Pentaho 5 + Saiku Plugin - Large datasets and cubes

Hi,


I have a dataset of 2 million records, when about 4 users use the cube (Saiku plugin) to assemble diverse views, after a time (2-3 hours), the memory becomes full, and is not freed more. The data returned by visions are limited by the plugin Saiku in 1000 records, but the raw data manipulated to aggregate the data is about 300,000 records, for example, if the Saiku not have the limitation of 1000, 300,000 records would be returned.


This aggravates further when I try to export to excel, I think he accumulates the data in memory for download, but does not release more.


However, during testing, following the memory, often
it was collected and returned, but after a while, does not return.


I followed the jstat, the Eden and Old memory,
there comes a time that the Old becomes full (100% use) and the Pentaho
goes to work only with the Eden, which is usually collected, however
some times the CPU hits 700%, trying to collect old, I believe,
this way a few moments the Pentaho is unavailable, does not quite catch, m
the is very slow.


- Set up the Mondrian to clear the cache every query
- I set the session-timeou apra 10 minutes
- Use Pentaho 5 + Saiku plugin, Xms 1G, Xmx 9G, 7 colors cpus
- I set the Mondrian memory limit for him abort the query
if you have more memory not to stay in line query.




I saw several posts about it, but no conclusive. I ask:


- Memory leak problem of Mondrian ?. Alias ​​because the Pentaho 5.1 uses
Mondrian 3.1, it's the latest?
- Problem of Memory Leak of pentaho Saiku plugin? Alias ​​he uses the Pentaho Mondrian, or raises an instance for it?
- It would be caching problem not released? and not necessarily memory leak?