It is possible to setup some hosts in watchdog.conf which are pinged on every watchdog loop cycle. There is an issue with the implementation which causes messages like Feb 20 16:39:57 [watchdog] network is unreachable (target: $some_ip) to appear in the logfiles. As far as I can see the problem lays the way the echo_reply packets are handled. The implementation is taking the first ICMP message available and tests for if (icp->type == ICMP_ECHOREPLY && icp->un.echo.id == pid) { If it happens three times that this ICMP message is not an ICMP_ECHOREPLY containing the watchdog pid the process assumes "network is unreachable" returning ENETUNREACH. I run some tests and the reason is always icp->type==8 or icp->un.echo.id==$random_number but never a network problem. Depending on how busy the host is this can happen quite often. Hence using watchdog to ping other hosts is pretty much useless. I don't think this could be easily fixed without changing the whole src/net.c modul. One more bug to be passed upstream ...
(In reply to comment #0) > One more bug to be passed upstream ... Since you already know this should be submitted upstream, why did you post it here? I've already told you how to handle a similar issue in bug #123404.
Because it helps other Gentoo users who may have the same trouble with this piece of software. Furthermore watchdog has no own bugtracking system and the maintainer didn't reply to my first email one month ago. I'm not sure if anyone actually cares to release a new version. I try to limit the bugs I report here to what may lead to false alarms/reboots caused by watchdog. Anyway - sorry for the trouble.
