I was wondering what sort of tuning is usually applied to the Java GC for the druid services.
In http://druid.io/docs/latest/operations/recommendations.html there are some general recommendations, thanks a lot for these!
We run a fairly compact setup, 2 nodes, each 128GB of RAM, 16Cores and don’t maintain a lot of data yet, about 60GB
I was a bit puzzled to see huge spikes in query latency into the multi-second range. I’m fairly sure that these are at least connected with a lot of garbage collection happening at that time.
The metrics log 10 to 40 gc runs in that timeframe, the longest taking ~400ms.
Do any of you have some recommendation regarding gc tuning to reduce its impact?
Please let me know if you need any info. Thanks a lot