This is an archived page covering the Mediawiki software and its use at Wikimedia. It covers the time of first use and issues with the initial use of Squid for caching of content served by MediaWiki. From 2004 Squid servers delivered in excess of 75% of all hits to the sites and made a massive difference in site scalability.
Since Monday, 2 Feb 2004 Larousse runs a en:Squid cache proxy. It's serving about 2/3 of all traffic straight from its ram and disk, this traffic never reaches the Apache and DB. Pages for logged-in users can't be cached because of user preferences, different links and so on. So no caching after login. Still the speed for logged-in users is much improved by lowering the load on Apache and DB.
In the first days a few bugs were found, most should be fixed now. We hope to get the new servers running soon, in the meantime Squid allows us to get away with just three servers running.
Open issues edit
Please add new issues here (or move solved ones up if it's not fixed for you).
Does Squid cached Accept-Language? edit
In wikis with variant languages like Chinese Wikipedia, LanguageConverter use $_SERVER['HTTP_ACCEPT_LANGUAGE'] to determine what variant was used by an anonymous user. But it often return a wrong variant. For example, a zh-cn user was redirected to a zh-tw page. I suppose it may be a Squid problem. Can someone help me? --PhiLiP 15:36, 20 August 2009 (UTC)
Solved issues edit
Have I been squidded? edit
This is the second time I tried to upload a sound clip and got this message (before the 100k warning, but after the upload appeared to be complete in the progress bar). The upload was apparently unsuccessful this time, though the first time it happened, last night, it did work (IIRC). After getting this message once today, I tried again with the same clip and the same thing happened. Tuf-Kat 19:05, Feb 9, 2004 (UTC)
ERROR The requested URL could not be retrieved
While trying to retrieve the URL: http://en.wikipedia.org/wiki/Special:Upload
The following error was encountered:
* Write Error
The system returned:
(32) Broken pipe
An error condition occurred while writing to the network. Please retry your request.
Your cache administrator is webmaster. Generated Mon, 09 Feb 2004 18:59:03 GMT by wikipedia.org (squid/2.5.STABLE4)
- Well, all requests pass through the Squid, but only anon requests are allowed to be cached. This could be an Apache error, possibly Apache was restarted during your request. How big is the file you uploaded? Afaik there's an upload size limit of 2Mb in php, and there was a limit at 1Mb in squid. I just increased this to 3Mb, that will kick in after reloading the squid config. Please let us know if this happens again. -- Gabriel Wicke 23:42, 9 Feb 2004 (UTC)
Limit is 7Mb now. -- Gabriel Wicke 22:06, 12 Feb 2004 (UTC)
Printable pages not up-to-date (solved 6 Feb '04) edit
Hehe- this problem didn't really exist in the first place, but i made a mistake while testing it. Tried "ab -v9 http://en.wikipedia.org/w/wiki.phtml?title=Squid&printable=yes" which caused ab to omit the &printable=yes...
"ab -v9 'http://en.wikipedia.org/w/wiki.phtml?title=Squid&printable=yes'" does the trick. Grepping the Squid log for 'printable' confirms this- print pages are not cached. -- Gwicke 16:22, 6 Feb 2004 (UTC)
Moved from the en:Wikipedia:Village Pump.
I just had a strange experience and I wonder if anyone else has had the same. I use Mozilla for browsing Wikipedia, and I have saved my User Page, Recent Changes and Watchlist as a group, so I click one button when I start Mozilla and all three pages come up (kind of) simultaneously. Well just now I did this, and whilst my User Page came up fine, the other two swore blind that I was no longer logged in. Is this some sort of strange cookie behaviour or is there a different explanation? (BTW rather faster today than yesterday it seems) --Phil 09:40, Feb 3, 2004 (UTC)
- We've just added a 'private' statement to the Cache-Control headers sent out by Squid. Some versions of IE apparently don't follow the http standard on a 'Cache-Control: max-age:0, must-revalidate, s-maxage=0' header. Hope this fixes it. -- Gabriel Wicke 16:14, 3 Feb 2004 (UTC)
- Sennheiser adds that IE doesn't usually follow standards.
