Jabber module uses 100% CPU

[spam-bitlbee@…]

Bitlbee version 1.1.1dev (1.1 did it too) starts 100% CPU from time to time.

I am not 100% sure but it seems it is when all my contacts are offline.

account off and then account on gets CPU utilisation back to normal.

[wilmer]

Ugh, I really hoped this wouldn't come back another time. So you can still give commands? I don't think the fact that your contacts are offline has anything to do with this, BitlBee doesn't really care about things like that.

Can you try to figure out what BitlBee's doing at those moments? Maybe using strace or gdb?

[anonymous]

yes I can still give commands, i'm going to recompile it with debug symbols to try to give you a backtrace or something usable

[wilmer]

Found something yet?

[wilmer]

BTW, can you give me some more info? Like, which event handling lib do you use (probably GLib)? Are you using SSL/TLS? If so, which SSL library?

This has to be fixed...

[anonymous]

i'm using glib/gnutls, ssl enabled yes.

the bug didn't show up yet. It is much less frequent since 1.1.1dev compared to 1.1dev.

Last time it did show up I did attach gdb to it and checked the backtrace: it was in glib but I had no debug symbols so I hardly could debug it.

I'll post more when it shows up next time.

[anonymous]

strace gives thousands of :

gettimeofday({1197371160, 331858}, NULL) = 0
gettimeofday({1197371160, 331915}, NULL) = 0
poll([{fd=0, events=POLLIN}, {fd=4, events=POLLIN}, {fd=7, events=POLLIN}, {fd=5, events=POLLIN, revents=POLLIN}, {fd=5, events=POLLOUT, revents=POLLOUT}], 5, 1686) = 2
gettimeofday({1197371160, 331995}, NULL) = 0
gettimeofday({1197371160, 332025}, NULL) = 0
poll([{fd=0, events=POLLIN}, {fd=4, events=POLLIN}, {fd=7, events=POLLIN}, {fd=5, events=POLLIN, revents=POLLIN}, {fd=5, events=POLLOUT, revents=POLLOUT}], 5, 1686) = 2

CPU utilisation is @ 100% with 40% system time. gdb backtrace doesn't give something meaningful. I'll let it running, so you can tell me if you want to do other things.

[wilmer]

We most likely tracked this down thanks to gdb. Will try to roll out a fix soon.

[wilmer]

I'm quite sure I fixed this now. Could you please try out r285 and reopen if it's still broken?