TykTechnologies / tyk / 2950 / 2

Ran 13 Sep 2017 09:42AM UTC

Files 84

Run time 3s

Badge

Embed ▾

Committed 13 Sep 2017 09:38AM UTC coverage: 44.033% (+0.1%) from 43.917%

Job # LATEST_GO=true

Build Type

push

travis-ci

Committed by

mvdan

Commit Message

Fix SkipUrlClean by not using DefaultServeMux

The issue itself happen because we wrap `gorilla.Mux` router through
`http.ServeMux`, which is kinda wrong, because `Mux` is already a
`ServeMux` implementation. So we processing routes twice.

Default `ServeMux` implementation have own route sanitization logic, and
can’t be configured with `SkipClean` option, like gorilla
implementation.

The solution here is to stop using ServeMux and use Gorilla directly.

Problem with this approach is that we need to reload, eg. replace, the
router inside `doReload` function. Neither gorilla or ServeMux support
replacing the routes, so replacing the whole router object is the only
solution. When we pass raw Gorilla Router object to the
`Server.Handler`, and replace `mainRouter` in `doReload`, server do not
see the changes: thats why we had issues when API reloads was not
working.

There are 2 ways to solve this issue: expose server object as a global
variable, and dynamically replace `Handler` attribute, or write simple
`http.Handler` wrapper around `mainRouter`, which always calls the
current router.

I’m not sure about safety of this approach but since previously we
replaced `DefaultServeMux` directly, and had no any issues, I may assume
that it should be thread safe.

Should fix #1093

Coverage Stats

5937 of 13483 relevant lines covered (44.03%)

0.48 hits per line

TykTechnologies / tyk / 2950 / 2

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 2950.2 (LATEST_GO=true)

TykTechnologies / tyk / 2950 / 2

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job 2950.2 (LATEST_GO=true)

README BADGES
x