- immich_api: .get("items") or [] handles {"items": null} without crashing len()
- immich_api: catch TypeError alongside ValueError in filter_recent_assets for
timezone-naive fileCreatedAt comparisons
- scheduler: catch SystemExit in addition to KeyboardInterrupt so cli.main()
cannot kill the long-running scheduler process
- scheduler: reseed croniter from wall-clock time after each run so overrunning
jobs don't schedule an immediate back-to-back rerun
- config: reject negative YEARS_FILTER values with a warning, reset to default 10
- frigate_api: return True (not False) for empty filenames list — callers cannot
distinguish no-op from network failure on False
- jobs: warn on unrecognised STRATEGY value instead of silently falling back
- jobs: casefold ONLY_PEOPLE / SKIP_PEOPLE matching so "john doe" matches "John Doe"
- executor: <= → < so a same-score candidate can fill a freed replacement slot
- embeddings: set _insightface_loaded=True on GPU+CPU double-failure to prevent
N re-init attempts (one per asset) when InsightFace is broken for a whole run
Use full_body for the 500 'could not process' permanent-rejection check,
consistent with the 400 'face' check on the line above. error_detail is
truncated to 100 chars via the fallback path, which could silently miss
the phrase in a long response body.
Remove | None from process_face_mode return type — every code path returns
tuple[int,int] or str; None is unreachable. Update docstring to match.
1. if saved: → if isinstance(saved, tuple): so string skip-reasons from
process_face_mode no longer register as successes and create phantom
asset_map entries with no JPEG on disk. Dead reason/fallback code in
the else branch now correctly handles str and None returns.
2. "could not process" permanent-rejection check now uses error_detail
(json message field, falling back to body[:100]) instead of full_body,
keeping the match consistent with what is displayed to the user.
3. CHANGELOG [Unreleased] breaking-change note for MAX_AUTO_IMAGES 20→5
so upgrading users know to set the env var if they want the old cap.
Smaller default cap is more conservative for new installs and better
reflects the minimum viable training set for Frigate face recognition.
Users who need more can set MAX_AUTO_IMAGES explicitly.
process_face_mode now returns a descriptive string instead of None for
filtered-out faces ("face too small 45x38px, min 90px", "no face
metadata"), so the executor can print a useful reason rather than the
generic "no usable face data".
Also suppresses the InsightFace norm_crop FutureWarning about deprecated
estimate usage, which was noisy at INFO level on every aligned crop.
Frigate returns HTTP 500 with 'Could not process' when its face detector
cannot find or embed a face in the uploaded crop — this will never succeed
on retry. Previously these were silently logged at DEBUG and retried on
every future run.
- Surface 500 error details inline (same display path as 400)
- Mark 500 + 'could not process' as a permanent rejection so the asset
is skipped on future runs instead of retried indefinitely
dev is a protected branch requiring PRs. The previous direct push caused
the lockfile update CI job to fail with 'protected branch hook declined'.
When the triggering branch is dev, the workflow now creates a side branch
and opens a PR; all other branches continue to push directly.
insightface depends on onnxruntime (CPU) as a direct dependency. During
uv sync --extra gpu, both onnxruntime (CPU, 24.6 MB binary) and
onnxruntime-gpu (GPU, 24.7 MB binary) are installed in parallel — both
claim onnxruntime/capi/onnxruntime_pybind11_state.so. The last writer wins,
which is non-deterministic in uv's parallel installer.
On GitHub Actions (no GPU, different scheduler ordering), the CPU binary
consistently wins, leaving onnxruntime-gpu's pybind11_state.so as the CPU
version. CUDAExecutionProvider then silently disappears because the CPU
binary's provider registration code has no CUDA EP.
