[Ocfs2-users] Big load and many error with ocfs2 1.3.3

Cyril FERAUDET ocfs2 at feraudet.com
Tue Sep 18 03:59:21 PDT 2007


Hello Sunil,

I've lot of problem with the patches on 2.6.22. I'm unable to mount  
the ocfs2 volume
shared by NFS on more than 2 NFS client when there are more than 1  
ocfs2 node.
I get an "NFS stale handle" ...

Without NFS with all node as ocfs2 node, it seems working fine. But  
I've more iowait
without NFS than with 1 ocfs2 node and the other by NFS ...

About lock, do you know if NFS need to have a filesystem with locking  
feature to purpose
locking to NFS client ?

Cyril

Le 17 sept. 07 à 19:15, Sunil Mushran a écrit :

> I'm sorry I could not understand your email. Are you saying that while
> most of the problems were solved with the patches, one wasn't?
>
> As far as cluster aware flock/lockf goes, it is still not  
> implemented in
> ocfs2.
>
> http://git.kernel.org/?p=linux/kernel/git/torvalds/ 
> linux-2.6.git;a=blob;f=Documentation/filesystems/ 
> ocfs2.txt;h=ed55238023a9843f7781ac6cfe94d174ba772072;hb=HEAD
>
> Cyril FERAUDET wrote:
>> Hello Sunil,
>>
>> Just back of holidays, I tested the 2.6.22 with all of your patches.
>>
>> No problem instead of without patches, put aside that I suspect a  
>> locking problem.
>>
>> Indeed, I've a script who replace a file with an flock (LOCK_EX)  
>> on the shared volume and
>> an other script who acces to the shared volume by NFS to copy this  
>> file to a local volume.
>>
>> This file is copied each minutes and regulary I've an empty or  
>> truncated file.
>>
>> Do you now a restriction about locking or something like that ?
>>
>> Thanks for all.
>>
>> Cyril
>>
>>
>> Le 30 août 07 à 22:29, Sunil Mushran a écrit :
>>
>>> Sorry I don't track the debian kernel. But typically they take
>>> the source for kernel.org and use it as is. But don't take my
>>> word for it.
>>>
>>> Cyril FERAUDET wrote:
>>>> There is actualy only one node who export the ocfs2 volume via  
>>>> NFS under the 2.6.18-4 on etch repository.
>>>> I've got no problem until I leave it alone. When I add an other  
>>>> node I got the error below.
>>>>
>>>> So if I understand the 2.6.22 provided by Debian haven't all  
>>>> patches listed below and
>>>> would be the cause of the truncated file of my ( a bad day to  
>>>> fix it ;-) ) ?
>>>>
>>>> Can I apply the missing patches to the debian package or I need  
>>>> to use a fresh sources ?
>>>>
>>>> Thanks in advance.
>>>>
>>>> Cyril
>>>>
>>>> Le 30 août 07 à 19:30, Sunil Mushran a écrit :
>>>>
>>>>> Are you exporting the ocfs2 volume via NFS? If so, then we did  
>>>>> fix this issue.
>>>>> http://git.kernel.org/?p=linux/kernel/git/torvalds/ 
>>>>> linux-2.6.git;a=commit;h=592282cf2eaa33409c6511ddd3f3ecaa57daeaaa
>>>>> It should be fixed in 2.6.22.
>>>>>
>>>>> Use 2.6.22 + patches as listed in the following link:
>>>>> http://oss.oracle.com/pipermail/ocfs2-users/2007-August/ 
>>>>> 001935.html
>>>>>
>>>>> Cyril FERAUDET wrote:
>>>>>> A response to my self ...
>>>>>>
>>>>>> Thanks Sunil Mushran, I haven't see your response in the  
>>>>>> thousand of mail at my return of holidays ...
>>>>>>
>>>>>> This patch seems to be in the 2.6.20.
