Template was missing theme.mkv/theme.mp4 and all trailer patterns,
including trailer-*.* for Trailarr's hyphenated TV trailer naming
(trailer-2.mkv) — a fresh install would hit the same false-positive
orphan bug just fixed live on prod.
Radarr 6+ supports multiple tracked files per movie (alternate editions/extras),
but the movie list's embedded movieFile.path only ever reflects the primary one.
Relying on it alone flagged legitimately-tracked secondary edition files as
orphans. Now batch-fetches /moviefile?movieId=X across all tracked movies to
catch every file, not just the primary.
Full-library scans are slow to reach any one show by chance — this lets a test
target a specific known-corrupt file/show without waiting on scan order.
Healarr does the same job but crashes on a Go concurrency bug (unsynchronized
map access) whenever multiple corruption events land close together — confirmed
via its own crash log, not fixable from our side. Processing one file at a time
here sidesteps the whole bug class instead of trying to work around it.
Delete + explicit EpisodeSearch rather than relying on Sonarr's own background
missing-search cycle, since that skips unmonitored episodes and this shouldn't.
Mirrors the Sonarr fix — same MoveMovieService one-at-a-time drain architecture,
never confirmed live on the Radarr side but the DB-instant/move-async split
is identical, so the same batch-verification race applies.
Confirmed live: DB fields flip instantly on the moveFiles=true PUT but the
physical move drains async, one command at a time — a batch could report
every series moved while most were still sitting at the old path.
arr_cache_prefill moved from critical (30min) to intermediate (4h) —
nothing in the critical tier reads this cache, and every actual
consumer already tolerates staleness well past 4h. Classification
scans run in daily, after the cleanup scripts (not before) — cleanup
frees disk space from orphans/junk first, so the searches --move
triggers for undownloaded relocated entries have headroom for the
resulting grabs.
Acts on forward misplacements and reverse-kids-leak (adult content in
the kids root — not a judgment call, same reasoning as the Radarr
side), moving to the new SONARR_GENERAL_ROOT config var. Reverse-anime
-leak stays report-only — deliberate style placements like Castlevania
are common and valid there. episodeFileCount is Sonarr's hasFile
equivalent: series with 0 files get relocated + an immediate
SeriesSearch instead of a file move. One series at a time, verified
after each — matches the race condition found doing this by hand for
the Fate/Zero and Fate/Stay Night moves earlier in the session.
Unlike reverse-anime-leak (deliberate style placements like Castlevania
are common and valid there), reverse-kids-leak's signal is specifically
"adult certification with zero Family/Animation genre" — confirmed live
the 4 titles it catches (Addams Family, Saving Mr. Banks, Dark Shadows,
The DUFF) are all genuinely non-kids content. Moves them to the new
RADARR_GENERAL_ROOT config var (container-path literal for the general
Movies root, needed since RADARR_MOVIES_ROOT is a filesystem path).
Junk entries have no release to blocklist and no file to delete —
only a bad monitored record with a thin/wrong TMDb match. Removes the
record and adds it to Radarr's import exclusion list, same mechanism
radarr_tmdb_removed.sh already uses, so the same bad match can't get
re-added by a future Overseerr request or list sync.
Instead of skipping monitored-but-undownloaded titles entirely,
correct their DB pointer (rootFolderPath/path) now so Radarr saves
under the right root whenever it grabs a release, and fire an
immediate MoviesSearch rather than waiting for the next scheduled
one. Junk entries stay excluded — nothing to search for there.
Acts only on forward misplacements (clear-cut anime/kids classified
content sitting outside its dedicated root) — reverse leaks and junk
entries stay report-only, since those need human judgment or a
different action (removal, not a move). One movie at a time with a
real re-fetch-and-verify after each, matching the race condition found
doing this by hand for Sonarr earlier in the session.
Overseerr lets users request content into the wrong root folder; these
new report-only scans classify every tracked movie/series (anime, kids,
regular) from metadata alone and flag mismatches against the actual
root folder, in both directions. Rules were validated against real
library data before being adopted — see the header comments in each
script and the master.conf notes above the curated lists.
Same fix as the arr cleanup scripts today: find -printf gets file size
directly from find's own stat() during the walk (mtime already handled
by -mmin), and the dry-run log line's basename call is replaced with
parameter expansion. Header notes this can matter under real load --
"thousands of HLS segment files" per the existing lsof design principle
this mirrors, even though the ramdisk is empty right now (no active
transcode session to benchmark against directly).
