| jMCg | I'm thinking out loud here, but would it make sense to support something like taskset(1) or sched_setaffinity(2) for that matter, in upstart? | 02:27 |
|---|---|---|
| phirox_ | looks like upstart follows fork or spawned children of a process it's monitoring, any way to turn that behaviour of? it leads to a lot of overhead coming from the init process | 09:40 |
| jhunt | out of interest, have you measured that overhead and found it to be too high for your environment? | 09:45 |
| phirox_ | it is the #1 process measured in cputime :) | 09:49 |
| jhunt | Upstart only uses ptrace for "expect fork" and "expect daemon". The overhead of using ptrace is negligible for starting such jobs. | 09:51 |
| phirox_ | ah ok, well I made a nagios init. what happens is that everytime nagios spawns a process, somehow init gets triggered | 09:54 |
| phirox_ | this is what init does in a strace: http://pastebin.com/JSkKFE7p | 09:55 |
| jhunt | upstart reaps zombie processes (as do all init systems). That is correct behaviour and not something you can (or would want to) turn off. | 09:58 |
| phirox_ | ok thanks, I'll look for another way then | 09:59 |
| jhunt | I'm confused. Is there a problem with upstart on your system? | 09:59 |
| phirox_ | actually no, I think I just understand what you mean. doesn't matter how I start these 1000's of processes. init still has to check them | 10:00 |
| phirox_ | so this isn't really an issue, probally more of a misunderstanding on my part :) | 10:00 |
| jhunt | yes. If init didn't do this, your process table would eventually fill up with zombies, which might eventually kill the system. | 10:01 |
| phirox_ | yeah, makes sense. thanks for explaining | 10:09 |
| jhunt | no problem | 10:09 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!