#mysql
03 January 2008
Total 3 pages. You are browsing page 2/3.
First :: Prev :: [1] [2] [3] [4] [5] [...][...] [-1] [0] [1] [2] [3] :: Next :: Last
00:34 <****> http://flickr.com/photos/jeremycole/395701609/
00:34 <****> maybe it's time I start thinking about colocation
00:34 <****> that's proven scaling :)
00:35 <****> ebergen: where do you colo at? or where is a good place to locate colo companies near me?
00:35 <****> that's at peer1 in LA
00:36 <****> the bottom 1425 is the box that hosts mirror.provenscaling.com if you want to test the speed
00:36 <****> I'm in boston hehe
00:36 <****> how do I do a select in the mysql command line tool without the ascii table borders around the results?
00:36 <****> xaxxon -B I think
00:36 <****> ebergen: heh. techno porn.
00:36 <****> http://flickr.com/photos/midom/2134993609/ these are some of our DBs \o/
00:36 <****> or \G for vertical
00:37 <****> im inserting data into a database but it inserts it in random places in the table was wondering how i can fix this?
00:37 <****> http://flickr.com/photos/midom/2134995929/in/set-72157603610967420/ one of four rack rows
00:37 <****> domas wikipedia?
00:37 <****> ebergen: yarrr
00:38 <****> you need some minis like us ;)
00:38 <****> look how awesome our datacenter technician is: http://flickr.com/photos/midom/2134991985/in/photostream/ ;-)
00:38 <****> mini = great test client
00:38 <****> nownot, use order by in your selects, strage oder in undefined
00:38 <****> ebergen: we want to put WRT54G as a router in some internet exchange
00:38 <****> cause people already do routing on minis
00:38 <****> ebergen, yeah -B did it. Took a second for me to figure out how to use it
00:38 <****> they're nice boxes
00:39 <****> big cpus and low power consumption
00:39 <****> and super cheap
00:40 <****> ebergen: cheers, hopefully this works
00:43 <****> nownot: the db engine will insert data previously freed by a delete if it will fit. Ignore it.
00:43 <****> in space*
00:45 <****> nownot it's not broken
00:47 <****> ebergen, test clients at Wikipedia? Don't be silly, if it breaks when we update we just get complaints and fix it. :)
00:47 <****> hahaha
00:47 <****> You think I'm joking?
00:47 <****> that's like a client I was working with the other day
00:48 <****> "We don't about capacity planning or monitoring. We just overbuild everything."
00:48 <****> but if you don't know what the capacity is then how do you know you're overbuilding?
00:48 <****> and my gmail box fill up with wikibug spam
00:48 <****> ...
00:49 <****> We sync the software to the servers every week or two. Whoever does it hangs out in #wikimedia-tech, and if anything broke that wasn't caught by developers testing stuff on their own boxes, it gets backed out or fixed.
00:49 <****> Like, just the other day I broke every single citation whose name contained a digit. Silly me, forgetting ^ and $ on my regex.
00:49 <****> even if the new build is so slow it overruns all your capacity?
00:49 <****> Those are generally caught before they go live.
00:49 <****> how?
00:49 <****> By people looking at the queries. :)
00:50 <****> even on the php side?
00:50 <****> heh you let the {{{{{{{{}}}}}}}} live though
00:50 <****> archivist, Tim just made that more efficient, or is trying to. \o/
00:50 <****> hehe
00:51 <****> ebergen, we don't generally overrun our capacity on the PHP side, really. Maybe slow down page loading a bit, I guess?
00:51 <****> It's not like we're doing heavy computation.
00:51 <****> (Except for in the parser.)
00:51 <****> rewrite it in assembler
00:51 <****> (Which does admittedly take ~20 seconds to parse some of the long pages, last I checked.)
00:52 <****> (But it doesn't matter so much, they're cached!)
00:52 <****> archivist, Tim hopes to rewrite parts of the parser in C++.
00:52 <****> lucky
00:52 <****> Eventually.
00:53 <****> there have been a few attempts over the years!
00:53 <****> ebergen, bah, it's nothing critical, just some silly encyclopedia project. Who cares if it's screwed up a bit for ten minutes?
00:53 <****> awesome
00:53 <****> No one will notice.
00:53 <****> so wikipedia can crash parsing our pages
00:53 <****> It doesn't crash, it just takes a long time.
00:53 <****> no, when you convert it to C++ :)
00:53 * Simetrical trusts Tim to write good C++ that won't crash
00:54 <****> We already use C(++) modules for generating diffs.
00:54 <****> And a couple other things.
00:54 <****> that won't cash
00:54 <****> man I had to write a apache core summarizer for yahoo so we could keep track of all the different crashes in modules
00:55 <****> it made a "signature" of a core based on the stack and tracked how many of each crash happened per day
00:56 <****> but that was mostly for stuff that hadn't been converted to php
00:57 <****> Interesting.
00:57 <****> C++ doesn't have to crash much, if you avoid pointers and arrays.
00:57 <****> I wonder how much different I'll see if I upgrade from a 7200RPM drive to a 10K or 15K one... hrm...
00:57 <****> Sembiance, you probably want RAID.
00:57 <****> about 2x if you go to 15k
00:58 <****> Well, that's good for throughput.
00:58 <****> ebergen: mod_what_killedus and mod_backtrace are nice things in apache
00:58 <****> For seek time you just want higher RPM, mainly.
00:58 <****> domas, and mod_proctitle?
00:58 <****> ah, yeah, mod_proctitle is mine
00:58 <****> I want a package. :(
00:58 <****> though we use zend module for that
00:58 <****> ebergen: one of nicest crashes lately was in PCRE library
00:58 <****> I did it with gdb batch mode
00:58 <****> though usually crashes happen due to APC corruption
00:59 <****> yeah those are fun
00:59 <****> APC and shm corruption
00:59 <****> yup
00:59 <****> then processes just crash all the time
00:59 <****> we had all kinds of screwed up shm processes
00:59 <****> things that were supposed to be read only would crash if you set the file read only :)
01:00 <****> good times
01:00 <****> =)
01:01 <****> yahoo makes consulting seem boring :)
01:01 <****> no wonder:)
01:01 <****> only at big shop you understand how opensource sucks, and how anything else would be even worse.
01:01 <****> now at least I don't have to know ASM to fix things :)
01:02 <****> it wasn't so much that open source sucks it was the development process
01:04 <****> Simetrical: Probably RAID 0 for speed right? I wonder how much a difference hardware vs. software makes a difference for RAID 0
Total 3 pages. You are browsing page 2/3.
First :: Prev :: [1] [2] [3] [4] [5] [...][...] [-1] [0] [1] [2] [3] :: Next :: Last