Antalya 26.3: Fix rescheduleTasksFromReplica

[zvonand]

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Fix rescheduleTasksFromReplica (#1568 by @ianton-ru).

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Performance tests
• All with ASAN
• All with TSAN
• All with MSAN
• All with UBSAN
• All with Coverage
• All with Aarch64
• All Regression
• Disable CI Cache

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

Cherry-picked from #1568.

---

Documentation entry for user-facing changes

Fix incorrect change from c523f29
getReplicaForFile uses replica_to_files_to_be_processed to find best replica for file, With removing lost replica after getReplicaForFile call, getReplicaForFile chooses the same replica, so rescheduling makes no sense, files will be choosen only in getAnyUnprocessedFile and executed on random replicas.
This PR fixes the order, now files are matched with new best replicas.

[github-actions]

Workflow [PR], commit [c2744ab]

[alsugiliazova]

Verification report: Altinity/ClickHouse PR #1747

---

Conclusion

PR is merged. PR test workflow is clean at the test level on head; only the chronic antalya-26.3 regression suites are red, all at baseline rates. No PR-caused regression found. The diff is 3 lines in rescheduleTasksFromReplica — minimal blast radius.

Caveat — partial frontport (same as the rest of the 26.3 cohort). Companion antalya-26.1 frontports are still missing on antalya-26.3; chronic regression failures here are branch-level missing-dependency symptoms. Iceberg sort key timezone continues to fail with the same UNRECOGNIZED_ARGUMENTS: '--iceberg_partition_timezone' from a binary that doesn't expose the option.

---

CI on head c2744abe — failures

PR test workflow

0 test-level FAILs in gh-data.checks on head SHA. All PR test jobs that run tests are green at the test level. (45 success / 50 skipped / 5 failure in the workflow rollup is dominated by Build/Workflow housekeeping, no actual test breakage.)

Regression workflow (10 failed checks)

Check	Top failing tests on PR-1747 builds (30d)	Baseline (antalya-26.3, 30d)	Class
Swarms (Release + Aarch64)	swarm joins / join clause, cluster discovery / multiple paths, node failure / network failure, node failure / cpu overload, swarm join sanity / join with clause (×2 each)	30–44% on every PR	Pre-existing broken
S3Export (partition) (Release + Aarch64)	sanity / no partition by (×2)	50%	Pre-existing broken
Iceberg (1) (Release + Aarch64)	rest catalog / sort key timezone / day transform utc (×2), rest catalog / iceberg iterator race condition (×2)	41% / 28%	Missing-dep + pre-existing flaky
Iceberg (2) (Release + Aarch64)	chronic glue-catalog / race-condition variants	chronic	Pre-existing flaky
Parquet (Release + Aarch64)	postgresql/mysql round-trip compression-type variants (×2 each)	~36%	Pre-existing flaky

Regression DB on /PRs/1747/ builds (30d): 152 Fail / 5,358 OK ≈ 2.8%. Every top failure matches the all-PR baseline fail rate on antalya-26.3.

---

Related to PR diff?

PR is a 3-line fix in rescheduleTasksFromReplica (1 file, replicated-task scheduling path).

Failing test	Diff overlap	Related?
swarms / *, parquet / *, s3_export_partition / *, iceberg / *	none — none of these suites exercise rescheduleTasksFromReplica	No

No failing test intersects the rescheduling code path.

---

Recommendations

1. No action on this PR. Merged and effectively clean — PR test workflow is green at the test level, and the regression failures are 100% chronic baseline.
2. Re-verify after the companion 26.1 → 26.3 frontports land — same list as the prior 26.3 verification reports.
3. Same chronic-baseline cleanup recommendation as VERIFICATION_PR_1640.md for swarms / parquet / s3_export_partition / iceberg scenarios.

---

Local checkout

cd /Users/alsugilyazova/workspace/altinity-clickhouse/ClickHouse
gh pr checkout 1747 --repo Altinity/ClickHouse
# HEAD: c2744abe886243da0b4e82745ba22dbd7c198c27

[alsugiliazova]

Audit: PR #1747 — Antalya 26.3: Fix rescheduleTasksFromReplica

AI audit note: This review comment was generated by AI (Cursor agent, audit-review skill).

Audit update for PR #1747 (rescheduleTasksFromReplica ordering)

Confirmed defects

No confirmed defects in reviewed scope.