Skip to content

Antalya 26.6: Bump to 26.6.4 - #2309

Open
zvonand wants to merge 324 commits into
antalya-26.6from
bump/antalya-26.6/26.6.4
Open

Antalya 26.6: Bump to 26.6.4#2309
zvonand wants to merge 324 commits into
antalya-26.6from
bump/antalya-26.6/26.6.4

Conversation

@zvonand

@zvonand zvonand commented Sep 3, 2026

Copy link
Copy Markdown
Member

Changelog category (leave one):

  • Not for changelog (changelog entry is not required)

CI/CD Options

Exclude tests:

  • Fast test
  • Integration Tests
  • Stateless tests
  • Stateful tests
  • Unit 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)
  • CAS (content-addressed storage; Antalya only)
  • 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)

robot-clickhouse and others added 30 commits August 10, 2026 13:12
…7366

Backport ClickHouse#107366 to 26.6: Reject TTL expressions referencing AggregateFunction columns at DDL time
… the split filter column name clashes with an input
…lineage field ids in native Parquet reader
…-index matching over a large dotted constant
…7377

Backport ClickHouse#107377 to 26.6: Tolerate Iceberg v3 reserved row-lineage field ids in native Parquet reader
…3390

Backport ClickHouse#113390 to 26.6: Fix ATTACH of Kafka tables with a large kafka_num_consumers
…gMetadata::iterate when datalake_table_state is missing
…le with an inconsistent bloom filter size
On 26.6, `EXPLAIN indexes = 1` does not print the combined
`Parts: 0 | Granules: 0` summary line that master's pretty plan
formatting emits, so the backported reference had one extra line.
The meaningful assertion — the JSON subcolumn index still prunes,
`Granules: 0/1000` — is unchanged.

CI: https://s3.amazonaws.com/clickhouse-test-reports/json.html?PR=114476&sha=9a63cb20d13792992441770495ca5610ca25ca62&name_0=BackportPR&name_1=Stateless%20tests%20%28amd_asan_ubsan%2C%20distributed%20plan%2C%20parallel%2C%202%2F2%29
PR: ClickHouse#114476

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two stateless tests from ClickHouse#113291 cannot run on 26.6 as cherry-picked, so
they fail before any assertion executes.

04672: join_runtime_filter_min_probe_rows does not exist on 26.6 (added by
ClickHouse#104860, settings history 26.8), so the SET at line 21 aborts the test with
UNKNOWN_SETTING. Dropping that one clause makes the line identical to the
merged 26.7 backport. max_bytes_ratio_before_external_join does exist here,
so it stays and 04029 needs no change.

04695: the t_utc block was removed from this branch but the trailing
DROP TABLE t_utc and the three matching reference lines were left behind,
so the test dies with UNKNOWN_TABLE.

Both files: EXPLAIN PLAN defaults to actions=1, compact=1, pretty=1 only
from 26.7 via explain_query_plan_default, which does not exist on 26.6, so
the references written for master do not match. Pinning the three options
explicitly restores the byte-exact output. All three are needed: compact
and pretty without actions collapse the plan and hide the sort description
rows.

Verified with clickhouse local on stock 26.6.3.30 and on a master build.
On 26.6 both tests now run to completion and every remaining difference is
the defect ClickHouse#113291 fixes. On master the same files match the references
byte for byte, so the pinning loses no coverage.
This is the only failing gtest on asan_ubsan and tsan. ClickHouse#113291 added the
TEST, but the monotonicity it asserts lives in src/Functions/toNullable.cpp
and came from ClickHouse#110121, which is labelled pr-performance and was not
backported: hasInformationAboutMonotonicity is absent from toNullable.cpp
on 26.6 and present on 26.7, which is why 26.7 went green.

