kobotoolbox / kpi / 23752379599 / 3

82%

master: 76%

push

github

fix(qual): fix flaky QA throttling tests in parallel CI using `LocMemCache` isolation (#6881)

### 📣 Summary
This PR resolves intermittent test failures in
`TestAutomaticQAThrottling` by isolating its cache to local memory. This
prevents a race condition where parallel workers with identical
test-database User IDs were colliding in the shared Redis cache.

### 📖 Description
Fixes the intermittent `200 != 429` and `429 != 200` assertion errors in
`TestAutomaticQAThrottling` during parallel GitHub Action runs.

**The Root Cause (Database ID Collision):**
The issue happens when tests run in parallel with multiple workers. DRF
throttling stores request history in the Django cache using a key
derived from the authenticated user's ID (e.g.
`throttle_<scope>_<user.pk>`).

When tests run with multiple workers, each worker creates its own test
database but they share the same cache backend. Since users in each
worker often have the same primary key (e.g. pk=1), they generate the
same throttle cache key.

As a result, requests from different workers collide in the same cache
entry, causing the throttle state to leak across tests. This leads to
nondeterministic failures where tests may receive 200 instead of 429 or
vice versa.

**The Fix:**
Applied the `@override_settings` decorator to the
`TestAutomaticQAThrottling` class to force it to use
`django.core.cache.backends.locmem.LocMemCache`.

This isolates the cache entirely to the local memory of the specific
worker process. Each worker now has its own private "cache bucket,"
preventing cross-worker interference and resolving the flakiness without
requiring us to patch application logic or hardcode unique cache keys.

3651 of 11800 branches covered (30.94%)

19392 of 35860 relevant lines covered (54.08%)

0.54 hits per line