Fix: after uv sync, reinstall onnxruntime-gpu explicitly using the already-
cached wheel. Since uv pip install is synchronous and runs after the parallel
sync completes, the GPU binary is guaranteed to be on disk when the build
layer commits.
cli.py:
- _smaller_duplicate_ids: walrus operator eliminates double p.get("id")
per element; truthiness check replaces dead "is not None" guard (all
persons in by_name are guaranteed to have a truthy id after the
line-75 gate)
- Extract _excl() helper inside _handle_duplicate_people — replaces 4
identical [p for p in lst if p.get("id") not in skip_ids] expressions
across all return paths
jobs.py:
- Extract _valid_people() — shared filter for interactive_configure and
auto_configure; uses (p.get("name") or "").strip() to match cli.py's
whitespace-strip gate, preventing whitespace-only Immich names from
reaching _build_job and creating blank Frigate person labels
- Hoist queued_ids set before the display loop in interactive_configure:
O(N) set lookup per render instead of O(N×|jobs|) linear scan
jobs.py:
- Add p.get("id") guard to valid_people filter in both
interactive_configure and auto_configure — id-less named persons
passed through by _handle_duplicate_people are now excluded before
any bare-subscript access in the configure paths
- Fix bare p["id"] → p.get("id") in the queued-marker check at line 214
(runs unconditionally on all valid_people during menu display, before
any user selection or fetch_all_assets guard)
cli.py:
- Remove dead-code survivor_id and merge_ids guards: after the by_name
fix (line 75 requires p.get("id")), all persons in any ordered list
have ids, so neither guard can ever fire; removing them prevents
misleading readers about what states are reachable
cli.py:
- Filter id-less persons from by_name at construction (root fix for all
bare-subscript crashes downstream — persons with a name but no id are
excluded from duplicate detection entirely)
- Belt-and-suspenders on warning-path display: p['id'] → p.get('id')
- Extract survivor_id with .get(); skip group if survivor has no id
- Guard merge_ids: skip API call when list is empty after id filtering
- Walrus operator in merge_ids comprehension: p.get("id") called once
per item instead of twice
executor.py:
- Add cross-reference comment at success-path reset so the for/else
rollback pairing is explicit for future maintainers
- executor.py: clear min_quality_score_for_slot alongside effective_count
restore in for/else block; leaving the stale floor from the deleted
file's score blocked the next candidate from filling the restored slot
- cli.py: guard merge_ids with p.get('id') is not None, consistent with
the _smaller_duplicate_ids fix; bare p['id'] raised KeyError on any
person dict missing the id field in the auto-merge path
- immich_api.py: replace bare data['major'/'minor'/'patch'] subscripts
with .get() in get_immich_version; KeyError was silently swallowed by
except Exception, causing version-gated flags to disable without warning
ORT 1.26.0 changed provider loading to gate on the presence of required
nvidia pip packages before attempting to load libonnxruntime_providers_cuda.so.
Without nvidia-cuda-runtime-cu12, nvidia-cufft-cu12, and nvidia-curand-cu12
installed as Python packages, ORT silently skips the CUDA EP plugin entirely
(confirmed via /proc/maps: the .so was never dlopen'd despite existing on disk
and all system CUDA libs being present in ldconfig).
nvidia-nvjitlink-cu12 pulled in as a transitive dependency.
- executor: restore effective_count when replacement upload fails all retries
(delete succeeded but slot was never filled, leaving cap undercount)
- diversity: skip zero-norm embeddings before dedup/FPS selection
(InsightFace zeros pass dedup with similarity 0 and score distance 1.0,
getting selected first as maximally diverse)
- cli: exclude None from skip_ids in _smaller_duplicate_ids
(p.get('id') without None guard lets None into the set, silently
dropping every other id-less person from the processed list)
- embeddings: select face nearest crop centre instead of largest by area
(25% margin can pull a bigger neighbouring face into the crop;
largest-face selection then embeds the wrong person)
Round 11's replace_all missed two occurrences:
- _smaller_duplicate_ids inner comprehension (line 84): p["id"] →
p.get("id") so a named person with a missing "id" field does not
crash skip_ids computation before any return path is reached
- all-merges-failed fallback return (line 157): same fix; the outer
indentation prevented replace_all from matching this occurrence
The intentional p["id"] in merge_ids (line 119) is kept: that ID is
passed directly to merge_people() where None would be a caller bug,
not a silent data corruption.
