chore(tests): harden migration-test connection reset and fix cohort optimize ordering

[Radu-Raicea]

Problem

Two unrelated flaky test failures on master, both test-harness issues (no product bug):

• products/warehouse_sources/.../test_migration_0052.py::...missing_a_lookback fails with django.db.utils.OperationalError: the connection is closed. The test never runs its own assertions — its DB connection is dead before setUp's first query.
• ee/clickhouse/queries/test/test_cohort_query.py::test_cohort_filter_with_extra intermittently returns the wrong set of people from a cohort query.

Follow-up to #72758, which added a first mitigation for the connection-closed case but did not stop it.

Changes

Migration connection is closed — root cause. The connection wrapper ends up holding a psycopg handle that was closed while inside a transaction (closed_in_transaction=True). In that state connection.close() is a no-op and Django never reconnects, so setUp's MigrationExecutor query hits the dead handle. A pre-test reset can't recover this — is_usable() fails, close() no-ops. The reason the dead-handle state survives into setUp at all is that the test runs inside a TestCase class-level transaction.

Fix: run the test on NonAtomicTestMigrations (already used by several other migration tests) so there is no class transaction. Without it, a connection severed mid-test is simply reconnected on next use instead of getting stuck. (Earlier commits in this PR tried a connection reset and then a signal-disconnect; CI showed neither stopped it, which is what pointed at the class-transaction as the structural cause — those attempts are reverted.)

Kept as a smaller, independent hardening: the setUpClass reset extracted into a reusable reset_unusable_db_connections() helper (now also catching errored connections), which still covers the separate case of a prior test leaving a dead connection behind; and the warehouse conftest now resets the connection after each test as well as before, containing the direct close_old_connections() calls the transaction=True v3 duckgres backfill suites make.

Cohort query flake. CI showed the real cause: the OR cohort query intermittently returned a matched person twice (a duplicate row), not stale or missing membership. That query is built with UNION DISTINCT, so it returns distinct persons by design — a duplicate row is a ClickHouse merge/dedup timing artifact that doesn't change which persons match. The test now asserts on the set of matched persons rather than a sorted list (the pattern several sibling tests already use); it verifies membership, not result cardinality. An earlier commit's person-table optimize (a membership-staleness theory) was reverted, since the observed failure had both persons present.

How did you test this code?

No database or flox stack is available in the agent environment, so I could not run the DB-backed tests locally — validation is by CI on this branch plus lint/format/syntax checks. The first two pushes reproduced the exact failure again on the warehouse-sources shard and, by elimination, established that the dead-handle state is tied to the class transaction; this push removes that transaction for the test. Recommend confirming with an N-run loop under CI shard conditions before relying on it. The cohort test passed on these CI runs (only the warehouse-sources shard failed), so its fix is validated analytically.

Automatic notifications

• Publish to changelog?
• Alert Sales and Marketing teams?

Docs update

Not needed, test infrastructure only.

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Diagnosed and authored via PostHog Code while triaging two flaky failures reported in a Slack thread, using the /fixing-flaky-tests skill. Two hypotheses were tried and falsified by CI (reset dead connections; disconnect the close_old_connections request-signal receiver) before reading the Django connection internals showed the failing state is a handle closed while in_atomic_block, which only persists because of the TestCase class transaction. Settled on the structural fix — the existing non-atomic migration base — rather than fighting Django's transaction bookkeeping. The cohort fix reorders the settle to before membership computation rather than loosening the assertion.

---

Created with PostHog from a Slack thread

[github-actions]

Hey @Radu-Raicea! 👋

It looks like your git author email on this PR isn't your @posthog.com address (radu@raicea.com). Since you're on the PostHog team, it's worth pointing your local git author email at your @posthog.com address. Why it matters:

• Consistent work identity in git history — internal tooling that attributes commits to team members keys off your @posthog.com address.
• Keeps team contributions easy to tell apart from external community ones when scanning history.

You can fix it for this repo with:

git config user.email "you@posthog.com"

Or set it globally with git config --global user.email "you@posthog.com". No need to redo this PR — just a nudge for next time. 🙂

[greptile-apps]

_{Reviews (1): Last reviewed commit: "chore(tests): harden migration-test conn..." | Re-trigger Greptile}

[github-actions]

🤖 CI report

✅ Playwright — all passed

All tests passed.

View test results →

⚠️ Backend coverage — 66.0% of changed backend lines covered — 5 uncovered

🧪 Backend test coverage

Patch coverage — changed backend lines (products + core): █████████████░░░░░░░ 66.0% (10 / 15)

File	Patch	Uncovered changed lines
posthog/test/base.py	16.7%	1557–1559, 1579, 1587

🤖 Agents: add a test covering the lines above, or note why under "How did you test this code?". Machine-readable gap list: the patch-coverage artifact on this run (gh run download 30103698640 -n patch-coverage), or the coverage-data block at the end of this comment.

