Mailing-List: contact chukwa-dev-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: chukwa-dev@hadoop.apache.org
Message-ID: <2043265627.1242618705842.JavaMail.jira@brutus>
Date: Sun, 17 May 2009 20:51:45 -0700 (PDT)
From: "Ari Rabkin (JIRA)" <jira@apache.org>
To: chukwa-dev@hadoop.apache.org
Subject: [jira] Reopened: (CHUKWA-229) ExecPlugin leaks a file descriptor
In-Reply-To: <1205793881.1242110565557.JavaMail.jira@brutus>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


     [ https://issues.apache.org/jira/browse/CHUKWA-229?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Ari Rabkin reopened CHUKWA-229:
-------------------------------


That fix didn't do the trick.  We're still leaking. What's more, it's sporadic; it doesn't happen every execution. So I suspect it's a race condition somewhere.  Instead of tracking it down immediately, I propose the following medium-term workaround:  If exec adaptor detects that we're out of file handles, we stop the agent process and wait for watchdog to respawn it.  

We went to a lot of trouble to implement robust checkpointing; I figure we might as well rely on it here.

> ExecPlugin leaks a file descriptor
> ----------------------------------
>
>                 Key: CHUKWA-229
>                 URL: https://issues.apache.org/jira/browse/CHUKWA-229
>             Project: Hadoop Chukwa
>          Issue Type: Bug
>          Components: data collection
>            Reporter: Ari Rabkin
>            Assignee: Ari Rabkin
>         Attachments: fixExecFDLeak.patch
>
>
> ExecPlugin never properly disposes of the subprocess's input fd. This means we run out of file descriptors eventually.  
> This only affects ExecAdaptor; if ExecPlugin is invoked by the inputtools framework, the process doens't stay running.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.