This turned out to be the actual dominant cost left in the cleanup
scripts' classification loop, hidden behind the stat/dirname forks fixed
earlier today -- called once per non-tracked file (most files in a
media library, since every protected sidecar counts). Measured: 47.3s
vs 1.65s for 10,000 calls (~28.6x), identical results confirmed via
spot check before switching. Only used by lidarr_cleanup.sh/
sonarr_cleanup.sh/radarr_cleanup.sh.
lidarr_cleanup.sh/sonarr_cleanup.sh/radarr_cleanup.sh each forked a
separate stat call per file during classification. find already has to
stat() every entry to know it's -type f, so -printf '%s %T@ %p' gets
size+mtime for free during the walk itself. Measured: 5.77s for all
175,954 files in the Lidarr music root (walk + stat combined) vs 85.98s
for stat alone on a 20K-file subset of the same library -- roughly 130x
faster per file, and collapses two passes into one. Verified path
parsing preserves spaces/parens/unicode exactly via read's trailing-
field capture before switching.
lidarr_missing_art.sh's album-directory-map loop (both the cache-hit and
live-fallback branches) and arr_profile_enforcer.sh's _is_kids_path()
called dirname/basename once per item -- 127K+ tracks and ~4000
series/movies respectively, each call forking a subprocess. Measured:
0.39s vs 72.2s for 20K calls, ~185x. Verified identical output against
real paths (including unicode/space/paren edge cases) before switching.
arr_cache_prefill.sh had no RUNTIME MODES section at all; arr_full_rescan.sh
had one but didn't mention --log despite supporting it via parse_args.
Comment-only.
arr_full_rescan.sh, arr_cache_prefill.sh, and arr_rescan_monitor.sh were
missing the standard SAFEGUARDS header section other Arrs_Stack/Tools
scripts have. Also: arr_cache_prefill.sh now checks for an active rescan
before fetching, instead of doing a live fetch that arr_cache_write()
would just refuse to persist anyway -- avoids wasted API calls every
30min during a long rescan. arr_rescan_monitor.sh was also missing an
actual root check despite writing cache files; added it to match
convention rather than just document a safeguard that wasn't there.
Comment-only. Headers on the scripts touched during today's caching work
(cache-first fetches, write-through per-item cache, single-walk
consolidation, movieFile-embedded fix) still described pre-change
behavior. Also brought common.sh's top-level cache doc block current --
it was written for the single-consumer 2026-07-16 state and didn't
mention the tmpfs move, the write guard, or the 15+ consumers that now
go through it.
lidarr_cleanup.sh and sonarr_cleanup.sh already walk every artist/series
individually (trackFile/episodefile) for their own cleanup decisions --
that walk now also writes the raw per-item data through to a short-lived
tmpfs cache (arr_item_cache_write, 4h freshness, no persistent backup
since every consumer already has its own live fallback). lidarr_missing_art.sh
runs later in the same nightly window and now reads that cache first,
skipping its own redundant per-artist walk entirely on a hit. Sonarr side
is write-through only for now -- no second consumer exists yet, but the
data's there for whenever one does. Future consumers: arr_get_cached_items()
first, live per-item fetch as fallback, same pattern as these two.
lidarr_cleanup.sh/sonarr_cleanup.sh/radarr_cleanup.sh each walked their
full media root twice per run: once to classify files and total the
deletion size for the safety-threshold check, then again to actually
delete. The size check needs to know the total before deleting anything,
not before knowing what to delete -- the classification pass now records
orphan/junk paths as it finds them, and the deletion pass just reads that
list instead of re-walking and re-classifying the whole tree again. Only
affects real (non-dry-run) executions, where the second walk used to
happen. Also merges two separate stat calls per file into one.
Radarr's movie list already embeds movieFile.path on every hasFile=true
entry -- confirmed live, zero exceptions across the full library. The
separate moviefile?movieId=X call per movie (2896 of them) was fetching
data already sitting in the list this script fetches anyway. One live
list fetch replaces up to 2896 per-movie calls, every time this function
runs including rescan-aware retries.