>>>>>>
>>>>>> Can this error cause my big load average ?
>>>>>>
>>>>>> No other patch was included till the 2.6.22. Some one have the  
>>>>>> same problem than me with the 2.6.22 (file truncated) ?
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Cyril
>>>>>>
>>>>>> Le 30 août 07 à 10:36, Cyril FERAUDET a écrit :
>>>>>>
>>>>>>> Hi all,
>>>>>>>
>>>>>>> I've one 3.5T Xserve with one LUN shared between 7 debian  
>>>>>>> nodes :
>>>>>>>
>>>>>>> # dmesg |grep -i ocfs
>>>>>>>
>>>>>>> OCFS2 Node Manager 1.3.3
>>>>>>> OCFS2 DLM 1.3.3
>>>>>>> OCFS2 DLMFS 1.3.3
>>>>>>> OCFS2 User DLM kernel interface loaded
>>>>>>> OCFS2 1.3.3
>>>>>>> ocfs2_dlm: Node 11 joins domain 7791C0F1366B4FAD99E46078BE5092A1
>>>>>>> ocfs2_dlm: Nodes in domain  
>>>>>>> ("7791C0F1366B4FAD99E46078BE5092A1"): 0 1 7 8 9 10 11
>>>>>>>
>>>>>>> # uname -a
>>>>>>> Linux img01 2.6.18-4-686 #1 SMP Wed May 9 23:03:12 UTC 2007  
>>>>>>> i686 GNU/Linux
>>>>>>>
>>>>>>> All nodes mount the shared partition without problem, but  
>>>>>>> after some minute of utilization I get these error :
>>>>>>>
>>>>>>> ...cut...
>>>>>>> (4191,0):ocfs2_populate_inode:236 ERROR: Invalid dinode:  
>>>>>>> i_ino=2832705, i_blkno=2832705, signature = INODE01, flags = 0x0
>>>>>>> (4191,0):ocfs2_read_locked_inode:393 ERROR: populate failed!  
>>>>>>> i_blkno=2832705, i_ino=2832705
>>>>>>> (4191,0):ocfs2_iget:131 ERROR: status = -116
>>>>>>> (4191,0):ocfs2_iget:141 ERROR: status = -116
>>>>>>> (4191,0):ocfs2_get_dentry:63 ERROR: status = -116
>>>>>>> (4192,1):ocfs2_populate_inode:236 ERROR: Invalid dinode:  
>>>>>>> i_ino=2832705, i_blkno=2832705, signature = INODE01, flags = 0x0
>>>>>>> (4192,1):ocfs2_read_locked_inode:393 ERROR: populate failed!  
>>>>>>> i_blkno=2832705, i_ino=2832705
>>>>>>> (4192,1):ocfs2_iget:131 ERROR: status = -116
>>>>>>> (4192,1):ocfs2_iget:141 ERROR: status = -116
>>>>>>> (4192,1):ocfs2_get_dentry:63 ERROR: status = -116
>>>>>>> (4191,0):ocfs2_get_dentry:69 ERROR: status = -116
>>>>>>> (4186,1):ocfs2_populate_inode:236 ERROR: Invalid dinode:  
>>>>>>> i_ino=8808616, i_blkno=8808616, signature = INODE01, flags = 0x0
>>>>>>> (4186,1):ocfs2_read_locked_inode:393 ERROR: populate failed!  
>>>>>>> i_blkno=8808616, i_ino=8808616
>>>>>>> (4186,1):ocfs2_iget:131 ERROR: status = -116
>>>>>>> (4186,1):ocfs2_iget:141 ERROR: status = -116
>>>>>>> (4186,1):ocfs2_get_dentry:63 ERROR: status = -116
>>>>>>> (4189,1):ocfs2_get_dentry:69 ERROR: status = -116
>>>>>>> ...cut...
>>>>>>>
>>>>>>> Then all node have a huge load more than 100 with more than  
>>>>>>> 50% of iowait.
>>>>>>>
>>>>>>> I've tested the kernel 2.6.22 (don't remember the ocfs  
>>>>>>> version), there are no ocfs2 error, but all opened to write  
>>>>>>> file are truncated to zero size !
>>>>>>>
>>>>>>> Have you an idea, do you know a stable version of ocfs2 build  
>>>>>>> into a debian kernel ?
>>>>>>>
>>>>>>> Thanks,
>>>>>>>
>>>>>>> Cyril
>>>>
>>>
>>
>>
>> _______________________________________________
>> Ocfs2-users mailing list
>> Ocfs2-users at oss.oracle.com
>> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>




More information about the Ocfs2-users mailing list