On 20/02/13 23:30, Thomas Dalton wrote:
On 20 February 2013 12:11, Andre Klapper aklapper@wikimedia.org wrote:
On Wed, 2013-02-20 at 12:08 +0000, Thomas Dalton wrote:
I've just had a colleague send me links to a couple of English Wikipedia articles that were displaying as complete garbage - it looked like corrupt character encoding or something (there was no UI - just a page full of random characters and boxes). Running ?action=purge on them sorted it out, but if he hit upon two corrupted pages in a few minutes, there are probably more.
Does anyone know anything about it?
Not without a testcase (URL) to start investigating. :)
I've fixed the ones I know about, so I don't know if they'll be much help (which is why I didn't specify them before). If it does help, they were:
http://en.wikipedia.org/wiki/Neil_Clark_Warren and http://en.wikipedia.org/wiki/Pepper_Schwartz
(You can draw your own conclusions from my colleague's office reading habits!)
It's not a test case after you've run action=purge on it. If you want to report things like this, it's best if you don't run action=purge, or even report it to anyone who might be inclined to do such a thing. Cache-related test cases are very fragile, so it takes some care to get them to a developer intact.
In the past, there have been problems with gzipped output being served without a Content-Encoding header, due to subtle Squid vary header bugs. But it's hard to tell if that's what happened here, just from your description.
-- Tim Starling