Environment Variables

A list of environment variables that control Ray and Anyscale behavior

ALLOW_MULTIPLE_CLIENTS

When set to 1, Anyscale allows multiple concurrent connections to a cluster. Multiple clients must use the same cluster environment, although they can use distinct runtime environments. Two connections in the same namespace can communicate with each other. If this variable is not set, or is set to 0, then Anyscale refuses subsequent connections while there is a single active client.

ANYSCALE_CLI_TOKEN

If set, this value contains the token that the current context uses to authentication to Anyscale. In circumstances where there is no user home directory or where it's not appropriate to use a normal credentials file, programs may leverage this environment variable's value for authentication.

ANYSCALE_CONNECT_SECURE

As of 7/14/2021 all connections to Anyscale clusters are encrypted and run over port 443. This environment variable has no effect.

ANYSCALE_DEBUG

Anyscale support might ask you to set this environment variable in a troubleshooting setting. When set, Anyscale emits lots of logging.

RAY_ADDRESS

If set, and an explicit connection URL is not supplied to ray.init("anyscale://"), then this address is used as the ray connection URL. These two are equivalent.

RAY_ADDRESS=anyscale://cluster_name python -c "import ray; ray.init()"

python -c "import ray; ray.init('anyscale://cluster_name')"

IGNORE_VERSION_CHECK

This environment variable overrides errors resulting from a detected version mismatch between client and server versions of Anyscale. Usually, upgrading the client environment with pip install -U anyscale or equivalent is the best course of action. If a developer wants for some reason to avoid upgrading, setting this environment variable to 1 will allow a script to continue execution.

ANYSCALE_CLOUD

If set, and cloud is not supplied to ray.init(), then this value is used to configure cloud. These two are equivalent:

ANYSCALE_CLOUD=my_cloud python -c "import ray; ray.init('anyscale://cluster')"

python -c "import ray; ray.init('anyscale://cluster', cloud='my_cloud')"

ANYSCALE_CLUSTER_COMPUTE

If set, and cluster_compute is not supplied to ray.init(), then this value is used to configure cluster_compute. These two are equivalent:

ANYSCALE_CLUSTER_COMPUTE=my_cluster_compute python -c "import ray; ray.init('anyscale://cluster')"

python -c "import ray; ray.init('anyscale://cluster', cluster_compute='my_cluster_compute')"

ANYSCALE_CLUSTER_ENV

If set, and cluster_env is not supplied to ray.init(), then this value is used to configure cluster_env. These two are equivalent:

ANYSCALE_CLUSTER_ENV=my_cluster_env:1 python -c "import ray; ray.init('anyscale://cluster')"

python -c "import ray; ray.init('anyscale://cluster', cluster_env='my_cluster_env:1')"

ANYSCALE_AUTOSUSPEND

If set, and autosuspend is not supplied to ray.init(), then this value is used to configure autosuspend. Example values of autosuspend: -1 (disabled), "2h" (2 hours), "20m" (20 minutes), 60 (60 minutes) These two are equivalent:

ANYSCALE_AUTOSUSPEND=20 python -c "import ray; ray.init('anyscale://cluster')"

python -c "import ray; ray.init('anyscale://cluster', autosuspend=20)"

ANYSCALE_JOB_NAME

If set, and job_name is not supplied to ray.init(), then this value is used to configure job_name. These two are equivalent:

ANYSCALE_JOB_NAME=my_job python -c "import ray; ray.init('anyscale://cluster')"

python -c "import ray; ray.init('anyscale://cluster', job_name='my_job')"

ANYSCALE_NAMESPACE

If set, and namespace is not supplied to ray.init(), then this value is used to configure namespace. These two are equivalent:

```text ANYSCALE_NAMESPACE=my_namespace python -c "import ray; ray.init('anyscale://cluster')"

python -c "import ray; ray.init('anyscale://cluster', namespace='my_namespace')"