Swap consumption after backup

[erwann]

Hi,

We have a 4 nodes hyperconverged cluster and we see a strange behavior:
During vm scheduled backup , swap is increasing significatly (about 500 Mo /node every night).

Ram is use about 50% on each node during the day.

We have to swapoff/swapon each week.

Why this behavior?
How could we fix it?

Regards,

 

[t.lamprecht]

We have to swapoff/swapon each week.

Why? do you see any actual impact of the increased SWAP usage, Linux is normally pretty good to decide what's worth to swap out and what can stay out without resulting in a constant performance impact. SWAP use is not something inherently bad, that's a misconception which needs to die.

The following "one liner" can shed some light on which processes use SWAP:

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

What's your swappiness: sysctl vm.swappiness

What storages do you use, ZFS, ceph?

 

[erwann]

Swapiness is set to 0.
Our storage is Ceph.

bash command on a node return this :



If I am right, swap is mainly used ceph and pveproxy ?

 

[erwann]

Hi,
this morning I had new alerts on swap.
Please see below


One node il alerting about swap utilisation. Its swap capacity is 16GB.
How could be sure it will not crash when utilisation come to its limit?