A simple script that an end-user can run on an existing cluster, that will:

1. Connect to a given node via JMX
2. Get a list of all caches on all nodes
3. Run JMX calls on each cache on each node to capture diagnostic data
4. Serialise this data (maybe JSON?), zip it up.

This will allow end-users to share such stats for debugging and perf tuning.

Diagnostic data to be captured would include all JMX info on hit/miss ratio, RPC performance, transaction commit/rollback rates, config details, time budgeting info, etc.

This tool would require that JMX statistics are enabled and running for a while before the snapshot is captured.