We’re trying to optimize our deployments, but I’m running into one thing that really appears to be bottlenecking us, at least on our relatively ancient Kubernetes 1.20 RKE deployment (Yeah I know.), where between the init and running phases, there’s a significant lag time. At least according to the events that are emitted, it appears that the init container completes, then the first non-init container is pulled, then the sandbox is created, then the next container is pulled and sandbox created, on and on down the list. Is this intended or even expected behavior from upstream Kubernetes, prior or current? If we could have the containers pull and sandboxes get created both in parallel, and/or during the course of the init container, it would speed up our deployments to a very large degree for large batches of jobs.
Is there a configuration thing that I’m missing, a possible upstream improvement that I could make to enable this, an already complete improvement that will help me push people to actually upgrade the systems, or is this something that’s intended and we’re just going to have to accept?