github junkerderprovinz/bombvault v8.2.0

latest releases: v10.6.3, v10.6.2, v10.6.1...
2 months ago

Schedules could silently run in the wrong hour, and this release makes that visible and fixable. It also replaces every "not verified against real hardware" caveat in the TrueNAS support with actual measurements, taken on a live TrueNAS SCALE box.

✨ Added

  • The resolved timezone is now stated in the boot log. One line at startup names the zone schedules are planned in. A TZ value that does not resolve gets its own, louder line: Go falls back to UTC for an unknown zone name, so a typo like Europe/Berln would otherwise look exactly like not setting TZ at all.
  • A guard against the TrueNAS catalog entry going stale. truenas-apps/'s advertised version is now pinned to the newest release note by a test, so the prepared listing cannot drift behind the published image the way it did once before.

⚡ Improved

  • TrueNAS support is now backed by measurements instead of reasoning. The zvol VM-disk backup path was written from ZFS and TrueNAS documentation and tested only against fakes, and every comment said so. It has now been exercised on a live TrueNAS SCALE 25.10 box against a zvol attached to a running VM: the domain XML carries the /dev/zvol/<pool>/<dataset> path verbatim rather than the /dev/zdN node it links to, zfs snapshot works on a disk in active use, and zfs send piped through restic backup --stdin and dumped back out is byte-identical, with the per-disk tag preserved. zfs receive into a fresh target and the cleanup both behave as designed.
  • Two TrueNAS requirements are now documented as requirements. LIBVIRT_URI must name the socket /run/truenas_libvirt/libvirt-sock, because the stock qemu:///system URI fails outright there. And bombvault.self_container_name must be set, because TrueNAS names an app's container after the app instance: with a pinned hostname, BombVault was measured identifying a different container as itself.
  • The README now says how far TrueNAS support is proven. The zvol chain is verified on hardware; a full restore driven end to end by BombVault's own orchestration is not, and the verified zvol was sparse, so multi-gigabyte throughput is untested. Both gaps are stated rather than implied.
  • The daylight-saving behaviour is documented as a decision. A wall-clock cadence skips one run in spring and runs twice in autumn. The reported fix for it is a placebo, measured: the library already defaults to local time, so passing the location explicitly returns identical fire times. Building a minimum-gap guard was considered and deliberately rejected, since it would sit on the path that starts every backup, where a bug suppresses runs that should have happened. A deployment that would rather have exact 24 hour intervals can simply leave TZ unset and run in UTC, which has no transitions.

🐛 Fixed

  • Schedules ran in UTC without anything saying so. With TZ unset, Go's local time is UTC, so a backup entered as 02:30 started at 02:30 UTC, which is 04:30 on a Central European summer clock. Nothing in the UI revealed the difference, and a backup running two hours late looks exactly like one running on time. Found on a real deployment where the container reported "Universal" and TZ was empty. Nothing about the scheduling logic changed: what changed is that the timezone is now stated at boot and documented, in English and in all 26 translations.
  • A Timezone field briefly appeared in the Unraid template, and has been removed again. It was wrong on the very platform it was added for: Unraid passes the system timezone into every container it starts, so the field produced a second TZ variable, and because the template left it blank by default it passed an EMPTY one, which overrode the system zone and put the scheduler back in UTC (#180). If you updated the template while the field existed, it is still in your container's configuration and a template update will not take it away. Open the container in Unraid, delete the Timezone variable, and apply. On Unraid nothing needs setting at all; a container still reporting UTC in the boot log predates the injection and only needs recreating.
  • A container was refused forever after deselecting all of its folders. Backing up a folder once and then clearing the selection left every further backup refused with "backup folders are not reachable", and the only apparent way out was to re-select the folder that had just been deliberately removed (#181). The guard asked whether the container "expects data", which reads the stored target and therefore stays true forever once a backup has captured anything. It could not tell a cleared selection from one that was never made, because the store holds both as an empty list meaning "use automatic detection". The guard now measures what its message actually claims: whether the paths a previous backup captured have disappeared from disk. An unmounted share still trips it, with or without a selection standing, and a first backup is still never refused.
  • Each domain's off-site destination can now use its own credentials. The off-site assistant is opened per domain, but its credentials step wrote the shared set, so filling them in under Containers rewrote every other domain's too (#176). The plumbing was already there: every destination carries a credential-set selector that replication resolves per destination. The assistant simply had no control for it, so the step now offers one and hides the shared fields while a set is chosen. The default stays "shared", so nothing changes unless you pick a set. Because the selector reads the destination, which is created when the repository path is saved, the assistant now re-reads it as soon as that save completes, so the control appears in the step it belongs to rather than after a reload.
  • A scheduler test failed for three hours every night, on correct code. It measured the interval one way while the gate it tests measures it another, and the two legitimately differ between midnight and the 03:00 fire. No user-visible behaviour was wrong, but on CI this would have surfaced as an unreproducible flake. The assertion now uses the gate's own measure, and the anchoring rule underneath it, which had never been tested directly, now has its own test with fixed dates.

🌐 Translations

  • The TZ row in all 26 configuration translations now explains what leaving the field empty does, instead of only naming the variable, and adds that on Unraid you never set it yourself because the system passes its own timezone into every container.

Don't miss a new bombvault release

NewReleases is sending notifications on new releases.