amaliujia opened a new pull request, #39057:
URL: https://github.com/apache/spark/pull/39057
<!--
Thanks for sending a pull request! Here are some tips for you:
1. If this is your first time, please read our contributor guidelines:
https://spark.apache.org/contributing.html
2. Ensure you have added or run the appropriate tests for your PR:
https://spark.apache.org/developer-tools.html
3. If the PR is unfinished, add '[WIP]' in your PR title, e.g.,
'[WIP][SPARK-XXXX] Your PR title ...'.
4. Be sure to keep the PR description updated to reflect all changes.
5. Please write your PR title to summarize what this PR proposes.
6. If possible, provide a concise example to reproduce the issue for a
faster review.
7. If you want to add a new configuration, please read the guideline first
for naming configurations in
'core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala'.
8. If you want to add or modify an error type or message, please read the
guideline first in
'core/src/main/resources/error/README.md'.
-->
### What changes were proposed in this pull request?
<!--
Please clarify what changes you are proposing. The purpose of this section
is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR. See the examples below.
1. If you refactor some codes with changing classes, showing the class
hierarchy will help reviewers.
2. If you fix some SQL features, you can provide some references of other
DBMSes.
3. If there is design documentation, please add the link.
4. If there is a discussion in the mailing list, please add the link.
-->
In current Spark optimizer, a single partition shuffle might be created for
a limit if this limit is not the last non-action operation (e.g. a filter
following the limit and the data size exceeds a limit). There is a possibility
that the previous output partitions before go into this limit are sorted. The
single partition shuffle approach has a correctness bug in this case: shuffle
read partitions could be out of partition order and the limit exec just take
the first limit rows which could lose the order thus result into wrong result.
This is a shuffle so it is relatively costly. Meanwhile, to correct this bug, a
native solution is to sort all the data fed into limit again, which is another
overhead.
So we propose a row count based AQE algorithm that optimizes this problem by
two folds:
So we propose a row count based AQE algorithm that optimizes this problem by
two folds:
Avoid the extra sort on the shuffle read side (or with the limit exec) to
achieve the correct result.
Avoid reading all shuffle data from mappers for this single partition
shuffle to reduce shuffle cost.
Note that 1. is only applied for the sorted partition case where 2. is
applied for general single partition shuffle + limit case
The algorithm works as the following:
1. Each mapper will record a row count when writing shuffle data.
2. Since this is single shuffle partition case, there is only one partition
but N mappers.
3. A accumulatorV2 is implemented to collect a list of tuple which records
the mapping between mapper id and the number of row written by the mapper (row
count metrics)
4. AQE framework detects a plan shape of shuffle plus a global limit.
5. AQE framework reads necessary data from mappers based on the limit. For
example, if mapper 1 writes 200 rows and mapper 2 writes 300 rows, and the
limit is 500, AQE creates shuffle read node to write from both mapper 1 and 2,
thus skip the left mappers.
6. This is both correct for limit with the sorted or non-sorted partitions.
This is the first step to implement the idea in
https://issues.apache.org/jira/browse/SPARK-41512, which is to implement a row
count accumulator that will be used to collect row
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
2. If you fix a bug, you can clarify why it is a bug.
-->
Implement the optimization algorithm for global limit with single partition
shuffle
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such as
the documentation fix.
If yes, please clarify the previous behavior and the change this PR proposes
- provide the console output, description and/or an example to show the
behavior difference if possible.
If possible, please also clarify if this is a user-facing change compared to
the released Spark versions or within the unreleased branches such as master.
If no, write 'No'.
-->
NO
### How was this patch tested?
<!--
If tests were added, say they were added here. Please make sure to add some
test cases that check the changes thoroughly including negative and positive
cases if possible.
If it was tested in a way different from regular unit tests, please clarify
how you tested step by step, ideally copy and paste-able, so that other
reviewers can test and check, and descendants can verify in the future.
If tests were not added, please describe why they were not added and/or why
it was difficult to add.
If benchmark tests were added, please run the benchmarks in GitHub Actions
for the consistent environment, and the instructions could accord to:
https://spark.apache.org/developer-tools.html#github-workflow-benchmarks.
-->
UT
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
