taskferry-sqs

Run Taskferry tasks on AWS SQS — pull-based delivery, consumed by a Lambda, an ECS service, or your own loop.

flowchart LR
    APP["Application"] --> TP["Taskferry"] --> AD["taskferry-sqs"] --> Q["SQS queue"] --> C["Lambda · ECS · Fargate"]

pip install 'taskferry-sqs[aws]'

Sending

runtime = Taskferry.from_mapping(
    {
        "backends": {
            "sqs": {
                "factory": "sqs",
                "queue_url": "https://sqs.eu-west-1.amazonaws.com/123456789/tasks",
            }
        },
        "defaults": {"task": "sqs"},
    }
)

runtime.tasks.submit("myapp.tasks:send_email", 42, queue="email")

Consuming

AWS Lambda, with partial batch responses enabled:

from taskferry import FunctionRegistry
from taskferry_sqs import process_event

REGISTRY = FunctionRegistry(allowed_modules=["myapp"])


def handler(event, context):
    return process_event(event, registry=REGISTRY)

ECS / Fargate / a plain process:

import boto3
from taskferry_sqs import poll_forever

poll_forever(boto3.client("sqs"), QUEUE_URL, registry=REGISTRY)

poll_forever long-polls, deletes on success, leaves failures for SQS's own redrive policy, and stops cleanly on SIGTERM. It is forty lines and deliberately not a worker framework — SQS already solves visibility timeouts and dead-lettering.

Capabilities

The capability set is computed per queue, because a .fifo queue genuinely differs from a standard one and one class serving both must say which it is.

Two limits are enforced rather than clamped:

• 900 seconds of delay, maximum. A longer delay raises with the limit named, instead of arriving 15 minutes from now when you asked for tomorrow.
• FIFO queues refuse per-message delay rather than dropping it silently.

Testing

The client is injectable, so the suite runs with no AWS account:

SQSTaskBackend(queue_url="https://sqs.eu-west-1.amazonaws.com/1/q", client=FakeSQS())

License

Apache-2.0.