I'm building a rust image for arm and amd, which is made ofapollo router and has a vast amount of dependencies (500+)

In the same github action, I have the build of a test image (amd only) and later I build the prod image (amd and arm). I use an amd machine with qemu, everything "by the book" and while looking at the log of the action, i can see the speed of compiling the dependencies of the arm build to slow down to the point where it times out ; i mean by that, that the amd build takes 20mn but the arm build goes to the timeout limit of the action set at 6hrs.

while it sounds like a rust/app issue, i think it is not: i have tried to spawn an arm machine and build the image with 2 separated machines and it worked without much delay and the build time between the 2 cpu was on par ; but manual things like the stitching of the manifest, and handling of the cache are things which i'd like to avoid (keep custom code away).

is there any advice to make things better? also if i go this way, i need the cache to be persistent across jobs and i wonder how to achieve that since the gha driver does not fit the usecase.