While `_ensure_indexes` was already secured, the `CREATE INDEX` for
`ix_saved_searches_user_id` was hardcoded. This commit explicitly
quotes it to unify our security posture against SQL injection
and keep static analyzers happy.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
This commit improves test coverage for the `register_settings_reload_signal` function in `app/utils/settings_sync.py`.
🎯 **What:** The testing gap addressed was that the `_reload_if_stale` inner Celery `task_prerun` signal handler was entirely untested, specifically around exception handling (e.g. Redis timeouts or OCR manager errors) and the code branch where Redis returns no version key.
📊 **Coverage:** The following scenarios are now tested:
- Redis returning `None` for the version.
- Redis throwing an exception (handled gracefully).
- `ensure_ocr_languages_async` throwing an exception (caught and logged without failing the task).
✨ **Result:** Test coverage for `register_settings_reload_signal` is now 100%. Total coverage for `app/utils/settings_sync.py` has been substantially improved.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
The previous commit introduced a benchmark test with unsorted imports inside the test method, which caused the Ruff Lint & Format CI check to fail with `I001 [*] Import block is un-sorted or un-formatted`. This commit runs `ruff format` and `ruff check --fix` on `tests/test_notifications_api.py` to fix the issue.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
- Added a benchmark script in tests/test_notifications_api.py that proved the N+1 issue issue.
- Replaced iterative DB lookups inside `for item in body.preferences:` with single pre-fetch query and local `prefs_dict` lookups.
- Verified test benchmark time drops from ~0.0964s to ~0.0141s for a batch of 100 items.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Replaced the synchronous `requests.post` calls in `app/api/dropbox.py` with asynchronous `httpx.AsyncClient().post` calls. This ensures that the FastAPI event loop is not blocked during network I/O, allowing better concurrent performance.
Also updated the `test_api_dropbox.py` tests to use `httpx.AsyncClient.post` in mocks and properly construct `httpx.RequestError` in exception handling tests.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Replaced the loop over `body.order` which generated an N+1 issue
with a single bulk query fetching all relevant `SubscriptionPlan`
records via the `.in_()` clause.
Added an in-memory dictionary map of `plan_id` to `SubscriptionPlan`
objects to allow `O(1)` lookups while updating the order.
Benchmark speedup: 14.71x faster on 500 records.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Added tests to `tests/test_api_advanced_filters.py` to cover missing edge cases and error handling for the `PUT /api/saved-searches/{id}` endpoint. New test coverage includes duplicate name conflicts (409), validation errors for names exceeding max length (422), empty names (422), empty filters (422), and payloads containing only invalid filter keys (422).
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Replaced the loop over duplicate hashes that resulted in O(N) database queries
per page with a single efficient `in_` batch query to retrieve both originals
and duplicates. The records are then grouped in memory using dictionaries.
This resolves the N+1 performance bottleneck and reduces response time from
an average of 1.65 seconds to ~0.45 seconds locally for 500 groups.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Adds unit tests for the notify_settings_updated function in
app/utils/settings_sync.py to verify that exceptions during Redis publish,
settings reload, and OCR language check are properly caught and logged as
warnings without raising up the call stack.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Replaced the N+1 query in `list_shared_links` which fetched `FileRecord` for each link. It now uses a single query with an `outerjoin` to fetch `original_filename` alongside the `SharedLink` object.
Measured a significant improvement from ~0.4547s to ~0.0579s per 1000 links.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
semantic-release