GC issues in historical node


Is there a reason we are not using G1GC as default GC for druid historical nodes? Recently we started noticing long Full GC pauses in our historical nodes. e.g: GC happens for ~30 - 40 seconds and recovers a lot of memory, but since the process is suspended for so long, the zk connectivity is lost and Druid think the node is unavailable etc.

Given that there is no real OOM issue here, we are trying to switch to G1GC, but i wanted to find out from the community if there’s a reason to still use CMS.