When rolling out an update to neonvm, we currently don't pre-load the neonvm-runner images. This has a couple effects:
For a brief period, new VM runner pods are additionally blocked on pulling the runner image (which can take a while); and
If there's issues with pulling the runner image, we only detect that once we're actually attempting to use it.
This PR adds a new daemonset to just keep the neonvm-runner image around on each node. Before rolling out a neonvm update, we can update the runner image in the daemonset to make sure that the new image is loaded on all the nodes.
When rolling out an update to neonvm, we currently don't pre-load the neonvm-runner images. This has a couple effects:
This PR adds a new daemonset to just keep the neonvm-runner image around on each node. Before rolling out a neonvm update, we can update the runner image in the daemonset to make sure that the new image is loaded on all the nodes.
Part of neondatabase/cloud#11232