Details
-
Improvement
-
Status: Resolved
-
Normal
-
Resolution: Fixed
-
None
-
None
Description
We have a cluster that takes alot of reads and we have some issues with the off-heap caching, seeing several G of Finalizer objects on the heap
My conclusion is that the finalizer thread is too slow to run the finalize method on the FreeableMemory (and Memory) objects.
Simply removing the finalize() methods improves performance alot (using finalizers is 50% slower in my micro benchmarks) and does not leak memory during "normal" use as far as i can see
im not sure about the implications for other use cases though