Unfortunately I have to swap out the switch that is used for the cluster traffic of our 4-node cluster for a new one. I'm hoping I can do this by connecting the new switch to the old switch and then moving the nodes over one by one.
Can I change the cluster configuration so that there is a longer grace period before a node is deemed 'lost' and gets fenced? The only line in my cluster.conf that looks like it has anything to do with it is this one:
<totem consensus="4800" join="60" token="10000" token_retransmits_before_loss_const="20"/>
I think that with faststart enabled the link with a node will be down for only a few seconds. I realize that this probably means the cluster will lock up during that period (since we use a lot of GFS), but it's still better than having to bring the entire cluster down.