Occasional problems joining the cluster

I am using ElasticSearch with my Grails application. When testing a
clustered configuration of the app, I noticed that occasionally the two ES
servers will not form a cluster but will be listed as two standalone
server. I am using a multicast configuration on both server with the
default group and custom port. Both instances are embedded with Grails app
and run in "dataNode" mode. Most of the times the two server connect to a
cluster (you can see through ElasticSearch Head the configuration of the
cluster). Sometimes, however, ES Head will show the two server as running
two separate instances. No error messages are visible on the console. What
may casue such erratic behavior? How to prevent it?

Thank you,