tensorchord/envd

feat: Support envd build from stdin

VoVAllen opened this issue · 1 comments

Description

In flyte, they use syntax like

@task(
        config=EnvdConfig(
        os="ubuntu20.04",
        pip_packages=["tensorflow"],
        language="python",
)

to configure the envd dependency. Then it needs to call envd build. it's better for envd build to support stdin input, which makes it easier for flyte to call envd


Message from the maintainers:

Love this enhancement proposal? Give it a 👍. We prioritise the proposals with the most 👍.

What about syntax like:

@task(
        config=EnvdConfig(
                script="path/to/*.envd/dir",
                tag="train:latest",
                func="filename:func"
        )
)

If users straightly give parameters to the stdin of build, without a entry script, they might be inaccessible to most envd features, like expose, or we would design a huge EnvdConfig.