KAMAL_VERSION, pushed docker images, and deploy hooks about kamal

[netshade]

I should note, btw, that the differences in KAMAL_VERSION is the component after the uncommitted marker. That is to say, the docker tag looks like:

MY_IMAGE:H1_uncommitted_H2

and the KAMAL_VERSION passed to the hook looks like:

MY_IMAGE:H1_uncommitted_H3

[netshade]

After some further work on this, I should note that this appears to happen during the deploy that occurs during kamal setup. kamal deploy appears? to be stable.

[yanguango]

I'm seeing the same issue, the KAMAL_VERSION is like xxx_uncommitted_xxx, my guess is the git directory has some changes are not committed, and kamal detects it and then adds it in the version tag. For my case, I see this message in the logs:

The following paths have uncommitted changes:
 ?? vendor/bundle/

kamal/lib/kamal/cli/build.rb

Line 23 in e8b9f89

say "The following paths have uncommitted changes:\n #{uncommitted_changes}", :yellow

[netshade]

I looked into this further, and it appears that when Kamal::Configuration encounters uncommitted changes, it tries to create a "random" hash via SecureRandom.hex(8). If the configuration was stable across the execution, this would probably be fine, but I think what appears to be happening during setup is that a different configuration is used to populate KAMAL_VERSION during the pre-deploy hook than what is used to actually push the image. I'm not sure the ordering on dependencies there, but that "feels" like what the bug probably is. :-p

Idk if it makes sense to fix the ordering problem here, or to change it such that the uncommitted hash is "honest"? Something like,

git status --porcelain | awk '{print $2}' | xargs -I{} md5sum {} | md5sum | awk '{print $1}' , computing a hash of hashes of all uncommitted files? It seems a bit heavyweight, but also, at least more true to what the intent is rather than a random number?

[morgoth]

@netshade There was a change #700 that may solve your issue as there should no more be "uncommited" part in the version.
Can you please verify and eventually close the issue?

[netshade]

That change works for me, w/ the absence of the uncommitted portion. Thanks.