Bentoml: Can't set bentoml deploy version to be based on a git url

[seantur]

Describe the bug
It looks like you can't use the configuration file to set the bentoml deploy version to be based on a git commit/ branch.

To Reproduce
I was trying to install a version of bentoml based off of a branch. I used bentoml config set core.bentoml_deploy_version=git+https://github.com/bentoml/BentoML.git@master to set the deploy version.

The generated requirements.txt then contains bentoml==git+https://github.com/bentoml/BentoML.git@master which is not how requirements.txt expects packages based on git urls to be formed.

If I removed the bentoml==, I am able to build the docker image. However, when I run the docker image, I get a "BentoService bundle version mismatch"

Expected behavior
I expect to be able to set the deploy version and have the requirements.txt file use that version, and to be able to run the docker image with that version.

Environment:

• Python 3.7.5 in a virtual environment, BentoML==0.6.1

@kevlar1818 @pdoyle5000

[parano]

Thanks for reporting the issue @seantur, the bentoml_deploy_version config does have some rough edges, it is currently only being used for bentoml devs to do deployment testing on dev branches. To make it usable for users I think 2 things need to be done:

1) as you suggested, when the bentoml_deploy_version is a link to git-repo, remove the bentoml== https://github.com/bentoml/BentoML/blob/master/bentoml/service_env.py#L101

2) SavedBundle loader should not raise an exception when conf["version"] is a link to git-repo
https://github.com/bentoml/BentoML/blob/master/bentoml/bundler/config.py#L111

I will try to follow up with a fix later this week(sorry it's a bit busy on our side this week), contributions are welcomed too!

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[parano]

hi @seantur, were you able to find a workaround for this issue? We are investigating this again and I'd like to hear more about your use case if you don't mind sharing, why a custom version of BentoML is needed for your deployment and/or if there's any way BentoML can make itself more customizable to fit that need. Feel free to ping me in the community slack channel as well if you prefer to discuss there.

[seantur]

@parano My use case was that I needed a fix that was in master, but not available yet as a release. I noticed that the functionality was supposed to be there, but that it didn't seem to work.

In my case, I didn't need this functionality once the fix was released as a version.

[parano]

@seantur got it - a workaround to that is to install BentoML from source with 'edit mode':

$ git clone [email protected]:bentoml/BentoML.git
$ cd BentoML
$ pip install -e .

This way, all the deployments created and docker images built will be using the version of BentoML that's in the source directory.

I guess the fix you were looking for has already been released now! Hope everything is working for you now, let me know if there's still anything blocking you!