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
The memory usage should not grow by over 80% after a Jersey update.
Current Behavior
We noticed a memory problem while testing the 2.2 beta. Nodes run into GC problems because the memory in the old gen is getting full. Comparing the 2.1 and 2.2 baseline looks like this:
The spikes in the second half of the graphs happens when processing the first messages. This might be a different problem.
So from 2.1 to 2.2 the old gen memory usage after starting Graylog went from around 170MB to over 300MB.
Possible Solution
The memory usage goes back to 2.1 levels when reverting the Jersey 2.25 update that was committed in db98621.
I suggest to downgrade to Jersey 2.24 for 2.2.
Steps to Reproduce (for bugs)
Start Graylog 2.1
Login to web interface and click around
Measure memory usage via jmap -heap <pid>
Start Graylog 2.2 beta
Login to web interface and click around
Measure memory usage via jmap -heap <pid>
Your Environment
Graylog Version: Graylog 2.2.0-beta.6
The text was updated successfully, but these errors were encountered:
When fixing the grizzly version to 2.3.26 and using Jersey 2.25.1, the memory usage goes down as well. The grizzly version 2.3.27 seems to introduce the memory problem.
This causes the memory usage to grow because of pre-allocated memory. Setting the -Dorg.glassfish.grizzly.DEFAULT_MEMORY_MANAGER=org.glassfish.grizzly.memory.HeapMemoryManager property brings the memory usage back to 2.1 levels.
* force grizzly to use the HeapMemoryManager and not its default
the default switched to PooledMemoryManager which uses 10% of the heap for its pools by default
this creates too much old gen pressure for our purposes, so we switch back to the previous setting
* clear messages in MessageEvent as soon as the processor is done with them
this allows the garbage collector to eagerly collect those references and makes it less likely that the messages are promoted to the old gen (where they typically only increase memory pressure)
in the case of the output buffer the last set of messages in the ring would be help in memory, which can needlessly increase the memory footprint as well
Fixes#3423
Expected Behavior
The memory usage should not grow by over 80% after a Jersey update.
Current Behavior
We noticed a memory problem while testing the 2.2 beta. Nodes run into GC problems because the memory in the old gen is getting full. Comparing the 2.1 and 2.2 baseline looks like this:
The spikes in the second half of the graphs happens when processing the first messages. This might be a different problem.
So from 2.1 to 2.2 the old gen memory usage after starting Graylog went from around 170MB to over 300MB.
Possible Solution
The memory usage goes back to 2.1 levels when reverting the Jersey 2.25 update that was committed in db98621.
I suggest to downgrade to Jersey 2.24 for 2.2.
Steps to Reproduce (for bugs)
jmap -heap <pid>
jmap -heap <pid>
Your Environment
The text was updated successfully, but these errors were encountered: