Add a pass in Analyzer for time filter optimization with preimage #52091

ZhiguoZh · 2023-07-13T15:32:32Z

The preimage has been identified as the general solution to the
optimization of predicates with time converters (#15257). PR #50951
implemented this solution via the AST rewrite.

As a follow-up, this commit extends the optimization to the experi-
mental analyzer by replacing the sub-QueryTree of the time filter
with its preimage-transformed one. The optimization is implemented
as a new pass in the Analyzer.

Changelog category (leave one):

Performance Improvement

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

The performance experiments of SSB on the ICX device (Intel Xeon Platinum 8380 CPU, 80 cores, 160 threads) show that this change could bring an improvement of 8.5% to the geomean QPS when the experimental analyzer is enabled. The details are shown below:

robot-ch-test-poll · 2023-07-14T02:33:10Z

This is an automated comment for commit 9ff409f with description of existing statuses. It's updated for the latest CI running
The full report is available here
The overall status of the commit is 🟢 success

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
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	🟢 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
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
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

ZhiguoZh · 2023-07-19T13:57:46Z

A new issue is observed in the functional test 00515_enhanced_time_zones when the Analyzer is enabled by default. I've basically identified the root cause and will suspend the code review temporarily until the issue is fixed.

The preimage has been identified as the general solution to the optimization of predicates with time converters (ClickHouse#15257). PR ClickHouse#50951 implemented this solution via the AST rewrite. As a follow-up, this commit extends the optimization to the experi- mental analyzer by replacing the sub-QueryTree of the time filter with its preimage-transformed one. The optimization is implemented as a new pass in the Analyzer.

Date/Date32/DateTime/DateTime64 columns are required for arguments of time converters, such as toYear and toYYYYMM.

ZhiguoZh · 2023-07-25T02:57:24Z

Hi @hanfei1991, with the failed cases fixed, I think this PR is ready for the code review. This PR basically follows the implementation of the AST rewrite (#50951), and could you kindly help me review this one as well? Thanks a lot for your continuous help!

ZhiguoZh · 2023-08-01T13:08:48Z

Thanks for your review and acceptance! @hanfei1991

fix master ci for #52091

novikd · 2023-08-02T15:42:27Z

Why is this optimization not under the setting? There is no way to disable it. If it's a performance improvement why is there no new performance test?
Changing the predicate may prevent the application of read-in-order optimization.

novikd

The whole cpp file is written not like in the style guide. Very hard to read, especially because of huge amount of the copy-paste.

novikd · 2023-08-02T15:50:27Z

src/Analyzer/Passes/OptimizeDateOrDateTimeConverterWithPreimagePass.cpp

+        size_t literal_id = 1 - func_id;
+        const auto * literal = function->getArguments().getNodes()[literal_id]->as<ConstantNode>();
+
+        if (!literal || literal->getValue().getType() != Field::Types::UInt64) return;


Why is the field type used?

The literal in a SQL query is parsed as one of UInt64, Int64, Float64, String, Null (as ParserLiteral described),
and in this case, where the literal is compared with results from toYear or toYYYYMM converters, we expect the type of the literal to be UInt64, and decide to return early elsewise.

novikd · 2023-08-02T15:52:49Z

src/Analyzer/Passes/OptimizeDateOrDateTimeConverterWithPreimagePass.cpp

+            const auto lhs = std::make_shared<FunctionNode>("less");
+            lhs->getArguments().getNodes().push_back(std::make_shared<ColumnNode>(column_node.getColumn(), column_node.getColumnSource()));
+            lhs->getArguments().getNodes().push_back(std::make_shared<ConstantNode>(start_date_or_date_time));
+            resolveOrdinaryFunctionNode(*lhs, lhs->getFunctionName());
+
+            const auto rhs = std::make_shared<FunctionNode>("greaterOrEquals");
+            rhs->getArguments().getNodes().push_back(std::make_shared<ColumnNode>(column_node.getColumn(), column_node.getColumnSource()));
+            rhs->getArguments().getNodes().push_back(std::make_shared<ConstantNode>(end_date_or_date_time));
+            resolveOrdinaryFunctionNode(*rhs, rhs->getFunctionName());
+
+            const auto new_date_filter = std::make_shared<FunctionNode>("or");
+            new_date_filter->getArguments().getNodes() = {lhs, rhs};
+            resolveOrdinaryFunctionNode(*new_date_filter, new_date_filter->getFunctionName());
+
+            return new_date_filter;


Almost a complete copy-paste of equals case.

Thanks for pointing out this flaw! Maybe I should implement a utility function for constructing FunctionNode and have it build the subtrees of equals, notEquals, etc. instead, which would make the code neater.

ZhiguoZh · 2023-08-04T15:21:54Z

Hi Dmitry,

I think it's required to add this optimization under the settings if it unintendedly breaks another optimization, and I will soon send out another PR for it. Meanwhile, I'd also like to explore if this preimage optimization could co-work with the read-in-order one.

And exactly, this optimization is supported more by the performance data of SSB than a CI performance test. I'm not sure if it is still necessary to add the performance test in this case, but I will try to add one if so.

The code style is also to be refined in the upcoming PR, where the code duplication problem will be further mitigated.

