On 3/7/23 00:36, Stuart Henderson wrote:
On 2023/03/07 07:10, Tom Smyth wrote:
I m running smokeping fcgi and rrdcached ontop of OpenbSD, to smokeping
about 150 devces
the page load times can take 30 seconds to 1 minute,
is there any way to speed this up.
im running 7.2 OpenBSD on amd64 vm on top of an SSD array
any tips tricks welccome ...
One quick thing to try is updating to -current, I made some changes to
the rrdtool port which may possibly help a little.
Check that smokeping is actually using rrdcached (watch top while
opening a page) - the pkg-readme only gives instructions for passing the
required fastcgi variable through for nginx, I don't know how to do that
for httpd (or whether it's actually possible).
FWIW I have been running smokeping and mrtg with about 200 targets using httpd
for quite some time.
Smokeping with 1 master and 2 slaves. fping probes, config w default
concurrentprobes (i.e., commented out in config).
Rest of config down to ***Targets*** looks like Tom's.
not running rddcached
We're a small wireless (5GHz backbone, 2GHz APs) co-op on private 10.42.0.0
net; ping to most distance site is ~30msec, but can climb substantially with
lousy weather and client contention.
I found using IPs for "host" entries helped by eliminating DNS lookups but
don't recall how much difference it made.
It takes <15sec to generate new page (/var/db/smokeping/* and
/var/www/htdocs/smokeping/cache/*), top indicates smokeping using 1 core.
Hardware is pcEngines apu2 (4 core amd64, 4Gb mem) with 16Gb mSATA; OpenBSD 7.2 release
To get this running cp'd perl (/usr/bin/perl) and relevant perl libs
(/usr/lib/[libs.so|libm.so|libperl.so] /usr/libexec/ld.so) to
/var/www/usr/[bin|lib|libexec]
httpd.conf:
ext_ip="[redacted]" # Change this value to your vultr IP
server "default" {
listen on $ext_ip port 80
}
types {
include "/usr/share/misc/mime.types"
}
server "[redacted]" {
listen on $ext_ip port 80
location "/smokeping/smokeping.fcgi*" {
fastcgi socket "/run/smokeping.sock"
root "/"
}
}
start smokeping and smokeping.fcgi with rcctl, (same for mrtg)
Cheers, Glen
Other than that, rrdtool/rrdcached is just slow on OpenBSD. If it's
anything like mine you'll see high cpu spin % in top while it's busy.
You can try changing the number of cores in the VM - if you've given it
lots of cores try *reducing* it a bit. To pick a number out of the air
I'd suggest probably 4-6. (mine is bare metal and I can't drop the
number short of kernel hacks to set more cores offline).
You can hide the slowness at the loss of dynamic functionality in the
web interface by pre rendering the html/graphs from a cron job rather
than using the fastcgi (see the pkg-readme). But other than the above
I'm out of ideas to actually make it run faster.
(If anyone interested in poking at kernel locks would like flamegraphs
from my monitoring box - librenms/smokeping/icinga2/mariadb with lots
of rrdtool/rrdcached - let me know. Spin is pretty brutal.)
