Image caching on compute nodes
I've noticed that frequently, the images cached on the compute nodes are ignored and new ones are downloaded from the Glance server. You can see the network traffic spike between the compute node and the Glance server when a VM is put into the spawning state, and the VM will not boot up until the network traffic returns to normal. If you have a proper network in place to handle image updates, this isn't much of an issue because most images are not that big.
However, if you have a snapshot that is tens of gigabytes in size, this can take a noticeable amount of time to the user. I was wondering, what setting controls this caching mechanism? Is it possible to keep things in the compute node's cache for a longer period of time?