In order to find the reason you need to debug it.
Power off one of the nodes and then wait for the failure , but don't clear it up.
Then Run the resource in debug mode - ou can check how to do it here: https://wiki.clusterlabs.org/wiki/De...ource_Failures