try to fix #1905 and add some notes

[pwli0755]

This tiny modification tries to avoid issue #1905 and add some notes about the global error handler.
If there are any better solutions, please ping me.

[codecov]

Codecov Report

Merging #1947 (c302290) into master (7d41537) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1947   +/-   ##
=======================================
  Coverage   91.03%   91.04%           
=======================================
  Files          32       32           
  Lines        2800     2802    +2     
=======================================
+ Hits         2549     2551    +2     
  Misses        160      160           
  Partials       91       91           

Impacted Files	Coverage Δ
middleware/timeout.go	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7d41537...c302290. Read the comment docs.

[aldas]

@pwli0755 could you rebase this pr?

[pwli0755]

@pwli0755 could you rebase this pr?

Done! But it seems codecov reports a warning, does that matter?

[aldas]

So it seems that order of middlewares with timeout middleware is important. If timeout is after logger then we have data race because we change that response Writer around and some middleware use it to get Headers for example.

My personal opinion:
sigh I really dislike this middleware because it so complex and has so many hidden "features" and I'm too dumb to fix all those problems in a reasonable way.

[aldas]

I'll merge this for now. @pwli0755 thanks for your PR and if you have ideas for this middleware let us know.

[pwli0755]

So it seems that order of middlewares with timeout middleware is important. If timeout is after logger then we have data race because we change that response Writer around and some middleware use it to get Headers for example.

My personal opinion:
sigh I really dislike this middleware because it so complex and has so many hidden "features" and I'm too dumb to fix all those problems in a reasonable way.

Yeah, I think the HTTP client is more responsible for the timeout things.

[aldas]

I totally agree and when semi "background processing" is needed this should be explicitly coded in handlers.