swhiteho at redhat.com
Wed Sep 5 09:56:17 UTC 2012
On Sun, 2012-09-02 at 02:11 +0200, Kveri wrote:
> we're using gfs2 on drbd, we created cluster in incomplete state (only 1 node). When doing dd if=/dev/zero of=/gfs_partition/file we get filesystem freezes every 1-2 minutes for 10-20 seconds, I mean every filesystem on that machine freezes, doing ls /etc hangs in D state for 10-20 seconds. Sometimes this hang last for more than 2 minutes and hung task message gets logged in dmesg.
> iotop shows gfs2_logd and flush-XXX:X kernel process taking 99% io resources.
> GFS is mounted with rw,noatime,nodiratime,hostdata=jid=0 options.
It sounds like the issue is related to the network and sending i/o to
the drbd device. If you've got a large backlog of cached and dirty data
(and that includes flushing the log) then it may take a while to send
that over the network. What is the speed of your network and what speed
doe the drbd device work at (i.e. can it accept data at line speed?)
It sounds to me as if the issue is that you are simply creating dirty
data at a rate far in excess of what the underlying hardware can cope
with, so it pauses now and then to catch up,
> gettune options:
> quota_warn_period = 10
> quota_quantum = 60
> max_readahead = 262144
> complain_secs = 10
> statfs_slow = 0
> quota_simul_sync = 64
> statfs_quantum = 30
> quota_scale = 1.0000 (1, 1)
> new_files_jdata = 0
> Server is kernel 3.2.0-25 64bit.
> What could be the problem?
> Thank you.
> Linux-cluster mailing list
> Linux-cluster at redhat.com
More information about the Linux-cluster