fix(watson): fix + batch full-text index updates on Celery worker tasks

[valentijnscholten]

Description

Celery workers serve no HTTP request, so AsyncSearchContextMiddleware never opens a watson search_context. Without an active context, django-watson's post_save receiver falls into its else-branch and indexes every saved object synchronously — one watson_searchentry DELETE + INSERT per object. Any bulk save that runs in a worker (e.g. a large async import) then re-indexes objects one at a time, flooding the DB with thousands of index writes.

This PR refactors the performance test to perform a real API call. This results in queries and celery tasks triggered by middleware is now also captured. This explains the small increase in counts in this PR. But this actually a decrease because of the watson fix bundled in this PR.

Approach

Phased, to confirm the problem before fixing it (regression test added downstream in DefectDojo Pro, where the async-import worker path lives — see the paired Pro PR):

1. Reproduce — a 1373-finding import through the worker path issued ~1000+ watson_searchentry INSERTs (query-count explosion). The failing test asserted the batched expectation.
2. Fix — batch the worker-side index updates the same way the request path already does.

Changes

• watson_search_context_for_task (dojo/middleware.py) — a context manager that opens a watson search_context for the duration of a Celery task so saves accumulate and drain in WATSON_ASYNC_INDEX_UPDATE_BATCH_SIZE-sized async batches on exit, mirroring AsyncSearchContextMiddleware for the request path. An empty context (task saved no indexed models) drains to nothing — a cheap no-op.
• CELERY_TASK_CONTEXT_MANAGERS default (dojo/settings/settings.dist.py) — wires the CM in; PluggableContextTask enters it around every task. Downstreams (e.g. Pro) extend this list rather than replacing it.
• Instance-bound intermediate-flush hook (dojo/middleware.py, dojo/tasks.py) — install_intermediate_flush_hook() now binds the flush wrapper to the module-global search_context_manager instance instead of monkeypatching SearchContextManager at class level. The terminal index task update_watson_search_index_for_model builds its own local SearchContextManager to index one already-bounded batch; with the class-level patch, a full 1000-PK batch re-triggered the flush from inside the task and re-dispatched a clone of itself — infinite recursion under eager celery, an infinite re-dispatch loop (nothing ever indexed) on a real worker. Binding to the singleton makes that unrepresentable: watson's post_save and the request/task paths all go through the module-global instance (which flushes), while private throwaway contexts keep the stock add_to_context and simply index their batch on end(). The same loop is fixed on the release line by the minimal identity guard in fix(watson): only intermediate-flush the global search context singleton #15187 (extracted from perf(importers): add defer_product_grading option + harden watson index-flush hook #15165); this PR supersedes the guard on dev with the structural fix.

Test results

Verified in the integrated (Pro) environment; a downstream regression test drives the worker path (AsyncImporter) end-to-end:

• 1373-finding JFrog import: watson_searchentry INSERTs ~1000+ → 14 (ceil(1000/100) + ceil(373/100)), with 1373/1373 findings still indexed exactly once (no findings dropped, no recursion).
• unittests/test_watson_intermediate_flush.py passes (5/5), including the new test_ad_hoc_context_manager_does_not_drain regression test: an ad-hoc SearchContextManager pushed past the flush threshold must not drain (it would re-dispatch a clone of itself and loop). The existing tests exercise the global manager, which still flushes.
• Loop verified fixed live: dispatching a full 1000-pk update_watson_search_index_for_model task on unguarded code self-replicates without bound (10 → 200+ tasks in ~3 min); on this branch the same dispatch executes exactly once and indexes its batch.

OSS-standalone CI will run the full unittests/ suite for final confirmation.

Labels

• performance, bugfix, settings_changes (new CELERY_TASK_CONTEXT_MANAGERS in settings.dist.py)

[mtesauro]

Approved