Three fixes, two of which could stop backups without reporting anything wrong.
Long server-side jobs no longer stop the scheduler (#424)
A large offsite sync or catalog rebuild ran inside the lock the scheduler's cron entry holds. For as long as it ran, every following minute's scheduler run was skipped — no backups were scheduled, no agents were marked offline, and no notifications, reports or post-backup scripts ran. The dashboard reported the scheduler as stopped at the same time, because the heartbeat is written at the start of a run.
Server-side jobs now run in their own process. A scheduler pass takes seconds, and jobs for different repositories can run at the same time instead of queueing behind each other.
The cron entry is unchanged, so there is nothing to reinstall.
Backups that store no files are now flagged (#429)
borg reports a warning when a source path is missing, and BBS passes that on. An empty directory is valid to borg, so a mount that fails to come up leaves the backup with nothing to archive and no warning to report. The job finished with zero files and showed as successful.
A backup that stores no files now completes with warnings and sends a notification saying what to check.
Folder picker showed stale directory listings (#429)
The cache of a client's directory listing was extended each time it was read, so it never expired while the picker was in use. The Refresh button returned the cached listing instead of asking the client.
The cache now expires 15 minutes after the listing was fetched, and Refresh asks the client for a new listing.