I have the following setup:
1 node, runs as GNBD server, it's export GFS (v1) partition, doesn't mount it localaly.
2 nodes (node1,node2), import the GFS (v1) partition from gnbd node and mount it as DocumentRoot for apache.
all the 3 nodes located on a Hosting Company, they have public IPs so i have ssh login for this servers without NATing or forwarding.., all located in one Rack with one subnet and one switch connecting them.
this setup was running for about 3 months, but unfortuntally it's now stopped for unknown resone,
The problem accour by starting the clustering manager (cman service), then i can not ping/access the gnbd server outside its subnet, becuase almost all of the packets were droped, although I can ping it from any server on its subnet without any packet lose. other nodes are accessible without any problem, also cman_tool status shows all nodes as connected without any problem (M).
I solved it by preventing INPUT/OUTPUT traffic for the mulitcast address (18.104.22.168
) from the server using its firewall (iptables) while keeping the cman service running, it is not the proper solution because this mulitcast address demand by clustering service to talk with the other clustering servers, which they are already using the same multicast address without any problem.
the hosting company say it's switch does not prevent multicasting.
any idea about this problem?
sorry for my english.
<cluster name="ex-cluster" config_version="6">
<device name="gnbd" ipaddr="gnbd1.example.com
<device name="gnbd" ipaddr="node1.example.com
<device name="gnbd" ipaddr="node2.example.com
<fencedevice name="gnbd" agent="fence_gnbd" servers="gnbd1.example.com
<clusterfs device="/dev/sdb1" force_unmount="0" fsid="5391" fstype="gfs" mountpoint="/www/www-data" name="www" options=""/>