Skip to content

MAINTENANCE.md

Latest commit

 

History

History
94 lines (75 loc) · 3.81 KB

MAINTENANCE.md

File metadata and controls

94 lines (75 loc) · 3.81 KB

Maintenance, Schedule, Stats, Troubleshooting

Maintenance

Settings → Maintenance has the manual tools:

  • Prune missing images - delete DB rows for files no longer on disk.
  • Prune orphaned thumbnails - delete thumbnail files whose image row is gone.
  • Rebuild thumbnails - regenerate every thumbnail. Useful after import or after a backup restore.
  • Compute perceptual hashes - backfill phash for every image that doesn't have one yet. Required reading for the Relations find-pairs job; see RELATIONS.md.
  • Rebuild pair queue - wipe potential_relation_pairs and rescan from scratch. Skipped pairs and any queue work in progress are dropped.
  • Recalculate tag counts - recompute usage_count on every tag (zero-usage rows persist; delete them individually on the Tags page when you want them gone). The watcher and bulk paths should keep counts correct in real time, so you only need this if you see discrepancies in usage count.
  • Re-extract metadata - re-runs SD/ComfyUI parameter extraction on every image and re-probes video duration via ffprobe.
  • Vacuum database - VACUUM plus WAL checkpoint to release space.
  • Free memory - shrink SQLite caches, return the Go heap, release the auto-tagger from RAM/VRAM.
  • Duplicates - list and remove duplicate file paths.

Each action reports how many rows it affected.

Schedule

Settings → Schedule turns on a daily run at a chosen time (default 01:00). Tick any of:

  • Sync gallery
  • Remove orphaned thumbnails
  • Run enabled auto-taggers (off by default)
  • Find relation pairs (phash near-duplicates) (off by default)

The scheduler runs through every gallery in turn. If a job is already running at fire time, the run is skipped silently.

Stats

Settings → Stats is a read-only diagnostic block:

  • Process memory - PSS broken into the Go heap, native heap (SQLite caches, CGO), and shared file-backed (DB pages, binary, shared libs) on Linux. Falls back to runtime.MemStats on platforms without /proc/self/smaps_rollup.
  • Auto-tagger - current mode (CPU / CUDA), load status, idle timer, and (when the subprocess worker is alive) its PSS and ONNX/CGO breakdown.
  • Database size per gallery - one row per gallery, sum of monbooru.db plus its -wal / -shm sidecars.
  • Filesystem free space - one row per distinct mount that holds the gallery / data / model directories. Mounts reporting the same totals collapse into a single row.

Troubleshooting

Degraded mode banner. The gallery folder is unreadable. Existing DB rows stay browsable but sync and watcher are off. Fix the path or permissions and restart.

Missing thumbnails. Run Maintenance → Rebuild thumbnails. For videos and animated GIF hover previews, ffmpeg must be installed (it ships with the Docker image).

Watcher hits inotify limit. Logged as no space left on device from inotify_add_watch. Raise fs.inotify.max_user_watches on the host (not inside the container) and restart. If you see too many open files instead, raise fs.inotify.max_user_instances. Or disable the watcher in Settings and use Sync manually.

Permission errors on delete or move. Monbooru takes ownership (chown) of files at ingest time so it can delete or move them later. If files were ingested before this was added, run a Sync - it re-chowns supported files it has just hashed.

Files added but not visible. Check that they're under gallery_path, that the file size is under gallery.max_file_size_mb, and that the extension is one of jpg/jpeg/png/webp/gif/mp4/webm/cbz/zip. Then run Sync.

Login locked out. The login backoff caps at 30s and entries are swept every 10 minutes. Wait it out, or restart the process to clear all sessions.

API returns 503 api_disabled. Generate a token in Settings → Authentication.