Fixes
-
Auto-retry no longer burns through attempts while the agent is offline (#281) — when an in-flight backup lost its heartbeat, the auto-retry path would re-queue the job, but the queue manager promoted it back to
sentregardless of agent state. The scheduler then immediately failed it as offline and re-queued, exhausting all three retries within ~3 minutes — well before the client could realistically reconnect. Queued agent-bound jobs (backup,restore*,update_borg,update_agent,plugin_test) now stayqueuedwhile the agent is offline and only promote once the heartbeat resumes. -
Light-mode styling on the redesigned dashboard tiles (#288) — the new Server Health and Backup Summary tiles assumed a dark background. Tile backgrounds, gauge tick marks, the memory thermometer outline + scale labels, and the disk-usage bar were all using rgba-white colors that rendered invisible (or close to it) in light mode. Switched to theme-aware Bootstrap tokens and added explicit light-mode color rules for the SVG elements.
-
Recovery Points tile cleanup (#275) — the sublabel "X protected · Y on disk" was a) misleading ("protected" is the pre-dedup original size, not bytes actually safe in backups) and b) wrapping on narrower viewports, making the tile taller than its row siblings. Now shows only "Y on disk".
-
rclone S3 sync needed
--s3-no-check-bucket— rclone's S3 backend issues a pre-flightCreateBucketcall per session as a 'does this exist?' probe. Least-privilege IAM policies that don't grants3:CreateBucket403 on that probe and the whole sync aborts before any object is written. The flag is now passed to every rclone invocation (sync / copyto / lsd) in bothS3SyncServiceandbbs-ssh-helper.