Skip to content

Add Apache Kylin 5.0.2-GA - #2384

Merged
alexey-milovidov merged 4 commits into
ClickHouse:mainfrom
KazukiKandaKK:add-kylin-5.0.2-ga-20261002
Oct 4, 2026
Merged

alexey-milovidov merged 4 commits into
ClickHouse:mainfrom
KazukiKandaKK:add-kylin-5.0.2-ga-20261002

Conversation

@KazukiKandaKK

Copy link
Copy Markdown
Contributor

Adds a kylin5/ entry: Apache Kylin 5.0.2-GA standalone (official Docker
image, Spark 3.3.0-kylin-5.2.2, Gluten 1.3.0-kylin-250110). No cube
is built. Every query is a raw pushdown query against the loaded table.

Ran the full load + 43×3 + concurrent-QPS on c6a.2xlarge against the
official 100M-row dataset. Load time was 421.406s; results are
tagged no-cold (no restart between tries).

38 of 43 queries completed on all 3 tries. The 5 failures are recorded
as null:

  • Q9, Q10: Gluten aborts with Memory limit exceeded ... maximum: 1.00 GiB
    on a grouped COUNT(DISTINCT UserID) with no WHERE filter.
  • Q19, Q29, Q40: Calcite rejects a SELECT-list alias referenced in
    GROUP BY (Column 'M'/'K'/'SRC' not found), at validation time
    before Gluten or Spark runs.

Queries containing OFFSET (Q39/41/42/43) completed; offset correctness
wasn't independently checked.

Concurrent-QPS (10 connections, 600s): 0.058 QPS / 0.167 error ratio
(7 errors: 3 match the sequential failures above, 4 are result-collection
interruptions seen only under concurrency).

A post-run COUNT(*) check returned the expected 99997497 rows.

apachekylin/apache-kylin-standalone:5.0.2-GA publishes an amd64 image
only; ./install fails outright on arm64 hosts.

Kylin's query cache stays on for lookup, but no successful pushdown
result is ever written to it at the shipped
kylin.query.pushdown.cache-enabled=false default; the
ResourceLimitExceededException cache path that bypasses the ordinary
exception-cache flag doesn't apply to this build's Gluten failures
either.

kylin5/start appends two canary config keys explicitly set to
false; checking the shipped source finds them likely redundant on
5.0.2-GA (one key matches no config in the source, the other already
defaults to false), not confirmed by a runtime test. The container runs at
Kylin's own documented -m 10G vendor default.

See kylin5/README.md for the full write-up.

Related to #1089.

Adds a kylin5/ entry: Apache Kylin 5.0.2-GA standalone (official Docker
image, Spark 3.3.0-kylin-5.2.2, Gluten 1.3.0-kylin-250110). No cube is
built; every query is a raw pushdown query against the loaded table.

Full load + 43x3 + concurrent-QPS run on c6a.2xlarge against the official
100M-row dataset. 38 of 43 queries completed on all 3 tries; 5 failures
are recorded as null (Gluten memory limit on Q9/Q10, Calcite GROUP BY
alias resolution on Q19/Q29/Q40).

See kylin5/README.md for the full write-up, including the known amd64-only
image limitation. The PR description covers additional -m 10G memory
testing across smaller instance sizes.
@alexey-milovidov alexey-milovidov added the machine:all PR benchmark on every machine type label Oct 4, 2026
@alexey-milovidov
alexey-milovidov deployed to benchmark-approval October 4, 2026 17:44 — with GitHub Actions Active
@github-actions

github-actions Bot commented Oct 4, 2026

Copy link
Copy Markdown
Contributor

The run of kylin5 on c8g.4xlarge did not produce results.
The run of kylin5 on c8g.metal-48xl did not produce results.

Logs:

@github-actions

github-actions Bot commented Oct 4, 2026

Copy link
Copy Markdown
Contributor

Results for kylin5 are ready for: c6a.2xlarge, c6a.4xlarge, c6a.large, c6a.metal, c6a.xlarge, c7a.metal-48xl.
The result files are committed as 7ba74a6.
Removed manually added result files: kylin5/results/20261001/c6a.2xlarge.json.

Logs:

@github-actions

github-actions Bot commented Oct 4, 2026

Copy link
Copy Markdown
Contributor

Results for kylin5 are ready for: t3a.small.
The result files are committed as e72911f.

Logs:

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@alexey-milovidov alexey-milovidov self-assigned this Oct 4, 2026
@alexey-milovidov
alexey-milovidov merged commit e56acc2 into ClickHouse:main Oct 4, 2026
1 of 2 checks passed

This branch is waiting to be deployed

1 waiting deployment
benchmark-approval — e691edc7 Waiting Oct 4, 2026 by alexey-milovidov via launch #660
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

machine:all PR benchmark on every machine type

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants