c21 opened a new pull request #29804:
URL: https://github.com/apache/spark/pull/29804
<!--
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'.
-->
### 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.
-->
This PR is to add support to decide bucketed table scan dynamically based on
actual query plan. Currently bucketing is enabled by default
(`spark.sql.sources.bucketing.enabled`=true), so for all bucketed tables in the
query plan, we will use bucket table scan (all input files per the bucket will
be read by same task). This has the drawback that if the bucket table scan is
not benefitting at all (no join/groupby/etc in the query), we don't need to use
bucket table scan as it would restrict the # of tasks to be # of buckets and
might hurt parallelism.
The feature is to add a physical plan rule right after `EnsureRequirements`:
The rule goes through plan nodes. For all operators which has "interesting
partition" (i.e., require `ClusteredDistribution` or
`HashClusteredDistribution`), check if the sub-plan for operator has `Exchange`
and bucketed table scan (and only allow certain operators in plan (i.e.
`Scan/Filter/Project/Sort/PartialAgg/etc`.), see details in
`PlanBucketing.canDisableBucketedScan`). If yes, disable the bucketed table
scan in the sub-plan.
Why the algorithm works is that if there's a shuffle between the bucketed
table scan and operator with interesting partition, then bucketed table scan
partitioning will be destroyed by the shuffle operator in the middle, and we
don't need bucketed table scan for sure.
The idea of "interesting partition" is inspired from "interesting order" in
"Access Path Selection in a Relational Database Management
System"(http://www.inf.ed.ac.uk/teaching/courses/adbs/AccessPath.pdf), after
discussion with @cloud-fan .
### 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.
-->
To avoid unnecessary bucketed scan in the query, and this is prerequisite
for https://github.com/apache/spark/pull/29625 (decide bucketed sorted scan
dynamically will be added later in that PR).
### 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'.
-->
A new config `spark.sql.sources.dynamic.decide.bucketing.enabled` is
introduced which set to true by default (the rule is enabled by default). User
can opt-out by disabling the rule, as we found in prod, some users rely on
assumption of # of tasks == # of buckets when reading bucket table to precisely
control # of tasks. This is a bad assumption but it does happen on our side, so
leave a config here to allow them opt-out for the feature.
### 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.
-->
Added unit tests in `BucketedReadSuite.scala`
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
