generate_presigned_url is 3x as slow as boto2 for s3 resources

When calling generate_presigned_url on a s3 resource, the resulting calls are 3x as slow as the boto2 equivalent…

In [1]: import timeit
In [7]: setup = '''
   ...: import boto3
   ...: from boto.s3.connection import S3Connection
   ...: b3_conn = boto3.client('s3')
   ...: b2_conn = S3Connection(None, None)
   ...: '''
In [13]: print min(timeit.Timer('b2_conn.generate_url(3600, "GET", bucket="foo-bucket", key="foo-key", query_auth=True)', setup=setup).repeat(7, 10000))
1.95865702629
In [14]: print min(timeit.Timer('b3_conn.generate_presigned_url("get_object", Params={"Bucket": "foo-bucket", "Key": "foo-key"}, ExpiresIn=3600, HttpMethod="GET")', setup=setup).repeat(7, 10000))
5.69024801254

This was run on c3.xlarge instances in us-west-2.

[Dizzzmas]

Ran into this problem today. Found it really frustrating, as it was increasing the response time of my API endpoint by 3-4 seconds which was rather significant. And I was generating URLs for only ~30 objects.

Any news on this being addressed in the near future?

[pabdelhay]

I had the same problem. Solved by setting AWS_ACCESS_KEY and AWS_SECRET_ACCESS_KEY.

When I migrate my servers to ECS it was no longer necessary to use AWS_ACCESS_KEY and AWS_SECRET_ACCESS_KEY, since everything was on the same security layer. But when I removed those credentials, my application starts performing low on generating_presigned_url from S3. The solutions was to rollback and set those credentials (AWS_ACCESS_KEY, AWS_SECRET_ACCESS_KEY) again even it is not mandatory on my new environment.