SWAP Usage

[TecScott]

Any insight as to why SWAP is being used when there's as low as 16% RAM usage on the node?

I have one host for example which is sat at 16% RAM usage and using 25MB SWAP, another at 62% RAM usage using 2.8GB SWAP, another at 64% RAM usage using 3.6GB SWAP, another at 55% RAM usage using 1GB SWAP, etc.

vm.swappiness is set to 10

 

[t.lamprecht]

Are those all hosts which are up for a long time?

SWAP only gets reclaimed (i.e., moved back to memory) if required, as else it has no use. So if those are a bit longer running hosts, with memory usage peaks in the past, the swapped out memory just stayed there if it's stale.

A simple test could be to now do a: swapoff -a on a node, then wait until all the swapped out memory then re-enable it with swapon -a and check if it grows and how fast.


You can also find the top processes using most SWAP with the following "one-liner":

for file in $(find /proc/*/status -exec grep -q 'VmSwap' {} \; -print); do awk '/^(VmSwap|Pid|Name)/{printf $2 " " $3 }END{ print ""}' $file; done | sort -k 3 -n -r | head

 

[TecScott]

All hosts have an uptime of around 10 days (in the process of upgrading from 5.3 to 6.1).

I suspect the swap would grow again given it's grown this much in only 10 days.

The top process for all servers is 4 KVM processes followed by 4 pvedaemon processes.