Let's assume I start a two node-cluster on Azure, with AZURE_PING configured, then kill the JVMs on both servers. AZURE_PING discovery file is going to be left in the container, there's nothing we can do about it. But let's say I start another EAP with the same AZURE_PING configuration.

This EAP will find the leftover file and will send join requests to the cluster members, which will fail. Then, the new members will become a singleton cluster, which is fine. The new cluster members then creates a new discovery file that contains all the records (the new cluster member and all the old dead records).

We've been discussing how to better handle this with josef.cacek@gmail.com, and there are some suggestions we have:

1, log a warn/error message that there's a discovery file inside the preconfigured container with unresponsive records in it and that the user should go and clean the container
2, add a configuration option to automatically delete files with unresponsive clusters
3, the new member probably shouldn't carry around the records of the old nodes when creating his new discovery file

WDYT?

Assigning to rhn-engineering-rhusar since he's the author or AZURE_PING, but feel free to reassign as you see fit.