Hi there Avinash,
Firstly I would change the critical-heap-percentage to something closer
to 70 or 80. This setting is a "stop-the-world" percentage, where the
amount of data inserted is greater than what can be evicted.
Also I would make sure that the CMSOccupancyPercentage is 3-5% less than
the eviction setting. This way the JVM will try and do a GC before
causing the cluster to evict data. Also depending on JVM size, you might
want to increase the eviction percentage to maybe around 60-70%.
The behavior of the OVERFLOW* files are not deleted is correct. When you
restart the server it will not use those files, but Geode does not
delete them on shutdown.
As for the overflow files for the client, it does sound like expected
behavior as well. I imagine that this behavior would be seen on the
server as well, but you might not have noticed it.
If a region is not defined to be persistent, you can be assured that
these files will not be once after a restart.
--Udo
On 11/06/2016 1:26 pm, Avinash Dongre wrote:
Hi All,
I have following scenario and would like to confirm the behavior
1. Region is PARTITION_OVERFLOW
2. Server is stated with --eviction-heap-percentage=50
--critical-heap-percentage=95
3. I started my puts into this region
4. After Some time Eviction starts and I see OVERFLOW* files are created
5. Delete the Region
Overflow files are not deleted , Is this expected Behavior ??
6. I start my client again ( I have not restarted the server ) , I see
immediately new OVERFLOW files are created
Is this expected behavior ?
Thanks
Avinash
