[
https://issues.apache.org/jira/browse/HDFS-5493?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13819173#comment-13819173
]
Chris Nauroth commented on HDFS-5493:
-------------------------------------
Below is the offending code. Thanks to [~avyakrita] for reporting the problem,
and thanks to [~szetszwo] for finding the potential leak in the code.
There was some initial suspicion that this could be related to HDFS-3373,
however the code is different on branch-1. This is likely to be a different
issue, though the symptoms are similar.
{code}
} catch (IOException ex) {
if (refetchToken > 0 && tokenRefetchNeeded(ex, targetAddr)) {
refetchToken--;
fetchBlockAt(target);
} else {
LOG.warn("Failed to connect to " + targetAddr
+ ", add to deadNodes and continue" + ex);
if (LOG.isDebugEnabled()) {
LOG.debug("Connection failure", ex);
}
// Put chosen node into dead list, continue
addToDeadNodes(chosenNode);
}
if (s != null) {
try {
s.close();
} catch (IOException iex) { }
}
s = null;
}
{code}
> DFSClient#DFSInputStream#blockSeekTo may leak socket connection.
> ----------------------------------------------------------------
>
> Key: HDFS-5493
> URL: https://issues.apache.org/jira/browse/HDFS-5493
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: hdfs-client
> Affects Versions: 1.2.1
> Reporter: Chris Nauroth
>
> {{DFSClient#DFSInputStream#blockSeekTo}} may handle {{IOException}} by
> refetching a new block access token and then reattempting {{fetchBlockAt}}.
> However, {{fetchBlockAt}} may then throw its own {{IOException}}. If this
> happens, then the method skips calling {{Socket#close}}. This is likely to
> manifest as a leak of sockets left in CLOSE_WAIT status.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