Per-product line coverage (touched products)

Product	Coverage	Lines
platform_features	██░░░░░░░░░░░░░░░░░░ 12.1%	7 / 58
batch_exports	████████░░░░░░░░░░░░ 39.6%	8,415 / 21,265
demo	███████████░░░░░░░░░ 56.2%	1,497 / 2,663
warehouse_sources_queue	████████████░░░░░░░░ 59.2%	148 / 250
tasks	██████████████░░░░░░ 69.4%	30,040 / 43,284
data_tools	██████████████░░░░░░ 70.0%	63 / 90
ai_gateway	███████████████░░░░░ 75.0%	9 / 12
signals	████████████████░░░░ 80.2%	22,273 / 27,784
cdp	████████████████░░░░ 81.0%	3,146 / 3,883
data_modeling	█████████████████░░░ 82.7%	5,556 / 6,722
wizard	█████████████████░░░ 84.8%	935 / 1,102
notebooks	█████████████████░░░ 85.6%	7,478 / 8,736
agent_platform	█████████████████░░░ 86.4%	3,807 / 4,405
actions	█████████████████░░░ 86.6%	717 / 828
cohorts	██████████████████░░ 87.8%	4,488 / 5,114
product_tours	██████████████████░░ 87.9%	1,303 / 1,482
exports	██████████████████░░ 88.4%	6,951 / 7,863
data_warehouse	██████████████████░░ 88.9%	11,905 / 13,396
business_knowledge	██████████████████░░ 89.0%	4,391 / 4,936
conversations	██████████████████░░ 89.3%	16,875 / 18,894
engineering_analytics	██████████████████░░ 89.4%	6,386 / 7,145
dashboards	██████████████████░░ 89.4%	5,983 / 6,693
visual_review	██████████████████░░ 89.4%	5,818 / 6,505
error_tracking	██████████████████░░ 89.7%	10,174 / 11,343
alerts	██████████████████░░ 90.0%	4,056 / 4,508
early_access_features	██████████████████░░ 90.1%	1,031 / 1,144
mcp_analytics	██████████████████░░ 90.3%	2,815 / 3,116
streamlit_apps	██████████████████░░ 90.4%	2,501 / 2,767
links	██████████████████░░ 90.6%	183 / 202
slack_app	██████████████████░░ 90.7%	9,028 / 9,951
marketing_analytics	██████████████████░░ 91.1%	11,903 / 13,072
stamphog	██████████████████░░ 91.1%	4,056 / 4,450
mcp_store	██████████████████░░ 91.9%	4,257 / 4,634
product_analytics	███████████████████░ 92.5%	5,848 / 6,319
notifications	███████████████████░ 92.6%	1,012 / 1,093
ai_observability	███████████████████░ 92.8%	15,132 / 16,313
surveys	███████████████████░ 93.0%	5,734 / 6,167
managed_migrations	███████████████████░ 93.1%	1,424 / 1,530
web_analytics	███████████████████░ 93.2%	14,671 / 15,744
posthog_ai	███████████████████░ 93.2%	1,326 / 1,422
approvals	███████████████████░ 93.3%	3,437 / 3,682
reminders	███████████████████░ 93.4%	468 / 501
workflows	███████████████████░ 93.6%	6,512 / 6,959
legal_documents	███████████████████░ 94.1%	1,568 / 1,667
endpoints	███████████████████░ 94.1%	8,640 / 9,177
messaging	███████████████████░ 94.3%	2,682 / 2,845
tracing	███████████████████░ 94.5%	2,670 / 2,826
review_hog	███████████████████░ 94.6%	6,898 / 7,289
skills	███████████████████░ 94.6%	3,158 / 3,337
logs	███████████████████░ 95.4%	10,012 / 10,498
experiments	███████████████████░ 95.8%	25,324 / 26,444
growth	███████████████████░ 96.1%	3,245 / 3,376
annotations	███████████████████░ 96.2%	732 / 761
replay_vision	███████████████████░ 96.2%	15,465 / 16,069
revenue_analytics	███████████████████░ 96.3%	1,887 / 1,960
feature_flags	███████████████████░ 96.5%	17,144 / 17,768
user_interviews	███████████████████░ 96.5%	2,638 / 2,734
access_control	███████████████████░ 96.9%	870 / 898
customer_analytics	███████████████████░ 97.1%	9,093 / 9,364
warehouse_sources	███████████████████░ 97.2%	340,184 / 350,022
data_catalog	███████████████████░ 97.5%	2,368 / 2,429
analytics_platform	████████████████████ 98.0%	2,153 / 2,197
metrics	████████████████████ 98.2%	2,491 / 2,536
pulse	████████████████████ 98.4%	2,017 / 2,049
live_debugger	████████████████████ 99.2%	613 / 618
field_notes	████████████████████ 99.4%	158 / 159

Report-only. Patch coverage = changed backend lines covered vs origin/master. Sorted lowest first.
Known gaps: lines covered only by Temporal tests show as uncovered; core line numbers may drift if master changed the same file.

[posthog-bot-comment-resolver]

🔀 Tried to auto-resolve conflicts with master but this one needs a human.

I won't retry until the branch or master moves.

[deployment-status-posthog]

Deploy status

Environment	Status	Deployed At	Workflow
dev	✅ Deployed	2026-07-24 16:24 UTC	Run
prod-us	✅ Deployed	2026-07-24 16:40 UTC	Run
prod-eu	✅ Deployed	2026-07-24 16:44 UTC	Run