When doing various operations on cluster mirrors with clvmd, the cluster mirror will be suspended on all nodes before performing LVM metadata operations. It is possible for the cluster log server to suspend all its mirror devices before the other machines have completed suspending. The other machines will issue their last requests to the log server before suspending, but the server will block trying to write to the log device because it has already suspended them. The server will not be able to continue until its log device is unsuspended - which from the point of clvmd, can not happen until all nodes are suspended. The result is an complete LVM lock-up. The fix is to allow the server to migrate during suspension to those nodes whose log device is not yet suspended.
Closing this bug...