We are running OCFS 1.6 on Oracle Linux Server release 5.5 kernel
2.6.32-100.24.1
a 3 node cluster.
The application creates and deletes many files throughout the day. We
are noticing that removing a file can sometime takes seconds, even
minutes. When the process issues the low level unlink and hangs
(observed by doing strace), it shows up in the "D" state while it
waits.
We can reproduce the result by a simple "rm".
Here are the options we used to make the OCFS2 filesystem:
> /sbin/tunefs.ocfs2 --query="M=%M H=%H O=%O T=%T\n" ?/dev/sdb2
M=backup-super strict-journal-super H=sparse extended-slotmap
inline-data metaecc xattr indexed-dirs refcount discontig-bg
O=unwritten usrquota grpquota
Block size=4096 Cluster size=4096 Number of node slots=8
Has anyone seen this phenomena? Can anything be done about it in terms
of tuning?
Thank you.