Antalya 26.6 Export partition - check name and pos of columns in partition key

[k-morozov]

From 26.3 #2134

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):

Added a check for the positions and names of columns that make up the partition key (including Tuple elements and nested Array/Map types) during ALTER TABLE ... EXPORT PART/PARTITION.

Documentation entry for user-facing changes

...

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Performance tests
• Aarch64 tests
• All with ASAN
• All with TSAN
• All with MSAN
• All with UBSAN
• All with Coverage
• 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)
• OAuth (5m)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

[github-actions]

Workflow [PR], commit [155107d]

[k-morozov]

Some regression tests failed - export_merge_tree_part_schema_mismatch_mode will be added in next PR.

[mkmkme]

The last CI run seemed stuck due to GH outage. I've merged the base into the PR to trigger the new CI run

[mkmkme]

@blau-ai

[blau-ai]

CI triage for #2218

Verdict: 6 red checks — 0 caused by this PR. All are pre-existing / infra / flaky / regression-suite-vs-build version skew. The PR's own new validation error strings (... is at position N in the source table, different Tuple element layout) appear 0 times across the failure logs, and this PR touches only docs/, the 4 EXPORT C++/header files, and 4 tests/integration/* files — none of the suites below.

I can't build/run ClickHouse here, so this is evidence-from-CI (downloaded the fails.log.txt / nice-new-fails.log.txt artifacts for each failing job).

Check	Class	Root cause
S3Export (part) / s3_export_part	pre-existing (stale test)	stale expected exit codes + REPLACE PARTITION guard
Iceberg (2) / iceberg_2	version skew / infra	suite uses a setting not in this build; stale exit codes; leftover tables
Swarms / swarms	flaky/infra	distributed node-restart timing
SQLLogic test	infra/flaky	ran below the minimum test-count threshold
Stateless tests (amd_debug, parallel)	infra/external	DeltaLake kernel + startup connection errors
PR	rollup	aggregate of the above, no independent error

---

RegressionTestsRelease / S3Export (part) / s3_export_part — 5 scenarios

Not PR-related. Two distinct pre-existing issues in the (separately-maintained) testflows suite:

• error handling/pending mutations and error handling/pending patch parts — the server does exactly the right thing:

  Code: 1009. DB::Exception: Part 1_1_1_0 can not be exported because there are pending mutations ... (PENDING_MUTATIONS_NOT_ALLOWED)
  

  Client exit code is 241 (= 1009 % 256), but the test asserts results[0].exitcode == 237. The semantics/message are correct; only the hardcoded expected exit code (237) is stale in the regression suite.
• concurrent alter/{before,after,stress} export → example=replace partition — fails on an unrelated REPLACE PARTITION safety guard:

  Code: 36. ... refusing `REPLACE PARTITION` because it would silently drop the destination partition's data.
  Set `allow_replace_partition_from_empty_source = 1` ... (BAD_ARGUMENTS)
  

  This is the empty-source REPLACE PARTITION guard, nothing to do with EXPORT schema validation.

RegressionTestsRelease / Iceberg (2) / iceberg_2 — 86/207 features, 239 scenarios

Not PR-related — regression-suite ahead of this build. Representative log lines:

Code: 552. DB::Exception: Unrecognized option '--export_merge_tree_part_schema_mismatch_mode' ... (UNRECOGNIZED_ARGUMENTS)
Expected duplicate-export client exit 36 or 238, got 242:
  Code: 1010. ... (EXPORT_PARTITION_ALREADY_EXPORTED)          # 1010 % 256 = 242, stale expected codes
Code: 57.  ... Iceberg table with path data/iceberg_.../ already exists. (TABLE_ALREADY_EXISTS)   # leftover state / isolation

I confirmed export_merge_tree_part_schema_mismatch_mode does not exist anywhere in src/ on this branch — the suite in Altinity/clickhouse-regression expects newer server features than this build carries. Even the basic "accepted"/"sanity" export cases fail on TABLE_ALREADY_EXISTS (test-isolation), not on export logic.

RegressionTestsRelease / Swarms / swarms — 1 scenario

Not PR-related. node failure/check restart swarm node: a long query issued while a swarm node is restarted expected client exit 138, got 0 (query completed). Timing-dependent distributed-resiliency test → flaky/infra.

SQLLogic test

Not PR-related. Threshold gate tripped: FAILED: total tests 4,468,123 < minimum 5,939,581. The run executed far fewer tests than the required minimum (partial run/timeout); the failed-test count was within tolerance. Infra/flaky, and SQLLogic doesn't exercise EXPORT PART.

Stateless tests (amd_debug, parallel) — Failed 1, Broken 3

Not PR-related. Failures are external-dependency/DeltaLake, plus a server-startup blip:

Code: 742. DB::Exception: Received DeltaLake kernel error    # 03441_deltalake_clickhouse_* tests
Code: 210. DB::NetException: Connection refused (localhost:9000). (NETWORK_ERROR)   # startup, retried

DeltaLake read tests + a startup connection retry — infra/external, unrelated to this PR's EXPORT PART changes.

---

Suggested next steps

• No code change is needed in this PR to address these — none are its fault.
• Safe to re-run Swarms, SQLLogic test, and Stateless tests (amd_debug, parallel) (flaky/infra/external).
• s3_export_part and iceberg_2 are failing on stale expected exit codes and a setting (--export_merge_tree_part_schema_mismatch_mode) that isn't in this build — these live in Altinity/clickhouse-regression and need to be updated there (or the build needs to catch up); they can't be fixed from this repo PR. Worth confirming with the regression maintainers that both are already red on antalya-26.6 for the same reasons.

Note: I inferred "pre-existing" from the root causes above (suite references a non-existent setting, stale hardcoded exit codes, external deps, and the PR's own error strings being absent from every failure log) rather than from a direct base-branch run comparison — I couldn't locate a matching base-branch antalya-26.6 run to diff against.