I'm working on modernizing Rosetta Code's infrastructure. Starting with communications. Please accept this time-limited open invite to RC's Slack.. --Michael Mol (talk) 20:59, 30 May 2020 (UTC)

User talk:Short Circuit/Problems with uploading new versions of existing images

From Rosetta Code

Problems with uploading new versions of existing images[edit]

Looks like there's a problem with uploading new versions of existing images. See File talk:Matlab-randomDisc-output.png and File talk:Eriksiers avatar.jpg. (Might this be related to the recent upstream problems? Seems pretty unlikely, but you never know...) -- Erik Siers 17:42, 22 September 2010 (UTC)

Might be related to the switch to the addition of Squid. Don't know. Don't have time to look at it right now. I'll check the logs this weekend, if I have time. --Michael Mol 18:29, 22 September 2010 (UTC)
As long as you're aware of it, I'm happy...ish. :-) -- Erik Siers 18:43, 22 September 2010 (UTC)
Looked at the PHP error log file. I'm not seeing any file I/O errors, or anything in the vicinity of your avatar replacement log entry's timestamp, but I am seeing a number of "Maximum execution time of 30 seconds exceeded" errors relating to SMW activity over the past few days. I might be able to resolve some of that. --Michael Mol 18:48, 22 September 2010 (UTC)
Tweaked a couple variables. Bumped max_execution_time to 45s, reduced the job run rate to 0.1. Not a permanent fix for the timeout issues. Need to fix squid, and deal with some serious performance issues in SMW. --Michael Mol 18:52, 22 September 2010 (UTC)
I've noticed some occasional-but-significant delays in loading pages today, but didn't really give it much thought... but it could be the cache, I guess. Would make sense, I suppose. -- Erik Siers 19:05, 22 September 2010 (UTC)
The cache won't catch nearly as much as would be ideal, because of complexities surrounding X-Vary-Options. The significant delays you see are going to be cache misses combined with other load on the server. (It looks like the SMW export functionality is getting hit pretty hard.) --Michael Mol 19:12, 22 September 2010 (UTC)
Ok. I'll hold off uploading again until you have time to sit and fiddle with everything (and I know how precious time can be). It's not very important to me but I can see it being a problem that might potentially affect everything... or... something. -- Erik Siers 19:44, 22 September 2010 (UTC)
TBH, I don't have a timeframe in mind for a proper fix for the caching issue. It'll involve using a patched version of Squid, and I really can't anticipate when I'll have time to deal with that. Apart from that, updating MediaWiki (I think I'm a version or two behind by now) may help.) A workaround might be to delete the page and recreate it. You'll have to poke someone with the relevant privs. --Michael Mol 19:51, 22 September 2010 (UTC)

Again[edit]

Me again. Looks like you never got around to a permanent fix on the cache problem; I uploaded a new version of my user avatar and it's still showing the old version on my user page (checked on a few machines on different networks). I know essentially nothing about caches so I can't offer any help beyond "hey, not working".

(My only thoughts on the matter may or may not be feasible; it would involve getting Mediawiki to somehow tell Squid to clear the cached version of an updated image, but that would likely require some hacking at the mw sources. Maybe the Squid sources, too, dunno.) -- Erik Siers 04:39, 4 February 2011 (UTC)

After a hard refresh (ctrl+R in Firefox) I saw the new version (the Feb 1 version). Have you tried that? --Mwn3d 04:47, 4 February 2011 (UTC)
I had done that without effect, right before I posted (and numerous times after uploading), but now that I look again, it's showing the correct one. WTF? -- Erik Siers 04:54, 4 February 2011 (UTC)
MediaWiki and Squid don't get along terribly well, I'm afraid. MediaWiki does have code for telling Squid to purge caches, but it requires a specially-patched version of Squid. The patch had been submitted to Squid devs ages ago, but they declined to include it in trunk. The key, though, is that the cache object is tied to a literal match of your browser's supported-encoding string. --Michael Mol 13:58, 4 February 2011 (UTC)
That's... kinda weird. Is it one of the patches in here? I might take a look at it (if I can understand it). -- Erik Siers 14:36, 4 February 2011 (UTC)
this one. --Michael Mol 14:58, 4 February 2011 (UTC)