Added client_max_body_size to authPath location

[olvesh]

Seems like nginx denies the request because it would be over the max body size,
even if proxy_pass_request_body is off.

This fixes #811

Not sure if it is straight forward to build tests for this. Hope this is ok as it is. For me this fixed my issues at least. Built a docker image and swapped the official one for one with this fix and it works.

[k8s-reviewable]

This change is 

[coveralls]

Coverage decreased (-0.03%) to 46.476% when pulling d4600a8 on olvesh:externalauth_client_max_body_size_811 into 4c868cf on kubernetes:master.

[aledbf]

@olvesh I think we need to fix the real issue here, i.e. why the auth block is passing the body.

[aledbf]

@olvesh can you test moving this line 383 to line 349 (before the auth section) and remove the line you added?

I think the issue here is that we need to accept the request first, that works because we are setting the max size before the proxy_pass directive but we need to move the configuration before the auth_request.

[olvesh]

Tried your suggestion and it does not work. Still this error in log file:

2017/06/06 13:49:34 [error] 25#25: *29 client intended to send too large body: 8964178 bytes, client: 82.134.26.130, server: rundeck.example.com, request: "POST /project/CI-env/importArchive HTTP/2.0", subrequest: "/_external-auth-Lw", host: "rundeck.example.com", referrer: "https://rundeck.example.com/project/CI-env/configure"
2017/06/06 13:49:34 [error] 25#25: *29 auth request unexpected status: 413, client: 82.134.26.130, server: rundeck.example.com, request: "POST /project/CI-env/importArchive HTTP/2.0", host: "rundeck.example.com", referrer: "https://rundeck.example.com/project/CI-env/configure"

Relevant config result here: https://gist.github.com/olvesh/42bf8a6769e6fc41d07bfb9a6c710fc6

Nginx preemptively stops the request since it knows it will be to big given that it is a sub request. I think the logic here is that since it is an internal call Nginx can see that it will exceed the max body size, but it does not know that send_body is off thus denying the subrequest.

It could be put outside of the location element, in the server group instead, but then I think more refactoring is needed.

[aledbf]

It could be put outside of the location element, in the server group instead, but then I think more refactoring is needed.

We cannot do that because it removes the feature of custom body size per path

[aledbf]

@olvesh thanks!

[aledbf]

/lgtm

[aledbf]

@olvesh I need to check if the auth_request is passing the body but I think this PR improves the current situation

[olvesh]

@aledbf Thanks!
I tried to see if the size of the post on the internal sub-request appeared, but couldn't see that the request was logged, even the setting log_subrequest on; is present.

[olvesh]

@aledbf btw, when will you (someone) cut a new beta of the ingress controller? Would like to drop the local snowflake we have now :-)

[aledbf]

@olvesh after completing this list https://github.com/kubernetes/ingress/projects/3