[SPARK-42794][SS] Increase the lockAcquireTimeoutMs to 2 minutes for acquiring the RocksDB state store in Structure Streaming #40425
Conversation
This file contains 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
…ring the RocksDB state store in Structure Streaming
sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/RocksDB.scala
Show resolved
Hide resolved
|
cc @HeartSaVioR |
|
Thanks! Merging to master. |
Kimahriman
pushed a commit
to Kimahriman/spark
that referenced
this pull request
Jun 20, 2023
…acquiring the RocksDB state store in Structure Streaming We are seeing query failure which is caused by RocksDB acquisition failure for the retry tasks. * at t1, we shrink the cluster to only have one executor ``` 23/03/05 22:47:21 INFO StandaloneAppClient$ClientEndpoint: Executor updated: app-20230305224215-0000/2 is now DECOMMISSIONED (worker decommissioned because of kill request from HTTP endpoint (data migration disabled)) 23/03/05 22:47:21 INFO StandaloneAppClient$ClientEndpoint: Executor updated: app-20230305224215-0000/3 is now DECOMMISSIONED (worker decommissioned because of kill request from HTTP endpoint (data migration disabled)) ``` * at t1+2min, task 7 at its first attempt (i.e. task 7.0) is scheduled to the alive executor ``` 23/03/05 22:49:58 INFO TaskSetManager: Starting task 7.0 in stage 133.0 (TID 685) (10.166.225.249, executor 0, partition 7, ANY, ``` It seems that task 7.0 is able to pass dataRDD.iterator(partition, ctxt) and acquires the rocksdb lock as we are seeing ``` 23/03/05 22:51:59 WARN TaskSetManager: Lost task 4.1 in stage 133.1 (TID 700) (10.166.225.249 executor 0): java.lang.IllegalStateException: StateStoreId(opId=0,partId=7,name=default): RocksDB instance could not be acquired by [ThreadId: Some(50), task: partition 7.1 in stage 133.1, TID 700] as it was not released by [ThreadId: Some(449), task: partition 7.0 in stage 133.0, TID 685] after 60003 ms. 23/03/05 22:52:59 WARN TaskSetManager: Lost task 4.2 in stage 133.1 (TID 702) (10.166.225.249 executor 0): java.lang.IllegalStateException: StateStoreId(opId=0,partId=7,name=default): RocksDB instance could not be acquired by [ThreadId: Some(1495), task: partition 7.2 in stage 133.1, TID 702] as it was not released by [ThreadId: Some(449), task: partition 7.0 in stage 133.0, TID 685] after 60006 ms. 23/03/05 22:53:59 WARN TaskSetManager: Lost task 4.3 in stage 133.1 (TID 704) (10.166.225.249 executor 0): java.lang.IllegalStateException: StateStoreId(opId=0,partId=7,name=default): RocksDB instance could not be acquired by [ThreadId: Some(46), task: partition 7.3 in stage 133.1, TID 704] as it was not released by [ThreadId: Some(449), task: partition 7.0 in stage 133.0, TID 685] after 60003 ms. ``` Increasing the lockAcquireTimeoutMs to 2 minutes such that 4 task retries will give us 8 minutes to acquire the lock and it is larger than connectionTimeout with retries (3 * 120s). ### What changes were proposed in this pull request? Increase the lockAcquireTimeoutMs to 2 minutes for acquiring the RocksDB state store in Structure Streaming ### Why are the changes needed? hanging the thread for lock acquisition rather than giving up easily ### Does this PR introduce _any_ user-facing change? NO ### How was this patch tested? Trivial change Closes apache#40425 from huanliwang-db/increase-timeout. Authored-by: Huanli Wang <huanli.wang@databricks.com> Signed-off-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
snmvaughan
pushed a commit
to snmvaughan/spark
that referenced
this pull request
Jun 20, 2023
