hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Erik Krogen (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-13265) MiniDFSCluster should set reasonable defaults to reduce resource consumption
Date Tue, 13 Mar 2018 17:35:00 GMT

    [ https://issues.apache.org/jira/browse/HDFS-13265?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16397335#comment-16397335

Erik Krogen commented on HDFS-13265:

Attached a test,  [^TestMiniDFSClusterThreads.java] , which tries to estimate how many threads
are used by NNs and DNs within a {{MiniDFSCluster}}. It is not perfect but will provide a
rough estimate.
Running this without any modifications results in the following:
MiniDFSCluster Thread Counts: Namenode 41 // Datanode 44 // Datanode #2 39

Most recently I was able to drop the thread counts down to:
MiniDFSCluster Thread Counts: Namenode 24 // Datanode 23 // Datanode #2 18
This was mostly done via (a) disabling unnecessary services on the MiniDFSCluster (b) leveraging
HDFS-13272 and HADOOP-15311 to decrease the number of threads consumed by Jetty. For a 1-DN
MiniDFSCluster (most common AFAIK), this would be a reduction of about 40 threads per MiniDFSCluster,
so hopefully pretty beneficial to the testing pipeline.

> MiniDFSCluster should set reasonable defaults to reduce resource consumption
> ----------------------------------------------------------------------------
>                 Key: HDFS-13265
>                 URL: https://issues.apache.org/jira/browse/HDFS-13265
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>          Components: datanode, namenode, test
>            Reporter: Erik Krogen
>            Assignee: Erik Krogen
>            Priority: Major
>         Attachments: TestMiniDFSClusterThreads.java
> MiniDFSCluster takes its defaults from {{DFSConfigKeys}} defaults, but many of these
are not suitable for a unit test environment. For example, the default handler thread count
of 10 is definitely more than necessary for (almost?) any unit test. We should set reasonable,
lower defaults unless a test specifically requires more.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org

View raw message