Skip to content

runtime-env-apis.md

Latest commit

 

History

History
56 lines (42 loc) · 1.42 KB

runtime-env-apis.md

File metadata and controls

56 lines (42 loc) · 1.42 KB

Runtime Environment APIs

APIs for configuring runtime environments for jobs, tasks, and actors.

RuntimeEnv

Class to define a runtime environment. Can be used interchangeably with a dictionary.

from ray.runtime_env import RuntimeEnv

# For entire job
ray.init(runtime_env=RuntimeEnv(
    pip=["numpy", "pandas"],
    working_dir="./my_code",
    env_vars={"MY_VAR": "value"}
))

# Per-task or per-actor
@ray.remote(runtime_env=RuntimeEnv(pip=["requests"]))
def task():
    import requests
    return requests.get("https://example.com")

actor = MyActor.options(runtime_env=RuntimeEnv(pip=["torch"])).remote()

Key Parameters:

  • pip: List of pip packages to install
  • conda: Conda environment specification
  • working_dir: Directory to use as working directory
  • py_modules: List of Python module paths
  • env_vars: Dictionary of environment variables
  • container: Container configuration
  • image_uri: Docker image URI

RuntimeEnvConfig

Configuration options for runtime environment setup.

from ray.runtime_env import RuntimeEnv, RuntimeEnvConfig

runtime_env = RuntimeEnv(
    pip=["numpy"],
    config=RuntimeEnvConfig(
        setup_timeout_seconds=300,
        eager_install=True
    )
)

Parameters:

  • setup_timeout_seconds: Timeout for runtime environment creation (default: 600, -1 to disable)
  • eager_install: Install at ray.init() time before workers are leased (default: True)