Reads/writes now hit tmpfs (ARR_CACHE_DIR) instead of the array disk --
a full rebuild for all three arrs measures ~12s live, so there's no real
cost to losing it on reboot. The existing on-disk file becomes a backup
that arr_cache_write() keeps in sync on every write, and
arr_cache_age_seconds() transparently restores it into tmpfs the moment
it notices tmpfs is missing -- so a cache that was fresh before reboot
reads as fresh after too, closing the cold-start gap without needing a
dedicated restore step anywhere else.
Every script that fetches the full Lidarr/Sonarr/Radarr tracked-library
list now goes through arr_get_tracked_data() instead of hitting the API
directly -- cache-first when fresh, live fetch as fallback when stale,
waits out an active rescan before either. Per-item file data (trackFile/
episodefile/moviefile) stays live-only everywhere, since that's the
actual disk-truth these scripts' decisions depend on and was never part
of what's cached.
Also adds arr_cache_prefill.sh to CRITICAL_MAINTENANCE_SCRIPTS (30min
tier) with a short 1min wait ceiling, so the cache stays consistently
fresh instead of only refreshing whenever some other script happens to
write through. A full cache refresh for all three arrs measured at ~12s
total live -- nothing like the multi-hour cost of an actual rescan.
A direct arr_cache_write() call mid-rescan wrote a partial snapshot that
looked like real data loss to every consumer of the cache. The guard now
lives in arr_cache_write() itself so every caller is protected, not just
arr_get_tracked_data(). arr_rescan_monitor.sh closes the resulting gap for
rescans triggered outside arr_full_rescan.sh's own trigger-and-wait path.
Organic scans only touch files actually involved in an import — an
artist/series/movie that already has files sitting untouched on disk
never gets its tracked-file stats refreshed on its own. Confirmed
2026-07-16: Lidarr reported ~23% of its true trackFileCount with no
scan running, for artists whose files were verified present and
readable the whole time. Downstream scripts trust these stats as
source of truth for the share, so drift needs to be caught before
someone notices a suspiciously low number.
Shared cache/rescan-duration logic in common.sh now takes an arr_type
param instead of being Lidarr-specific, so Sonarr and Radarr cleanup
scripts get the same cache-first fetch + rescan-aware retry Lidarr had.
Avoids redundant full-library API calls across scripts run back to back,
and stops false failures when a fetch lands mid-rescan.
Daily maintenance jobs alone now take ~4h with rsync disabled. 19h cap
leaves enough room in the 24h cycle for maintenance to still run
same-day before the next 1am fire, even if a share hits the cap.
Shared cache (lidarr_get_tracked_data() in common.sh) so scripts stop
hitting Lidarr's live API for tracked counts every run, and stop
treating a mid-rescan dip as a genuine problem — a whole-library
RescanFolders legitimately makes trackFileCount read far below normal
while it re-verifies every file (confirmed 2026-07-16: 22% of normal
mid-scan). Cache reads fresh-if-recent, waits out an active rescan
(calibrated to that command's own historical duration, tracked per
command name since RescanFolders and DownloadedAlbumsScan take wildly
different amounts of time), then falls back to a stale cache rather
than hard-failing after a few strikes.
lidarr_cleanup.sh: no longer stacks a fresh DownloadedAlbumsScan on
top of one already running, and the tracked-count floor check now
waits out a genuine rescan instead of aborting on every overlap.
lidarr_duplicate_artist_cleanup.sh (new): finds case-insensitive
duplicate artist entries — same display name, different MusicBrainz
ID, added when a search/list-sync matches the wrong same-named artist.
Deletes the empty phantom side and blocks it from Import List
Exclusions, leaves genuinely-different-real-artists alone (checked by
album title overlap, deduped per-artist first so a legitimate reissue
under an artist's own catalog doesn't false-flag as cross-artist
overlap), and only notifies for the rare case where both sides have
real, overlapping content.
lidarr_cache_prefill.sh (new): warms the cache at array start so nothing
reads it cold after boot.
lidarr_missing_art.sh, lidarr_release_fixer.sh: write-through the cache
as a side effect of fetches they already needed for their own purposes.
Most importBlocked downloads are junk/duplicates and still fall straight
through to blocklist+research unchanged. But some are releases arr already
correctly parsed (episode/movie, quality, language all known) that just
trip the title-vs-grab-history safety net — those get imported directly
if the target has no file yet, or the candidate is a same-language
resolution upgrade over what's already there. Sonarr/Radarr only; Lidarr's
manual-import matching doesn't reliably resolve album/track context.
