[
https://issues.apache.org/jira/browse/SPARK-48586?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Riya Verma updated SPARK-48586:
-------------------------------
Description: Currently the lock of the *RocksDB* state store is acquired
when uploading the snapshot inside maintenance tasks when change log
checkpointing is enabled, which causes lock contention between query processing
tasks and state maintenance thread. We want to eliminate lock contention to
decrease latency of streaming queries so lock acquisition inside maintenance
tasks should be avoided. This can introduce race conditions between task and
maintenance threads. By making a deep copy of {{versionToRocksDBFiles}} and
{{localFilesToDfsFiles}} in {*}RocksDBFileManager{*}, we can ensure that the
file manager state is not updated by task thread when background snapshot
uploading tasks attempt to upload a snapshot. (was: Currently the lock of the
*RocksDB* state store is acquired when uploading the snapshot inside
maintenance tasks when change log checkpointing is enabled, which causes lock
contention between query processing tasks and state maintenance thread. To
eliminate the lock contention, lock acquisition inside maintenance tasks should
be avoided. To prevent race conditions between task and maintenance threads, we
can ensure that *RocksDBFileManager* has a linear history by ensuring a deep
copy of *RocksDBFileManager* every time a previous version is loaded. The
original file manager is not affected by future state update. The new file
manager is not affected by background snapshot uploading tasks that attempt to
upload a snapshot.)
> Remove lock acquisition in doMaintenance() by making a deep copy of file
> mappings in RocksDBFileManager in load()
> -----------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-48586
> URL: https://issues.apache.org/jira/browse/SPARK-48586
> Project: Spark
> Issue Type: Improvement
> Components: Structured Streaming
> Affects Versions: 3.4.3
> Reporter: Riya Verma
> Priority: Major
> Labels: pull-request-available
>
> Currently the lock of the *RocksDB* state store is acquired when uploading
> the snapshot inside maintenance tasks when change log checkpointing is
> enabled, which causes lock contention between query processing tasks and
> state maintenance thread. We want to eliminate lock contention to decrease
> latency of streaming queries so lock acquisition inside maintenance tasks
> should be avoided. This can introduce race conditions between task and
> maintenance threads. By making a deep copy of {{versionToRocksDBFiles}} and
> {{localFilesToDfsFiles}} in {*}RocksDBFileManager{*}, we can ensure that the
> file manager state is not updated by task thread when background snapshot
> uploading tasks attempt to upload a snapshot.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
