This repository has been archived by the owner on Sep 13, 2023. It is now read-only.
Set docker build args and similar for deployment targets
#647
Labels
breaking-change
This is going to break users code
build
Exporting models (building model packages)
deploy
Related to model deployment
🏡 housekeeping
Fighting technical debt and improving dev process
Comments
aguschin
added
🏡 housekeeping
|
Request from customer: add ability to set |
aguschin
added a commit
that referenced
this issue
Mar 31, 2023
this is for #647 (comment) This work similarly to #645: `mlem build docker` works the same, but I'll provide an example for k8s deploy here: ``` $ mlem declare deployment kubernetes deployer \ --image_name myimage --service_type loadbalancer --registry remote \ --env docker --env.registry remote --registry.host localhost --namespace myns \ --set_env.0 VAR1 --set_env.1 VAR2=aguschin 💾 Saving deployment to deployer.mlem ``` ```yaml # deployer.mlem env: object_type: env registry: type: remote type: kubernetes image_name: myimage namespace: myns object_type: deployment registry: host: localhost type: remote service_type: type: loadbalancer set_env: - VAR1 - VAR2=aguschin type: kubernetes ``` Then in Dockerfile for your image you'll get: ``` ... ENV VAR1=VALUE1 ENV VAR2=aguschin ``` `VALUE1` will be taken from shell/env vars with `os.getenv("VAR1")`. If it's not set, an Exception will be raised by MLEM.
# for free
to subscribe to this conversation on GitHub.
Already have an account?
#.
Right now
mlem build docker --helpprovide control over manydocker buildoptions via using class DockerBuildArgs(BaseModel).Ideally, these options should be available when running
mlem deploy $TARGET, but they're not. One reason for this implementation was that platforms set their own limitations for these options (such as Heroku allows only someplatformsfor docker build to be used). However, this requires us to introduce each option needed manually, example being:build_argin Resolve requirements installed from Git or local source #645daemonandregistryintroduced much earlierIdeally, all deployments should inherit class
DockerModelDirectoryorDockerImageBuilderinstead of re-implementing functions likebuild_heroku_dockerorbuild_k8s_dockerorbuild_sagemaker_docker. This should extend to CLI so users be able to set those options in CLI.This calls for refactoring with breaking changes, since options added manually (mentioned above) could change their locations, and start looking just like those in
build dockernow.Keeping it "as is" instead will lead to manually work (e.g. I've added
build_argto k8s deployment only, but it should be added to each deployment target, and it should be done for each param), so refactoring and breaking changes are inevitable if we're going to improve the tool.cc @omesser FYI
The text was updated successfully, but these errors were encountered: