Cache RPMs between builds

[agracey]

Builds can take a while on dependency resolution and package download. This seems like it's network bound so caching the packages would likely provide a better experience when iterating on an image

[jdob]

For clarification, you mean RPMs specifically right? As compared to container images for k8s workloads.

[agracey]

Yes, RPMs

[pbronneberg]

Recognise the long build duration. For a project tasked to build Edge images using EIB, this results in rather high cycle teams when editing configuration. (> 20 minutes)

We observe that Resolving package dependencies... is consuming the beef of the build time (we pull in large packages from Nvidia RPM registries during this step). To assess whether this is CPU bound, we tested our CI on 2 different Amazon instances leading to the following results:

Timing results:

instance-type	m5a.large	m5ad.xlarge
vCPU	2	4
Memory (GB)	8	16
EIB time	21:44	17:43
time savings	-	-4:01

Based on this, CPU does not appear the largest contributing factor and we also suspect the resolving to be network bound.

Hope this helps