Optimize array element function when input is array(map)/array(array(num)/array(array(string))/array(bigint)/array(decimal)

[taiyang-li]

Changelog category (leave one):

• Performance Improvement

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

Optimize array element function when input is array(map)/array(array(num)/array(array(string))/array(bigint)/array(decimal). Current implementation causes too many reallocs. The optimization speed up by ~6x especially when input type is array(map)

[robot-ch-test-poll1]

This is an automated comment for commit 231de4a 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
ClickBench	Runs [ClickBench](https://github.com/ClickHouse/ClickBench/) with instant-attach table	✅ 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
Docs check	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ 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
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
Stateless tests	Runs stateless 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
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	❌ failure
Stateful tests	Runs stateful functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	❌ error

[taiyang-li]

Some tests

Optimize when input type is array(array(num))

Q: select materialize(array(array(1,2,3,4)))[1] from numbers(10000000) format Null;

Before:

0 rows in set. Elapsed: 0.551 sec. Processed 10.01 million rows, 80.06 MB (18.18 million rows/s., 145.41 MB/s.)
Peak memory usage: 4.26 MiB.
0 rows in set. Elapsed: 0.528 sec. Processed 10.01 million rows, 80.06 MB (18.95 million rows/s., 151.63 MB/s.)
Peak memory usage: 4.26 MiB.
0 rows in set. Elapsed: 0.566 sec. Processed 10.01 million rows, 80.06 MB (17.67 million rows/s., 141.35 MB/s.)
Peak memory usage: 4.25 MiB.

After

0 rows in set. Elapsed: 0.464 sec. Processed 10.01 million rows, 80.06 MB (21.55 million rows/s., 172.41 MB/s.)
Peak memory usage: 4.25 MiB.
0 rows in set. Elapsed: 0.455 sec. Processed 10.01 million rows, 80.06 MB (21.99 million rows/s., 175.93 MB/s.)
Peak memory usage: 4.25 MiB.
0 rows in set. Elapsed: 0.470 sec. Processed 10.01 million rows, 80.06 MB (21.28 million rows/s., 170.22 MB/s.)
Peak memory usage: 4.25 MiB

Optimize when input type is array(array(string))

Q: select materialize(array(array('hello', 'world')))[1] from numbers(10000000) format Null

Before:

0 rows in set. Elapsed: 0.824 sec. Processed 10.01 million rows, 80.06 MB (12.14 million rows/s., 97.16 MB/s.)
Peak memory usage: 12.07 MiB.
0 rows in set. Elapsed: 0.827 sec. Processed 10.01 million rows, 80.06 MB (12.10 million rows/s., 96.78 MB/s.)
Peak memory usage: 12.07 MiB.
0 rows in set. Elapsed: 0.842 sec. Processed 10.01 million rows, 80.06 MB (11.88 million rows/s., 95.06 MB/s.)
Peak memory usage: 12.07 MiB. 

After:

0 rows in set. Elapsed: 0.722 sec. Processed 10.01 million rows, 80.06 MB (13.87 million rows/s., 110.94 MB/s.)
Peak memory usage: 12.06 MiB.
0 rows in set. Elapsed: 0.712 sec. Processed 10.01 million rows, 80.06 MB (14.05 million rows/s., 112.37 MB/s.)
Peak memory usage: 12.06 MiB.
0 rows in set. Elapsed: 0.722 sec. Processed 10.01 million rows, 80.06 MB (13.86 million rows/s., 110.86 MB/s.)
Peak memory usage: 12.06 MiB.

Optimize when input type is map with its key/value type string/num

Q: select materialize(array(map('hello', 1, 'world', 2)))[1] from numbers(10000000) format Null;

Before:

0 rows in set. Elapsed: 6.504 sec. Processed 10.01 million rows, 80.06 MB (1.54 million rows/s., 12.31 MB/s.)
Peak memory usage: 12.58 MiB.
0 rows in set. Elapsed: 6.394 sec. Processed 10.01 million rows, 80.06 MB (1.57 million rows/s., 12.52 MB/s.)
Peak memory usage: 12.58 MiB.
0 rows in set. Elapsed: 6.398 sec. Processed 10.01 million rows, 80.06 MB (1.56 million rows/s., 12.51 MB/s.)
Peak memory usage: 12.58 MiB.

After:

0 rows in set. Elapsed: 1.029 sec. Processed 10.01 million rows, 80.06 MB (9.72 million rows/s., 77.79 MB/s.)
Peak memory usage: 12.95 MiB.
0 rows in set. Elapsed: 1.010 sec. Processed 10.01 million rows, 80.06 MB (9.91 million rows/s., 79.26 MB/s.)
Peak memory usage: 12.95 MiB.
0 rows in set. Elapsed: 1.032 sec. Processed 10.01 million rows, 80.06 MB (9.70 million rows/s., 77.58 MB/s.)
Peak memory usage: 12.95 MiB.

[taiyang-li]

@alexey-milovidov can you help review it, thanks very much!

[taiyang-li]

@Avogar can you help review it, thanks very much !

[alexey-milovidov]

It looks very promising; I didn't have time to review it yet.

[taiyang-li]

It looks very promising; I didn't have time to review it yet.

It doesn't matter. I'll wait for you..

[taiyang-li]

Do we have any progress recently ?

[taiyang-li]

Do we have any progress recently ?

[Avogar]

I will review this PR this week. Sorry for long wait.

[Avogar]

Actually, I think we can generalize new implementations a bit to support any nested arrays, not only 2 level arrays. We will just need to use a vector of nested and result offsets. Not sure if there will be an overhead for 2 level arrays because of iterating through vector with single element. WDYT?
Current code LGTM, only one small comment.

Also need to make new perf tests faster, they are too slow: https://s3.amazonaws.com/clickhouse-test-reports/56403/4bbc0db1540bae020fecc1f1d8ff3bcc4eb191c2/performance_comparison_[1_4]/report.html

[taiyang-li]

Actually, I think we can generalize new implementations a bit to support any nested arrays, not only 2 level arrays. We will just need to use a vector of nested and result offsets. Not sure if there will be an overhead for 2 level arrays because of iterating through vector with single element. WDYT? Current code LGTM, only one small comment.

Also need to make new perf tests faster, they are too slow: https://s3.amazonaws.com/clickhouse-test-reports/56403/4bbc0db1540bae020fecc1f1d8ff3bcc4eb191c2/performance_comparison_[1_4]/report.html

I'll commit again, and refersh the perf tests(current result seems not precise, similar situations happended in other prs before). You thoughts about support any nested arrsys is interesting, but I'm consused about how to implement it, do you have any example.

[Avogar]

You thoughts about support any nested arrsys is interesting, but I'm consused about how to implement it, do you have any example.

I started to write an explanation, but it turned out to be difficult to explain without just writing this code :)
Let's merge current implementation, it's also really good. I will try to implement optimization for arrays with any level by myself.

[Avogar]

Perf test still complains about array_element tests, they are too long, we should make them run under 1 second

[taiyang-li]

[taiyang-li]

@Avogar I don't known why there are always two errors about parallel_index in perf tests. I'll merge master and observe it again.

[Avogar]

Error in parallel_index - #58593

[Avogar]

I tried to implement my idea for any nested arrays, and it's not worth it, almost no improvement.