Maintenance tasks
Purpose: Reference for routine platform health tasks: what each does, its cadence, and what happens when skipped.
Who this page is for
| Audience | Why it matters to you |
|---|---|
| Platform engineers | Ops routine |
| Backend engineers | Understanding weekly slowness patterns |
The scheduled set (Operations → Maintenance)
| Task | What it does | Cadence | Skipped ⇒ |
|---|---|---|---|
| Revision GC / compaction (TarMK) | Reclaims old segment revisions | Daily (online) | Disk growth → outage |
| Datastore GC | Deletes unreferenced binaries | Weekly | Storage bloat |
| Workflow purge | Removes completed/stale workflow instances | Weekly | /var growth, slow workflow console |
| Audit log purge | Trims audit events | Weekly | Repo growth |
| Version purge | Prunes old page versions per policy | Scheduled/config | Version explosion, slow ops |
| Lucene binaries cleanup | Index housekeeping | Daily windows | Index bloat |
| Project/Task purge | Housekeeping | As used | Clutter |
6.5 runs most of these online in maintenance windows configured in the Operations console — verify they actually run (each shows last-run status) rather than assuming.
Purge policies are business decisions
Version retention ("keep 30 versions / 90 days") and audit retention trade rollback/compliance ability against repo size. Get them agreed with content ops and compliance, then encode them — see Archiving & housekeeping.
Health signals that maintenance is failing
- Disk usage climbing steadily with flat content volume (revision GC not running/completing).
/var/workflow/instancescount in the tens of thousands.- Backup durations growing week over week.
- Compaction logs reporting repeated "cancelled" or forced-stop.
These feed monitoring (monitoring endpoints), not quarterly surprises.