named
Cameron Simpson
cs at zip.com.au
Wed Apr 4 22:00:33 UTC 2012
On 04Apr2012 16:29, Scott van Looy <scott at ethosuk.net> wrote:
| Hoping this might help someone debug my issue…
|
| I attached a strace:
| strace -f -p 1747 -o trace.txt
|
| Left it running for around 10 seconds and got 34Mb of log that's all pretty much this:
|
| 1751 epoll_wait(9, <unfinished ...>
| 1750 futex(0x7f8148e22028, FUTEX_WAKE_PRIVATE, 1 <unfinished ...>
| 1749 futex(0x7f8148e22028, FUTEX_WAIT_PRIVATE, 2, NULL <unfinished ...>
| 1748 futex(0x7f8148e2108c, FUTEX_WAIT_PRIVATE, 4692676, NULL <unfinished ...>
| 1747 rt_sigsuspend([] <unfinished ...>
| 1750 <... futex resumed> ) = 0
| 1749 <... futex resumed> ) = -1 EAGAIN (Resource temporarily unavailable)
[...]
Looks like heavy contention for something.
Try:
strace -e trace=!futex -f -p 1747 -o trace.txt
and see what's left over. And get the output of:
lsof -p 1747
too so you can match of file descriptor numbers.
This may shed a little light on what named is working on at the time.
Does it settle down if you leave it alone?
Is it easy to reproduce? For you I mean; for example, if it _has_ settled
down, can you make it happen again by stopping and restarting named? If so,
does HUPping named trigger it also?
Cheers,
--
Cameron Simpson <cs at zip.com.au> DoD#743
http://www.cskk.ezoshosting.com/cs/
Can't talk. Eating. - Homer Simpson
More information about the users
mailing list