Leave legacy Lance directory removal to the user rather than deleting it
automatically on first run. Beta policy: user is expected to do a clean
re-embed anyway; no need for the system to silently delete their data.
Remove _cleanup_legacy_lance_index(), the forced-rebuild path that called
it, and the associated tests.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Same treatment as title/tags/correspondent in #12944: excluded from
the embedded text, visible to the LLM via metadata prepend. Changes
embedded text for every document, so it ships inside the sqlite-vec
transition, whose forced rebuild re-embeds everything anyway.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Fixes the non-AVX2 SIGILL class (#12970) at the root: lancedb is no
longer imported. sqlite-vec 0.1.9 wheels carry no baked SIMD, vec0
metadata columns give parameterized EQ/IN filtering, WAL preserves the
lock-free-reader model, and compact() rebuilds the table because vec0
DELETEs never reclaim space.
Implementation notes vs. the Task 3A draft:
- compact() uses a file-swap approach (new db file + Path.replace) rather
than ALTER TABLE RENAME, which does not cascade to shadow tables in
sqlite-vec 0.1.9 (upstream limitation).
- Bloat is tracked via a cumulative total_inserts counter in index_meta
because the _rowids shadow table does not accumulate deleted rows in
0.1.9 (contrary to the design doc assumption from #54).
- None distances from the zero-vector cosine edge case are mapped to
similarity 0.0 rather than raising TypeError.
- Test suite updated accordingly: _bloat_ratio reads index_meta instead
of _rowids; seed collision in force-compact test fixed (seed=100.0).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: Move LLM index lock outside index dir and skip per-doc tasks on bulk update
Two concurrency bugs from #12893:
[P1] Lock file lived inside LLM_INDEX_DIR. A rebuild calls
shutil.rmtree(LLM_INDEX_DIR), deleting the lock while a worker still
held it. A second worker then acquired a fresh lock on the new path and
ran concurrently, defeating serialisation. Move the lock to
DATA_DIR/locks/llm_index.lock (a new settings constant LLM_INDEX_LOCK)
so rmtree cannot touch it. The locks/ dir is created at settings load
time, matching the existing pattern for LOGGING_DIR.
[P2] document_updated was connected to add_or_update_document_in_llm_index
in apps.py. bulk_update_documents() emits document_updated for every
document in the batch, queuing N per-document LLM tasks, and then also
calls update_llm_index(rebuild=False) once at the end. Pass
skip_ai_index=True when sending document_updated from the bulk path so
the handler skips the per-document enqueue; the existing batch call at
the end of bulk_update_documents is the only LLM update for that path.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: ghost vectors leave KeyError-prone nodes_dict entries after deletion
docstore.delete_document() removes a node from the docstore but leaves its
entry in index_struct.nodes_dict (the FAISS positional-id to node-UUID map).
A subsequent similarity query resolves the ghost position to the deleted UUID,
finds nothing in fetched_nodes_by_id, and raises KeyError inside
_insert_fetched_nodes_into_query_result.
Purge stale nodes_dict entries after each docstore deletion and re-sync the
mutated index_struct into the kvstore so persist() writes the updated mapping.
Dead FAISS vectors remain in the flat index until the next full rebuild
(IndexFlatL2 is append-only); add a try/except KeyError around
retriever.retrieve() as a defensive fallback for any residual ghost positions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: acquire index lock in query_similar_documents
query_similar_documents() loaded the index and ran the FAISS retriever
without holding the file lock. All write paths (update_llm_index,
llm_index_add_or_update_document, llm_index_remove_document) hold
FileLock(_index_lock_path()), so a concurrent rebuild calling
shutil.rmtree(LLM_INDEX_DIR) while a read is mid-load produces an IOError
or corrupt partial state.
