Celery-S3

Celery-S3 is a simple S3 result backend for Celery.

If used in conjunction with the SQS broker, it allows for Celery deployments that use only distributed AWS services -- with no dependency on individual machines within your infrastructure.

This backend probably isn't suitable for particularly high-traffic Celery deployments, but it works just fine in general -- and imposes no limits on the number of workers in the pool.

Installation

Install via pip:

pip install celery-s3

Then configure Celery to use the S3Backend:

CELERY_RESULT_BACKEND = 'celery_s3.backends.S3Backend'

CELERY_S3_BACKEND_SETTINGS = {
    'aws_access_key_id': '<your_aws_access_key_id>',
    'aws_secret_access_key': '<your_aws_secret_access_key>',
    'bucket': '<your_bucket_name>',
}

Configuration

To use a folder within the specified bucket, set the base_path in your CELERY_S3_BACKEND_SETTINGS:

CELERY_S3_BACKEND_SETTINGS = {
    ...
    'base_path': '/celery/',
    ...
}

To use a region other than the default (us-east-1), set the aws_region parameter:

CELERY_S3_BACKEND_SETTINGS = {
    ...
    'aws_region': 'us-east-1',
    ...
}

To use reduced redundancy storage, set the reduced_redundancy parameter:

CELERY_S3_BACKEND_SETTINGS = {
    ...
    'reduced_redundancy': True,
    ...
}

To use server-side encryption, set the encrypt_key parameter:

CELERY_S3_BACKEND_SETTINGS = {
    ...
    'encrypt_key': True,
    ...
}

Notes

Storing Celery results with this backend will obviously result in API calls being made to Amazon S3. For each result, at least one PUT request will be made (priced at $0.01 per 1,000 requests at the time of writing). Also, the data contained within the result object will be stored indefinitely, unless otherwise specified.

To fetch a result for a task that has already finished, at least two requests will be made (one HEAD and one GET). If you use Celery's result.get() to wait for a task to finish, S3 will be polled continuously until the task has finished.

By default, the poll interval is set to 0.5 seconds, which could result in a lot of requests (two HEAD requests per second until the task has finished, then one GET request to fetch the result). If you need to use result.get(), consider increasing the interval and using a timeout to prevent polling forever: result.get(interval=5, timeout=600).

Also, for tasks whose result you don't need, be sure to use ignore_result:

@celery.task(ignore_result=True)
def process_data(obj):
    obj.do_processing()

Once task results have been used and are no longer needed, be sure to call result.forget() to delete the corresponding S3 key. Otherwise, old results will remain forever and contribute to storage costs (storage is priced at $0.095 per GB per month at the time of writing).

Also, the S3 lifecycle can be used to archive or delete old keys after a certain period of time.