Release validation — 27–28 October 2026#
The PostgreSQL resolve/reopen blocker is fixed. Linux/POSIX checks and the Python 3.10–3.14 CI matrix still need to pass before release. The documentation and OpenAPI update is complete. Legacy AlertGroup responder and stakeholder API/storage removal is approved as a 2.5 breaking change.
The 27 October checks used 7738f25 plus the workspace and documentation changes on release/2.5.0, subsequently committed as 7cf123d. The 28 October follow-up used 7cf123d plus the uncommitted resolve/reopen fix. No commit or push was made for this fix. All databases were tested in isolated copies; the development checkout's database was not used.
Results — 27 October#
| Check | Result |
|---|---|
| SQLite suite, Python 3.12.10 on Windows | 2,275 passed; 78 skipped; 2 deselected |
| SQLite coverage, including branches | 73.46% (required: 50%) |
Full PostgreSQL 16.15 suite, including tests/postgresql | 2237 passed, 5 failed, 17 errors, 77 skipped in the initial run; see follow-up below |
| Final OpenAPI contracts and environment-related PostgreSQL retries | 62 passed |
| Final OpenAPI/combined activity checks on SQLite | 33 passed; the additional same-user/different-role scenario also passed |
| Workspace in real Chrome, Playwright 1.62.1 | Passed |
| Workspace Node state/navigation checks | Passed |
| Helm 3.19.0 suite | 51 passed; covers the 47 Helm skips from the general suite |
helm lint | Passed; informational icon recommendation only |
| Real 2.4 → 2.5 upgrade, SQLite and PostgreSQL | Both passed |
| Python compilation, schema checks, templates | Passed |
| Complete OpenAPI 3.0.3 validation and SLO examples | Passed |
Ruff for OpenAPI/contract changes, duplicate test check, documentation links/front matter, workflow YAML, git diff --check | Passed |
The PostgreSQL follow-up reran the affected tests with UTF-8 enabled and a writable temporary directory. It also used the corrected activity test, which checks the built specification rather than source formatting. Those failures/errors cleared. Two initial failures require POSIX/Linux (container bootstrap and file mode 0600); the remaining functional failure was the race fixed on 28 October below. The initial PostgreSQL run remains a historical failing run.
Resolve/reopen blocker fixed — 28 October 2026#
The original regression test reproduced TimeoutError before the fix. A manual resolve previously waited for an uncommitted reopen's row lock and could then resolve its new occurrence. It now returns without a write when it observes an already resolved or merged group. Active groups still acquire the row lock and recheck their state before writing, so simultaneous resolves remain idempotent.
Follow-up results on Python 3.12.10 / Windows:
| Check | Result |
|---|---|
| PostgreSQL 16.15: alerts, PostgreSQL-specific, incidents, notifications, notifiers | 748 passed, 1 failed, 0 skipped |
| PostgreSQL smoke retry with its expected database name | 1 passed, 0 failed, 0 skipped |
| SQLite: alerts, incidents, notifications, notifiers | 737 passed, 0 failed, 0 skipped |
Ruff for changed Python files, Python syntax, duplicate tests, git diff --check | Passed |
The only PostgreSQL failure was the connection smoke test's hard-coded expectation of database name incidentrelay_test; the main run used the isolated database incidentrelay_resolve_reopen_test. The unchanged smoke test passed when rerun against the disposable database with its expected name. No functional failures remain in these suites.
The tests cover both committed and rolled-back reopen, a fresh resolve after reopen commits, simultaneous manual resolves with one event/audit/message update, and rollback plus retry when event or audit persistence fails. Existing recovery delivery/reopen and Incident concurrency checks also pass. These are targeted regression suites; the full release suite was not rerun in this follow-up.
Run against a disposable database using its test configuration:
pytest tests/alerts tests/postgresql tests/incidents tests/notifications tests/notifiers \
-o 'norecursedirs=.git .venv .tmp __pycache__' -qFor SQLite, omit tests/postgresql. The original regression remains at tests/postgresql/test_alert_reopen_concurrency.py::test_manual_resolve_during_uncommitted_reopen_is_noop, now parameterized for commit and rollback.
Upgrade evidence and platform limits#
The upgrade source was verified against release 2.4 commit 21ea27d92927242ed499b7584a276f47cb1c68cd, allowing Git line-ending normalization. For each legacy table, five rows produced one retained row, one merged row, two unmapped skips and one ambiguous skip on both databases. Technical comments, notification flags, merge history, new inserts/sequence state and a second no-op migration run were checked. Both upgraded schemas match the current models.
The 78 general-suite skips comprise 47 Helm checks (subsequently passed), the browser check (subsequently passed), one PostgreSQL-only assertion (covered on PostgreSQL), and 29 Linux container checks. Ten POSIX multiprocess metrics tests were excluded from collection; the two deselected tests also require POSIX/Linux. An import-only fcntl shim allowed Windows to load the application; it does not validate POSIX locking. These checks still require Linux CI.
The temporary PostgreSQL server used loopback only. WAL synchronization/fsync were disabled for test speed; these runs validate SQL transactions and concurrency, not persistence across an OS crash or power loss. The server was stopped after verification. The supported Python 3.10–3.14 Linux matrix and actual container startup were not run locally. The updated CI includes PostgreSQL-specific tests and a Chromium workspace run on Python 3.12; it has not been dispatched for these uncommitted files.