The exclusion assistant suggests the folders well-known apps fill again by themselves, ZFS backups keep each dataset's properties and can restore into a new dataset, a single plugin comes back from a flash backup, and archives of the Appdata.Backup plugin can be imported as restore points. Scheduled backups can wait until an app is idle, and off-site copies slow down while a media server streams. Anomaly findings from history read back after an upgrade point at the right backup, and a finding can compare its backup with the last good one, folder by folder. Each source can keep its local and off-site backups by rules of its own. Settings, schedules and repositories carry over as they are, and the new switches start off.
✨ Added
Sources and imports
- Recommended excludes per app. For Plex, Jellyfin, Emby, Sonarr, Radarr, Lidarr, Readarr, Prowlarr, Immich, Nextcloud, PhotoPrism and Tautulli, from linuxserver, hotio, binhex or the official publisher, the exclusion assistant offers the folders the app fills again by itself: caches, logs, preview images and posters. Each entry says what it holds. Downloaded artwork starts switched off, since images you uploaded yourself sit there too, and nothing is excluded until you press Exclude selected.
- ZFS properties come back. Every ZFS backup keeps each dataset's locally set properties, such as compression, record size, quota and case sensitivity. Into a new dataset creates the dataset with them and restores into it. A restore into an existing dataset lists them and sets them only with Also set these properties. Either way, quotas and reservations are set after the files, so they cannot refuse them. The properties live in BombVault's own database, so after a rebuild they come back with the self-backup.
- The checks before a restore cover a new dataset too. They create nothing: the plan counts the files as they will land in the new, empty dataset, and the space check looks at where it will be mounted, or stays grey while that cannot be told yet. A dataset above it that does not exist turns the check red and locks Start, with its name.
- One plugin from a flash backup. Plugins on a flash backup lists the plugins in it with their version and size, and puts a single one back into the running flash: its
.plgfile, its folder underconfig/pluginsand the package files the backup holds. Nothing else on the flash changes. Unraid installs the plugin on the next boot, or right away under Plugins, Install Plugin. - Import from the Appdata.Backup plugin. On the Recovery page, point BombVault at the plugin's backup folder. Each container archive becomes a restore point of its container, dated when the plugin made it, and shows as an import in the activity log and the run lists. Retention leaves these restore points alone, so delete one yourself when you no longer need it. The archives are only read, ones imported before are skipped, and the container needs one backup in BombVault first, so the restore has its definition. Links inside an archive are never followed, so nothing in it can write outside BombVault's work folder.
Backups that make way
- Streaming first. While Plex, Jellyfin, Emby or another media server you pick streams, off-site copies upload at a lower limit (512 KiB/s unless you change it) and return to the normal one a few minutes after the stream ends. REST, S3, B2, Azure, Google Cloud, Swift and rclone over HTTP slow down in the middle of a copy; SFTP and local or mounted folders take the lower limit from their next copy step. A media server on the host network cannot be measured. Under Settings, Off-site in the advanced view, off by default.
- Wait until the app is idle. A container (advanced view, and in its detail on a phone) can let its scheduled backup wait while its app is busy, for at most the hours you set, and start as soon as the app is idle. A media server is idle when it isn't streaming, any other container when its CPU and traffic stay below the limits in Idle before backup under Settings, Schedules for a few minutes. The wait shows in the activity log and on the container with its reason and deadline. It holds no lock, so the other containers go ahead.
- Stacks wait together, and a wait survives a restart. The members of a compose stack due in the same run wait as one and back up together with their project folder. A wait carries on with its deadline after BombVault restarts. Switching the containers off drops every waiting backup, switching their schedule off drops the ones its runs held back, and lowering the hours shortens a wait that has already begun. Manual backups never wait.
Retention
- Keep rules per source. Under Settings, Retention, the Local retention and Off-site retention cards each have the shared keep rules and a switch per source (containers, VMs, flash, folders, ZFS, self-backup) that gives the source rules of its own, say 7 daily backups of containers that change every day and fewer of VMs that rarely change, or a longer off-site archive for one source. A source's off-site rules apply to its off-site repository, and additional off-site targets keep the rules set for them on the Off-site page. Retention after a backup, the off-site copy, Prune and the retention preview all follow the rules of the source, and the settings export carries them. When several sources replicate into one off-site repository, each one ages only its own backups there. In the advanced view, the Preview card shows what the next run would remove from the local and the off-site copy of a source. A source you leave alone keeps the shared rules, so nothing changes until you switch one on. Asked for in #293.
