You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
pytest tests/unit/ on pristine origin/main (d6243bb, 2026-08-09) fails 7 tests; 357 pass. The Test Suite Validation workflow is intermittently red on main pushes, and because these failures are ambient, real regressions get lost in the noise.
Failing tests
test_zot_cache_policy.py::test_zot_sync_is_on_demand_and_bounded — asserts no "**" sync prefix, but zot-cache.yaml uses "**" for fedora/k8s/ecr/lscr registries. Test and manifest disagree on the bounded-prefix policy.
test_build_metrics.py::test_zot_metrics_are_enabled_on_both_registries — expects annotation sync-policy-v2, manifest has sync-policy-v4.
For each test, decide whether the test or the manifest is the current truth (e.g. was widening to ** prefixes deliberate? was the sync-policy-v4 bump deliberate?) and update the losing side. The page-dataset tests likely need counts derived from fixtures instead of hardcoded.
Related follow-up (from the 2026-08 bandwidth session)
The ~1 TiB spike days (2 of last 5) were never definitively attributed — prime suspect is cold BuildStream/BST source fetches. Verify with du -s on the buildbarn CAS after the next spike; zot exports pull counts, not bytes.
What
pytest tests/unit/on pristineorigin/main(d6243bb, 2026-08-09) fails 7 tests; 357 pass. TheTest Suite Validationworkflow is intermittently red on main pushes, and because these failures are ambient, real regressions get lost in the noise.Failing tests
test_zot_cache_policy.py::test_zot_sync_is_on_demand_and_bounded— asserts no"**"sync prefix, butzot-cache.yamluses"**"for fedora/k8s/ecr/lscr registries. Test and manifest disagree on the bounded-prefix policy.test_build_metrics.py::test_zot_metrics_are_enabled_on_both_registries— expects annotationsync-policy-v2, manifest hassync-policy-v4.test_page_dataset_collector.py::test_tests_matrix_derives_rows_from_surface_and_results—rows_with_completed_runsexpected 14, got 15 (data drift vs. hardcoded counts).test_page_dataset_collector.py::test_applications_matrix_keeps_bazaar_fallbacks_explicit— same class of data drift.test_recc_runner_seam.py×3 — recc overlay/runner seam assertions no longer match the manifests.Reproduction
Fix
For each test, decide whether the test or the manifest is the current truth (e.g. was widening to
**prefixes deliberate? was the sync-policy-v4 bump deliberate?) and update the losing side. The page-dataset tests likely need counts derived from fixtures instead of hardcoded.Related follow-up (from the 2026-08 bandwidth session)
scripts/traffic_report.py --window 24ha day after PR chore(lab): cut uplink bandwidth — upstream digest polls, suspend unreviewed crons #632 merged; baseline was ~170–225 GiB/day and should collapse.du -son the buildbarn CAS after the next spike; zot exports pull counts, not bytes.Automatable: yes