Hard to tell what exactly is happening just from the load - it may be some operation that is causing it, might be a bug that only occurs under certain conditions etc. We have seen a few reports of higher memory consumption since the switch to puma, for example in Foreman 2.0.0 - memory leak?. Perhaps playing with the number of workers and threads can lead to an improvement, but it would be good in any case to get the logs when there is an issue to identify if there is some operation that is gobbling up memory/cpu and can be optimized.
