Re: NFS Failover
I successfully run nfsv4 and drbd in clustered mode.
The main thing to do wrt config files for nfs is pin down port numbers
to specific (rather than dynamic ones) at startup for the rpc suite.
And also switch to UDP rather than transport (solves session issues
during failover) - your clients all need to explicitly ensure they are
mounting with udp options.
Also you need to have the rpc socket file handles on a clustered
filesystem somewhere mounted on both nodes (I use GFS2 for this
purpose as it's easier).
I have heard great things about ceph instead of drbd but haven't tried
it myself yet.
On 27 June 2013 09:06, Stan Hoeppner <stan@hardwarefreak.com> wrote:
> On 6/26/2013 2:54 PM, David Parker wrote:
>
>> As you both pointed out, it
>> would be easier and safer to use a clustered filesystem instead of NFS for
>> this project. I'll check out GlusterFS, it looks like a great option.
>
> It may be worth clarification to note GlusterFS is not a cluster
> filesystem. It is a distributed filesystem. There is a significant
> difference between clustered and distributed.
>
> A distributed filesystem such as Gluster is applicable to your needs as
> you can add/remove clients in an ad hoc manner without issue. A cluster
> filesystem is probably not suitable, because you simply can't connect
> new nodes in a willy nilly fashion. None of OCFS, GFS, GPFS, CXFS, etc
> handle this very well, if at all. Cluster filesystems require hardware
> fencing between nodes. One doesn't setup hardware fencing willy nilly.
>
> --
> Stan
>
>
> --
> To UNSUBSCRIBE, email to debian-user-REQUEST@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
> Archive: [🔎] 51CB57D6.20306@hardwarefreak.com">http://lists.debian.org/[🔎] 51CB57D6.20306@hardwarefreak.com
>
Reply to: