[
https://issues.apache.org/jira/browse/HADOOP-15358?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mikhail Pryakhin updated HADOOP-15358:
--------------------------------------
Description:
Methods of SFTPFileSystem operate on poolable ChannelSftp instances, thus some
methods of SFTPFileSystem are chained together resulting in establishing
multiple connections to the SFTP server to accomplish one compound action,
those methods are listed below:
# mkdirs method
the public mkdirs method acquires a new ChannelSftp [from the
pool|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L658]]
and then recursively creates directories, checking for the directory existence
beforehand by calling the method
[exists|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L321]
] which delegates to the getFileStatus(ChannelSftp channel, Path file)
[method|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L202]]
and so on until it ends up in returning the [FilesStatus
instance|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L290]].
The resource leakage occurs in the method getWorkingDirectory which calls the
getHomeDirectory
[method|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L640]]
which in turn establishes a new connection to the sftp server instead of using
an already created connection. As the mkdirs method is recursive this results
in creating a huge number of connections.
# open
[method|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L504]].
This method returns an instance of FSDataInputStream which consumes
SFTPInputStream instance which doesn't return an acquired ChannelSftp instance
back to the pool but instead it
[closes|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPInputStream.java#L123]]
it. This leads to establishing another connection to an SFTP server when the
next method is called on the FileSystem instance.
was:
Methods of SFTPFileSystem operate on poolable ChannelSftp instances, thus some
methods of SFTPFileSystem are chained together resulting in establishing
multiple connections to the SFTP server to accomplish one compound action,
those methods are listed below:
# mkdirs method
the public mkdirs method acquires a new ChannelSftp from the pool [1]
and then recursively creates directories, checking for the directory existence
beforehand by calling the method exists[2] which delegates to the
getFileStatus(ChannelSftp channel, Path file) method [3] and so on until it
ends up in returning the FilesStatus instance [4]. The resource leakage occurs
in the method getWorkingDirectory which calls the getHomeDirectory method [5]
which in turn establishes a new connection to the sftp server instead of using
an already created connection. As the mkdirs method is recursive this results
in creating a huge number of connections.
# open method [6] This method returns an instance of FSDataInputStream which
consumes SFTPInputStream instance which doesn't return an acquired ChannelSftp
instance back to the pool but instead it closes it[7]. This leads to
establishing another connection to an SFTP server when the next method is
called on the FileSystem instance.
[1]
https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L658
[2]
https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L321
[3]
https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L202
[4]
https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L290
[5]
https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L640
[6]
https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L504
[7]
https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPInputStream.java#L123
> SFTPConnectionPool connections leakage
> --------------------------------------
>
> Key: HADOOP-15358
> URL: https://issues.apache.org/jira/browse/HADOOP-15358
> Project: Hadoop Common
> Issue Type: Bug
> Components: fs
> Affects Versions: 3.0.0
> Reporter: Mikhail Pryakhin
> Priority: Critical
>
> Methods of SFTPFileSystem operate on poolable ChannelSftp instances, thus
> some methods of SFTPFileSystem are chained together resulting in establishing
> multiple connections to the SFTP server to accomplish one compound action,
> those methods are listed below:
> # mkdirs method
> the public mkdirs method acquires a new ChannelSftp [from the
> pool|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L658]]
> and then recursively creates directories, checking for the directory
> existence beforehand by calling the method
> [exists|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L321]
> ] which delegates to the getFileStatus(ChannelSftp channel, Path file)
> [method|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L202]]
> and so on until it ends up in returning the [FilesStatus
> instance|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L290]].
> The resource leakage occurs in the method getWorkingDirectory which calls
> the getHomeDirectory
> [method|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L640]]
> which in turn establishes a new connection to the sftp server instead of
> using an already created connection. As the mkdirs method is recursive this
> results in creating a huge number of connections.
> # open
> [method|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPFileSystem.java#L504]].
> This method returns an instance of FSDataInputStream which consumes
> SFTPInputStream instance which doesn't return an acquired ChannelSftp
> instance back to the pool but instead it
> [closes|[https://github.com/apache/hadoop/blob/736ceab2f58fb9ab5907c5b5110bd44384038e6b/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/sftp/SFTPInputStream.java#L123]]
> it. This leads to establishing another connection to an SFTP server when the
> next method is called on the FileSystem instance.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