- Thanks for the info, but since when has Mozilla been bothered by IE problems? I was under the impression that Mozilla is related to Netscape. --Phil 10:59, Feb 4, 2004 (UTC)
- Well, Gecko is the HTML (etc) rendering part of Mozilla. It switches from "standards compliant" to "quirks" mode if a page is not written correctly, and tries to behave how the page's author probably intended (i.e. it emulates oft-exploited harmless IE bugs). Cookies and caching would have nothing to do with this anyway, as far as I know, they'd be covered elsewhere in the code (Necko?) - IMSoP 15:41, 4 Feb 2004 (UTC)
- You're right of course. I posted my reply after discussing a similar report (IE in that case) on irc, oviously i didn't read very careful.
- Caching is only on if no cookie is sent (anons), and even then the client is forced to revalidate the page on every request. When the browser sends a cookie it's the same as before (no caching on the Squid). Monday was a rough start, especially with Squid3 exhibiting unexpected bugs. It's not unlikely that some pages were sent out with improper header for short times. Now the dust has settled, the setup seems to work rather well. Is this problem still reproducible? -- Gabriel Wicke 20:08, 4 Feb 2004 (UTC)
- Update: i've tested this again with firebird, a set of pages opens logged- in for me. -- Gabriel Wicke 12:00, 5 Feb 2004 (UTC)
VP text not appearing edit
Probably a dumb question, but why is it that text that I can see in the history of this page is not visible in the page itself (e.g. Adam Carr's lates addition)? Bmills 10:52, 2 Feb 2004 (UTC)
Yes, our interesting debate this afternoon (my time) is now all screwed up. Parts of it are repeated several times and the later part of it is missing. Either someone has (deliberately or acceidentally) messed it up, or a bug has crept it. There seem to have been a few of them loose recently. Adam 12:46, 2 Feb 2004 (UTC)
I suspect that Save bug that's been around. I have restored the text manually from Page history. Bmills 14:09, 2 Feb 2004 (UTC)
I see that there's still some missing--specifically, my original post about a rocks & minerals table, to which there are responses now but they look like they're a continuation of the Copyright 2000... discussion. I'll try finding the original & putting it back... Elf 16:27, 2 Feb 2004 (UTC)
I'd guess that we have here the same clock-skew issue that seems to have caused my complaint about http handling, a few topics above here. Dandrake 01:56, Feb 3, 2004 (UTC) For the record, something is still odd: after checking in that remark, I had to do shift-Reload before my new text would show up. Dandrake 01:58, Feb 3, 2004 (UTC)
Should be fixed since Tuesday. Same bug as elsewhere. -- Gabriel Wicke 12:09, 5 Feb 2004 (UTC)
Since the last day or two, Wikipedia has become very buggy on Mozilla 1.6 (Mac 10.3). Right now and occasionally on other pages, I see the outline to part of a large grid in the upper left-hand corner of the screen (appears here, in the edit window, and on the displayed pump). The grid seems to be the normally invisible lines in between the sections of links on the left side of the page, but extended outward into the edit window and title. I also keep getting "connection was refused" and timeout errors, and sometimes pages just do not load, with no error message at all. About half the time I try to load a page, it doesn't appear. Tuf-Kat 20:57, Feb 2, 2004 (UTC)
- I've often had pages load with lines of a table spread across the article (not just in the last day) and sometimes no left hand menu. I associate it with times the servers are failing to cope, and you only get part of the html of the page downloaded. Hopefully the newly bought servers (Wikipedia:Announcements) will be tested and installed soon. (I'm sure someone will correct me if I'm wrong about the cause). fabiform | talk 21:40, 2 Feb 2004 (UTC)
- I also get this almost every day. Usually a warning to paste my text into a text editor as it does seem to represent server load problems. Bmills 10:32, 3 Feb 2004 (UTC)
Server change to http handling? edit
Currently when I check in a page edit, I'm getting back the pre-edit version; I have to do a shift-Reload on the browser (Mozilla) to get the correct version. Without expertise in writing http handlers, I'd guess that the server used to send the updated version with a command to override cached versions, but now the order to override is omitted. Of course, my browser (Mozilla) and my proxy servers have not changed at all. Has a new and better server gone in with, this little difference in progra (quote truncated somehow by someone) Tim Starling 04:51, Feb 2, 2004 (UTC)
