openstack team mailing list archive

[Ji Zhang <zhangji87@xxxxxxxxx>]

Hi,

One of my compute node eats up swap, and it turns out to be a kvm
process. So I have two questions:

1. How to release the swap? I tried terminating other instances on the
same node, but it doesn't help. So can I use the normal approach like
"sync; echo 3 > /proc/sys/vm/drop_caches, swapoff -a && swapon -a" on
compute node? Will it be too dangerous?

2. What will cause the kvm process eating up swap? I don't think it
possible to launch new instance in a cloud that is already full. Are
there any other causes? Something I should avoid in the future?

Thanks.

PS: I'm coping with a cloud of Diablo release, which is handed over
from a previous Ops.