Perl 6 - the future is here, just unevenly distributed

IRC log for #rosettacode, 2012-01-19

| Channels | #rosettacode index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
01:24 mwn3d_phone joined #rosettacode
01:32 nimdAHK joined #rosettacode
01:32 nimdAHK joined #rosettacode
01:32 nimdAHK it's down :(
01:35 nimdAHK is it down for anyone  else?
01:35 nimdAHK I keep getting Error 500's
01:37 nimdAHK only on Chrome :/
01:38 nimdAHK nope, now IE too
01:38 nimdAHK HTTP 500s
01:54 mwn3d_phone I got 500 too
01:54 nimdAHK :(
01:54 nimdAHK I just finished 99 bottles of beer in LabVIEW
01:54 nimdAHK http://www.autohotkey.net/~crazyfi​rex/LabVIEW_99_Bottles_of_Beer.png
01:54 fedaykin http://rldn.net/C8h
02:31 nimdAHK still down :(
03:00 mwn3d_phone I got absolutely nothing now
03:00 mwn3d_phone No 500
03:00 mwn3d_phone Just a blank screen
03:46 AndroUser joined #rosettacode
03:52 mwn3d_phone Back to a 500
03:53 mwn3d_phone This isn't a SOPA protest is it?
04:06 mikemol ?
04:06 mikemol No
04:07 mikemol I wasn't aware it was down.
04:07 mikemol ah
04:08 mikemol Should be fixed
04:11 mwn3d_phone Yep
04:11 sorear What happened?
04:13 sorear OOM killer ate apache again because there's no mechanism to limit conc. requests?
04:28 mikemol No; things are configured enough that the OOM killer can't kick in.
04:28 mikemol But part of that was that I set PHP's memory_limit too low.
04:30 mikemol Originally, it was at 128M. I cut that in half so I'd have enough RAM to support 10 apache processes. Then I cut it in half again so I could support more apache processes.
04:31 mikemol It worked fine for several hours, but that was apparently because some intermediate, memory-intensive things were well-cached.
04:31 mikemol Then it broke for all pageloads.
04:33 mikemol I increased memory_limit to 48M. Between the configuration of apache, PHP, squid, memcached and mysql, I don't believe the box can run out of RAM unless some additional factor comes into play, such as an expensive CGI script at the exact wrong time, or a bad coincidental run of some cron'd process.
04:35 sorear So... how did this cause a 500?
04:35 sorear does php send 500 to all requests once the memory_limit is hit once by any thread?
04:37 * sorear is trying to learn about webserver administration, not actually trying to be difficult/harrassing
05:37 Util joined #rosettacode
06:23 NevilleDNZ joined #rosettacode
07:52 GlitchMr joined #rosettacode
08:04 mischi joined #rosettacode
12:34 mikemol sorear: the server is configured to use mod_php, which means there's one Apache process per PHP interpreter, and the Apache process is single-threaded.
12:35 mwn3d_phone joined #rosettacode
12:36 mikemol The only way to explain it that makes sense to me is that there is some memory-intensive calculation whose results are needed by every run of the MediaWiki code, and the outcome of that calculation was cached in memcached.
12:36 mikemol Actually, there's one other possibility I can think of.
12:37 mikemol Pageviews on MediaWiki aren't committed to pageview counters immediately. The events are queued up in a table in MySQL, and are eventually committed to incrememnting their related counters.
12:40 mikemol This is done because those commits can be time-expensive in MySQL. Viewcounting is one of the very few parts of MediaWiki which write to the database when a viewer hasn't explicitly modified something. (The other example which comes to mind is where MediaWiki tells you you have new messages on your talk page, and viewing your talk page clears its insistence on telling you that.)
12:46 mwn3d_phone1 joined #rosettacode
13:13 mwn3d_phone joined #rosettacode
17:48 pat_js joined #rosettacode
19:14 maustin joined #rosettacode
22:17 pat_js joined #rosettacode
23:46 mwn3d_phone joined #rosettacode
23:48 maustin joined #rosettacode

| Channels | #rosettacode index | Today | | Search | Google Search | Plain-Text | summary