[Ocfs2-users] OCFS2 slow with multiple writes

[Prakash Velayutham]

Hi Sunil,

Thanks for the response. Do you mean OCFS2 is blocking writes from multiple clients? Is that how OCFS2 works? I can understand that writing the (2) 20G files might take longer with "ordered" option as data needs to be flushed to the FS before journal commit, but why is that blocking a new separate file from being written to the file system?

Regards,
Prakash

On Oct 20, 2011, at 6:25 PM, Sunil Mushran wrote:

> Use writeback. Ordered data requires the data to be flushed
> before journal commit. And flushing 40G takes time.
> 
> mount -t data=writeback DEVICE PATH
> 
> On 10/20/2011 03:05 PM, Prakash Velayutham wrote:
>> Hi,
>> 
>> OS - SLES 11.1 with HAE
>> OCFS2 - 1.4.3-0.16.7
>> Cluster stack - Pacemaker
>> 
>> I have Heartbeat Filesystem monitor that monitors the OCFS2 file system for availability. This monitor kicks in every minute and tries to write a file using dd as below.
>> 
>> dd of=/var/lib/mysql/data1/.Filesystem_status/default_bmimysqlp3 oflag=direct,sync bs=512 conv=fsync,sync
>> 
>> If the OCFS2 file system is busy, like when I try to create 2 large files (20GB each) in the OCFS2 directory, I see that the above monitor process hangs until the 2 files are created. But this causes Pacemaker to fence the node as the RA is configured for a timeout of 45secs and the 2 file creations do take more than that. The OCFS2 file system is mounted as below.
>> 
>> /dev/mapper/bmimysqlp3_p4_vol1 on /var/lib/mysql/data1 type ocfs2 (rw,_netdev,nointr,data=ordered,cluster_stack=pcmk)
>> 
>> Is there something wrong with the file system itself that a small file creation hangs like that? Please let me know if you need any more information.
>> 
>> Thanks,
>> Prakash