[
https://issues.apache.org/jira/browse/HDFS-2810?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13189377#comment-13189377
]
Todd Lipcon commented on HDFS-2810:
-----------------------------------
I think I see the bug here. DFSClient.renewLease has:
{code}
void renewLease() throws IOException {
if (clientRunning && !isFilesBeingWrittenEmpty()) {
namenode.renewLease(clientName);
}
}
{code}
and LeaseRenewer only calls {{renewLease()}} on the first DFSClient with any
given name.
So, if there are multiple DFSClients with the same clientName, but the first of
them has no files open, then the lease renewer will stop actually renewing the
lease.
> Leases not properly getting renewed by clients
> ----------------------------------------------
>
> Key: HDFS-2810
> URL: https://issues.apache.org/jira/browse/HDFS-2810
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: hdfs client
> Affects Versions: 0.23.0, 0.24.0
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Priority: Critical
>
> We've been testing HBase on clusters running trunk and seen an issue where
> they seem to lose their HDFS leases after a couple of hours of runtime. We
> don't quite have enough data to understand what's happening, but the NN is
> expiring them, claiming the hard lease period has elapsed. The clients report
> no error until their output stream gets killed underneath them.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
