Support Interval data types in Arrow Format

[petern48]

Changelog category (leave one):

• Improvement

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

Users can now write Clickhouse interval datatypes to the Arrow Format

Documentation entry for user-facing changes

• Documentation is written (mandatory for new features)

Previously, queries that try to write interval data into Arrow format would encounter the following error.

Error on processing query: Code: 50. DB::Exception: The type 'IntervalNanosecond' of a column 'CAST('3', 'IntervalNanosecond')' is not supported for conversion into Arrow data format. (UNKNOWN_TYPE) (version 26.1.3.52 (official build))
(query: SELECT 3::IntervalNanosecond FORMAT ArrowStream;)

Now, these queries work successfully.

SELECT CAST('3', 'IntervalNanosecond')
FORMAT Arrow

Closes: #97849

[alexey-milovidov]

LGTM

[clickhouse-gh]

Workflow [PR], commit [10cdbe9]

Summary: ✅

---

AI Review

Summary

This PR adds Arrow support for ClickHouse Interval types in both directions: sub-second intervals (Nanosecond/Microsecond/Millisecond/Second) are exported as Arrow DURATION and imported back as interval types, while larger intervals remain INT64. The implementation and tests are coherent with that design, and I did not find correctness, safety, concurrency, or performance issues requiring changes.

Missing context

• ⚠️ Full CI evidence is not yet available (Style check, Build (arm_tidy), Fast test, and Code Review are still in progress), so this review is based on code and tests in the diff.

ClickHouse Rules

Item	Status	Notes
Deletion logging	➖
Serialization versioning	➖
Core-area scrutiny	✅
No test removal	✅
Experimental gate	➖
No magic constants	✅
Backward compatibility	✅
SettingsChangesHistory.cpp	➖
PR metadata quality	✅
Safe rollout	✅
Compilation time	✅

Final Verdict

• Status: ✅ Approve

[clickhouse-gh]

⚠️ This test only checks round-trip values, so it would still pass if the Arrow schema/type mapping regressed (for example, if IntervalNanosecond stopped being exported as Arrow DURATION(NANO) and became plain INT64).

Could we also assert the inferred types when reading back (e.g. via toTypeName/DESCRIBE on file('-', '$fmt')) so the test guards the new type mapping behavior, not only numeric payload preservation?

[petern48]

This was a really great catch. There was existing code that made it return int64 for all interval types. Just fixed it.

[petern48]

@alexey-milovidov I believe the remaining failures are unrelated. Can we move forward regardless, or do you need me to keep retrying until it's all green?

The test_polymorphic_parts failures are occurring in other PRs. The test failures for Integration tests (amd_llvm_coverage, 5/5) are all from timeouts, and are also occurring in other PRs.

[alexey-milovidov]

Sorry, can't merge with failed checks.

[alexey-milovidov]

But the PR is good and approved, so as soon as we will fix them, we can merge.

[clickhouse-gh]

LLVM Coverage Report

Metric	Baseline	Current	Δ
Lines	84.10%	84.10%	+0.00%
Functions	24.40%	24.40%	+0.00%
Branches	76.60%	76.70%	+0.10%

PR changed lines: PR changed-lines coverage: 92.59% (75/81, 0 noise lines excluded)
Diff coverage report
Uncovered code

[alexey-milovidov]

Thanks, this is a great change!

[EmeraldShift]

Any chance this can be backported to 26.3 LTS? It will be quite sad to not have this in the LTS release for a long time.