Make 408's not result in orphan mitigation

[duglin]

/cc @nilebox

Closes #449

Signed-off-by: Doug Davis dug@us.ibm.com

[cfdreddbot]

Hey duglin!

Thanks for submitting this pull request! I'm here to inform the recipients of the pull request that you and the commit authors have already signed the CLA.

[duglin]

While we could technically merge the two 4xx lines, I think its important to call out the 408 by itself so we can be clear that this is a "client side" timeout and not a "server side" one - which are very very different w.r.t. orphan mitigation

[nilebox]

TBH I would prefer to remove the special mention of 408 completely, as it's not actually special as we discussed in #449. All 4xx are client side errors, and there are no exceptions in regards to orphan mitigation.
We also don't mention the 504 Gateway Timeout explicitly anywhere apart from 5xx, for example.

[duglin]

@nilebox my original version of the PR did merge them and within 5 minutes I got a ping about it because the distinction between client-side vs server-side timeout wasn't clear - so I added it back in, just for the sake of clarity. We don't need to call-out 504 because it results in orphan mitigation which is what most people would expect from a "timeout".

[nilebox]

@duglin I see your concern, but the distinction between client-side and server-side errors is an HTTP standard's problem, OSB doesn't bring anything additional there. Furthemore, HTTP standard makes the distinction very clear - 4xx are client-side errors, and 5xx are server-side errors. If we needed to alter this distinction (which OSB spec currently does), it would obviously require a special wording in the OSB spec. But since we don't (anymore), I don't see the value of keeping 408 separate from 4xx.

[n3wscott]

I don't understand the usage of 'server' here. A client could get a 408 by having it's local timeout setting very small. Or you could be talking to a broker proxy and the second connection could timeout. Both cases could result in orphans. I am not sure how a broker author could prevent their implementation from returning a 408 AND/OR make sure that orphans are not possible.

[duglin]

perhaps "receiver" would be better, but in the other cases you mentioned I'm not sure the Platform would see a 408. As long as the message makes it to the next hop I don't think 408 would be appropriate. And I would think that hop before that one would return a 5xx since from its POV it did receive the entire request so it can't be a 408. Right?

[n3wscott]

Looking up 408 in the rfc...

Ok I agree. In this case, server is fine with me.

[n3wscott]

LGTM

[duglin]

reviews needed

[fmui]

LGTM

[mattmcneeney]

LGTM

[zrob]

lgtm

[n3wscott]

LGTM

[duglin]

2 more reviews needed
@fmui @vaikas-google @mattmcneeney

[fmui]

LGTM

[vaikas]

LGTM