[SPARK-49673][CONNECT] Increase CONNECT_GRPC_ARROW_MAX_BATCH_SIZE to 0.7 * CONNECT_GRPC_MAX_MESSAGE_SIZE #48122
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
dillitz
changed the title
kamcheungting-db
approved these changes
Sep 16, 2024
HyukjinKwon
reviewed
Sep 17, 2024
...connect/server/src/main/scala/org/apache/spark/sql/connect/planner/SparkConnectPlanner.scala
Outdated
Show resolved
Hide resolved
dillitz
changed the title
nija-at
reviewed
Sep 17, 2024
Contributor
nija-at
left a comment
nija-at
left a comment
There was a problem hiding this comment.
Any existing tests that use the new limits to ensure that everything still works?
Contributor
Author
I've added an E2E test that explicitly tests multiple batches and changed the static limit to 10MiB there to not create too much memory pressure. |
nija-at
approved these changes
Sep 17, 2024
connector/connect/client/jvm/src/test/scala/org/apache/spark/sql/ClientE2ETestSuite.scala
Show resolved
Hide resolved
HyukjinKwon
changed the title
HyukjinKwon
approved these changes
Sep 18, 2024
Member
|
Merged to master. |
attilapiros
pushed a commit
to attilapiros/spark
that referenced
this pull request
Oct 4, 2024
…0.7 * CONNECT_GRPC_MAX_MESSAGE_SIZE ### What changes were proposed in this pull request? Increases the default `maxBatchSize` from 4MiB * 0.7 to 128MiB (= CONNECT_GRPC_MAX_MESSAGE_SIZE) * 0.7. This makes better use of the allowed maximum message size. This limit is used when creating Arrow batches for the `SqlCommandResult` in the `SparkConnectPlanner` and for `ExecutePlanResponse.ArrowBatch` in `processAsArrowBatches`. This, for example, lets us return much larger `LocalRelations` in the `SqlCommandResult` (i.e., for the `SHOW PARTITIONS` command) while still staying within the GRPC message size limit. ### Why are the changes needed? There are `SqlCommandResults` that exceed 0.7 * 4MiB. ### Does this PR introduce _any_ user-facing change? Now support `SqlCommandResults` <= 0.7 * 128 MiB instead of only <= 0.7 * 4MiB and ExecutePlanResponses will now better use the limit of 128MiB. ### How was this patch tested? Existing tests. ### Was this patch authored or co-authored using generative AI tooling? No. Closes apache#48122 from dillitz/increase-sql-command-batch-size. Authored-by: Robert Dillitz <[email protected]> Signed-off-by: Hyukjin Kwon <[email protected]>
himadripal
pushed a commit
to himadripal/spark
that referenced
this pull request
Oct 19, 2024
…0.7 * CONNECT_GRPC_MAX_MESSAGE_SIZE ### What changes were proposed in this pull request? Increases the default `maxBatchSize` from 4MiB * 0.7 to 128MiB (= CONNECT_GRPC_MAX_MESSAGE_SIZE) * 0.7. This makes better use of the allowed maximum message size. This limit is used when creating Arrow batches for the `SqlCommandResult` in the `SparkConnectPlanner` and for `ExecutePlanResponse.ArrowBatch` in `processAsArrowBatches`. This, for example, lets us return much larger `LocalRelations` in the `SqlCommandResult` (i.e., for the `SHOW PARTITIONS` command) while still staying within the GRPC message size limit. ### Why are the changes needed? There are `SqlCommandResults` that exceed 0.7 * 4MiB. ### Does this PR introduce _any_ user-facing change? Now support `SqlCommandResults` <= 0.7 * 128 MiB instead of only <= 0.7 * 4MiB and ExecutePlanResponses will now better use the limit of 128MiB. ### How was this patch tested? Existing tests. ### Was this patch authored or co-authored using generative AI tooling? No. Closes apache#48122 from dillitz/increase-sql-command-batch-size. Authored-by: Robert Dillitz <[email protected]> Signed-off-by: Hyukjin Kwon <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What changes were proposed in this pull request?
Increases the default
maxBatchSizefrom 4MiB * 0.7 to 128MiB (=CONNECT_GRPC_MAX_MESSAGE_SIZE) * 0.7. This makes better use of the allowed maximum message size.
This limit is used when creating Arrow batches for the
SqlCommandResultin theSparkConnectPlannerand forExecutePlanResponse.ArrowBatchinprocessAsArrowBatches. This, for example, lets us return much largerLocalRelationsin theSqlCommandResult(i.e., for theSHOW PARTITIONScommand) while still staying within the GRPC message size limit.Why are the changes needed?
There are
SqlCommandResultsthat exceed 0.7 * 4MiB.Does this PR introduce any user-facing change?
Now support
SqlCommandResults<= 0.7 * 128 MiB instead of only <= 0.7 * 4MiB and ExecutePlanResponses will now better use the limit of 128MiB.How was this patch tested?
Existing tests.
Was this patch authored or co-authored using generative AI tooling?
No.