systemctl is-active returns "activating" (exit 3) for Type=oneshot
services while ExecStart is running. Match both "active" and
"activating" states so the deploy script correctly stops the index
service. Also stop the cluster service to break the index↔cluster
cycle cleanly during deploys.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace OnSuccess with ExecStartPost so cluster only starts after
index genuinely completes, not when stopped by a deploy. Cluster's
ExecStartPost restarts index, creating a continuous cycle that breaks
cleanly when either service is stopped externally.
Add Conflicts= on both units so they never run simultaneously.
Change index to Type=oneshot and remove Restart=always.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Show how many galleries have been checked, how many were skipped vs
indexed, and the total so overall progress is visible at a glance.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Emit an info log with the running skip count every 100 unchanged
galleries so long runs show progress rather than silence.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Configure sqlx pool with min_connections = max_connections so all
connections are established at startup, avoiding slow-acquire warnings
from lazy mTLS handshakes. Add idle_timeout (5 min) to recycle stale
connections from prior runs, and reduce acquire_timeout to 10s for
faster failure.
Size the pool to io_concurrency + ml_concurrency + 2 to accommodate
the worst case where all IO tasks call image_exists concurrently.
Reduce default io_concurrency from 4× to 2× ML concurrency to keep
pool size within PostgreSQL's default max_connections.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Skip paths in process_image no longer call upsert_gallery_image
individually. Instead they return the GalleryImage and the gallery
loop batch-upserts all skipped rows in a single query. This eliminates
~100 concurrent DB calls per gallery during the common case where most
images are unchanged.
Pool size reduced from io_concurrency + ml_concurrency + 4 to
ml_concurrency + 6, since only ML transactions and the occasional
image_exists fallback need connections during task execution.
Reduce service concurrency from 32 to 24 for index and cluster.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Move IO semaphore acquisition to the task spawn site so at most
io_concurrency tasks are active at any time. Previously all ~100
images per gallery were spawned eagerly and raced for DB connections
on their skip-path upserts outside any semaphore, causing pool
timeouts when tasks exceeded the connection limit.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Drop FilterConfig and the glob-based include/exclude filtering that was
never used in practice and complicated gallery stats computation.
Remove 17 unused crate dependencies flagged by cargo-machete across
rbv-data, rbv-cluster, rbv-ingest, rbv-auth, rbv-search, rbv-cli, and
rbv-api. Remove glob, hyper-util, tokio-rustls, and tower from the
workspace root as no crate references them.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Gallery-level skip: store file_count/total_bytes per gallery, skip
entirely when unchanged. Batch existence check: one query per gallery
instead of per image. File-size fast path: stat-only skip when filename
and size match stored values. Two-stage pipeline: separate IO and ML
concurrency (--io-concurrency flag).
Adds `rbv backfill` to populate file_size and gallery stats from disk
without ML, so skip optimisations are effective immediately after upgrade.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
QueryBuilder::push_values() already prepends VALUES, but the initial
query string also included it, producing invalid SQL: FROM (VALUES VALUES ...).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Tags and subjects were mixed in SQL queries (tags || subjects,
ANY(tags) OR ANY(subjects)) and UI rendering. Now each has its own
data queries, API endpoints, browse pages (/tags, /subjects), detail
pages (/tag/:tag, /subject/:subject), and add buttons in the gallery
footer. The gallery footer shows two labelled rows for subjects and
tags with independent autocomplete and add controls.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add usePageParam hook that reads/writes ?page=N via useSearchParams,
making all paginated pages bookmarkable and shareable. Applied to
People, Person (gp/fp params), MergePerson, Tag, and Tags pages.
