Autosave-like environment versions?

[jarandaf]

I am trying to use a custom environment (docker-based) within a PythonScriptStep. The environment exists (version=1) in the workspace and has already been successfully built with Docker (build status is succeeded).

Whenever I try to run the pipeline and use this environment AzureML ignores the environment version and injects a version named Autosave_{YYYY}-{MM}-{DD}T{HH}:{MM}:{SS}Z_{some_hash}.. The pipeline run then fails saying this version does not match the one provided (version=1). This environment version then appears in the environment versions list but I did not create it. There are no builds available for such environment version.

Is this expected? What might be causing this?

[v-a-s-a]

I started seeing this issue yesterday as well. I've rebuilt my environment and republished my AML pipeline, but a new "Autosave..." version keeps getting picked up as the latest build. I am unable to specify the environment version explicitly for a pipeline run.

[jarandaf]

@v-a-s-a do you have any update by any chance?

[RezaMadi]

Hi @jarandaf : Are you sure you are passing the right environment to your pipeline? make sure you are using env = Environment.get(....), and not directly env = Environment(....). The latter one will create a new environment with the same name as the one you have registered, but using the default image, and of course it will add a version with the Autosave... pattern.

[jarandaf]

Hi @RezaMadi, thank you for your answer. Yes, I do use the Environment.get method.

[v-a-s-a]

@v-a-s-a do you have any update by any chance?

I don't have a satisfying answer. Eventually, the autosave environment had picked up the changes I needed, and I moved on.

I can also confirm that I am using the Environment.get method in my pipeline creation/update script.