You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Stavros Kontopoulos (JIRA)" <ji...@apache.org> on 2019/06/12 18:18:00 UTC
[jira] [Comment Edited] (SPARK-28025) HDFSBackedStateStoreProvider
should not leak .crc files
[ https://issues.apache.org/jira/browse/SPARK-28025?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16862339#comment-16862339 ]
Stavros Kontopoulos edited comment on SPARK-28025 at 6/12/19 6:17 PM:
----------------------------------------------------------------------
There is a workaround by setting `--conf spark.hadoop.spark.sql.streaming.checkpointFileManagerClass=org.apache.spark.sql.execution.streaming.FileSystemBasedCheckpointFileManager` when using local fs
and modifying the FileSystemBasedCheckpointFileManager to run `fs.setWriteChecksum(false)` after fs is created.
Reason is the FileContextBasedCheckpointFileManager will use ChecksumFS ([https://github.com/apache/hadoop/blob/73746c5da76d5e39df131534a1ec35dfc5d2529b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/ChecksumFs.java]) under the hoods which will ignore "
CreateOpts.checksumParam(ChecksumOpt.createDisabled())" passed. These settings will only avoid creating checksums for the checksums themshelves.
FileSystemBasedCheckpointFileManager uses [https://github.com/apache/hadoop/blob/73746c5da76d5e39df131534a1ec35dfc5d2529b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/ChecksumFileSystem.java] which allows to avoid checksums when creating files.
Note that the crc is created when the tmp file is created not during rename or mv.
I will create a PR shortly.
was (Author: skonto):
There is a workaround by setting `--conf spark.hadoop.spark.sql.streaming.checkpointFileManagerClass=org.apache.spark.sql.execution.streaming.FileSystemBasedCheckpointFileManager` when using local fs
and modifying the FileSystemBasedCheckpointFileManager to run `fs.setWriteChecksum(false)` after fs is created.
Reason is the FileContextBasedCheckpointFileManager will use ChecksumFS ([https://github.com/apache/hadoop/blob/73746c5da76d5e39df131534a1ec35dfc5d2529b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/ChecksumFs.java]) under the hoods which will ignore "
CreateOpts.checksumParam(ChecksumOpt.createDisabled())" passed. These settings will only avoid creating checksums for the checksums themshelves.
FileSystemBasedCheckpointFileManager uses [https://github.com/apache/hadoop/blob/73746c5da76d5e39df131534a1ec35dfc5d2529b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/ChecksumFileSystem.java] which allows to avoid checksums when creating files.
Note that the crc is created when the tmp file is created not during rename or mv.
> HDFSBackedStateStoreProvider should not leak .crc files
> --------------------------------------------------------
>
> Key: SPARK-28025
> URL: https://issues.apache.org/jira/browse/SPARK-28025
> Project: Spark
> Issue Type: Bug
> Components: Structured Streaming
> Affects Versions: 2.4.3
> Environment: Spark 2.4.3
> Kubernetes 1.11(?) (OpenShift)
> StateStore storage on a mounted PVC. Viewed as a local filesystem by the `FileContextBasedCheckpointFileManager` :
> {noformat}
> scala> glusterfm.isLocal
> res17: Boolean = true{noformat}
> Reporter: Gerard Maas
> Priority: Major
>
> The HDFSBackedStateStoreProvider when using the default CheckpointFileManager is leaving '.crc' files behind. There's a .crc file created for each `atomicFile` operation of the CheckpointFileManager.
> Over time, the number of files becomes very large. It makes the state store file system constantly increase in size and, in our case, deteriorates the file system performance.
> Here's a sample of one of our spark storage volumes after 2 days of execution (4 stateful streaming jobs, each on a different sub-dir):
> #
> {noformat}
> Total files in PVC (used for checkpoints and state store)
> $find . | wc -l
> 431796
> # .crc files
> $find . -name "*.crc" | wc -l
> 418053{noformat}
> With each .crc file taking one storage block, the used storage runs into the GBs of data.
> These jobs are running on Kubernetes. Our shared storage provider, GlusterFS, shows serious performance deterioration with this large number of files:
> {noformat}
> DEBUG HDFSBackedStateStoreProvider: fetchFiles() took 29164ms{noformat}
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org