You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Oct 26, 2022. It is now read-only.
An Exo controller slowdown triggered investigation into memory use. Heap dump showed that active slices were consuming large amounts of RAM (on the order of 100M each) by storing in-memory ad models acquired from the broker. The problem appears in Domain.java constructors that create a model in-memory.
An initial trivial attempt to switch those to TDB Ephemeral resulted in
Jenkins running out of memory on tests (!)
and
Tests by @hinchliff showing a significant degradation in performance (a slowdown factor of 3-4 in tests).
We should investigate this further, this thread is the place to dump findings and conclusions
A related issue in controller JVM configuration was discovered that limited actual heap to 4G (see RENCI-NRIG/exogeni#143) using Tanuki wrapper.
Running ac7d9d locally without any Heap size modifications strangely resulted in the tests effectively never finishing. I think I saw some test failures, but without explicit messages as to what caused them. My laptop became very unresponsive, and after over an hour I killed the test. I haven't been able definitively confirm, but I think the tests were limited to 4GB of Heap.
I changed the maximum heap size to 8GB, and ran the tests again. The tests were still slower than normal, but ended in failure after 25 minutes with Heap and GC errors.
can you try setting uniqueDM to true (create a document manager for each model separately)? It's parameter on a call to getModelFromXXX currently set to false.
An Exo controller slowdown triggered investigation into memory use. Heap dump showed that active slices were consuming large amounts of RAM (on the order of 100M each) by storing in-memory ad models acquired from the broker. The problem appears in Domain.java constructors that create a model in-memory.
An initial trivial attempt to switch those to TDB Ephemeral resulted in
and
We should investigate this further, this thread is the place to dump findings and conclusions
A related issue in controller JVM configuration was discovered that limited actual heap to 4G (see RENCI-NRIG/exogeni#143) using Tanuki wrapper.
@YufengXin
The text was updated successfully, but these errors were encountered: