[
https://issues.apache.org/jira/browse/SPARK-59907?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jasraj Dhanoa resolved SPARK-59907.
-----------------------------------
Resolution: Won't Fix
Resolving as Won't Fix. The three paths described can leave a buffer
unreleased, but a fix would only matter if the failure were recoverable (the
executor keeps running tasks afterward). With UnsafeRowSerializer, the
realistic failure on these paths is a direct-memory allocation failure
(OutOfDirectMemoryError), which is fatal: the executor exits through
SparkUncaughtExceptionHandler, and its memory is freed on restart. A persistent
leak would require a separate non-fatal bug, so the added defensive handling
isn't worth it.
> StreamingShuffleWriter leaks in-flight off-heap buffer when write fails
> -----------------------------------------------------------------------
>
> Key: SPARK-59907
> URL: https://issues.apache.org/jira/browse/SPARK-59907
> Project: Spark
> Issue Type: Bug
> Components: Structured Streaming
> Affects Versions: 5.0.0
> Reporter: Jasraj Dhanoa
> Priority: Minor
> Labels: pull-request-available
>
> When StreamingShuffleWriter.write() fails partway through, the in-flight
> off-heap buffer can be left unreleased: either it is referenced only by a
> local variable, or the send completion callback that releases it never runs.
> In both cases task completion cleanup does not release it. By default these
> buffers have no cleaner, so GC does not reclaim them either.
> This only leaves a lasting leak when the failure is a non-fatal exception: a
> fatal error such as OutOfDirectMemoryError makes the executor exit, which
> frees its memory anyway. With UnsafeRowSerializer these paths are not
> expected to throw a non-fatal exception in normal operation, so this is a
> defensive fix. If one did occur, the buffer would stay allocated until the
> executor restarts.
> Failure scenarios:
> # The serializer throws while writing a record.
> # Closing the serialization stream fails before a buffer is sent.
> # Allocating or encoding the outgoing network frame fails before it is handed
> to the client, so the send completion callback never runs.
> Expected behavior: when write() fails, StreamingShuffleWriter should release
> the in-flight buffer and return its memory budget before rethrowing the
> exception.
> Reproduction (tested on 5.0.0-SNAPSHOT, master at commit 44929778453): run a
> streaming shuffle write with a serializer that throws on a given record
> (scenario 1) or on stream close (scenario 2). After the task fails, the
> captured buffer still has refCnt 1 and one buffer's worth of memory budget is
> not returned. For scenario 3, send a DataMessage whose data buffer throws
> while being encoded: the send completion callback is never invoked, so the
> buffer and its memory budget are never returned. With the fix, the buffer is
> released and the budget is fully returned in all three cases. Unit tests
> reproducing all three scenarios are included in the PR.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
