Canonicalize domains and redirect in proxito #6905
Conversation
- Only redirect if the appropriate flag is set on the Domain (default is not to redirect) - Use a 302 redirect for now. Ideally this is a 301 redirect, but redirecting other people's domains with a 301 is always a bit risky.
- Do HTTP -> HTTPS redirects if domain.https is set - Redirect public domain to canonical CNAME if domain.canonical - Redirect CNAME -> public domain if the custom domain is not canonical - Set additional data on the X-RTD-Redirect header whether the redirect is to canonicalize, a system redirect, or a user redirect
| r = self.client.get('/', HTTP_HOST=self.domain.domain) | ||
| self.assertEqual(r.status_code, 302) | ||
| self.assertEqual( | ||
| r['Location'], f'https://{self.domain.domain}/en/latest/', |
There was a problem hiding this comment.
We are doing only 1 redirect now, instead of 2. I think we were doing,
- http -> https
- https -> /en/latest/
Now we are doing http -> https /en/latest/ in one step. Is this correct?
Example,
$ curl --silent -IL http://docs.readthedocs.io/ | grep Location
Location: https://docs.readthedocs.io/
Location: https://docs.readthedocs.io/en/stable/
There was a problem hiding this comment.
Ideally there are two redirects. I actually think that for *.readthedocs.io there will be two since the http -> https redirect will be handled by nginx before it hits proxito. For custom domains that need to both change domain and scheme, this change does use a single redirect.
| @@ -176,7 +176,26 @@ def system_redirect(self, request, final_project, lang_slug, version_slug, filen | |||
| ) | |||
| log.info('System Redirect: host=%s, from=%s, to=%s', request.get_host(), filename, to) | |||
| resp = HttpResponseRedirect(to) | |||
| resp['X-RTD-System-Redirect'] = True | |||
| resp['X-RTD-Redirect'] = 'system' | |||
This is a tricky one. I'm not sure to understand if this is something wanted at all, though. Why would you create only one custom domain in RTD but expecting to keep serving the docs under I can see the use case of having multiple-domains for your site and wanted one to be the canonical, though. Some quick numbers in .org |
I don't know. Perhaps historical reasons? Regardless, I don't think we should serve the same docs from multiple domains. We currently do that and I'd like to get away from it if possible. |
I think we should comment this code out with a TODO & xfail the tests for it. It seems less valuable than the rest of this PR which seems 👍. We can address the canonical domain stuff later -- I think we could probably just only respect the |
|
I removed the redirect from custom domains -> public domain if the domain is not canonical for now. This leaves only:
|
|
We should definitely plan on QA'ing this one |
domain.httpsdomain.canonicalRedirect CNAME -> public domain (or another canonical domain) if the requested custom domain is not canonicalX-RTD-Redirectheader whether the redirect is to canonicalize, a system redirect, or a user redirectIf this is merged, #6882 can be closed as it won't be needed. This is an alternative that uses the same code from the resolver to perform the redirect to the appropriate domain.
Fixes: #4641
Concerns
I don't think the HTTP -> HTTPS redirect is a concern. Users who set
domain.httpswill be happy with the change and users who didn't won't see a change.The canonicalizing domain change is a little more worrying. Users who don't havedomain.canonicalset might be surprised their domain is now redirecting toslug.readthedocs.io/...(or the equivalent on readthedocs.com). Maybe we need a big note in the changelog about settingdomain.canonical?