[DRBD-user] Xen block-drbd and fencing

Sauro Saltini saltini at shc.it
Tue Oct 5 05:16:16 CEST 2010


Looking through the git changelog for crm-fence-peer.sh I've found some 
notes that pointed me out in the right direction (I hope).

It seems that I need to call crm-fence from drdb config helpers section 
of each resource like this: crm-fence-peer.sh -i <cib id of Xen 
resource> -r Started

This way the script doesn't try to guess the cib id from drbd resource 
name but takes the given id and searches for "Started" role (instead of 
"Master" for the drbd resource type).

Still waiting to listen your opinions regarding how drbd is supposed to 
manage failover of xen resource in case of io-error on Primary node with 
"on-io-error detach".

Sauro.

On 04/10/2010 15:51, Sauro Saltini wrote:
> Hello everybody,
> has someone used crm-fence-peer.sh for resource fencing with 
> block-drbd managed resources ?
>
> Looking at the changelog I can see that in 8.3.7 it's supposed to work :
>
> * crm-fence-peer.sh is now also usable if DRBD is managed from the xen 
> block helper script
>
> But I've tried to figure out how this could be done looking at the 
> scripts (both block-drbd and crm-fence-peer) finding nothing that 
> seems related to this feature.
>
> My drbd version is 8.3.7 (from sources)
> crm is corosync 1.2.1 (from sources)
> I have configured 'Xen only' resources at crm level, while drbd is 
> indirectly managed via block-drbd scripts.
>
> my drbd resources are configured this way:
>
> resource res1 {
>   device    /dev/drbd1;
>   disk      /dev/vhosts/res1;
>   meta-disk internal;
>   disk {
>     on-io-error detach;
>     fencing resource-only;
>   }
>   handlers {
>     split-brain "/usr/lib/drbd/notify-split-brain.sh root";
>     fence-peer  "/usr/lib/drbd/crm-fence-peer.sh";
>     after-resync-target  "/usr/lib/drbd/crm-unfence-peer.sh";
>   }
>   syncer {
>     rate 40M;
>   }
>   net {
>     allow-two-primaries;
>     max-buffers 8000;
>     max-epoch-size 8000;
>     sndbuf-size 0;
>   }
>   on cl1 {
>     address   10.20.30.41:7789;
>   }
>   on cl2 {
>     address   10.20.30.42:7789;
>   }
> }
>
> The question is : is this supposed to work with drbd 8.3.7 ?
> Or the crm-fence-peer scripts needs an explicit drbd resource in cib 
> to operate with ?
> In second case can I still use Xen resources with drbd-block.sh adding 
> to cib the relative drbd resources or I need to go for a completely 
> crm-managed resource group?
>
> Lastly, with this setup in case of io-error on a host with primary 
> state resource the manual says the resource will be detached and will 
> "run diskless", but I can't figure out if the process of promoting the 
> drbd resource to primary on the peer and starting the xen host is 
> intended to be automatically managed by crm-fence scripts or if I must 
> configure some other scripts to obtain the following result :
>
> 1) io failure on host cl1 (local disk failure) with res1=Primary on it
> 2) on-io-error detach detaches resource from device on cl1 - go Diskless
> 3) cl2 notifies crm of cl1 failure (how?)
> 4) crm starts res1 on cl2
>
>
> Many thanks in advance.
> Sauro Saltini.
>
> _______________________________________________
> drbd-user mailing list
> drbd-user at lists.linbit.com
> http://lists.linbit.com/mailman/listinfo/drbd-user


More information about the drbd-user mailing list