I made further tests and tried to use same approach of tc_aws to re-use s3_client.
I changed s3_client S3Client class to:

class S3Client:
    _client = None
    _instances = {}

    @staticmethod
    def __new__(cls, context):
        key = (context.config.AWS_LOADER_BUCKET_NAME, context.config.AWS_LOADER_REGION_NAME, context.config.AWS_LOADER_S3_ENDPOINT_URL)

        if not cls._instances.get(key):
            cls._instances[key] = super(S3Client, cls).__new__(cls)

        return cls._instances[key]

and checked if self._client does exist before creating a new one.

As a result performance are now comparable to the tc_aws one!

Mine was only a test, but i can confirm you that the difference in response time is caused by the "multiple" s3 client initialization compared to initialize it just one time (for bucket, region, endpoint).

That fix work only if you do not put code in an async with context and so do not use aiobotocore > 0.12.

Just for info the PR #19, at the end re-use only the session not the client, and the bigger effort is on creating the whole s3 client.

Currently we have a fast tc_aws but that to re-use aiobotocore s3 client need to use an old (very) version.
And an up to date thumbor-aws that is not usable in production, compared to tc_aws, because increase reponse time of 0.2 sec (and under an high number of request, also memory usage).

Thanks for this heads up

just for info i have mixed tc_aws solutions amanagr/aws#1 and thumbor-community/aws@b5058e6 in thumbor-aws
(for the second one i had to put _client = None in class definition root and not in class __init__)

This way i have thumbor-aws that use a persistent s3 client and that work using an up to date aiobotocore [2.x].
(i have it running in production without any problem so far)

This issue is stale because it has been open 30 days with no activity. Remove the stale label or add a comment, or this issue will be closed in 5 days. You can always re-open if you still feel this is still an issue. Tag @heynemann for more information.

This issue was closed because it has been stale for 5 days with no activity.