Thanks for your comments! @novikd

Algunenano · 2024-02-27T13:01:09Z

The fuzzer is reporting a problem: https://s3.amazonaws.com/clickhouse-test-reports/0/f69bfa452633d2fcc2052a258f7202666053c787/ast_fuzzer__debug_.html

The comments from @novikid still stands by the way. The format of the file is bad, and there is no setting to disable this optimization. I'm having a look to see if it's something easy or I'll simply disable this completely until fixed

ZhiguoZh · 2024-02-27T14:01:29Z

Extremely sorry for failing to make progress in this code refinement in the last months due to lots of urgent internal tasks. I will try to revise the code accordingly in the coming days to 1) fix the reported problem 2) refine the code style 3) add new setting to turn the optimization on optionally. But it's still OK if it needs to be disabled temporarily before I could completely resolve the known issues.

Thanks for your comments and picking up this thread.

Algunenano · 2024-02-27T15:34:16Z

All 3 items should be covered by #60453

ZhiguoZh · 2024-02-28T11:09:52Z

Thanks for the fixes from which I've learned a lot. And I also noticed that the implementation of generating FunctionNode is not as elegant as expected. I'm working on the code refactoring locally and plan to send it out for your review after your PR gets merged.

ZhiguoZh · 2024-03-05T15:32:15Z

#60875 has been created for refactoring this part of code. Could you help me review it? Thanks in advance for your help!

@novikd @Algunenano

ZhiguoZh force-pushed the 20230713-analyzer-preimage branch from bfa81bb to c933d6a Compare July 13, 2023 15:34

alexey-milovidov added the can be tested Allows running workflows for external contributors label Jul 14, 2023

robot-ch-test-poll added the pr-performance Pull request with some performance improvements label Jul 14, 2023

ZhiguoZh force-pushed the 20230713-analyzer-preimage branch from c933d6a to 4b54ba8 Compare July 18, 2023 01:54

ZhiguoZh marked this pull request as draft July 19, 2023 13:52

ZhiguoZh added 2 commits July 24, 2023 13:48

Rename: extend the tests to QueryTree validation

eb8e7f4

ZhiguoZh force-pushed the 20230713-analyzer-preimage branch 2 times, most recently from 2195b70 to 9ff409f Compare July 24, 2023 12:58

Enforce type check for function arguments in preimage optimization

9ff409f

Date/Date32/DateTime/DateTime64 columns are required for arguments of time converters, such as toYear and toYYYYMM.

ZhiguoZh marked this pull request as ready for review July 24, 2023 14:20

hanfei1991 self-assigned this Jul 25, 2023

hanfei1991 approved these changes Aug 1, 2023

View reviewed changes

hanfei1991 merged commit b25847e into ClickHouse:master Aug 1, 2023
272 checks passed

hanfei1991 mentioned this pull request Aug 1, 2023

Revert "Add a pass in Analyzer for time filter optimization with preimage" #52872

Closed

hanfei1991 added a commit to hanfei1991/ClickHouse that referenced this pull request Aug 1, 2023

fix ci for ClickHouse#52091

4e849a8

hanfei1991 mentioned this pull request Aug 1, 2023

fix master ci for #52091 #52873

Merged

1 task

tavplubix added a commit that referenced this pull request Aug 1, 2023

Merge pull request #52873 from hanfei1991/hanfei/fix-master-ci

911a931

fix master ci for #52091

novikd reviewed Aug 2, 2023

View reviewed changes

canhld94 mentioned this pull request Aug 8, 2023

Query with array join and toYYYYMM in predicate doesn't work as expected #53152

Closed

ZhiguoZh mentioned this pull request Mar 5, 2024

Refactor OptimizeDateOrDateTimeConverterWithPreimageVisitor #60875

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add a pass in Analyzer for time filter optimization with preimage #52091

Add a pass in Analyzer for time filter optimization with preimage #52091

ZhiguoZh commented Jul 13, 2023

robot-ch-test-poll commented Jul 14, 2023 •

edited by robot-ch-test-poll4

ZhiguoZh commented Jul 19, 2023

ZhiguoZh commented Jul 25, 2023 •

edited

ZhiguoZh commented Aug 1, 2023

novikd commented Aug 2, 2023

novikd left a comment

novikd Aug 2, 2023

ZhiguoZh Aug 4, 2023

novikd Aug 2, 2023

ZhiguoZh Aug 4, 2023

ZhiguoZh commented Aug 4, 2023

Algunenano commented Feb 27, 2024

ZhiguoZh commented Feb 27, 2024

Algunenano commented Feb 27, 2024

ZhiguoZh commented Feb 28, 2024

ZhiguoZh commented Mar 5, 2024

Add a pass in Analyzer for time filter optimization with preimage #52091

Add a pass in Analyzer for time filter optimization with preimage #52091

Conversation

ZhiguoZh commented Jul 13, 2023

Changelog category (leave one):

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

robot-ch-test-poll commented Jul 14, 2023 • edited by robot-ch-test-poll4

ZhiguoZh commented Jul 19, 2023

ZhiguoZh commented Jul 25, 2023 • edited

ZhiguoZh commented Aug 1, 2023

novikd commented Aug 2, 2023

novikd left a comment

Choose a reason for hiding this comment

novikd Aug 2, 2023

Choose a reason for hiding this comment

ZhiguoZh Aug 4, 2023

Choose a reason for hiding this comment

novikd Aug 2, 2023

Choose a reason for hiding this comment

ZhiguoZh Aug 4, 2023

Choose a reason for hiding this comment

ZhiguoZh commented Aug 4, 2023

Algunenano commented Feb 27, 2024

ZhiguoZh commented Feb 27, 2024

Algunenano commented Feb 27, 2024

ZhiguoZh commented Feb 28, 2024

ZhiguoZh commented Mar 5, 2024

robot-ch-test-poll commented Jul 14, 2023 •

edited by robot-ch-test-poll4

ZhiguoZh commented Jul 25, 2023 •

edited