temporalio.contrib.aws.lambda_worker

Create a Temporal worker Lambda handler.

Calls the configure callback to collect workflow/activity registrations and option overrides, then returns a Lambda handler function. On each invocation the handler connects to the Temporal server, starts a worker with Lambda-tuned defaults, polls for tasks until the invocation deadline approaches, and then gracefully shuts down.

The configure callback is invoked once per invocation and may be synchronous or asynchronous:

• Synchronous def configure(config) -> None — runs per invocation. Use for static worker definition (task queue, registrations, option tuning) and resources that are not bound to an event loop.
• Async async def configure(config) -> None — awaited per invocation. Use when setup must await (for example, opening an async client). Pair with shutdown_hooks for teardown.
• Async generator async def configure(config): ...; yield; ... (or an equivalent @contextlib.asynccontextmanager-decorated function) — entered per invocation. Statements before the single yield run before the client connects; the worker runs while the generator is suspended at the yield; statements after the yield run as teardown once the worker has stopped. Any shutdown_hooks registered before the yield run after the worker stops but before the post-yield teardown, so this resource outlives the hooks (e.g. a telemetry flush hook can still emit before the resource is closed). This is the recommended shape for event-loop-bound resources that must live for the duration of the invocation, such as an aioboto3 S3 client backing the external-storage data converter (see the async example below).

The callback runs per invocation (not once at cold start) because event-loop-bound resources cannot be created at cold start (there is no running loop) and cannot be shared across invocations (each invocation runs under a fresh asyncio.run loop).

The version parameter identifies this worker's deployment version. run_worker always enables Worker Deployment Versioning (use_worker_versioning=True). To provide a default versioning behavior for workflows that do not specify one at registration time, set deployment_config in worker_config in the configure callback.

The returned handler has the signature handler(event, context) and should be set as your Lambda function's handler entry point.

from temporalio.common import WorkerDeploymentVersion
from temporalio.contrib.aws.lambda_worker import (
    LambdaWorkerConfig,
    run_worker,
)

def configure(config: LambdaWorkerConfig) -> None:
    config.worker_config["task_queue"] = "my-task-queue"
    config.worker_config["workflows"] = [MyWorkflow]
    config.worker_config["activities"] = [my_activity]

lambda_handler = run_worker(
    WorkerDeploymentVersion(
        deployment_name="my-service",
        build_id="v1.0"),
    configure,
)

import aioboto3
import dataclasses
from temporalio.contrib.aws.s3driver import S3StorageDriver
from temporalio.contrib.aws.s3driver.aioboto3 import new_aioboto3_client
from temporalio.converter import DataConverter, ExternalStorage

session = aioboto3.Session()

async def configure(config: LambdaWorkerConfig):
    config.worker_config["task_queue"] = "my-task-queue"
    config.worker_config["workflows"] = [MyWorkflow]
    async with session.client("s3") as s3_client:
        driver = S3StorageDriver(
            client=new_aioboto3_client(s3_client), bucket="my-payloads",
        )
        config.client_connect_config["data_converter"] = dataclasses.replace(
            DataConverter.default,
            external_storage=ExternalStorage(drivers=[driver]),
        )
        yield

lambda_handler = run_worker(
    WorkerDeploymentVersion(
        deployment_name="my-service",
        build_id="v1.0"),
    configure,
)