I added support to LJ::TextMessage for arch.com paging and put up a new release.
I got sick of all the work required to clean, package, tar, and copy around packages after I changed them so I wrote a little script to do it all for me. Releasing stuff is really easy now, and I have my build directory all setup nicely. Little things are amusing, ya know?
I have the spell-checking stuff cleaned and packaged up and ready to go, but I want to actually use it on LiveJournal (instead of the old code) so I know it works before I release it. I'll do that tomorrow afternoon.
(55)No buffer space available: FastCGI: failed to connect to server "/home/lj/cgi-bin/log.cgi": ap_psocket() failed
does this have anything to do with it?
[lj@lj-kenny logs]$ netstat -m 4410/7968/16384 mbufs in use (current/peak/max): 2407 mbufs allocated to data 2003 mbufs allocated to packet headers 2326/4096/4096 mbuf clusters in use (current/peak/max) 10184 Kbytes allocated to network (82% of mb_map in use) 45643 requests for memory denied 51 requests for memory delayed 0 calls to protocol drain routine
load's high, too:
[lj@lj-kenny logs]$ w 6:45PM up 17 days, 6:46, 1 user, load averages: 7.18, 7.58, 7.81
I did order the new servers ... due to arrive next week sometime. feb 7th or 8th, perhaps.
the more i use unix, the less i realize i know it.
btw, we did solve the mail problem ..... dormando pointed out that the installation was all fucked and i then saw that /usr/sbin/sendmail was sendmail and /usr/local/sbin/sendmail was postfix's sendmail. i changed a $SENDMAIL constant in ljlib.pl and everything's dandy now. fuck freebsd's ports tree ... i'm installing everything important by hand from now on.
i'm looking at our server specs.. and it seems that our current servers shouldn't be cringing so hard. what's the bottleneck here? kenny and cartman's stats are pretty beefy, and they should be flying.
Evan and I rebuilt the kernel and rebooted the machine. (I typed... he watched my screen session... and then he hit enter when we went to reboot so I could blame him if anything went wrong. :P)
The new kernel config, postfix config, and whatever else you guys ask for is at /misc/conf/. I have a script to keep that area updated now. (I'd symlink, but I have reservations about symlinking some stuff in some areas.... like htdocs/ areas)
I'm working on getting some web server reports up soon, so you guys can see that the server is in fact working, not just poorly misconfigured. ("what? you mean forking a new perl process for each request is bad?")
I don't have time to keep working on this, but I do have some preliminary stats:
Lately we're doing about 650,000 hits per day.... excluding images, which I don't log. So, that includes pages and protocol hits. That's 7.5 hits/second ... but in reality that means 15-20 hits/second during peak times, and less at night.
Also, remember that these are all dynamic pages ... and nearly always hitting the database.
Does this suck? Should we be able to do more than 20 hits/second on these two servers?
Like several other users, I have found myself somewhat frustrated by the search capabilities of LiveJournal. Searching takes up a lot of server resources, etc... so any searches have to be well developed.
The kind of search I'd like to see is being able to find all people interested in __________ within __ miles of where I live. This would be very cool, but then it occurred to me how hard such a search could be. Still, I know that a lot of people who do regional communities would love to have this feature.
Well, to facilitate things, I have poked around for solutions. There is a good article on how to do this here and here... and even free latitude/longitude data here to help out. Yes, the data is only for the U.S., but I suspect that we could get similar data for other countries and add support over time. Infact, there are lists of such data available here.