On Sat, Sep 02, 2006 at 08:19:01AM +0100, Laurence Skinner wrote: > I can see the packets coming in and going out with tcpdump, but they don't > reach the monitoring server: > > This exchange just got logged as "server is unreachable":
Your tcpdump log suggests that your network stack is generating invalid UDP check-sums. This should cause the receiving system to discard the packet, which would cause your box to appear unreachable. The interesting question is why they are sometimes valid and other times not -- you said that it sometimes works. If you've got another box available that you can plug into your network it may be good to see if that box sees the same thing happening. > I wonder what the timeout is on the get_ntp_response function deciding no > packet was received..? When there were the monitoring issues yesterday the > non-American servers seemed to be affected more than American ones. Maybe > the timeout needs to be increased? Yesterdays badness was caused by a typo, it was just random that this bug happened to prefer closer hosts (as given by network latency). The next bug will probably work differently so you can't really protect for it. My personal preference is for software like this, that breaks in obvious ways, as it tends to get fixed faster than software that tries to be too clever and ends up causing more trouble by masking the real problems. Sam _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