Wrap the load_or_build_index() call and all subsequent retriever work inside
FileLock. The early-return guards (vector_store_file_exists check, empty
allowed_document_ids) remain outside the lock; the DB query for the final
result set also stays outside.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: skip LLM index enqueue on document_updated during version addition
When a document is consumed as a new version of an existing document, the
consumer fires document_consumption_finished (which triggers
add_or_update_document_in_llm_index) and then document_updated for the root
document. Both signals are connected to the same handler, so the root document
was enqueued for LLM indexing twice per version-addition event.
Pass skip_ai_index=True on the consumer's version-addition document_updated
send so the handler's existing guard suppresses the duplicate enqueue.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Test: bulk_update_documents must not enqueue per-doc LLM tasks
With AI enabled, bulk_update_documents() sends document_updated for every
document in the batch. The skip_ai_index=True kwarg (added in the P2 fix)
prevents add_or_update_document_in_llm_index from enqueuing a per-document
task for each one. Only the single update_llm_index call at the end should run.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Debug level log sure
* Update src/paperless_ai/indexing.py
Co-authored-by: shamoon <4887959+shamoon@users.noreply.github.com>
* Apply suggestion from @shamoon
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: shamoon <4887959+shamoon@users.noreply.github.com>
* Fix: Remove all nodes for multi-chunk documents in update_llm_index incremental path
The existing_nodes dict comprehension keyed on document_id silently dropped all
but the last node per document, so only that one node was deleted when a
modified document was re-indexed, leaving all other chunks as ghost vectors in
the FAISS index. Switch to a defaultdict(list) that collects every node per
document_id, then iterate and delete all of them before inserting fresh nodes.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: Wire document_updated signal to LLM index update handler
Connect document_updated to add_or_update_document_in_llm_index in
DocumentsConfig.ready() so REST API edits (PATCH /api/documents/{id}/)
enqueue an LLM vector store update, matching the existing
document_consumption_finished behavior.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: Add file lock around FAISS index mutations to prevent concurrent write corruption
Two concurrent Celery workers calling llm_index_add_or_update_document or
llm_index_remove_document each loaded the same on-disk index independently,
made their own change, and the last writer silently overwrote the first's
update. Wrap both functions and the rebuild/persist body of update_llm_index
in a filelock.FileLock keyed on LLM_INDEX_DIR/index.lock. Add a TOCTOU
comment on queue_llm_index_update_if_needed explaining the residual risk
(duplicate rebuild tasks are wasteful but not corrupting because the lock
serialises the actual write).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: Apply _normalize() in extract_unmatched_names to prevent duplicate suggestions
extract_unmatched_names was using .lower() while _match_names_to_queryset
uses _normalize() (which also strips punctuation). A name like "J. Smith"
matched to existing correspondent "J Smith" would still appear in the
unmatched list, causing duplicate object creation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: Skip LLM index update gracefully when document has no indexable content
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: Persist empty index when all documents are deleted to clear stale FAISS vectors
The early-return guard in update_llm_index fired before persist() when no
documents existed, leaving a stale on-disk FAISS index that returned phantom
hits for deleted document IDs. Now the guard only returns early for the
incremental (rebuild=False) path when no index exists on disk; the rebuild
path always continues through to persist(), producing an empty clean index.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Chore: Simplify incremental index update — use docs.values() and deduplicate node extend
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: Validate and limit chat question input in ChatStreamingView
Add max_length=4000 to ChatStreamingSerializer.q and replace the bare
request.data["q"] read with proper serializer.is_valid(raise_exception=True)
so oversized or missing questions are rejected with HTTP 400 before
reaching the LLM.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix: Add defensive prompt framing to mark document content as untrusted
* Also adds a system prompt which is treated higher that this is untrusted stuff
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(tasks): replace PaperlessTask model with structured redesign
Drop the old string-based PaperlessTask table and recreate it with
Status/TaskType/TriggerSource enums, JSONField result storage, and
duration tracking fields. Update all call sites to use the new API.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(tasks): rewrite signal handlers to track all task types
Replace the old consume_file-only handler with a full rewrite that tracks
6 task types (consume_file, train_classifier, sanity_check, index_optimize,
llm_index, mail_fetch) with proper trigger source detection, input data
extraction, legacy result string parsing, duration/wait time recording,
and structured error capture on failure.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test(tasks): add traceback and revoked state coverage to signal tests
* refactor(tasks): remove manual PaperlessTask creation and scheduled/auto params
All task records are now created exclusively via Celery signals (Task 2).
