[TOPIC] Snapshot data collection definition - what data to collect

[Steve Poole <sp...@googlemail.com>]

Andreas posted some interesting information here
http://mail-archives.apache.org/mod_mbox/incubator-kato-spec/200911.mbox/%3C6EB74B946FF4CF4BAA39332269A659D5024AE296@DYNASERV.dynatrace.local%3E

This thread is to discuss the topic of  what sort of data is collected in a
snapshot dump.    Since snapshots are intended to be small and frequent
there is a need to expose statistical data as well as detailed information.
We have to consider very carefully how this information would be gathered.
The Reference Implementation must be able to demonstrate the capabilty but
its likely that a faster performing implementation could be done by a JVM
directly or by enhancing existing API's

To start Andreas has said that he would want to see

1)  A  count of instances of a class
2)  Instance data including related heap information

Collecting statistics about instances is going to be dependent on either
byte code insertion or by asking the JVM to do the equivilent.  Collecting
creation counts is fairly straightforward but decrementing is not.  The JVM
does know when an object is garbage but there could be significant overhead
in having a global counter of this form.     I'm also thoughtful of the
synchronisation that may be required to do incrementing and decrementing

Question :Is the count the number of objects created or the number of live
objects or do we need both?

JVM use heaps in different ways -  its likely that all that could be
returned (if anything)  in the way of heap information would be some vm
specific arbitary label.  That's probably ok but it would mean that the
consumer of the data would have to have jvm specific knowledge on how to
interpret it.

Question:   Do we need to do more than provide this arbitary label?

Question:   What other data is required?


-- 
Steve