Anomalies
- Compare with the backup before. Findings about how much a container, VM or folder set holds or stores have a Compare with the backup before button. It compares the last good backup with the first affected one and lists the folders where files were removed, added or changed. If nearly all of the change is in a search index, a cache or thumbnails, it says so. Apps rebuild those folders on their own, and excluding such a folder keeps the rebuilds out of your backups and out of the findings.
🎨 Design
- One eye for show and hide. Show phrase on the pairing card, Show what would be removed in the retention preview and the reveal toggle in every password field use KnightLoader's eye, and its struck-through version for hiding.
- An archive box for Retention. The Retention tile in Settings shows an archive box. The shredder it wore stays with Prune.
- New README screenshots. Each one shows a page of the current interface in a browser window next to a short caption. They come from a demo instance running invented services.
⚡ Improved
- The MCP server knows the waiting backups.
get_statuslists the scheduled backups waiting for an idle app, with the stack, the container that is busy, the reason and the time they start at the latest. - The settings export carries Streaming first and Idle before backup. The hours a container waits are set on the container and stay out of the file, like its other settings.
- Questions have a place of their own. The GitHub repository has Discussions with a Q&A category, and the README and the issue form point there. A plain issue still works.
🐛 Fixed
- A container on its own schedule gets its compose project folder backed up. Only a run of the whole domain backed that folder up, so a stack member with a schedule of its own never had it saved.
- Findings from history read back after an upgrade are dated to the backup they began on. BombVault reads that history in one go. A source that had already shrunk several backups earlier was dated to the newest backup, and the last good backup it offered was the one just before, which could be affected too. The finding now starts at the first affected backup and offers the one before it for a restore. Reported in #281.
- A finding names the backup it began on. An opened finding shows its first affected backup. First seen, when BombVault noticed the problem, moves to the details, because after an upgrade that can be days after the backup.
- A restored container keeps its limits, log driver, DNS and GPU. A restore, and the update after a backup, recreated a container without its memory, CPU, PID and disk I/O limits, shared memory size, OOM settings, log driver, DNS servers, links and its GPU or runtime (
--gpus,--runtime=nvidia). A container started with--memory 16m --cpus 0.05 --log-driver nonecame back unlimited and logging to json-file. On a host without that GPU driver or runtime, every restore says so and offers Restore without GPU and runtime for the containers concerned, and the run history shows which of the two happened. A link to a container that is missing, or stopped when the restored one starts, is left out and noted instead of failing the restore, and a stack restore recreates dependencies first. Backups made with this version carry these settings; an older backup restores with Docker's defaults, as before. - Measuring a large off-site copy no longer puts errors in the log. After each off-site copy, BombVault measures the size of the copy in the background, and it allowed five minutes for that. On a large remote repository that was too short. The measurement was stopped, and the log showed restic's own lines from the stopped run in red plus a warning, even though every backup was fine. Off-site and remote repositories now get an hour. If a measurement still runs out of time, the log gets one plain line saying the next backup or off-site copy measures again. Reported in #295.
- The activity log keeps earlier runs while the same job runs again. While an off-site copy of a domain was running, the log hid every earlier off-site copy of that domain, and while a container was being backed up, every earlier backup of that container. The log now hides only the one finished run that the running line already shows. Reported in #294.
🌐 Translations
- Every new card, hint and message: 147 new strings in all 42 languages, and the flash restore note that now says a single plugin goes into the running flash.
- The new parts of the features, configuration, MCP server and ZFS datasets pages in all 26 documentation languages.
- The documentation in every language calls pages, buttons and cards by the names the app shows, and the Receiver switch is found under Settings, General. The Anomalies section describes where an item's settings open: under Monitoring on its card, or from the Nothing open card.
Upgrading
- Nothing has to be changed. The database is upgraded on the first start, settings export files from older versions still import, and no setting or environment variable was removed or renamed. restic stays at 0.17.3 and the repositories are left as they are.
- New switches start off: Streaming first, waiting for an idle app and keep rules per source. Every source keeps the shared keep rules until you give it its own. No recommended exclude is applied until you pick it.
- ZFS properties are kept from the first ZFS backup after the update. Older restore points have none, so a new dataset made from one of them starts with the pool's defaults.
- Restoring into a new dataset as a user other than root needs
createon the dataset above it, and permission for each property BombVault sets.