envd Concepts

This doc describes concepts in the envd. The name envd is inspired by systemd. ^[1]

Concepts

The core abstractions in envd are environments and images. Images are built from the given build.envds and follow docker image spec v1.2. Thus you can use the images built by envd with Docker directly.

Environments are containers run by docker, Kubernetes, or some other OCI runtime spec-compatible runtimes (e.g. runc, crun, containerd).

Execution Model

build.envd is written in Starlark, a simplified dialect of Python 3. The default function build() will be invoked on startup.

# build.envd
# The function `build` will be evaluated by default.
def build():
    # envd instructions here.

Functions like install.python_packages and config.jupyter can be used to register information. The information will be kept in a static graph in the memory. envd uses the static graph to build the environment at the envd of execution.

envd evaluates the build function with the starlark interpreter, thus you can use loops, if/else conditional execution or other statements in build.envd.

WARNING

Code below is for conceptual demonstration purposes only. Some features are not yet implemented.

# build.envd
# The function `build` will be evaluated by default.
def build():
    version = "2.9.1"
    dev_env(version, "zsh")

# Here is a new function and it is used in `build()`
def dev_env(version, custom_shell):
    install.python_packages(name=[
        "tensorflow=="+version,
        "tensorboard"
    ])
    config.jupyter()
    # Configure zsh if it is specified.
    if custom_shell == "zsh":
      shell("zsh")

---

1. Thread about the name on GitHub. ↩︎