You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@livy.apache.org by "Dustin Koupal (Jira)" <ji...@apache.org> on 2020/04/24 22:03:00 UTC

[jira] [Created] (LIVY-763) FD leak in InteractiveSession and PythonInterpreter

Dustin Koupal created LIVY-763:
----------------------------------

             Summary: FD leak in InteractiveSession and PythonInterpreter
                 Key: LIVY-763
                 URL: https://issues.apache.org/jira/browse/LIVY-763
             Project: Livy
          Issue Type: Bug
          Components: Interpreter, Server
    Affects Versions: 0.7.0, 0.6.0
            Reporter: Dustin Koupal


We recently upgraded from Livy 0.4 to Livy 0.6 and started to run out of FDs (we fire up thousands of InteractiveSessions a day). Looking at the open FDs, they were all DIR handles for `/usr/lib/spark/python/lib`.

I believe this change was introduced in https://github.com/apache/incubator-livy/commit/1bbefe601641aa4db74dbed1f7faf458b9a70a63#diff-7649a51ad4bddc91b6f1038e06479d41R259-R264

The behavior without this change and the default config is Livy will open 2 FDs for each InteractiveSession, and they will never be closed (I think because `DirectoryStream` is 
`AutoCloseable` and try-with-resource doesn't work with Scala's `Try`.

As a workaround, in our production clusters, we have set `livy.rsc.pyspark.archives` to an empty string in `livy.conf`



--
This message was sent by Atlassian Jira
(v8.3.4#803005)