Merge remote-tracking branch 'origin/master' into fix-convertToFullIfNeeded-recursive

alexey-milovidov · alexey-milovidov · commit b039c035224b · 2026-02-21T13:12:00.000+01:00
diff --git a/.github/workflows/retry_infra_failures.yml b/.github/workflows/retry_infra_failures.yml
@@ -52,7 +52,9 @@ jobs:
             # Collect per-job verdicts across all pages (runs can have >100 jobs).
             # Each failed job emits "true" (infrastructure) or "false" (real failure).
             # A job is considered an infrastructure failure if:
-            # - it is the "Config Workflow" job (pipeline configuration, not actual tests), or
+            # - it is the "Config Workflow" or "Finish Workflow" job (pipeline plumbing,
+            #   not actual tests; "Finish Workflow" runs in ~15s so the <120s heuristic
+            #   would always misclassify it as infra), or
             # - it never reached its main "Run" step (failed during checkout/setup), or
             # - the "Run" step was skipped, or
             # - the "Run" step failed almost immediately (under 2 minutes), indicating
@@ -61,7 +63,7 @@ jobs:
             verdicts=$(gh api "repos/$GH_REPO/actions/runs/$run_id/jobs?per_page=100" \
               --paginate --jq '
                 .jobs[] | select(.conclusion == "failure") |
-                if .name == "Config Workflow" then true
+                if .name == "Config Workflow" or .name == "Finish Workflow" then empty
                 else
                   [.steps[] | select(.name == "Run")] |
                   if length == 0 then true
diff --git a/tests/queries/0_stateless/03644_cancel_huge_mutation_subquery.sh b/tests/queries/0_stateless/03644_cancel_huge_mutation_subquery.sh
@@ -25,6 +25,12 @@ done
 
 $CLICKHOUSE_CLIENT  --query "SYSTEM STOP MERGES cancel_huge_mutation_subquery"
 
+# SYSTEM STOP MERGES prevents scheduling of new mutations, but if the background pool
+# hasn't picked up the mutation task yet, the mutation entry stays with is_done=0 and
+# no fail reason, causing mutations_sync=2 to wait forever. KILL MUTATION removes
+# the entry from the mutation list and notifies the wait event.
+$CLICKHOUSE_CLIENT --query "KILL MUTATION WHERE database='${CLICKHOUSE_DATABASE}' AND table='cancel_huge_mutation_subquery'" > /dev/null
+
 wait
 
 $CLICKHOUSE_CLIENT -q "DROP TABLE IF EXISTS cancel_huge_mutation_subquery"
