sailfork / t3code-github / .repos/ alchemy-effect/ website/ src/ content/ docs/ cloudflare/ compute/ container-lifecycle.mdx

A Container runs only when you start it. You start it, stop it, and restore it from the Durable Object that owns it.

The examples use this container:

// src/Sandbox.ts
export class Sandbox extends Cloudflare.Container<Sandbox>()("Sandbox", {
  schedulingPolicy: "durable_object",
  image: "node:24-slim",
}) {}

Bind the container

yield* Sandbox in a Durable Object's Construction phase gives you the container's handle:

export default class Workspace extends Cloudflare.DurableObject<Workspace>()(
  "Workspace",
  Effect.gen(function* () {
    const sandbox = yield* Sandbox;

    return Effect.gen(function* () {
      return {};
    });
  }),
) {}

Binding does not start the container.

Start it before you use it

Call start at the top of each operation that needs the container:

const start = sandbox.start({
  entrypoint: ["sleep", "infinity"],
  enableInternet: false,
});

return {
  run: (cmd: string[]) =>
    Effect.gen(function* () {
      yield* start;
      // the container is running
    }),
};

start does nothing when the container is already running, and concurrent calls start it once. A container that stopped or crashed since the last request starts again on the next one.

The options only apply when start actually starts the container. To run with different options, stop the container first.

Choose a size per start

A container with schedulingPolicy: "durable_object" picks its size when it starts:

yield* sandbox.start({
  instance: "standard-1",
  entrypoint: ["sleep", "infinity"],
  enableInternet: false,
});

start starts the Container's own image. To run another image in the same Durable Object, bind another Container — see Several images in one Durable Object. An async Worker attaches Containers to a class with Cloudflare.DurableObject(className, { containers }) — see Several images on an async Worker.

Start it from a Layer

A Layer that depends on the container starts it the same way:

export const ShellLive = Layer.effect(
  Shell,
  Effect.gen(function* () {
    const sandbox = yield* Sandbox;
    const start = sandbox.start({ enableInternet: false });

    return {
      run: (cmd: string[]) =>
        start.pipe(
          Effect.andThen(sandbox.exec(cmd)),
          Effect.flatMap((child) => child.output()),
          Effect.scoped,
        ),
    };
  }),
);

Every Layer in the Durable Object shares one container, so a Layer never stops it.

Wait for a port

getTcpPort waits until the server inside the container accepts connections:

const { fetch } = yield* sandbox.getTcpPort(8080);

const health = Effect.gen(function* () {
  yield* start;
  const response = yield* fetch(HttpClientRequest.get("http://container/health"));
  return yield* response.text;
});

A port never starts the container. If the container is stopped, the request fails with ContainerNotRunningError.

Run a command

exec runs a command in a started container:

const output = yield* Effect.scoped(
  Effect.gen(function* () {
    const child = yield* sandbox.exec(["npm", "test"]);
    return yield* child.output();
  }),
);

exec runs the command without a shell. Pass ["sh", "-c", command] when you need one. Closing the scope kills a process that is still running.

Check whether it is running

if (yield* sandbox.running) {
  // the process is up; its ports may not be ready yet
}

Stop it

destroy stops the container immediately:

yield* sandbox.destroy();

signal sends a signal to its main process instead, so it can shut down on its own:

yield* sandbox.signal(15); // SIGTERM

Let it sleep

When nothing calls the Durable Object, it goes idle and its container stops. Change how long it waits with setInactivityTimeout:

yield* sandbox.setInactivityTimeout(10 * 60 * 1000); // 10 minutes

The next operation that calls start starts the container again.

Watch for a crash

monitor waits until the container exits:

yield* sandbox.monitor().pipe(
  Effect.catchTag("ContainerError", (error) =>
    Effect.logError(`container exited: ${error.message}`),
  ),
);

A crash fails with ContainerError; a clean exit succeeds.

Save a snapshot

snapshotContainer saves the container's filesystem:

const snapshot = yield* sandbox.snapshotContainer({ name: "workspace" });
yield* state.storage.put("snapshot", snapshot);

Store the snapshot in the Durable Object's storage so you can restore it after the container stops. A snapshot holds the writable root filesystem: files, installed packages, build caches. It does not hold memory, running processes, or mounted filesystems.

Snapshots need schedulingPolicy: "durable_object".

Restore a snapshot

Pass the snapshot to start to begin from that filesystem:

const snapshot = yield* state.storage.get<Cloudflare.Containers.ContainerSnapshot>("snapshot");

yield* sandbox.start({
  ...(snapshot ? { containerSnapshot: snapshot } : {}),
  entrypoint: ["sleep", "infinity"],
  enableInternet: false,
});

Without a snapshot, start uses the Container's image. The entrypoint starts fresh. A snapshot can start any number of containers, and expires after 30 days without a restore.

Snapshot before you stop

Save the filesystem, then stop the container, so the next start picks up where this one left off:

const suspend = Effect.gen(function* () {
  const snapshot = yield* sandbox.snapshotContainer({ name: "workspace" });
  yield* state.storage.put("snapshot", snapshot);
  yield* sandbox.destroy();
});

Where next