…acquiring the RocksDB state store in Structure Streaming We are seeing query failure which is caused by RocksDB acquisition failure for the retry tasks. * at t1, we shrink the cluster to only have one executor ``` 23/03/05 22:47:21 INFO StandaloneAppClient$ClientEndpoint: Executor updated: app-20230305224215-0000/2 is now DECOMMISSIONED (worker decommissioned because of kill request from HTTP endpoint (data migration disabled)) 23/03/05 22:47:21 INFO StandaloneAppClient$ClientEndpoint: Executor updated: app-20230305224215-0000/3 is now DECOMMISSIONED (worker decommissioned because of kill request from HTTP endpoint (data migration disabled)) ``` * at t1+2min, task 7 at its first attempt (i.e. task 7.0) is scheduled to the alive executor ``` 23/03/05 22:49:58 INFO TaskSetManager: Starting task 7.0 in stage 133.0 (TID 685) (10.166.225.249, executor 0, partition 7, ANY, ``` It seems that task 7.0 is able to pass dataRDD.iterator(partition, ctxt) and acquires the rocksdb lock as we are seeing ``` 23/03/05 22:51:59 WARN TaskSetManager: Lost task 4.1 in stage 133.1 (TID 700) (10.166.225.249 executor 0): java.lang.IllegalStateException: StateStoreId(opId=0,partId=7,name=default): RocksDB instance could not be acquired by [ThreadId: Some(50), task: partition 7.1 in stage 133.1, TID 700] as it was not released by [ThreadId: Some(449), task: partition 7.0 in stage 133.0, TID 685] after 60003 ms. 23/03/05 22:52:59 WARN TaskSetManager: Lost task 4.2 in stage 133.1 (TID 702) (10.166.225.249 executor 0): java.lang.IllegalStateException: StateStoreId(opId=0,partId=7,name=default): RocksDB instance could not be acquired by [ThreadId: Some(1495), task: partition 7.2 in stage 133.1, TID 702] as it was not released by [ThreadId: Some(449), task: partition 7.0 in stage 133.0, TID 685] after 60006 ms. 23/03/05 22:53:59 WARN TaskSetManager: Lost task 4.3 in stage 133.1 (TID 704) (10.166.225.249 executor 0): java.lang.IllegalStateException: StateStoreId(opId=0,partId=7,name=default): RocksDB instance could not be acquired by [ThreadId: Some(46), task: partition 7.3 in stage 133.1, TID 704] as it was not released by [ThreadId: Some(449), task: partition 7.0 in stage 133.0, TID 685] after 60003 ms. ``` Increasing the lockAcquireTimeoutMs to 2 minutes such that 4 task retries will give us 8 minutes to acquire the lock and it is larger than connectionTimeout with retries (3 * 120s). ### What changes were proposed in this pull request? Increase the lockAcquireTimeoutMs to 2 minutes for acquiring the RocksDB state store in Structure Streaming ### Why are the changes needed? hanging the thread for lock acquisition rather than giving up easily ### Does this PR introduce _any_ user-facing change? NO ### How was this patch tested? Trivial change Closes apache#40425 from huanliwang-db/increase-timeout. Authored-by: Huanli Wang <huanli.wang@databricks.com> Signed-off-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
We are seeing query failure which is caused by RocksDB acquisition failure for the retry tasks.
It seems that task 7.0 is able to pass dataRDD.iterator(partition, ctxt) and acquires the rocksdb lock as we are seeing
Increasing the lockAcquireTimeoutMs to 2 minutes such that 4 task retries will give us 8 minutes to acquire the lock and it is larger than connectionTimeout with retries (3 * 120s).
What changes were proposed in this pull request?
Increase the lockAcquireTimeoutMs to 2 minutes for acquiring the RocksDB state store in Structure Streaming
Why are the changes needed?
hanging the thread for lock acquisition rather than giving up easily
Does this PR introduce any user-facing change?
NO
How was this patch tested?
Trivial change