Michael Schwartzkopff wrote:
> Am Freitag, 14. August 2009 16:47:52 schrieb Harald Dunkel:
>> Hi folks,
>>
>> If I understood the Linux-HA concept correctly, then it
>> works because the services (NFS, Samba, EMail, Apache,
>> etc.) share state information between the cluster hosts
>> somehow. For NFS and Samba this state information is often
>> stored in a file system on a shared block device, e.g.
>> using drbd.
>>
>> But does this concept work for the file system itself?
>> AFAICS most Linux file systems keep state information in
>> RAM. When the backup host takes over because the master
>> died of something, does it really have consistent
>> information on file system, service and application
>> level?
>>
>> Sorry for asking. I am highly concerned because I saw
>> 2 fatal file system failures within the last 7 days (xfs).
>>
>>
>> Regards
>>
>> Harri
> 
> For NFS this is no problem. NFS is in charge for the file locks. It on the 
> disk. so if the other node wants to write to a locked file, NFS says: "no, 
> no".
> 
> CIFS; I don't know.
> 
> Do NOT use normal file systems like ext3 on shared media in clusters like 
> DRBD 
> or iSCSI. This will lead to data corruption. Use filesystems that are cluster 
> aware: OCFS2, GFS and GPFS.
> 

Probably a misunderstanding. I wasn't talking about using
shared file systems, but about a "single host" file system
on a shared block device in a traditional master-backup
scenario. XFS, for example.

The point is: I am not convinced that a fallback host has
always consistent and up-to-date information in the file
system on the shared block device.

On filesystem level the master keeps a part of the file
system data or meta data in RAM instead of the block device.
When it dies, then this part could be lost. The file system
on the shared block device could be corrupt or incomplete.
Maybe the ocf "Filesystem" script on the fallback can try
to repair the file system by running fsck, or it can rely on
mount to fix the inconsistency, but in the worst case you
could get some new entries in lost+found (or nothing).

Is this correct?


Regards

Harri

_______________________________________________
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