Unicode headers name + SSL + big bodies

[lasote]

When a body is very big, using SSL, the unicode header names filled by aws4auth makes the requests crash:

OpenSSL.SSL.Error: [('SSL routines', 'SSL3_WRITE_PENDING', 'bad write retry')]

More info: https://www.bountysource.com/issues/27417596-ssl3_write_pending-error-from-urllib3-contrib-pyopenssl-sendall

I've solved making a wrapper that converts to str the header names:

    class AWS4AuthNotUnicode(AWS4Auth):
        def __call__(self, req):
            req = super(AWS4AuthNotUnicode, self).__call__(req)
            req.headers = {str(name): value for name, value in req.headers.items()}
            return req

[jamshid]

Interesting -- I wonder if #29 is related to this. Does removing this line from /usr/lib/python2.7/site-packages/requests_aws4auth/aws4auth.py also workaround the error you see with SSL?

from __future__ import unicode_literals

[jminuscula]

@jamshid after much debugging with the help from folks over at urllib3, the unicode headers are definitely the issue.

Commenting out the unicode_literals import fixed the problem for me. From the discussion on
urllib3/urllib3#855 (comment) I take that headers should not be provided as unicode strings.

Hope this helps.

[DavidLemayian]

@lasote Thank you! Solved with your workaround.

[cristianocca]

How to work around this without patching the library?

EDIT: Monkey patch on the first post solves the issue.

[baragona]

The AWS4AuthNotUnicode fix worked for me. Wondering if this is also an issue on python3? (For future reference, this code was on python2.)