I'm running version 2.4.1. I have confirmed the bootstrap.mlockall setting is true for the nodes in my cluster. I've been monitoring swap space usage on two servers (which run the ES nodes). I've found that the swap space usage for several nodes seems to continue to grow. For one of the nodes, the swap space grew to around 1.6GB over the span two months time. For another node, it grew to about 734MB over the same time period.
I found this posting which I thought was helpful:
It seems to indicate mlockall isn't the most an effective way to prevent swapping. Would you agree with that statement? Also, can you provide recommendations for what to try next in terms limiting swapping? I'm thinking the next step is set the vm.swappiness setting to a lower value?
Overall the two servers are approaching the total defined swap space. I can restart nodes, but would like to avoid doing that if I can. Any thoughts are appreciated.