cluster_erp/server/app.py:1246-1263).

Control-plane retention¶

sweep_control_plane runs at start-up and daily on a multi-tenant node, and on demand with python -m cluster_erp.server.membership retention (src/cluster_erp/server/membership.py:137-142). Every pass writes one row to control_retention_runs (src/cluster_erp/server/control_retention.py:170-171).

Active or suspended customers, their memberships, and provisioned tenants are not touched by this pass (src/cluster_erp/server/control_retention.py:30-31).

Discrepancies¶

Terms introduced¶

Related pages¶

• Operations
• Backups and restore
• Records module
• Multi-tenancy

Verified against commit 92f8051 on 2026-11-13.