Optimization possible to avoid pulling each container image twice

[dziemba]

I think there’s an optimization that can be done in the agent-stack-k8s regarding image pulling. I understand the reasoning for the init-containers/preFlightContainers (see

agent-stack-k8s/internal/controller/scheduler/scheduler.go

Line 510 in 00a778d

// Init containers. These run in order before the regular containers.

) and I think that’s a reasonable solution.

Then however we do another pull for each image here:

agent-stack-k8s/internal/controller/scheduler/scheduler.go

Line 357 in 00a778d

c.ImagePullPolicy = corev1.PullAlways

which I believe is not necessary since we just pulled all these images via the init-containers.

In our clusters a no-op pull takes roughly a second, so this adds a second for each container used. When using many containers (we often have 5+), this adds up to quite a lot of extra time waiting. I believe it’s possible to just change the PullPolicy to Never to avoid this extra pulling, which is fine since we just pre-pulled the images in the init containers.

[DrJosh9000]

Thanks for raising this @dziemba! This proposal seems sensible to me.

[DrJosh9000]

I wonder if its worth choosing PullIfNotPresent over PullNever, on the (admittedly very very unlikely) chance that the image disappears in between the init container pulling and the regular containers starting?

[dziemba]

Thanks for the quick fix! Yes, PullIfNotPresent is probably the better choice.