On Mon, Apr 04, 2005 at 04:09:34PM -0400, Schwimmer, Eric E *HS wrote:
I have a bb-hosts file with 1229 hosts in it, but no TCP
(http/ssh/ftp) enabled.
In my hobbitlaunch.cfg, I have bbnet configured as follows:
CMD bbtest-net --report --ping --checkresponse --dns=ip --concurrency=1500 --dns-timeout=5 --timeout=5
The bbtest output for my hobbit server shows the following snippets:
DNS statistics:
# hostnames resolved : 1220
# succesful : 25
# failed : 975
# calls to dnsresolve : 0
I thought that the --dns=ip option in the hobbitlaunch.cfg precluded
any DNS resolution, but the bbtest output above seems to indicate
the opposite. Does anybody know what those DNS tests are the result
of?
The heading for that bit of statistics is slightly misleading. It
really is just a summary of how many hostnames were converted into
IP-adresses; this can happen via DNS, but when you use --dns=ip it is
done entirely by using the IP's in the bb-hosts file. (The clue here
is that the "# calls to dnsresolve" is 0). As your timing statistics
show:
DNS lookups completed 1112644659.080039 0.000387
those 1000 IP's were found in less than one millisecond - that would
be quite a feat if any DNS was involved.
Also, I'm a bit curious as to the chronlogical breakdown of the
bbtest events;
Service definitions loaded 1112644638.545180 0.002916
Tests loaded 1112644659.079652 20.534472
specifically the "Tests Loaded" section. Does 20 seconds seem
reasonable for a bb-hosts file our our size?
No, it seems a bit much. My main server has about 1500 hosts in it,
and spends 4 seconds loading that file (actually, it is split into
about 30 files). If you look at the "bbgen" status, what's the time
reported for the "Load bbhosts done" line ?
I've looked over the code, but can't really spot anything that would
explain why it takes so long.
(I'm trying desperately to get below the 60 second mark :)
Well, in that case perhaps you should lower the timeout on your
ping-tests - they account for 70% of the total time:
PING test completed (1229 hosts) 1112644712.816387 53.726504
TIME TOTAL 74.562619
The ping tests are performed using "fping", so you may want to try and
play with fping options to control the timeout and # of retries it
does. You can add those to the FPING setting in hobbitserver.cfg.
Any particular reason you want to get below 60 seconds ? Are you aware
that Hobbit has a "re-test" script that performs more frequent tests
on hosts that go down ? When a network test begins to fail, Hobbit
puts that host on "frequent-test" list meaning that for up to 30
minutes that host will be tested once a minute rather than once every
5 minutes; so recoveries should be picked up faster.
Regards,
Henrik