CI: remove leftover standalone arm_ubsan build by Algunenano · Pull Request #109048 · ClickHouse/ClickHouse · GitHub
Skip to content

CI: remove leftover standalone arm_ubsan build#109048

Merged
Algunenano merged 2 commits into
ClickHouse:masterfrom
Algunenano:remove-standalone-arm-ubsan-build
Jul 2, 2026
Merged

CI: remove leftover standalone arm_ubsan build#109048
Algunenano merged 2 commits into
ClickHouse:masterfrom
Algunenano:remove-standalone-arm-ubsan-build

Conversation

@Algunenano

@Algunenano Algunenano commented Jul 1, 2026 •
edited by robot-ch-test-poll4
Loading

Copy link
Copy Markdown
Member

The standalone ARM UBSan-only build (-DSANITIZE=undefined) is a leftover. It was added in a5fca9d ("add all aarch64 variants") shortly before 774a9ed combined ASan+UBSan into single builds and removed the standalone UBSan/ASan builds. That commit missed arm_ubsan.

UBSan on ARM is already covered by arm_asan_ubsan, so the standalone build and its stress test were redundant CI cost with no test coverage depending on them. This removes the ARM_UBSAN build type, its CH_ARM_UBSAN/DEB_ARM_UBSAN artifacts, the build job, and the arm_ubsan stress test, and regenerates the workflow YAMLs.

Changelog category (leave one):

  • CI Fix or improvement

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

Not for changelog (changelog entry is not required)

Version info

  • Merged into: 26.7.1.422

Algunenano and others added 2 commits July 1, 2026 13:33
@Algunenano @claude
CI: remove leftover standalone arm_ubsan build
2a2a6cd 
The standalone ARM UBSan-only build (`-DSANITIZE=undefined`) was a leftover.
It was added in a5fca9d ("add all aarch64 variants") shortly before
774a9ed combined ASan+UBSan into single builds and removed the standalone
UBSan/ASan builds; that commit missed `arm_ubsan`. UBSan on ARM is already
covered by `arm_asan_ubsan`, so the standalone build and its stress test were
redundant CI cost with no test coverage depending on them.

Remove the `ARM_UBSAN` build type, its `CH_ARM_UBSAN`/`DEB_ARM_UBSAN`
artifacts, the build job, and the `arm_ubsan` stress test. Regenerate the
workflow YAMLs.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@Algunenano
Merge remote-tracking branch 'blessed/master' into remove-standalone-…
facf048 
…arm-ubsan-build
@clickhouse-gh

clickhouse-gh Bot commented Jul 1, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

@clickhouse-gh clickhouse-gh Bot added the pr-ci label Jul 1, 2026
@Algunenano Algunenano requested review from fm4v and maxknv July 1, 2026 14:38
@clickhouse-gh

clickhouse-gh Bot commented Jul 1, 2026

Copy link
Copy Markdown
Contributor

📊 Cloud Performance Report

🔴 AI verdict: degradation — 2 regressed, 1 improved out of 38 analysed

This PR only removes the arm_ubsan build and stress-test jobs from the GitHub Actions CI workflows; there are no server or query-execution code changes. Because nothing in the diff can touch how queries run, the flagged clickbench deltas (Q4, Q27, Q30, Q31, Q33, Q34) and the tpch Q8 change are run-to-run variance and have been downgraded to inconclusive. Two TPC-H queries (Q4 +193%, Q7 +786%) and Q16 (-50%) show very large shifts that the automated trust gate preserves, but their baselines were extremely noisy (base variation of roughly 46%, 103%, and 23%), so a rerun is recommended before treating them as real.

clickbench

⚠️ 7 inconclusive

Flagged queries (7 of 43)
Query Verdict Baseline median (ms) PR median (ms) Change q-value Hint
⚠️ 4 not_sure 259 215 -17.0% <0.0001 PR only deletes an arm_ubsan CI build/stress job; nothing touches query execution. -17% is run-to-run variance.
⚠️ 27 not_sure 192 158 -17.4% <0.0001 CI-only diff (drops arm_ubsan jobs); execution path untouched. -17% not attributable to this PR.
⚠️ 30 not_sure 240 275 +14.6% <0.0001 CI-only diff removes an arm_ubsan build; cannot slow a SELECT. Treat +15% as variance.
⚠️ 31 not_sure 308 358 +15.9% <0.0001 Same CI-only change; no server code touched. +16% is run-to-run variance.
⚠️ 32 not_sure 1306 1226 -6.1% 0.0018 Already inconclusive and the two tests disagree; consistent with a CI-only diff.
⚠️ 33 not_sure 1452 1376 -5.2% <0.0001 CI-only diff; -5% cannot come from removing an arm_ubsan CI job. Run-to-run variance.
⚠️ 34 not_sure 1455 1376 -5.4% <0.0001 CI-only diff; -5% is variance, not a PR effect.

