Actions
Bug #7206
closedCeph MDS Hang on hadoop workloads
Status:
Can't reproduce
Priority:
Normal
Assignee:
-
Category:
-
Target version:
-
% Done:
0%
Source:
other
Tags:
Backport:
Regression:
Severity:
3 - minor
Reviewed:
Affected Versions:
ceph-qa-suite:
Component(FS):
Client, Common/Protocol, Hadoop/Java, MDS
Labels (FS):
Java/Hadoop
Pull request ID:
Crash signature (v1):
Crash signature (v2):
Description
With blatant disregard for posted advice about the stability of ceph-fs, I attempted to use it for testing out some hadoop workloads.
When the hadoop job gets to roughly 30% complete the MDS appears to go out for lunch locking up.
This is with ubuntu 13.10, kernel 3.11.0, ceph emperor, hadoop cdh 4.x.
I had to update the ceph filesystem hook for hadoop 2.x, that might be the cause, but I think that even if the updates are invalid a buggy client should not really be able to freak out the MDS.
The machines are on AWS so Xen might be part of the issue
Ceph FS is mounted on the machines as well.
I could not coax a perf report out of the machine (that also locks up)
I am more than willing to help track this down.
Files
Actions