Page MenuHomePhabricator

[Regression] doc.wikimedia.org is unreliable (Redirect loop, DNS problems)
Closed, ResolvedPublic

Description

https://doc.wikimedia.org/mediawiki-core/master/php/html/ linked from https://doc.wikimedia.org/ is giving me 'infinite redirect loop'.


Version: wmf-deployment
Severity: major

Details

Reference
bz60822

Event Timeline

bzimport raised the priority of this task from to High.Nov 22 2014, 2:52 AM
bzimport set Reference to bz60822.

Works for me too. Browser info etc needed.

Worked once for me when by-passing cache, but it is there again. Seems to be intermittent?

Pau was able to reproduce. We are both using Chrome.

(In reply to comment #3)

Worked once for me when by-passing cache, but it is there again. Seems to be
intermittent?

(In reply to comment #4)

Pau was able to reproduce. We are both using Chrome.

Sounds similar to bug 56842. FWIW, I've never seen such problems with Chromium and I can't reproduce in this case either (only tried a dozen times though).

Cannot reproduce either with Google Chrome 32.0 on Fedora 20.

Created attachment 14482
Proof in two browsers

It's not a browser related.

It doesn't make a sense unless there is a broken server behind load balancing to explain why this problem is intermittent.

Attached:

hfiwurhgiwurhgiwahi2u4htazuhigah3qiaehi28yf33f.png (629×1 px, 64 KB)

The integration websites have been migrated behind the misc-varnish box so pages ends up being cached there.

Using Chrome development tools, can you grab the response headers?

Also if you try the page and append a cache breaker (i.e. ?sometimestamp), that should serve the page.

FWIW I hit this issue this morning, but was unable to reproduce after opening Firefox's web console once.

This has been happening for little over a week now. Ever since it was changed from being served from misc-web instead of gallium, something got messed up in the dns or apache config that causes it to get stuck in a redirect loop sometimes.

It seems to usually be mitigated by chance (e.g. change the url with a bogus query param, or flush local dns cache).

Recent changes that might have contributed to this problem;

operations/dns:

doc and integration.mediawiki.org go to text varnish, not misc

Change-Id: I62d29d9c552acd86b29663eb961d67b21fddf1e5

  • fix cname for doc.wikimedia.org

    Change-Id: Ie483903b6f77270a4295672b5a0a5f6a2375e3e4 RT: 6594
  • moving doc.wikimedia.org to misc-web-lb

Change-Id: Ie80551a532dbc32cbc9c0d73c396e88f63fcb0b3
RT: 6594

operations/puppet:

doc.wikimedia.org and doc.mediawiki.org to use misc-web-lb

moving https termination from gallium to the misc-web-lb.
this change will go live at the same time as the dns change

Change-Id: I2fe62629f4deaf4c1ace4345ba7c743b84469fd9
RT: 6594

Change 111917 had a related patch set uploaded by BryanDavis:
Send Vary header on http to http redirect

https://gerrit.wikimedia.org/r/111917

Change 111917 merged by Faidon Liambotis:
Send Vary header on http to http redirect

https://gerrit.wikimedia.org/r/111917

The patch proposed by Bryan Davis and later amended by me fixed up the issue :-]