q-value = BH-FDR adjusted p; smaller is stronger evidence. MIRAI flags a query when q < fdr_q (default 0.10) — the value the verdict is based on.

tpch_adapted_1_official

🔴 2 regressed · 🟢 1 improved · ⚠️ 1 inconclusive

Flagged queries (4 of 22)
Query Verdict Baseline median (ms) PR median (ms) Change q-value Hint
🔴 4 regression 326 956 +193.1% <0.0001 Large +193% delta stands per trust gate, but baseline was very noisy (base variation ~46%) and diff is CI-only; rerun ad
🔴 7 regression 71 629 +785.9% <0.0001 +786% over a 71ms baseline with extremely noisy base measurements (~103%); delta stands per gate but rerun advised.
🟢 16 improvement 100 50 -50.0% <0.0001 -50% large delta stands per trust gate, but base was noisy (~23%) and a CI-only diff can't explain it; rerun advised.
⚠️ 8 not_sure 168 209 +24.4% <0.0001 CI-only diff (removes arm_ubsan jobs); server execution untouched. +24% not attributable to this PR.

q-value = BH-FDR adjusted p; smaller is stronger evidence. MIRAI flags a query when q < fdr_q (default 0.10) — the value the verdict is based on.

Debug info
  • StressHouse run: 1c0246ce-bd82-4432-9584-c76033377b33
  • MIRAI run: ea2faf19-d632-4475-b1ef-fdcfe496317e
  • PR check IDs:
    • clickbench_631908_1782919325
    • clickbench_631914_1782919324
    • clickbench_631924_1782919324
    • tpch_adapted_1_official_631932_1782919324
    • tpch_adapted_1_official_631944_1782919324
    • tpch_adapted_1_official_631957_1782919324

@Algunenano

Algunenano commented Jul 1, 2026

Copy link
Copy Markdown
Member Author

CI failure investigation

This PR is CI-only (removes the redundant standalone arm_ubsan build; touches only .github/workflows/*.yml and ci/defs/*.py), so neither failure can be a regression from it. Both are known TSan-build flakies with open tracking issues and remediation in progress.

Failure (check) Verdict Master freq 7/14/30/90d Last master fail Tracking issue In-flight fix
ThreadSanitizer: data race (STID: 3336-2c6d) (BuzzHouse amd_tsan) FLAKY 3 / 4 / 8 / 17 2026-06-30 #101319 (open) #103158 (open)
Hung check failed, possible deadlock found (Stress arm_tsan) FLAKY 152 / 235 / 430 / 1096 2026-07-01 #107941 (open) #107640, #108212, #108351 (merged); #105905 (open)

1. Data race STID 3336-2c6d

Tracked by #101319 (exact STID match). Root cause: Context::setCurrentTransaction (src/Interpreters/Context.cpp:7215) writes merge_tree_transaction unsynchronized while other threads read it under Context::mutex via Context::createCopy; the std::swap<DB::MergeTreeTransaction*> in the TSan SUMMARY is that write. Fixed by open PR #103158, which stops the server-side AST fuzzer from clearing the caller's session/query transaction without a lock.

2. Hung check (possible deadlock)

Tracked by #107941, an umbrella for stress/fuzzer hangs and extremely common on master (1096/90d), so a single hit is noise. Root cause: the server-side AST fuzzer generating unbounded/heavy in-flight queries that ignore cancellation. Fixes already merged (#107640, #108212, #108351); diagnostic #105905 still open. The remaining tail is a single in-flight write/generate that can't be interrupted mid-task.

Recommendation: re-run the two failed checks. Both are pre-existing flakies unrelated to this PR.

@Algunenano Algunenano enabled auto-merge July 2, 2026 10:04
@Algunenano Algunenano added this pull request to the merge queue Jul 2, 2026
Merged via the queue into ClickHouse:master with commit c3f8ef3 Jul 2, 2026
342 of 344 checks passed
@Algunenano Algunenano deleted the remove-standalone-arm-ubsan-build branch July 2, 2026 10:20
@robot-ch-test-poll3 robot-ch-test-poll3 added the pr-synced-to-cloud The PR is synced to the cloud repo label Jul 2, 2026
@alexey-milovidov alexey-milovidov mentioned this pull request Jul 2, 2026
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@fm4v fm4v fm4v approved these changes
@maxknv maxknv maxknv approved these changes

Assignees

@fm4v fm4v

Labels

pr-ci pr-synced-to-cloud The PR is synced to the cloud repo

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@Algunenano @fm4v @maxknv @robot-ch-test-poll3