I have a collector that eats memory and in 24 h goes OOM.
The GC cannot free memory also with a huge full GC, forced by me with jmap
It works with 10GB Heap and has 320 agents connected but live metrics is < 200K and historic is <390K
I have produced an heap dump and I have found that a lot of memory is eaten by java/util/concurrent/LinkedBlockingDequeue$Node (4GB and 21million instances)
These objects have as a parent object the class com/ca/apm/baseline/em/BaseLineEngine
After browsing the Internet with this calss name I have solved.
I have solved the problem moving the variance.db file and restarted EM
I hope that this thread can be useful for other people facing the same issue.