SigV4Auth should not use cached timestamps for signing requests.

[willmcg]

Versions

Python Interpreter Version: 2.7.5
Botocore Version: 18.0

Issue Description:

The SigV4Auth class generates a signature timestamp in its init and uses that timestamp for signing all requests for the endpoint. This means that an endpoint can only generate correctly signed requests for 5 minutes before the requests are rejected by AWS services as having an expired timestamp.

For example:

Signature expired: 20131008T001036Z is now earlier than 20131008T001112Z (20131008T001612Z - 5 min.)

Use Case

Ran into this problem when using botocore to construct complex CloudFormation stacks that took extended periods of time to complete. I was getting signature expired errors from CloudFormation when polling for completion of the stack construction after 5 minutes because it was using the same endpoint object to poll as it used to issue the initial CreateStack.

Reproducing the Issue:

session = botocore.session.get_session()
service = session.get_service('ec2')
endpoint = service.get_endpoint('us-east-1')
<wait more than 5 minutes>
operation = svc.get_operation(<method>)
operation.call(endpoint, ...)

This will result in a 403 error with a signature expired error like this:

Signature expired: 20131008T001036Z is now earlier than 20131008T001112Z (20131008T001612Z - 5 min.)

Suggested Fix

Instead of generating a cached signature timestamp in init for SigV4Auth, it should be generated dynamically for each signature operation.

Workaround

The issue can be worked around by calling get_endpoint() for the service before each operation call which will cause a new signature timestamp to be regenerated for requests on the new endpoint and thus avoid the expired timestamp issue.