Gated by ARR_SMART_IMPORT_ENABLED (default true) and
ARR_SMART_IMPORT_PREFERRED_LANGUAGE (default English).
page=1&pageSize=200 silently truncated anything past record 200. Sonarr's
queue currently runs 1700+ during a large search campaign, which pushed
every importBlocked/warning item past page 1 — the script logged 'clean'
every run while 52 stuck imports sat completely unseen despite yesterday's
importBlocked fix matching them correctly once actually queried.
The pull step (merge mode) had no timeout or bandwidth limit at all,
so a slow/stalled pull could block the entire per-share sync
indefinitely — and since daily_sync_maintenance.sh calls rsync.sh
synchronously per share, that blocked every later share and all
post-sync maintenance jobs too. RSYNC_MAX_RUNTIME_HOURS was only ever
wired into the push half.
Recovery script only matched importFailed/importPending/error/stalled, so
releases matched to the wrong media by grab-history ID (Sonarr/Radarr's
importBlocked state) sat forever, invisible to automated cleanup.
Both had a basic HTTP check (Emby) or no coverage at all (Jellyfin), but
neither would have caught today's real incident: Jellyfin's SQLite
database locked up hard (repeated 'database table is locked' errors,
30s+ query timeouts) while its own /System/Info endpoint kept responding
200 the whole time — a basic HTTP check on that endpoint would never have
tripped. /Users forces an actual DB round-trip and was confirmed live to
hang during the exact incident.
Generalized the API check's success condition to also accept array-shaped
responses (/Users returns an array; the existing check only recognized
object fields like .ServerName/.Id/.Version, which would error when
applied to an array) — benefits any future array-returning endpoint, not
just this one. Also corrected the host.conf.template's API_CHECKS format
comment, which described a 3-field format the code never actually used.
blink-182 kept getting rescored and re-added every week despite already
being in the library (id=155). Root cause: MusicBrainz's canonical name
is 'blink‐182' using a Unicode hyphen (U+2010), while Last.fm's candidate
list returns the plain ASCII hyphen — the exact-string _in_lidarr() /
_in_emby_library() checks never matched, so it was treated as a new
artist every run, scored, accepted, and its add attempt correctly failed
against Lidarr's duplicate-MBID rejection.
Added _normalize_dashes() to collapse Unicode hyphen/dash variants (U+2010
through U+2014) to ASCII '-' before comparing, applied to both the stored
library name lists and each candidate name at match time. Fixes this for
any artist with a stylized dash in their canonical name, not just this one.
No 'already running' guard existed — a relaunch (array stop/start that
doesn't kill the old node process) would exec straight into node, hit
EADDRINUSE on the port, and exit 1 within ~1s. array_started.sh correctly
reported that as a failure, even though the prior listener instance was
still healthy and serving webhooks the whole time. Confirmed live: the
node process from 2026-06-23 (PID 25977) is still running today, and the
2026-07-03 array start logged this exact false failure.
Added acquire_lock "continuous" before the exec — an existing, documented
common.sh mode (skip gracefully if a healthy instance is running) that
wasn't actually used anywhere in the codebase yet.
Some items (e.g. an album missing 1-2 tracks where no available release
matches the existing edition/track count) can never resolve via blind
retry. Without a limit, the same media ID gets blocklisted and re-searched
every 4 hours forever — confirmed live on ~19 Lidarr albums cycling
identically across five consecutive runs today, each one downloading a
fresh release, failing import for the same structural reason, and
starting over.
Tracks consecutive failures per (arr_type, media_id) in a persisted state
file. After ARR_RECOVERY_MAX_ATTEMPTS (default 3) failures, the item is
still blocklisted and removed from the queue, but auto re-search stops —
notified once when it crosses the threshold, then left for manual review
instead of retried forever.
The memory warning used plain system-wide 'free' RAM (via free -g), a
carryover from before this script was ZFS-specific. ZFS ARC deliberately
consumes most otherwise-unused RAM, so 'free' being low is normal and not
a meaningful signal — it fired a false alarm on 2026-07-05 (free 5.8Gi,
but available a healthy 45Gi).
Replaced ZFS_REPORT_FREE_WARN_GB with ZFS_REPORT_ARC_FREE_WARN_GB: warns
when ARC headroom (ARC_MAX - ARC_CURRENT) drops below threshold, which is
what actually indicates ARC is running out of room to grow. Available RAM
check is unchanged — it's a legitimate system-pressure signal on its own.
Gmer4Lfe