Dynamically size ESP partition to support images > 512MB

[alexhulbert]

@0x416e746f6e Mentioned that we're going to have a couple images that go just over 512MB soon we need to support images over 512MB, so I modified the gcp postoutput script to dynamically round up the disk size to the nearest GiB in a way that's deterministically computable based on the efi size alone. This PR should be merged with flashbots/dstack-mr-gcp#12

[0x416e746f6e]

@0x416e746f6e Mentioned that we're going to have a couple images that go just over 512MB soon, so I modified the gcp postoutput script to dynamically round up the disk size to the nearest GiB in a way that's deterministically computable based on the efi size alone. This PR should be merged with flashbots/dstack-mr-gcp#12

no, I never said we will have a couple of images going over 500MiB soon.

I said I had a couple of corner-cases when I needed an image to grow over that size. (to be precise I needed a bunch of debug tools + rust toolchain + debug symbols + whatnot to be readily present on the instance to troubleshoot some infrequent condition).

[0x416e746f6e]

however, on the subject of the PR: I understood that the measurement script is going to be broken if the image is over 500MiB.

see: #115 (comment)

with this PR, how is that going to be addressed?

[alexhulbert]

gotcha. so this PR adjusts two things which affect the measurement:

• the size of the partition that holds the .efi file and
• the .raw disk image that gets put in the final GCP .tar.gz

Both of these are sized deterministically based on the size of the .efi file, so the corresponding measurement script PR (which I linked in the first post) can reconstruct the partition table header and generate valid measurements no matter how large the images are.