make nulls direction configuable for FullSortingMergeJoin

[loudongfeng]

Changelog category (leave one):

• Improvement

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

Add a flag for SMJ to treat null as biggest/smallest. So the behavior can be compitable with other SQL systems, like Apache Spark.

Documentation entry for user-facing changes

• Documentation is written (mandatory for new features)

Information about CI checks: https://clickhouse.com/docs/en/development/continuous-integration/

[CLAassistant]

All committers have signed the CLA.

[vdimir]

Thanks for your work on ClickHouse!

Could you please add more details about use-case?
Generally, NULLs should not affect join, because they are never joined.
In case you would like to have specific NULLs direction after join, you may just add ORDER BY with specific modifier, but join itself does not guarantee an order

[loudongfeng]

Thanks for you attention.
Actually, the use case is not for ClickHouse user, but for developers who want to use ClickHouse as a vectorization library.
For example, in Apache Gluten, we accelerate spark's Sort Merge Join by offloading it into ClickHouse 's SortingStep and JoinStep with FullSortingMergeJoin. The SortingStep offloading is optional(which means sorting can be done in spark's SortExec in nulls smallest order and then pass data to CH's FullSortingMergeJoin), so we need assign null orders for FullSortingMergeJoin in that case.
From ClickHouse's point of view, it's just an internal API richness.

[vdimir]

Let's then simply add an argument to the MergeJoinAlgorithm and define it as a constant somewhere outside. This will meet your requirements without introducing significant changes. Especially, adding it as a template argument is questionable since it increases the binary size, but nullableCompareAt already has a runtime argument now.

[vdimir]

Can't we pass direction as a value?

[loudongfeng]

fixed

[vdimir]

Let's not expose it to user

[loudongfeng]

fixed

[loudongfeng]

Hi vdimir, thanks for your review.
I have remove the template argument as you suggested.
But i'm not sure how to fix this one:

Let's then simply add an argument to the MergeJoinAlgorithm and define it as a constant somewhere outside.

Do you mean no changes in JoinStep,QueryPipelineBuilder,MergeJoinTransform etc and only add a runtime argument for MergeJoinAlgorithm?
But how to use a constant to set null_direction_hint, since it is unchangable.
Sample code that gluten uses to offloading SMJ to ClickHouse library:

JoinPtr smj_join = std::make_shared<FullSortingMergeJoin>(table_join, right->getCurrentDataStream().header.cloneEmpty());
QueryPlanStepPtr join_step
                 = std::make_unique<DB::JoinStep>(left->getCurrentDataStream(), right->getCurrentDataStream(), smj_join, 8192, 1, false);

[robot-ch-test-poll2]

This is an automated comment for commit 96e9043 with description of existing statuses. It's updated for the latest CI running

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

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
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	❌ failure
Integration tests	The integration tests report. In parenthesis the package type is given, and in square brackets are the optional part/total tests	❌ failure
Mergeable Check	Checks if all other necessary checks are successful	❌ failure
Stateless tests	Runs stateless functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	❌ failure

Successful checks

Check name	Description	Status
A Sync	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ success
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
Compatibility check	Checks that clickhouse binary runs on distributions with old libc versions. If it fails, ask a maintainer for help	✅ success
Docker keeper image	The check to build and optionally push the mentioned image to docker hub	✅ success
Docker server image	The check to build and optionally push the mentioned image to docker hub	✅ success
Docs check	Builds and tests the documentation	✅ 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
PR Check	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ 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
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

[vdimir]

Yes I mean just like that

auto null_direction = getNullDirection();

JoinPtr smj_join = std::make_shared<FullSortingMergeJoin>(table_join, right->getCurrentDataStream().header.cloneEmpty(), null_direction);
QueryPlanStepPtr join_step
                 = std::make_unique<DB::JoinStep>(left->getCurrentDataStream(), right->getCurrentDataStream(), smj_join, 8192, 1, false);

Also, to be honest I don't fully understand how it can be useful, because handling nulls is FullSortingMergeJoin implementation detail, and if you need particular order after that you may want add specific step further.

[loudongfeng]

Particular null order after join does not matter, but null order before Join matters.
I find nulls smaller/bigger matters after digging a result mismatch issue in gluten. Here is the case

-- a simple table
CREATE TABLE item (
  i_current_price DECIMAL(7,2),
  i_category STRING)
USING parquet;

-- data with nulls
insert into item values
(null,null),
(null,null),
(0.63,null),
(0.74,null),
(null,null),
(90.72,'Books'),
(99.89,'Books'),
(99.41,'Books');


-- the following SQL will use SMJ with join type Left Outer
-- expected result is 2, but gluten returns 0 when offloading to ClickHouse
-- notice there are nulls in join key 'i_category'
SELECT  count(*) cnt
FROM item i
where
i.i_current_price > 1.0 *
  (SELECT avg(j.i_current_price)
  FROM item j
  WHERE j.i_category = i.i_category
 )

The reason is sort node's order is asc and nulls first (both in Spark and ClickHouse's SortStep), but FullSortingMergeJoin joins streams treating data in asc order and nulls last. FullSortingMergeJoin will stop when find left/right table start with null, I think. Simply changing null direction in FullSortingMergeJoin can fix it.

[loudongfeng]

CI failures seems unrelated. Ready for review.
Thanks, @vdimir

[vdimir]

Thanks for your effort on the PR!