Timeout makes the next request to the same domain fail

[honux]

Heya.

Whenever a request fail with a socket.timeout, the following request to the same domain fails. Example:

import httplib2
h = httplib2.Http(timeout=3)
h.force_exception_to_status_code = True

head_1, cont_1 = h.request('https://httpbin.org/delay/60', headers={'cache-control':'no-cache'})
# head_1: {'status': '408', 'content-type': 'text/plain', 'content-length': 15}
# cont_1: b'Request Timeout'

# right after the timeout, request anything to the same domain
head_2, cont_2 = h.request('https://httpbin.org/delay/2', headers={'cache-control':'no-cache'})
# Instead of a 2 second delay, it will reuse the connection made on the previous request, making the response timeout / have the response of the previous request.

I'm using the latest httplib2 and python3

---

If this is expected, is there any way to really 'drop' the connection? Forcing a new connection instead.

[justinfx]

Out of curiosity, if I am going to handle this case myself, is it sufficient to catch the socket.timeout exception, and then try and pop and close the connection from h.connections?

[temoto]

@justinfx sounds good. I'd start from a test looking like this:

  def handler(request):
    if request.number == 1:
      time.sleep(0.6)
      return tests.http_response_bytes(status=500)
    return tests.http_response_bytes(status=200)

  http = httplib2.Http(timeout=0.5)
  with tests.server_request(handler, request_count=2) as uri:
    with tests.assert_raises(...):
      http.request(uri)
    response, _ = http.request(uri)
    assert response.status == 200

[justinfx]

@temoto I have a fix and a test locally. The fix works when I manually verify it, but I am having some trouble getting it to replicate the broken behaviour in the test because of the way it seems a new connection is being swapped in already by the test behaviour. I will end up submitting a wip MR for this, but thought in the meantime I would ask to see if you know off the top of your head why the test would be side stepping the standard logic of the same connection being reused on the second request?

[temoto]

@justinfx I don't see your code.

[thehesiod]

nice! I was just about to fix this myself ;)

[ClamsTheCat]

@justinfx commits fixed my issue as well

[temoto]

Please try latest release v0.12.0 with Justin fix.

[justinfx]

@temoto I didn't see that this was merged. There is no mention of it in the changelog and this issue is still open.
Are you referring to this other thing that came from a suggestion I had made?
211d6f0

[temoto]

@justinfx yes I confused things up, sorry.

[temoto]

Fix by Justin is merged in master now.

[justinfx]

Thanks!

[max-sixty]

Can this be closed now?

[temoto]

@max-sixty thanks