Removed PaperlessTask creation/update from train_classifier, sanity_check,
llmindex_index, and check_sanity. Removed scheduled= and auto= parameters
from all 7 call sites. Updated apply_async callers to use trigger_source
headers instead. Exceptions now propagate naturally from task functions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(tasks): auto-inject trigger_source=scheduled header for all beat tasks
Inject `headers: {"trigger_source": "scheduled"}` into every Celery beat
schedule entry so signal handlers can identify scheduler-originated tasks
without per-task instrumentation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(tasks): update serializer, filter, and viewset with v9 backwards compat
- Replace TasksViewSerializer/RunTaskViewSerializer with TaskSerializerV10
(new field names), TaskSerializerV9 (v9 compat), TaskSummarySerializer,
and RunTaskSerializer
- Add AcknowledgeTasksViewSerializer unchanged (kept existing validation)
- Expand PaperlessTaskFilterSet with MultipleChoiceFilter for task_type,
trigger_source, status; add is_complete, date_created_after/before filters
- Replace TasksViewSet.get_serializer_class() to branch on request.version
- Add get_queryset() v9 compat for task_name/type query params
- Add acknowledge_all, summary, active actions to TasksViewSet
- Rewrite run action to use apply_async with trigger_source header
- Add timedelta import to views.py; add MultipleChoiceFilter/DateTimeFilter
to filters.py imports
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(tasks): add read_only_fields to TaskSerializerV9, enforce admin via permission_classes on run action
* test(tasks): rewrite API task tests for redesigned model and v9 compat
Replaces the old Django TestCase-based tests with pytest-style classes using
PaperlessTaskFactory. Covers v10 field names, v9 backwards-compat field
mapping, filtering, ordering, acknowledge, acknowledge_all, summary, active,
and run endpoints. Also adds PaperlessTaskFactory to factories.py and fixes
a redundant source= kwarg in TaskSerializerV10.related_document_ids.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test(tasks): fix two spec gaps in task API test suite
Move test_list_is_owner_aware to TestGetTasksV10 (it tests GET /api/tasks/,
not acknowledge). Add test_related_document_ids_includes_duplicate_of to
cover the duplicate_of path in the related_document_ids property.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test(tasks): address code quality review findings
Remove trivial field-existence tests per project conventions. Fix
potentially flaky ordering test to use explicit date_created values.
Add is_complete=false filter test, v9 type filter input direction test,
and tighten TestActive second test to target REVOKED specifically.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(tasks): update TaskAdmin for redesigned model
Add date_created, duration_seconds to list_display; add trigger_source
to list_filter; add input_data, duration_seconds, wait_time_seconds to
readonly_fields.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(tasks): update Angular types and service for task redesign
Replace PaperlessTaskName/PaperlessTaskType/PaperlessTaskStatus enums
with new PaperlessTaskType, PaperlessTaskTriggerSource, PaperlessTaskStatus
enums. Update PaperlessTask interface to new field names (task_type,
trigger_source, input_data, result_message, related_document_ids).
Update TasksService to filter by task_type instead of task_name.
Update tasks component and system-status-dialog to use new field names.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* chore(tasks): remove django-celery-results
PaperlessTask now tracks all task results via Celery signals. The
django-celery-results DB backend was write-only -- nothing reads
from it. Drop the package and add a migration to clean up the
orphaned tables.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test: fix remaining tests broken by task system redesign
Update all tests that created PaperlessTask objects with old field names
to use PaperlessTaskFactory and new field names (task_type, trigger_source,
status, result_message). Use apply_async instead of delay where mocked.
Drop TestCheckSanityTaskRecording — tests PaperlessTask creation that was
intentionally removed from check_sanity().
