On 11/22/2010 at 09:27 PM, "Alain.Moulle" <[email protected]> wrote: 
> Hi 
>  
> Finally I use the o2cb and dlm resources in Pacemaker , so the o2cb-pcmk  
> stack 
> to manage the ocfs2 cluster via Pacemaker. 
> But it seems there is a bug somewhere :  one resources are started and   
> one or several 
> ocfs2 FS mounted as clone on both nodes, if one node is powered off, the  
> other remaining 
> node can't access to the data of OCFS2 FS anymore (i.e a ls command  
> remains stalled) , 
> until killed node is rebooted and Pacemaker started and again (then the  
> previous stalled ls gives 
> information again). 

Is that a clean shutdown?  Or are you just pulling the plug?

Have you configured STONITH?  If a node "goes away", DLM will stall until
the node is successfully fenced.

Regards,

Tim

> It seems that locks in dlm are not well managed at the Pacemaker level : 
> cat /proc/23921/stack 
> [] ocfs2_wait_for_recovery+0x77/0x8f [ocfs2] 
> [] ocfs2_inode_lock_full_nested+0x160/0xb8d [ocfs2] 
> [] ocfs2_inode_revalidate+0x163/0x25c [ocfs2] 
> [] ocfs2_getattr+0x8b/0x19c [ocfs2] 
> [] vfs_getattr+0x4c/0x69 
> [] vfs_fstatat+0x50/0x67 
> [] vfs_stat+0x1b/0x1d 
> ] sys_newstat+0x1f/0x39 
> [] system_call_fastpath+0x16/0x1b 
> [] 0xffffffffffffffff 
>  
> So, one lock is not released. 
> But locks are managed in this case by pacemaker , aren't they ? 
>  
> Could someone tell me if I can expect a patch ? or if there is already  
> an available patch somewhere ? 
>  
> Release informations: 
> RHEL6 with : 
> pacemaker-1.1.2-2 
> dlm-pcmk-3.0.12-23 
> cluster-glue-1.0.5-1 
> ocfs2-tools-pcmk-1.6.3-2 
>  
> Thanks a lot 
> Alain 
> _______________________________________________ 
> Linux-HA mailing list 
> [email protected] 
> http://lists.linux-ha.org/mailman/listinfo/linux-ha 
> See also: http://linux-ha.org/ReportingProblems 
>  



_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to