Page 1 omits the param for clean default URLs.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New /tags route shows a grid of tag cards, each with a randomly
selected thumbnail from a gallery that has that tag. Uses a single
LATERAL join query paginated to 24 tags at a time to avoid timeout
on large tag sets. Refresh re-rolls the random images. Tags link
added to nav header.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds --cdn-map CLI arg (e.g. --cdn-map /tank/data/rbv/vault=/vault)
to configure filesystem-to-URL prefix mappings. A GET /api/config
endpoint exposes these mappings to the frontend. Gallery viewer,
home page, and tag page resolve image/thumbnail URLs through nginx
when a matching CDN map exists, falling back to the API otherwise.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Tags and subjects in the gallery view now link to /tag/:tag, which
shows a paginated grid of all galleries with that tag. A + button
lets users add new or existing tags to a gallery, with autocomplete
from all known tags. Backend adds by-tag query, tag listing, and
add-tag endpoints.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds a favourites table, API endpoints (list/add/remove), and UI
integration: heart icons on gallery cards (home page) and in the
gallery viewer header, with a favourites section on the home page.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Remove MlClient HTTP client and reqwest dependency; all inference now runs
locally via rbv-infer (CLIP visual+textual, SCRFD, ArcFace)
- Make --model-dir required, remove --ml-uri from CLI and API server
- Add denormalized image_count column to persons table with trigger to keep
it updated; order /people route by image count descending
- Add GET /galleries/:id/persons endpoint and display person links on gallery page
- Fix face crop endpoint to resize source image to match detection coordinates
- Add --ml-purge flag to index command for wiping all ML data before re-index
- Add --face-score-thresh CLI arg (default 0.7, was hardcoded 0.5) to tune
face detection sensitivity
- Lower default cluster threshold from 0.65 to 0.55 for more aggressive grouping
- Include gallery ID in pipeline log output and image path in error messages
- Update docs and systemd service files
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Generate all PersonIds upfront, then bulk-insert all persons with a
chunked QueryBuilder INSERT and bulk-update all face assignments with a
chunked QueryBuilder UPDATE FROM VALUES. Reduces the 40-minute write
phase to seconds. Also fixes NaiveDateTime/TIMESTAMPTZ decode in person.rs.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Manual install from GitHub release zip using exact paths confirmed from
zip contents (sharedir/extension/* → /usr/pgsql-17/share/extension/,
pkglibdir/vchord.so → /usr/pgsql-17/lib/). Adds shared_preload_libraries
and CREATE EXTENSION for both rbv and immich databases.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Leaves certs root-owned; ACL grants postgres read access so renewals
take effect without re-copying. Use absolute paths in postgresql.conf.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace scram-sha-256 with mTLS cert auth using the step CA convention
already in use on all infra hosts. Documents cert installation, pg_hba
hostssl/cert rules, pg_ident CN mapping, passwordless role creation,
and cert-auth connection strings for rbv and immich.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Documents Phase 1 (PG17 + pgvector standalone on frankie.hanzalova.internal)
and outlines Phase 2 (Patroni three-node cluster) for future reference.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Mirrors the index command convention: defaults to 4, tunable per host in
the service unit. The inner similarity loop is parallelised via a rayon
thread pool of the configured size; union-find remains sequential.
rayon added to workspace and rbv-cluster dependencies.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Emits pct/unions/eta_secs at INFO so large runs show meaningful progress
rather than silence. Also makes union() return bool so merges are counted.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Each time rbv-index@.service completes successfully, systemd fires
rbv-cluster.service (a oneshot) to assign unprocessed faces to persons.
Keeping them as separate units gives each its own status and journal,
and allows cluster to be triggered independently. deploy_index now
deploys and password-substitutes both unit files.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Smoothly scroll the thumb strip to center the active thumbnail when
navigating between images. Remove unused variables that broke the build.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Zoom now recenters around the mouse pointer position instead of the
image center. When zoomed in, click-drag pans the image. Cursor
changes to grab/grabbing when zoomed. Reset clears both zoom and pan.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The wheel event listener was attached on mount when the thumb strip
hadn't rendered yet (early return before gallery data). Re-run the
effect when gallery loads so the ref is available.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Split deploy.sh into deploy_db, deploy_index, deploy_api, deploy_ui
functions. Components are selected via script args (e.g. "db index"),
"all" deploys everything, and no args defaults to index+api+ui.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Scroll to zoom in/out (0.5x–5x) on the main gallery image, with a
reset button that appears when zoomed. Zoom resets automatically on
image navigation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
A sorted root-level index always replays already-processed galleries
before reaching new ones, delaying their availability in the UI.
Shuffling (the new default) distributes the probability evenly so new
galleries surface early in any run. Pass --sort-galleries to restore
deterministic ordering.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
With --concurrency 24 the hardcoded pool of 10 meant 14 workers always
queued for a connection, causing >2s acquire warnings. Pool size is now
derived from concurrency (concurrency + 4 spare) for the index command
so it always matches the workload. Other commands use fixed small values
appropriate to their access patterns.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
'Processing gallery: ...' was a generic pre-action statement repeated
thousands of times at INFO. Replace with a post-completion summary line
carrying actual data: images processed/skipped, faces detected, errors.
Falls back to the filesystem path when source_name is empty.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
id defaults to 0, collection/source/source.name/source.url default to ""
so a gallery with an incomplete or partially valid index.json still has its
images ingested rather than being skipped entirely.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
