zhuzhurk commented on a change in pull request #9797: [FLINK-14059][core]
Introduce option allVerticesInSameSlotSharingGroupByDefault in ExecutionConfig
URL: https://github.com/apache/flink/pull/9797#discussion_r340561238
##########
File path:
flink-table/flink-table-planner-blink/src/test/scala/org/apache/flink/table/planner/runtime/utils/BatchTestBase.scala
##########
@@ -479,6 +479,6 @@ object BatchTestBase {
conf.getConfiguration.setString(TABLE_EXEC_RESOURCE_HASH_JOIN_MEMORY,
"2mb")
conf.getConfiguration.setString(TABLE_EXEC_RESOURCE_SORT_MEMORY, "1mb")
conf.getConfiguration.setString(TABLE_EXEC_RESOURCE_EXTERNAL_BUFFER_MEMORY,
"1mb")
- conf.getConfiguration.setString(TABLE_EXEC_SHUFFLE_MODE,
ShuffleMode.PIPELINED.toString)
+ conf.getConfiguration.setString(TABLE_EXEC_SHUFFLE_MODE,
ShuffleMode.BATCH.toString)
Review comment:
Yes, we can remove it.
The deadlock happens in `RankITCase.testRankValueFilterWithRange` and
`DistinctAggregateITCaseBase.testTwoDistinctAggWithGroupByAndCountStar` due to
a combination of multiple reasons.
1. blink batch executor disables allVerticesInSameSlotSharingGroupByDefault
(in this PR)
2. batch jobs in the given cases has multiple pipelined regions, it is a
bug(FLINK-13708)
3. multiple pipelined regions lead to multiple slot sharing groups if
allVerticesInSameSlotSharingGroupByDefault is disabled, as in #10007
I synced with @KurtYoung offline and the conclusion is blink will not
generate a test job with multiple regions in the near future. So the tests
would be Ok if FLINK-13708 is fixed. And we can drop this change in this way.
----------------------------------------------------------------
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]
With regards,
Apache Git Services
