Fix backup check notifications and workload recovery; release 2026.09.21.r001

2026-09-21 20:36:31 +02:00
parent 8f75a04db6
commit 5009dc202b
14 changed files with 295 additions and 38 deletions
@@ -44,3 +44,16 @@ The daemon has no TCP listener. Restic passwords are passed through root-only te
 ## Current scope

 The code implements the MVP control plane and adapters. Docker and VM operations require the standard Unraid `docker` and `virsh` commands. Managed SMB/NFS mounts require the corresponding Unraid mount helpers. Hardware and end-to-end compatibility must be validated on supported Unraid 7 releases before a stable publication.
+
+
+### Backup safety (2026.09.21.r001)
+
+Repository checks report success/failure through configured notification targets. The
+standard check verifies repository structure; it is not a full data-read or restore test.
+Docker jobs discover bind mounts and named volumes. Live backups do not guarantee
+application/database consistency. Use stop mode for coordinated offline backups.
+Stop mode restarts only workloads that were running before preparation; ambiguous
+paused/restarting states fail safely. Each restart has a 120-second deadline and
+recovery continues after individual failures. Shutdown waits for cleanup; the rc
+script refuses a forced kill or overlapping restart if cleanup exceeds five minutes.
+Hard crashes or power loss still require checking workload state and backup results.