Clebert;

I am still running a bit high level on the invocation call graph reporting and it's relationship to the actual invocation metrics collection, but here is my take on how they could be supported based on my assumption that graphing will have some higher overhead than simple metrics collection and aggregation (and jump in and correct me where I am off.......)

The configuration meachanism to select between metrics and graphing/metrics can be in the interceptor adaptors. We can define two sets of adaptors, one that will generate stats only and one that will generate a call graph with embedded stats.

Metrics only adaptors report simple flat collected stats on invocations (with the exception of the JDBC Proxy Driver which dives a bit deeper).

The Call Graph/Metrics adaptors will generate call graphs with embedded metrics with the entry point for the call graph starting at the intercepting point of the configured adaptor.

Does this make sense ?

//Nicholas