[Linux-cluster] problem with GNBD device

[Lon Hohberger]

On Fri, 2007-10-12 at 10:18 +0800, Changer Van wrote:
> Hi all,
> I set up a http HA cluster consist of 3 nodes.
> Node 1 is set to gnbd server for fencing.
> Node 2 and node 3 are set to http HA.
> In case the http service is running on node 3.
> Once the network cable of node 3 was unplug, 
> the service would shift to node 2 properly,
> but cman service on node 3 was killed after the catble was plugged in,
> and cman's pid file was still there. 

After a node is evicted from the cluster, you generally need to reset it
before it can rejoin the cluster.


> partial log messages on node 3:
> openais[6621]: [CPG  ] got joinlist message from node 1 
> openais[6621]: [CPG  ] got joinlist message from node 2 
> openais[6621]: [CMAN ] cman killed by node 3 for reason 2 
> gnbd_import: ERROR [../../utils/gnbd_utils.c:78] cman_init failed :
> Connection refused 
> gfs_controld[6648]: cman_start_notification error -1 104
> dlm_controld[6641]: cluster is down, exiting
> fenced[6635]: cluster is down, exiting 
> fence_node[6645]: agent "fence_gnbd" reports: gnbd_import: ERROR
> cannot get node name : Connection refused gnbd_import: ERROR If you
> are not planning to use a cluster manager, use -n failed: fence_gnbd,
> node03 

This is weird...  CMAN killed itself on node 3?

-- Lon