Multithreading after window functions

[frinkr]

Changelog category (leave one):

• Performance Improvement

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

Add option query_plan_preserve_num_streams_after_window_functions to preserve the number of streams after evaluating window functions to allow parallel stream processing.

Detail

The WindowTransform needs to merge streams into one, but the new option will split the result from WindowTransform to multiple streams, which enables the parallel processing of down-streams functions/aggregates.

Benchmark

Test data

CREATE TABLE window_test(id Int64, value Int64, partition Int64, msg String) Engine=MergeTree ORDER BY id;

INSERT INTO window_test SELECT number, rand(1) % 500, number % 3000,randomPrintableASCII(2) FROM numbers(50000000);

Test queries

-- query0: sipHash64 over window
SELECT id, AVG(value) OVER (ORDER BY id ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING) AS frame, sipHash64(frame) FROM window_test FORMAT Null;

-- query1: aggregate (by id) over window
SELECT id AS key, sipHash64(sum(frame)) AS value FROM (SELECT id, AVG(value) OVER (ORDER BY id ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING) AS frame FROM window_test) GROUP BY key ORDER BY key, value FORMAT Null;

-- query2: aggregate (by id % 100000) over window
SELECT id % 100000 AS key, sipHash64(sum(frame)) AS value FROM (SELECT id, AVG(value) OVER (ORDER BY id ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING) AS frame FROM window_test) GROUP BY key ORDER BY key, value FORMAT Null;

-- query3: aes enc/dec over window
WITH 'xxxxyyyyxxxxyyyyxxxxyyyyxxxxyyyy' AS cipherKey SELECT id, AVG(value) OVER (ORDER BY id ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING) AS frame, toString(frame) AS str,encrypt('aes-256-ofb', str, cipherKey) AS enc, decrypt('aes-256-ofb', str, cipherKey) AS dec FROM window_test FORMAT Null;

-- query4: window with partitions
SELECT id, AVG(value) OVER (PARTITION by partition ORDER BY id ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING) AS frame FROM window_test ORDER BY id FORMAT Null;

-- query5: window with partitions
SELECT DISTINCT AVG(value) OVER (PARTITION by partition ORDER BY id ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING) AS frame FROM window_test ORDER BY frame FORMAT Null;

Test Env

• CPU: AMD EPYC 7K62 8 Cores
• Memory: 32G
• Disk: 512G SSD
• OS: Ubuntu 22.04.2 LTS

Test Result

Query	OPTION	Elapsed	Memory	RPS
0	OFF	3.07s	257.98M	16.27M
0	ON	2.38s	359.06M	20.99M
1	OFF	10.67s	463.51M	4.78M
1	ON	6.618s	592.35M	7.56M
2	OFF	3.20s	15.99M	15.62M
2	ON	2.52s	18.98M	19.78M
3	OFF	13.918s	164.49M	3.59M
3	ON	3.902s	554.46M	12.81M
4	OFF	9.602s	138.74M	5.21M
4	ON	5.404s	102.53M	9.25M
5	OFF	3.89s	20.17M	12.85M
5	ON	3.55s	88.29M	14.07M

[robot-ch-test-poll2]

This is an automated comment for commit ac07b34 with description of existing statuses. It's updated for the latest CI running

