Bastian Bittorf wrote: > I forgot that to mention on wcw2008/berlin, but IMHO those > hardware-watchdog are a nice-to-have but not really necessary. > Practically i have never seen an router which totally hangs, > so the hw-watchdog would'nt react... > Well, I had a quite few situations where I've shot a running system and was lucky to have it or had crashes in MadWifi, etc.... I also never had a situation where the HW watchdog didn't react, but thats not surprising given that thats the whole point of having one. > In Weimar we had _some_ situations, were do to memory-problems > nearly all tasks were killed. Even sshd/dropbear and cron were > killed. So sven-ola added an kernel-module named "crondog", which > is early loaded and not killed in lowmem situations. > A cron-called-script has to do this regulary: > > test -e /dev/misc/crondog && echo > /dev/misc/crondog # sign of life > > If crond is'nt running anymore (and so the checkers cant check: > "is sshd running?", "is httpd running?",...), then the device > reboots hard. > > Maybe we can include this as a package in openWRT. > Does anyone have enough skill the enhance the crondog.o-driver > to be more platform independent, so we can include this as a > package in menuconfig? > First, you can't guarantee which processes are killed by the oom-killer, so you might end up with dropbear killed and cron still running. I think what you need is softdog. It is in the current kernels and does just that. Just configure the watchdog program to check for whatever condition (i.e. pid files) you want.
- Axel _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org http://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel
