[Xymon] Xymon disruption every night!

[J.C. Cleaver]

On Wed, April 20, 2016 12:55 pm, Greg Earle wrote:
>
>> On Feb 17, 2016, at 12:50 AM,"J.C. Cleaver" <cleaver at terabithia.org>
>> wrote:
>>
>>> - Why I have a "DNS error" here ?  I set up the IP yesterday to this
>>> host
>>> to solve the issue.  The "conn" error disappeared since yesterday
>>> evening
>>> but the http error still remains.
>>
>> All signs do point to an issue with DNS resolution here.
>>
>> Was this a custom compile or are you using a package?  If custom, what
>> version of c-ares is on your system?  That's the underlying resolution
>> library that xymonnet is using by default to handle DNS lookups.  The
>> fact
>> that the 'conn' test remained good after you added the local hosts entry
>> matches that, since HTTP tests are performed using their own secondary
>> DNS
>> lookup (to deal with vhosts, etc.) unless the IP is specified there as
>> well.
>
> J.C.,
>
> I just stumbled across this thread from 2 months ago.  We're having DNS
> glitches at my work and it's causing a flood of <hostname>:http "DNS
> error"
> alerts in Xymon, which is becoming a real problem.
>
> But here's what I don't understand.  All of our HTTP-tested hosts are
> in the "hosts.cfg" file with their short names (instead of FQHNs).  So I
> couldn't understand why DNS was involved since the IP addresses and names
> were right there in "hosts.cfg" for Xymon to use.
>
> Your response - specifically "unless the IP is specified there as well" -
> implies that there might be another location where I could load the
> names and addresses of our HTTP-tested hosts, to avoid this problem.
>
> (Yes, I know - hosts and IP addresses can change.  But I'm in control of
> that so I can deal.)
>
> If this is the case, where is that other location where I can specify the
> short names/IP addresses for the HTTP tests?
>
> Thanks,
>
> 	- Greg

Yep: https://xymon.com/help/manpages/man5/hosts.cfg.5.html#lbAR

192.168.0.10 mywebserver # http://www.sample.com=192.168.0.10/


For HTTP tests, an IP override is put in for each URL you're using. This
will prevent a DNS lookup at http-test time for this URL, and when
combined with combined with testip (for any other TCP checks here) and
noconn (for ping->fping resolution), it should prevent any DNS lookups
from being done for the string "mywebserver".

Unless there's been a regression, that should be sufficient. It's
definitely worked at scale for mass testing of "hosts" that are not being
referred to by a valid DNS name and simply listen on a distinct port.


HTH,
-jc