sudden peak in load average

[Zbigniew Szalbot]

Hello,

2008/2/27, Kris Kennaway <kris at freebsd.org>:
> Zbigniew Szalbot wrote:
>  > Hello,
>  >
>  >>From the MRTG data I can see that suddenly on a currently not very
>  > busy machine the load averege went over 15 or more. This happened
>  > around 10 in the morning. Not many entries in httpd access log, smtp
>  > server was not too much loaded (at that time it generally produces a
>  > load average of about 1).
>  >
>  > But this time:
>  > 2008-02-27 10:03:36 1JUICO-0002Bs-23 no immediate delivery: load average 15.88
>  >
>  > Where can I look for more information about what happened. The server
>  > was rebooted early in the morning so its swap consumption was almost
>  > non-existent. Around 10 it suddenly consumed probably around 200MB of
>  > swap, now it stands at 135MB which is pretty normal with this machine
>  > (it only has 512 RAM). SWAP usage is normal with this machine but it
>  > usually is a matter of 2-3 days before it gradually takes over 10-13%
>  > of swap space.
>  >
>  > I tried by looking at:
>  > http access log
>  > auth.log
>  > maillog
>  > messages
>  >
>  > Nothing in there indicating an outburst of sudden activities.
>  >
>  > heh - I now think I may be wrong. The load average did not necessarily
>  > produce so much swap consumption so fast. At 10:13 I run a cron job
>  > optimising all mysql tables. So maybe in fact the swap was used by
>  > mysql operations, although I am still interested to now what casued
>  > load average to go above 15 at 10:03.
>  >
>  > 10:00AM  up  3:14, 0 users, load averages: 0.40, 0.13, 0.05
>  > 10:01AM  up  3:15, 0 users, load averages: 0.23, 0.14, 0.05
>  > 10:02AM  up  3:16, 0 users, load averages: 0.08, 0.11, 0.05
>  > 10:03AM  up  3:17, 0 users, load averages: 11.54, 3.09, 1.15
>  > 10:04AM  up  3:18, 0 users, load averages: 13.26, 5.69, 2.28
>  > 10:05AM  up  3:19, 0 users, load averages: 4.98, 4.69, 2.14
>  > 10:06AM  up  3:20, 0 users, load averages: 1.79, 3.80, 1.99
>  >
>  >
>  > Thanks for any pointers.

> No way to know.  You'll have to set up more detailed logging, e.g. a
>  script that runs ps if load is over a certain limit.

Thank you Kris! Is anybody willing to share such a script (if there is one)?

Thanks!

-- 
Zbigniew Szalbot