❌ Click here to open a full report in a separate page

Successful checks

Check name	Description	Status
AST fuzzer	Runs randomly generated queries to catch program errors. The build type is optionally given in parenthesis. If it fails, ask a maintainer for help	✅ success
ClickHouse build check	Builds ClickHouse in various configurations for use in further steps. You have to fix the builds that fail. Build logs often has enough information to fix the error, but you might have to reproduce the failure locally. The cmake options can be found in the build log, grepping for cmake. Use these options and follow the general build process	✅ success
Compatibility check	Checks that clickhouse binary runs on distributions with old libc versions. If it fails, ask a maintainer for help	✅ success
Docker image for servers	The check to build and optionally push the mentioned image to docker hub	✅ success
Fast test	Normally this is the first check that is ran for a PR. It builds ClickHouse and runs most of stateless functional tests, omitting some. If it fails, further checks are not started until it is fixed. Look at the report to see which tests fail, then reproduce the failure locally as described here	✅ success
Flaky tests	Checks if new added or modified tests are flaky by running them repeatedly, in parallel, with more randomization. Functional tests are run 100 times with address sanitizer, and additional randomization of thread scheduling. Integrational tests are run up to 10 times. If at least once a new test has failed, or was too long, this check will be red. We don't allow flaky tests, read the doc	✅ success
Install packages	Checks that the built packages are installable in a clear environment	✅ success
Integration tests	The integration tests report. In parenthesis the package type is given, and in square brackets are the optional part/total tests	✅ success
Mergeable Check	Checks if all other necessary checks are successful	✅ success
Performance Comparison	Measure changes in query performance. The performance test report is described in detail here. In square brackets are the optional part/total tests	✅ success
Push to Dockerhub	The check for building and pushing the CI related docker images to docker hub	✅ success
SQLTest	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ success
SQLancer	Fuzzing tests that detect logical bugs with SQLancer tool	✅ success
Sqllogic	Run clickhouse on the sqllogic test set against sqlite and checks that all statements are passed	✅ success
Stateful tests	Runs stateful functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	✅ success
Stress test	Runs stateless functional tests concurrently from several clients to detect concurrency-related errors	✅ success
Style Check	Runs a set of checks to keep the code style clean. If some of tests failed, see the related log from the report	✅ success
Unit tests	Runs the unit tests for different release types	✅ success
Upgrade check	Runs stress tests on server version from last release and then tries to upgrade it to the version from the PR. It checks if the new server can successfully startup without any errors, crashes or sanitizer asserts	✅ success

Check name	Description	Status
CI running	A meta-check that indicates the running CI. Normally, it's in success or pending state. The failed status indicates some problems with the PR	⏳ pending
Stateless tests	Runs stateless functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	❌ failure

[nickitat]

also would be good to add perf tests

[nickitat]

01568_window_functions_distributed - I suppose we need similar logic to what we have for aggregation, when remote nodes pass buckets to initiator in order through a single stream.

[nickitat]

@frinkr a few tests left to fix. take a look pls

[frinkr]

@nickitat I didn't see how the performance failures (zeros_mt, hasAll, etc.) related to the PR.

[nickitat]

maybe getNumStreams()?

[frinkr]

getNumStreams() return 1 if there is a ORDER BY in the window,

[nickitat]

so, we don't preserve the number of stream, but rather always fan out after the WindowStep

[frinkr]

exactly.

[nickitat]

then I think we should name the setting accordingly

[nickitat]

pls create a perf test with the queries you've provided in the description

[frinkr]

@nickitat a new perf test 'window_functions_downstream_multithreading' was added, but the CI reports run errors. How do I know the error details? I run the perf with command performance-comparison/perf.py in my localhost, and no error. Maybe the test dataset is too large (50M rows)?

[ucasfl]

LGTM. @nickitat How is it going on?

[nickitat]

apart from the setting name lgtm

[frinkr]

How about query_plan_enable_multithreading_after_window_functions? @nickitat could you propose a proper name, please?

[nickitat]

How about query_plan_enable_multithreading_after_window_functions? @nickitat could you propose a proper name, please?

I'm ok with query_plan_enable_multithreading_after_window_functions

[frinkr]

@nickitat the setting was renamed.

[alexey-milovidov]

Ok, let's fix clang-tidy:

Oct 17 09:20:17 /build/src/Processors/QueryPlan/WindowStep.h:17:14: error: function 'DB::WindowStep::WindowStep' has a definition with different parameter names [readability-inconsistent-declaration-parameter-name,-warnings-as-errors]
Oct 17 09:20:17    17 |     explicit WindowStep(const DataStream & input_stream_,
Oct 17 09:20:17       |              ^
Oct 17 09:20:17 /build/src/Processors/QueryPlan/WindowStep.cpp:46:13: note: the definition seen here
Oct 17 09:20:17    46 | WindowStep::WindowStep(
Oct 17 09:20:17       |             ^
Oct 17 09:20:17 /build/src/Processors/QueryPlan/WindowStep.h:17:14: note: differing parameters are named here: ('streams_fan_out_'), in definition: ('preserve_num_streams_')
Oct 17 09:20:17    17 |     explicit WindowStep(const DataStream & input_stream_,
Oct 17 09:20:17       |              ^
Oct 17 09:20:17    18 |             const WindowDescription & window_description_,
Oct 17 09:20:17    19 |             const std::vector<WindowFunctionDescription> & window_functions_,
Oct 17 09:20:17    20 |             bool streams_fan_out_);
Oct 17 09:20:17       |                  ~~~~~~~~~~~~~~~~
Oct 17 09:20:17       |                  preserve_num_streams_

and it will be ready for merging.

[frinkr]

@alexey-milovidov the clang-tidy was fixed.