[zvonand]

Backport of ClickHouse#86933

Changelog category (leave one):

• Performance Improvement

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

Improves performance of fast queries with lots of parts in table (by optimizing MarkRanges by using devector over deque) (ClickHouse#86933 by @azat)

[github-actions]

Workflow [PR], commit [e93b59b]

[zvonand]

Backporter: CI failures at e93b59b

Failed checks:

• SourceUpload
• Stress test (amd_msan)
• Stateless tests (amd_ubsan, parallel)
• RegressionTestsAarch64 / S3 (azure, 1) / s3_azure_1

Unrelated to this PR: The PR only swaps the MarkRanges container from deque to devector; the failures are a Hybrid-table reload problem from an existing test, a known flaky ZooKeeper DROP_RANGE error, and infrastructure jobs.

• Stress test (amd_msan): not caused by the PR. The server failed to restart because the Hybrid table from the existing test 03644_hybrid_unqualified_table points at a dropped segment table, which is a Hybrid-engine reload problem.
• Stateless tests (amd_ubsan, parallel): not caused by the PR. The server aborted on the flaky ReplicatedMergeTree error "Part … remains in ZooKeeper after DROP_RANGE", and the tests after that failed because the server had died.
• SourceUpload and RegressionTestsAarch64 S3 (azure): not caused by the PR. Neither report has details, and both look like infrastructure or external-storage failures.
• Changes: none. The PR's MarkRanges container swap built successfully on every configuration.

[zvonand]

@blau-ai

[blau-ai]

CI triage for #2638 (backport of upstream ClickHouse#86933, MarkRanges: devector over deque)

Verdict: 6 red checks — 0 caused by this code change. 1 is a DCO/sign-off formality, the rest are pre-existing/infra/flaky failures that also hit the antalya-25.8 base branch (every recent merge into it is red). The actual diff is clean: it only touches the MergeTree read path (swaps the MarkRanges base container std::deque → boost::container::devector and adjusts three iterators to auto * since devector iterators are raw pointers). All 11 builds, all 4 unit-test variants, the Fast test, and every stateless suite including all S3-storage ones passed.

Check	Classification	Cause
DCO	⚠️ process	Missing Signed-off-by on the upstream merge commit
Stateless tests (amd_ubsan, parallel)	✅ not PR	Pre-existing ReplicatedMergeTree assertion crash
Stress test (amd_msan)	✅ not PR	Hybrid engine startup dependency-order bug
RegressionTestsAarch64 / S3 (azure, 1)	✅ not PR (infra/flaky)	Could not retrieve failing test; no S3 report
SourceUpload	✅ not PR (infra)	Known-flaky source-upload step on this branch
PR	—	Aggregate gate; just rolls up the above

Details

DCO — the head commit e93b59b ("Merge pull request ClickHouse#86933 from azat/…", author Azat Khuzhin) has no Signed-off-by: line, so cncf/dco2 fails. This is the only item the author needs to act on, and it's purely administrative — not a code problem. Add a sign-off (e.g. git commit --amend -s / re-push with sign-off, per Altinity's DCO policy).

Stateless tests (amd_ubsan, parallel) — reported as "test runner terminated unexpectedly". The server aborted on a logical error, and the test 02344_distinct_limit_distiributed was merely the victim ("Reason: server died / Connection refused"). Decisive line:

<Fatal> : Logical error: 'Part all_110_110_0 from table test_8ew2jlfi.alter_table0 (...) remains in ZooKeeper after DROP_RANGE all_0_110_999999999_999999999'.

The abort is in StorageReplicatedMergeTree::paranoidCheckForCoveredPartsInZooKeeper (StorageReplicatedMergeTree.cpp:2775, via executeDropRange) — a replicated-table ZooKeeper sanity check in a background queue thread. This code path has nothing to do with MarkRanges/read-path marks. Not PR-caused; safe to re-run.

Stress test (amd_msan) — "Cannot start clickhouse-server". The server fails to load metadata because the Altinity-specific Hybrid engine can't attach a table whose segment table doesn't exist yet at load time:

Load job 'load table test_13.test_hybrid_unqualified' failed: Code: 36 ...
table 'test_13.test_hybrid_unqualified_segment' does not exist:
Cannot attach ... ENGINE = Hybrid(remote('localhost:9000', 'system.numbers'), number = 0, test_13.test_hybrid_unqualified_segment, number >= 10)

This is a load-ordering/dependency bug in the Hybrid engine (Altinity-specific), unrelated to this diff. Not PR-caused.

RegressionTestsAarch64 / S3 (azure, 1) — the Actions annotation is only "Process completed with exit code 1" and the job log isn't downloadable via gh (stream CANCEL); no praktika/regression report was published at the expected S3 path (404), so I could not identify the failing test name and won't over-claim. That said: all S3-storage stateless suites passed, this Azure suite depends on external cloud connectivity, and the base branch is red on every merge — so this is most likely infra/flaky rather than from a read-path container swap. Suggest a re-run; if it persists, it needs a human to pull the regression logs.

SourceUpload — infra step (source tarball upload); log not downloadable. It's a known-flaky area on antalya-25.8 (recent base merges include #2635 25.8-antalya-sourcefix and #2642 submodule-cache targeting exactly this). Not PR-caused; re-run.

Bottom line

Nothing in the C++ change needs fixing — it builds everywhere and passes all functional/unit/S3 tests. The only author action is adding the DCO sign-off; the remaining red checks are pre-existing/infra and can be re-run (ideally after the base branch's own CI is green again).