- Personally I think users should ber%h its own domain and URLs. Still others may be stand-alone CD-based versions that will run on a 386 in the 3rd world. It's all possible. This is what the . :)
- What you experienced may be due to a clock skew issue; it should be fixed now if so. --Brion 07:38, 2 Feb 2004 (UTC)
Deleted pages reappearing (solved) edit
- Solved - it was a namespace conflict in the doDeleteArticle() function that was triggered if the Squid purging was switched on. Renaming the variable solved it. Sorry for the trouble. -- Gabriel Wicke 16:58, 3 Feb 2004 (UTC)
Page moves edit
Something's wrong. All page moves that I have done recently has timed out or resulted in a blank page while the rest of WP is speedy enough. BL 01:58, Feb 4, 2004 (UTC)
Yes, I just tried to delete User:GenePoole/Self-aggrandizement, an obnoxious page that Wik had created, and move it back to George Francis Cruickshank, but somehow, the delete/move got timed out and now we don't have the George Francis Cruickshank data anywhere. Can somebody help me to get it back? RickK 04:05, 4 Feb 2004 (UTC)
- I've usually found that the move goes ahead, despite not returning the confirmation page. Apparenly working around the bug, Wik replaced George Francis Cruickshank with a redirect -- the original text was in the history of that article. And fixing Special:Movepage is right at the top of my priority list, after real life and procrastination. OK, maybe 3rd. -- Tim Starling 04:21, Feb 4, 2004 (UTC)
- As far as I can tell, the moves still go through, even if the page times out. Wait a minute or so and check Recent Changes. --Delirium 04:26, Feb 4, 2004 (UTC)
This was the bug that was fixed already on Tuesday (the loop), but for some reason the old code reappeared. -- Gwicke 02:19, 6 Feb 2004 (UTC)
Zero Sized Reply edit
- Happens when the Apache is restarted while you're trying to get a page out of it through Squid- not much Squid can do about that. -- Gabriel Wicke 16:19, 3 Feb 2004 (UTC)
I've been seeing these a lot in the last few weeks. It hits me persistently at home, so I've been mailing edits to work and committing them there, where I've had no problems. The proxy involved is sq29.wikimedia.org. Buffyg 19:51, 22 December 2006 (UTC)
Current problems edit
Why do I have difficulty loading Recent changes and the page histories? I remember an announcement saying hardware problems would be over by the end of January? (I won't ask why it's me again and no one else who asks these questions.) <KF> 23:56, 2 Feb 2004 (UTC)
There is definitely something strange going on this morning. I keep being told that pages don't exist and that edits can't be saved. There do seem to be a lot of bugs in the system recently. Adam 23:59, 2 Feb 2004 (UTC)
- The version that was put into production yesterday morning (Squid3) wasn't tested properly under high load and showed bugs. Now 2.5 is running without problems so far. In benchmarks the squid currently serves pages at up to 800 requests/second to anon visitors without hammering the DB or the Apache, so this helps a lot.
- Edit/update: All wikipedias are currently running on 1 DB server, 1 Apache and 1 Squid. Geoffrin and some other server (don't recall which ;-) are broken. Without the Squid (running on old Larousse, a pentium3) wikipedia would be hardly reachable. The occasional 'Can't reach parent server' messages indicate that the Apache is overloaded and not reachable, so instead of a timeout you get a message now. Should make it a nicer and more explaining page though.
- Yesterday evening Larousse ran out of disk space for the log which caused Squid to crash. Brion has set up log rotation and compression now, and there's a lot of space now on the drive. -- Gabriel Wicke 15:46, 3 Feb and 16:02, 4 Feb (UTC)
- Some new hardware was put in, and there are some problems with the squid setup. Being on Slashdot didn't help things much. Dori | Talk 00:40, Feb 3, 2004 (UTC)
Can someone explain "squid" and "slashdot" to non-technopersons? Adam 00:52, 3 Feb 2004 (UTC)
- The end of January date was when the new servers were being delivered. Well, they've been delivered, but they're still being tested, they're not doing anything useful yet. We are, however, being more creative in our use of the hardware we have. This creativity leads to improved speed especially for anonymous users, but lots of bugs and the occasional bit of downtime. A working squid server will be useful when the new hardware is ready for service. -- Tim Starling 04:22, Feb 3, 2004 (UTC)