Plane HTTP API

The controller in Plane serves an HTTP/JSON API, which is used both by other Plane components and other trusted components (such as your own application web server) to interact with Plane.

The Plane API is partitioned into two parts:

  • A control API which trusted clients can use to control aspects of Plane, such as starting and terminating backends.
  • A public API which can be used by trusted or untrusted clients alike (including your app end-users’ browsers).

The control API is served with the /ctrl prefix, and the public API is served with the /pub prefix. For flexibility, Plane itself does not implement any access control or TLS termination on these endpoints. If you intend to expose Plane's API over the open internet, you should use a reverse proxy (like Nginx, Caddy, or Envoy) to terminate TLS and implement access control for each path prefix.

Connect API

One way to think about Plane is that it acts like a distributed key-value store, except that instead of storing data associated with each key, it manages a running process for each key. When you ask Plane for the process associated with a key, Plane ensures that a process is running for that key, and returns a URL that points to it.

In the key-value map analogy, the default behavior of Plane is comparable to “get or create” semantics that some key-value maps implement (such as setdefault (opens in a new tab) in Python).

The connect API is the endpoint that implements this behavior.

The connect request is located at the path /ctrl/connect. The connect request is a POST request that takes a JSON body. An example payload might look like this:

{
    "key": {
        "name": "room-abcde",
        "namespace": "games",
        "tag": "drop-four"
    },
    "spawn_config": {
        "executable": {
            "image": "ghcr.io/drifting-in-space/demo-image-drop-four",
        },
        "lifetime_limit_seconds": 3600,
        "max_idle_seconds": 60,
    },
    "pool": "example-pool",
    "user": "user-123",
    "auth": {
        "any arbitrary JSON object": "can go here",
        "Plane does not read it": "it is just passed through to your backend",
        "even lists and numbers": ["like", 3.14, "are", "fine"]
    }
}
  • key: Optional object describing the key to connect to. If not provided, one will be generated randomly (forcing a new backend to start).
  • spawn_config: Optional configuration that is used to start a new backend if necessary.
  • pool: Optional string specifying the resource pool where the backend should be spawned. If not provided, the default pool is used.
  • user: Optional string to associate with the user on whose behalf this request is being made.
  • auth: Optional key-value map of unforgeable data (such as claims) that you would like to pass to the backend about this user.

At least one of key or spawn_config must be provided. If only spawn_config is provided, the connect call will always attempt to spawn the backend. If only key is provided, the connect call will attempt to connect to an existing backend, and will return an error if one does not exist.

Key configuration object

The key configuration passed as key refers to an object, with a required name field and two optional fields.

  • name: The name of the key. Keys with the same name in the same namespace are considered the same key, meaning that only one backend will run for them at a time.
  • namespace: The namespace of the key. Keys with the same name in different namespaces are considered different keys,
  • tag: If provided, only a backend with the same tag as requested will be returned, but if there is a backend with the same name and namespace but a different tag, an error will be returned instead.

It is expected that most users of Plane will only need to care about the name field; the others are provided for users who need more advanced control.

Spawn configuration

The spawn configuration tells Plane how to spawn a new backend for this request if necessary (i.e. if the key does not match an existing backend). It is an object with the following fields:

  • max_idle_seconds: An optional numeric field which, if provided, creates a limit for how long (in seconds) a backend can have no inbound connections to it before it is terminated. If not provided, there is no limit.
  • lifetime_limit_seconds: An optional numeric field which, if provided, creates a deadline (in seconds from now) that the backend will be terminated regardless of whether it has inbound connections.
  • executable: An object containing configuration of the backend process itself.

Both max_idle_seconds and lifetime_limit_seconds are optional; if neither is provided, the backend will continue running until it is either terminated through the control API, or exits on its own accord.

If both max_idle_seconds and lifetime_limit_seconds are provided, the backend will be terminated when either limit is reached.

Executable configuration

The executable field of the spawn configuration is an object with the following fields. Only image is required; the others are optional.

  • image: The Docker image to use to run the backend.
  • pull_policy: Optional string specifying the Docker pull policy to use when pulling the image. Valid values are Always, IfNotPresent, and Never. If not provided, IfNotPresent is used.
  • credentials: Optional object containing credentials to use to connect to the Docker registry. Currently, only username/password credentials are supported, by providing an object with the fields username and password.
  • env: Optional object containing environment variables to pass to the backend. The keys and values of this object are passed directly to the backend as environment variables.
  • resource_limits: Optional object containing resource limits to apply to the backend.

TODO: Document resource limits.

TODO: Document return value.

Terminate API

The terminate API is mainly provided to support manual termination from the Plane CLI.

You are free to call it from application code as well, but consider using max_idle_seconds, lifetime_limit_seconds, or exiting from within your session backend instead.

To “soft-terminate” a backend, send a POST request with an empty body to:

/ctrl/c/:cluster/b/:backend/soft-terminate

Where :cluster is the name of the cluster the backend is running on, and :backend is the name of the backend.

Soft-terminating first sends a SIGTERM signal to the backend, and then waits for 10 seconds for the backend to exit gracefully before force-terminating it.

To “hard-terminate” a backend, send a POST request with an empty body to:

/ctrl/c/:cluster/b/:backend/hard-terminate

Hard-terminating does not send a SIGTERM signal to the backend, and instead immediately force-terminates it.

Status API

The status API tells you the status of a given backend. Unlike the connect and terminate APIs, it is considered a “public” API, meaning that it is safe to expose on the open internet without authentication.

To get the status of a backend, send a GET request to:

/pub/c/:cluster/b/:backend/status

Where :cluster is the name of the cluster the backend is running on, and :backend is the name of the backend.

The response is a JSON object with the following fields:

  • status: The status of the backend, such as ready. See backend lifecycle for a list of possible values.
  • time: The time at which the backend entered its current status, in milliseconds since the Unix epoch.

Streaming status API

A streaming variant of the status API is also available. To use it, send a GET request to:

/pub/c/:cluster/b/:backend/status-stream

Provided that the client supports it, this returns a server-sent event (opens in a new tab) stream that emits a JSON object every time the backend changes status. Each JSON object emitted has the same fields as the non-streaming status API.

The streaming API will also replay past state changes on connection. It supports reconnects without duplication as specified by the server-sent event protocol.