twb | Is it because upstart gets confused when it's not PID 1 that Ubuntu's bootchart starts within initramfs (rather than with init=/sbin/bootchartd)? | 02:56 |
---|---|---|
jdong | twb: no, it's useful to factor in the time spent in initramfs too | 05:07 |
twb | initramfs runs before init? | 05:07 |
jdong | twb: as we found a few weeks ago, for example, one developer's laptop spends nearly half its bootup in initramfs | 05:07 |
jdong | twb: correct. initramfs runs before / is even mounted, much less an init daemon exists | 05:07 |
twb | You're right, I was confused | 05:07 |
keesj | hello twb | 05:44 |
keesj | I am VERY interested in profiling the bootsequence, die you get bootchart runnning? | 05:45 |
twb | keesj: I have it running on Debian Lenny... | 05:49 |
keesj | and that uses upstart? | 06:02 |
keesj | let me put it differently , can i "bootchart" upstart? | 06:03 |
twb | keesj: that's broadly my question. | 06:16 |
keesj | twb: My feeling is no. | 06:43 |
Amaranth | keesj: i don't see why not, you just have to start it in initramfs and have a stop task at the end of boot | 08:06 |
Amaranth | keesj: jdong's upstart scripts for ubuntu work with bootchart | 08:06 |
AlexExtreme | you can bootchart upstart with native scripts in the normal way you would start bootchart (init=/path/to/bootchart on kernel command line), i've done it | 08:09 |
Amaranth | AlexExtreme: but then you don't chart initramfs time | 08:11 |
AlexExtreme | mmm, true | 08:11 |
Amaranth | which for me is 10 seconds of a 32 second boot :P | 08:12 |
AlexExtreme | if you're using it with initramfs you don't need to have a stop task for it, bootchart will stop in it's normal way | 08:12 |
AlexExtreme | hehe | 08:12 |
AlexExtreme | i wasn't using initramfs with mine though | 08:12 |
Amaranth | no, you still need a stop task | 08:12 |
AlexExtreme | really? why wouldn't the way bootchart stops itself work? | 08:13 |
Amaranth | me and jdong found that out when i got a 12 minute bootchart that took 10 minutes to convert from svg to png :P | 08:13 |
AlexExtreme | it waits until gdm, kdm, xdm or getty starts | 08:13 |
AlexExtreme | mmm... :P | 08:13 |
Amaranth | http://www.realistanew.com/random/hardy-20080317-3.png <--initramfs time is important to measure :P | 08:14 |
AlexExtreme | mm, yeah | 08:14 |
Amaranth | after that boot we spent awhile trying to maximize disk throughput throughout the process as you can see it seeking most of the time there | 08:15 |
AlexExtreme | http://alex-smith.me.uk/files/bootchart-ud.png << insane boot time using upstart native scripts on a custom (LFS based) distro with a custom kernel ;) | 08:15 |
Amaranth | ended up with a boot that has a more consistent throughput but a lower max throughput and thus takes the same amount of time to boot :P | 08:16 |
Amaranth | wow | 08:16 |
Amaranth | except you don't start anything :P | 08:16 |
Amaranth | would still take my system 13 seconds or so though, since 10 seconds is lost in initramfs :P | 08:17 |
AlexExtreme | well, it booted up to a getty with networking enabled | 08:17 |
Amaranth | yeah | 08:17 |
AlexExtreme | http://alex-smith.me.uk/files/bootcharts/bootchart-upstart3.png << that's a better one from when I was getting upstart on frugalware, which doesn't use initramfs | 08:17 |
AlexExtreme | so stick about 10 secs on for initramfs and you a bit less than your bootchart | 08:18 |
AlexExtreme | *get a | 08:18 |
Amaranth | that's a weird chart | 08:18 |
Amaranth | your high disk usage was at the end of boot | 08:18 |
AlexExtreme | about where gdm starts | 08:19 |
Amaranth | http://www.realistanew.com/random/hardy-20080317-10.png | 08:19 |
Amaranth | that boot is with readahead disabled and i still peak at beginning | 08:19 |
AlexExtreme | probably something screwed up that i've done then :p | 08:19 |
Amaranth | also i seem to start about 3x as many things :P | 08:20 |
AlexExtreme | pff, i like my simplicity ;) | 08:20 |
Amaranth | i suspect i can get this thing down to 20 seconds easily if i fix my initramfs problems and get rid of stuff i don't use from startup | 08:21 |
Amaranth | which is really exciting because with sysvinit 45 seconds was crazy fast :P | 08:21 |
AlexExtreme | with sysv on that system i was getting 29 seconds | 08:22 |
Tv | hi. how do i handle a package that's removed but not purged? | 14:28 |
ion_ | tv: This is #upstart, btw. :-) | 15:14 |
Tv | yeah i mean it's event.d | 15:40 |
Tv | i don't want to keep respawning on exec /usr/bin/does-not-exist-anymore | 15:40 |
ion_ | Well, you could always purge the package. ;-) | 16:07 |
keesj | thanks for the help guys . i was not able to read earlyer | 19:07 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!