Uh oh!
There was an error while loading. Please reload this page.
[SPARK-26941][YARN]Fix incorrect computation of maxNumExecutorFailures in ApplicationMaster for streaming - #23845
Conversation
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
There was a problem hiding this comment.
For other reviewers -- this is the fix itself.
Uh oh!
There was an error while loading. Please reload this page.
SparkQA
commented
Feb 22, 2019
Test build #4565 has finished for PR 23845 at commit
|
srowen
commented
Feb 25, 2019
@liupc could you have a look at the build failure? |
SparkQA
commented
Feb 26, 2019
Test build #4575 has finished for PR 23845 at commit
|
liupc
commented
Mar 1, 2019
retest this please! |
liupc
commented
Mar 1, 2019
@srowen UT seems failed with some unrelated tests, can you help to retest this please? |
SparkQA
commented
Mar 1, 2019
Test build #4583 has finished for PR 23845 at commit
|
liupc
commented
Mar 5, 2019
retest this please |
1 similar comment
liupc
commented
Mar 6, 2019
retest this please |
liupc
commented
Mar 6, 2019
SparkQA
commented
Mar 6, 2019
Test build #4598 has started for PR 23845 at commit |
SparkQA
commented
Mar 9, 2019
Test build #4603 has finished for PR 23845 at commit
|
SparkQA
commented
Mar 11, 2019
Test build #4608 has finished for PR 23845 at commit
|
Hm, that's weird. The failure is not directly related, and I thought we fixed this in #23887 Not sure what's going on. |
SparkQA
commented
Mar 11, 2019
Test build #4609 has started for PR 23845 at commit |
SparkQA
commented
Mar 12, 2019
Test build #4612 has finished for PR 23845 at commit
|
vanzin
left a comment
There was a problem hiding this comment.
I'm really not familiar with streaming dynamic allocation, but I assume you can't enable both that and regular dynamic allocation together?
If for some reason that is allowed, then the fix should take the max of both values.
To fix the tests, probably needs a merge with master. Maybe github is confused.
There was a problem hiding this comment.
Could you add these to a new Streaming.scala instead? We shouldn't be adding more stuff to package.scala.
There was a problem hiding this comment.
If for some reason that is allowed, then the fix should take the max of both values.
This is not allowed to enable both of them. This check is done in the scheduler/ExecutorAllocationManager
I agree to put these configs to Streaming.scala, it's more clear. I will update.
f825118 to
882bd35Comparevanzin
commented
Mar 13, 2019
ok to test |
SparkQA
commented
Mar 13, 2019
Test build #103459 has finished for PR 23845 at commit
|
SparkQA
commented
Mar 15, 2019
Test build #4625 has finished for PR 23845 at commit
|
srowen
commented
Mar 15, 2019
@liupc can you take a look at ... |
SparkQA
commented
Mar 16, 2019
Test build #103570 has finished for PR 23845 at commit
|
srowen
commented
Mar 17, 2019
merged to master |
What changes were proposed in this pull request?
Currently, when enabled streaming dynamic allocation for streaming applications, the maxNumExecutorFailures in ApplicationMaster is still computed with
spark.dynamicAllocation.maxExecutors.Actually, we should consider
spark.streaming.dynamicAllocation.maxExecutorsinstead.Related codes:
spark/resource-managers/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala
Line 101 in f87153a
How was this patch tested?
NA
Please review http://spark.apache.org/contributing.html before opening a pull request.