Since upgrading from ELK 7.17.1 to ELK 8.6.2 (and even with ELK 8.7.1) we are experiencing OOMKilled on filebeat and metricbeat pods. We had no issues with ELK 7.17.1. Increasing the resources allocations does not resolve the issue and simply delays the crash. This appears to be a memory leak issue with beats.
State: Running
Started: Thu, 25 May 2023 15:18:43 +0000
Last State: Terminated
Reason: OOMKilled
Exit Code: 137
Started: Thu, 25 May 2023 02:53:22 +0000
Finished: Thu, 25 May 2023 15:18:41 +0000
Ready: True
This is an example of our filebeat pod memory in the past 24 hours
Can you post your complete Filebeat configuration?
What we need is heap profiles from Filebeat which should tell us what is using the memory. The instructions to do this are:
Start the Beat process with httpprof (profiling) enabled. This allows us to easily extract memory profiles of the running process. Add these configuration options:
Once Beats is started and is done initializing (after 5-10 minutes), you can collect the first memory dump via a simple curl command like this: curl -s -v http://localhost:8080/debug/pprof/heap > heap_normal.bin.
Once you start noticing that the process is taking excessive amounts of memory, a second dump needs to be generated like curl -s -v http://localhost:8080/debug/pprof/heap > heap_high.bin.
If you attach the .bin files we can analyze them to see what is going on. The profile taken when the memory usage is excessive is the most important one.
Here's the case where the heap is high, the heap is dominated by allocations from the Kubernetes watch API. This could mean a few things, one is that it could be receiving very large responses from the Kubernetes API, but it could also mean that we are subscribing to too many events and the rate of allocations is too high.
This isn't something I have an immediate answer for. It seems similar to other cases I have seen before but it looks like we are still working on a fix.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.