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test(tasks): improve test_api_tasks.py structure and add api marker
- Move admin_client, v9_client, user_client fixtures to conftest.py so
they can be reused by other API tests; all three now build on the
rest_api_client fixture instead of creating APIClient() directly
- Move regular_user fixture to conftest.py (was already done, now also
used by the new client fixtures)
- Add docstrings to every test method describing the behaviour under test
- Move timedelta/timezone imports to module level
- Register 'api' pytest marker in pyproject.toml and apply pytestmark to
the entire file so all 40 tests are selectable via -m api
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* refactor(tasks): simplify task tracking code after redesign
- Extract COMPLETE_STATUSES as a class constant on PaperlessTask,
eliminating the repeated status tuple across models.py, views.py (3×),
and filters.py
- Extract _CELERY_STATE_TO_STATUS as a module-level constant instead of
rebuilding the dict on every task_postrun
- Extract _V9_TYPE_TO_TRIGGER_SOURCE and _RUNNABLE_TASKS as class
constants on TasksViewSet instead of rebuilding on every request
- Extract _TRIGGER_SOURCE_TO_V9_TYPE as a class constant on
TaskSerializerV9 instead of rebuilding per serialized object
- Extract _get_consume_args helper to deduplicate identical arg
extraction logic in _extract_input_data, _determine_trigger_source,
and _extract_owner_id
- Move inline imports (re, traceback) and Avg to module level
- Fix _DOCUMENT_SOURCE_TO_TRIGGER type annotation key type to
DocumentSource instead of Any
- Remove redundant truthiness checks in SystemStatusView branches
already guarded by an is-None check
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* refactor(tasks): add docstrings and rename _parse_legacy_result
- Add docstrings to _extract_input_data, _determine_trigger_source,
_extract_owner_id explaining what each helper does and why
- Rename _parse_legacy_result -> _parse_consume_result: the function
parses current consume_file string outputs (consumer.py returns
"New document id N created" and "It is a duplicate of X (#N)"),
not legacy data; the old name was misleading
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(tasks): extend and harden the task system redesign
- TaskType: add EMPTY_TRASH, CHECK_WORKFLOWS, CLEANUP_SHARE_LINKS;
remove INDEX_REBUILD (no backing task — beat schedule uses index_optimize)
- TRACKED_TASKS: wire up all nine task types including the three new ones
and llmindex_index / process_mail_accounts
- Add task_revoked_handler so cancelled/expired tasks are marked REVOKED
- Fix double-write: task_postrun_handler no longer overwrites result_data
when status is already FAILURE (task_failure_handler owns that write)
- v9 serialiser: map EMAIL_CONSUME and FOLDER_CONSUME to AUTO_TASK
- views: scope task list to owner for regular users, admins see all;
validate ?days= query param and return 400 on bad input
- tests: add test_list_admin_sees_all_tasks; rename/fix
test_parses_duplicate_string (duplicates produce SUCCESS, not FAILURE);
use PaperlessTaskFactory in modified tests
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(tasks): fix MAIL_FETCH null input_data and postrun double-query
- _extract_input_data: return {} instead of {"account_ids": None} when
process_mail_accounts is called without an explicit account list (the
normal beat-scheduled path); add test to cover this path
- task_postrun_handler: replace filter().first() + filter().update() with
get() + save(update_fields=[...]) — single fetch, single write,
consistent with task_prerun_handler
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(tasks): add queryset stub to satisfy drf-spectacular schema generation
TasksViewSet.get_queryset() accesses request.user, which drf-spectacular
cannot provide during static schema generation. Adding a class-level
queryset = PaperlessTask.objects.none() gives spectacular a model to
introspect without invoking get_queryset(), eliminating both warnings
and the test_valid_schema failure.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test(tasks): fill coverage gaps in task system
- test_task_signals: add TestTaskRevokedHandler (marks REVOKED, ignores
None request, ignores unknown id); switch existing direct
PaperlessTask.objects.create calls to PaperlessTaskFactory; import
pytest_mock and use MockerFixture typing on mocker params
- test_api_tasks: add test_rejects_invalid_days_param to TestSummary
- tasks.service.spec: add dismissAllTasks test (POST acknowledge_all +
reload)
- models: add pragma: no cover to __str__, is_complete, and
related_document_ids (trivial delegates, covered indirectly)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Well, that was a bad push.
* Fixes v9 API compatability with testing coverage
* fix(tasks): restore INDEX_OPTIMIZE enum and remove no-op run button
INDEX_OPTIMIZE was dropped from the TaskType enum but still referenced
in _RUNNABLE_TASKS (views.py) and the frontend system-status-dialog,
causing an AttributeError at import time. Restore the enum value in the
model and migration so the serializer accepts it, but remove it from
_RUNNABLE_TASKS since index_optimize is a Tantivy no-op. Remove the
frontend "Run Task" button for index optimization accordingly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tasks): v9 type filter now matches all equivalent trigger sources
The v9 ?type= query param mapped each value to a single TriggerSource,
but the serializer maps multiple sources to the same v9 type value.
A task serialized as "auto_task" would not appear when filtering by
?type=auto_task if its trigger_source was email_consume or
folder_consume. Same issue for "manual_task" missing web_ui and
api_upload sources. Changed to trigger_source__in with the full set
of sources for each v9 type value.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tasks): give task_failure_handler full ownership of FAILURE path
task_postrun_handler now early-returns for FAILURE states instead of
redundantly writing status and date_done. task_failure_handler now
computes duration_seconds and wait_time_seconds so failed tasks get
complete timing data. This eliminates a wasted .get() + .save() round
trip on every failed task and gives each handler a clean, non-overlapping
responsibility.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tasks): resolve trigger_source header via TriggerSource enum lookup
Replace two hardcoded string comparisons ("scheduled", "system") with a
single TriggerSource(header_source) lookup so the enum values are the
single source of truth. Any valid TriggerSource DB value passed in the
header is accepted; invalid values fall through to the document-source /
MANUAL logic. Update tests to pass enum values in headers rather than raw
strings, and add a test for the invalid-header fallback path.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(tasks): use TriggerSource enum values at all apply_async call sites
Replace raw strings ("system", "manual") with PaperlessTask.TriggerSource
enum values in the three callers that can import models. The settings
file remains a raw string (models cannot be imported at settings load
time) with a comment pointing to the enum value it must match.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test(tasks): parametrize repetitive test cases in task test files
test_api_tasks.py:
- Collapse six trigger_source->v9-type tests into one parametrized test,
adding the previously untested API_UPLOAD case
- Collapse three task_name mapping tests (two remaps + pass-through)
into one parametrized test
- Collapse two acknowledge_all status tests into one parametrized test
- Collapse two run-endpoint 400 tests into one parametrized test
- Update run/ assertions to use TriggerSource enum values
test_task_signals.py:
- Collapse three trigger_source header tests into one parametrized test
- Collapse two DocumentSource->TriggerSource mapping tests into one
parametrized test
- Collapse two prerun ignore-invalid-id tests into one parametrized test
All parametrize cases use pytest.param with descriptive ids.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Handle JSON serialization for datetime and Path. Further restrist the v9 permissions as Copilot suggests
* That should fix the generated schema/browser
* Use XSerializer for the schema
* A few more basic cases I see no value in covering
* Drops the migration related stuff too. Just in case we want it again or it confuses people
* fix: annotate tasks_summary_retrieve as array of TaskSummarySerializer
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix: annotate tasks_active_retrieve as array of TaskSerializerV10
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Restore task running to superuser only
* Removes the acknowledge/dismiss all stuff
* Aligns v10 and v9 task permissions with each other
* Short blurb just to warn users about the tasks being cleared
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Trenton Holmes
Moritz StücklerandClaude Sonnet 4.6
Sebastian Steinbeißer
dependabot[bot]andshamoon