- upload_tracker: revert data[flat_key] = [] from round 8; clearing the
entire shared legacy flat list on a corrupt value wipes all persons'
IDs, not just the one being reset; since a corrupt non-list value is
already unreadable by load_uploaded_ids, leaving it in place is safer
than a mass-wipe; update warning message to note the field is unaffected
but unreadable so the corruption is still observable
- cli: use p.get("id") instead of p["id"] in both people-list fallback
returns (_handle_duplicate_people lines 144 and 157) for consistency
with the success path at line 149; bare subscript crashes on malformed
unnamed persons that bypass _smaller_duplicate_ids
- immich_api: use 'or []' instead of .get("people", []) in get_people
so {"people": null} responses (some Immich versions with zero people
enrolled) return [] rather than None; .get() default only fires when
the key is absent, not when its value is null
- embeddings: log OSError from os.dup2 restore at DEBUG rather than
silently swallowing it; if a C extension (CUDA/onnxruntime) invalidates
the saved fd, the restore fails silently and stdout stays wired to
/dev/null — logging makes the event observable without changing the
swallow-and-continue semantics
- cache: remove MemoryError re-raise from EmbeddingCache.get(); a cache
read OOM aborted the entire diversity-selection batch for the person
rather than falling back to a fresh embedding computation, which is
the more appropriate OOM gate; broadening back to except Exception
restores the pre-round-5 fallback behavior
A KeyboardInterrupt raised inside the saved_out cleanup block would
propagate past the saved_err and devnull_fd blocks, leaking those fds.
In CPython this race is not realistically triggerable — KI is delivered
between bytecodes and os.dup2 is a single atomic C syscall — so we
accept the theoretical risk rather than silencing BaseException in a
finally block.
- executor: revert person_has_fscores=True back into try/except else
branch; moving it outside in round 8 was a regression — when the
tracker write fails on the first-ever upload (no prior frigate_scores
in tracker), setting the flag True prematurely switches at-cap
replacement into fscore mode, get_most_redundant_mapped_file returns
None (no entries), and all replacements are silently skipped;
the flag must only be set when the score is actually written
- diversity: remove dead face_crop-None guard; any face that passes
assess_quality (≥90 px MIN_FACE_WIDTH) produces a crop ≥135 px
(face + 25% margin), which is always above the 30 px crop minimum,
making the guard unreachable; _crop_face_from_thumbnail also calls
_get_face_bbox internally, so face_bbox is not None guarantees the
inner bbox check also passes
- diversity: fix hard_count regression from round 7 — revert to
'is not None and < 0.85' so only images that actually receive a
FPS boost (confirmed low confidence) are counted as hard examples;
None-confidence images use conf_array=1.0 (no boost) and should
not appear in the hard-example log count
- diversity: fix _scale_bbox_to_thumbnail to use explicit zero-guard
for imageWidth/imageHeight (meta_w or 0; scale = img_w/meta_w if
meta_w else 1.0) — mirrors image_processing.py pattern; prevents
`or img_w` from silently treating imageWidth=0 as missing and
returning scale=1.0 without surfacing the zero-metadata case
- embeddings: wrap all three os.close calls in _suppress_output
finally block with try/except OSError: pass so a failed close
in one branch cannot abort the outer finally and leak devnull_fd
or the saved_err/saved_out fds
- upload_tracker: clear corrupt flat-list key (data[flat_key] = [])
after the isinstance warning instead of leaving the corrupt value
in place — prevents stale IDs persisting across reset_person calls
and future load_uploaded_ids() from seeing a non-list value
- executor: move 'if pre_fscore is not None: person_has_fscores = True'
out of the try/except else branch so it fires even when mark_uploaded
raises; Frigate scores exist once measured regardless of tracker
write success, and replacement strategy should reflect that
- diversity: revert conf_array default from 0.5 back to 1.0 (np.ones);
the 0.5 default caused None-confidence images to receive a 1.7× FPS
boost and beat high-confidence detections — counter-productive for
Frigate training data quality
- diversity: fix hard_count to include None-confidence images (count
images where score is None or < 0.85, not only confirmed < 0.85);
the previous check systematically undercounted boosted images when the
Immich faces API omits the score field
- executor: fix garbled comment fragment "Skipped on / skipped when"
left by a partial edit in round 4; merge into a single coherent sentence
- executor: expand actually_uploaded trade-off comment to document all
three consequences of a tracker write failure (Frigate duplicate,
quality-replacement exclusion, cap-slot consumption), not only the
duplicate risk mentioned previously
- upload_tracker: add person_ids guard to reset_person isinstance check
so the non-list warning only fires when cleanup would actually have run,
not on no-op calls where person_ids is empty
flan
dependabot[bot]