Separate commit so it can be dropped if ClickHouse#110121 is backported instead.
…2601

Backport ClickHouse#112601 to 26.6: Iterate ColumnObject subcolumns in sorted path order
…icas read when the requested stage cannot absorb finalized data
alexey-milovidov and others added 27 commits August 29, 2026 12:07
… over a `Merge` table returning no rows for children that read from other tables
…serialization

`TextIndexPostingsDeserializationTest` constructed `PostingsSerialization`
with a null codec, but in 26.6 the constructor has `chassert(posting_list_codec)`,
so the test aborted in builds with sanitizers, where `chassert` is active
(on `master` the test goes through `PostingListCodecNone` directly and never
constructs `PostingsSerialization`). Pass the `None` codec instead: a posting
list without the `IsCompressed` flag never consults it, so the tested code
path is the same.

Unit tests (tsan): https://s3.amazonaws.com/clickhouse-test-reports/json.html?PR=116535&sha=e71cd5c2e4bb1ed299b4fb5feaaf87db7a350d94&name_0=BackportPR&name_1=Unit%20tests%20%28tsan%29
Related: ClickHouse#116535
…5334

Backport ClickHouse#115334 to 26.6: Do not let SET <name> = DEFAULT bypass the settings constraints
…8256

Backport ClickHouse#108256 to 26.6: Do not use projections for distributed reads in make_distributed_plan
…6371

Backport ClickHouse#116371 to 26.6: Fix a `_table`/`_database` filter over a `Merge` table returning no rows for children that read from other tables
…7511

Backport ClickHouse#107511 to 26.6: Fix LOGICAL_ERROR on UNION subquery with INTERSECT/EXCEPT children (old analyzer)
…9747

Backport ClickHouse#109747 to 26.6: Avoid server abort on type mismatch during partial result evaluation
…3478

Backport ClickHouse#103478 to 26.6: Push equi-key filter into the left input of RIGHT JOIN for cross-type USING keys
…5700

Backport ClickHouse#115700 to 26.6: Bound the recursion in several parsers and schema readers
… on an empty IN set over a Nullable column
…: guard count compare against empty read
…5222

Backport ClickHouse#115222 to 26.6: Bound text index posting list deserialization
…6538

Backport ClickHouse#116538 to 26.6: Restore short-circuit of a filter on an empty IN set over a Nullable column
Pull in the 26.6 fix for the flaky `test_kafka_formats_with_broken_message` (ClickHouse#116741).
…0480

Backport ClickHouse#110480 to 26.6: Fix flaky 03151_unload_index_race: guard count compare against empty read
… a subquery, a CTE, or a view to the settings constraints
…7011

Backport ClickHouse#117011 to 26.6: Do not allocate from sizes a peer declares in the Native protocol
…5397

Backport ClickHouse#115397 to 26.6: Clamp a SETTINGS clause nested in a subquery, a CTE, or a view to the settings constraints
…6851

Backport ClickHouse#116851 to 26.6: `formatDateTime`: Cache resolved time zones locally
…5701

Backport ClickHouse#115701 to 26.6: Validate aggregate function states on deserialization
…9764

Backport ClickHouse#109764 to 26.6: fix partition pruning for timestamp type
@github-actions

github-actions Bot commented Sep 3, 2026

Copy link
Copy Markdown

Workflow [PR], commit [6a9583a]

Adapt Antalya-specific code to upstream API changes brought by the
26.6.4 merge:

* `StorageObjectStorageConfiguration::partition_strategy_type` is now
  `private`, so `StorageObjectStorage::read` uses the public accessor
  `getPartitionStrategyType` instead.
* `DatabaseDataLake` keeps its settings in
  `MultiVersion<DatabaseDataLakeSettings> database_settings` rather than
  a plain `settings` member. `getCatalog` and `getConfiguration` still
  referred to `settings`; they now take a version and alias it the same
  way the rest of the file does.
* The new `gtest_rest_catalog` tests construct `RestCatalog` and
  `OneLakeCatalog` with the upstream signatures, which lack the Antalya
  `namespaces_` argument, and call `tryGetTableMetadata` without the
  Antalya `ContextPtr`. Pass both.

Only the first error appeared in the CI report because `ninja` stopped
before reaching the others.

CI report: https://altinity-build-artifacts.s3.amazonaws.com/json.html?PR=2309&sha=feba0a23cfd061a10d92b9b53056e7efbf3188cf&name_0=PR&name_1=Fast%20test
Related: #2309
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.