HTTPProxy cannot be aware of timeout of waiting response

[silentred]

In case using httputil.ReverseProxy, the error returned by RoundTrip(), is only recorded by Logger. https://github.com/golang/go/blob/master/src/net/http/httputil/reverseproxy.go#L203 .

Therefore, it seems the HTTPProxy cannot distinguish whether it is a timeout or a normal response, when it tries to gather the metrics. https://github.com/fabiolb/fabio/blob/master/proxy/http_proxy.go#L134

Also, the backend service timeout causes the response have 502 code and without body. And the body cannot be customized right now. The worst is timeout error is not logged with enough context information, for example, which service or which URI gets timeout.

[magiconair]

I can always fork httputil.ReverseProxy and hack this in. Then file an issue with the Go team and hope for a future fix.

[magiconair]

What is the behavior you would like to see?

[magiconair]

Also, I'm already injecting a single-use round tripper which captures the *http.Response so that I can use it for access logging. I've just pushed a change that captures the error there as well. This allowed me to simplify that code.

https://github.com/fabiolb/fabio/blob/master/proxy/http_handler.go#L31-L43

[silentred]

Thanks. Another problem is ReverseProxy returns 502 without body to the client. I am hoping to customize the body according to each err type. It seems I have to copy ReverseProxy and make my own version. I wonder if there is a simpler way to achieve that.

[silentred]

I guess I found a work-around. Replace the out writer of ErrorLog in ReverseProxy with a custom one, which writes a JSON format body to the client. I am working on it.
Update:
It works fine for me.