I don't know if this behaviour has been changed, but sometimes, when
you run out of swap, you will not get a error messages like "not enough
virtual memory", but the system will silently hang.
has this problem been touched with the recent 1.99.? kernels ?
I'm using perl-mirror to mirror a site (> 1G) at night, and perl-mirror
is definitely a memory hog. Yet I don't know what's causing the hang,
but perl-mirror looks very suspicious to me.
The system is not completely dead. TCP is dead, but ICMP still lives,
i.e. you can ping the system and will get echo-replies, but you cannot
telnet to the system. It will either not respond at all, or your client
will simply write "Connected to [...]" and then you'll wait infinitely
for a login-prompt.
I have not tried the software watchdog yet (will soon), but, since, as
the last case shows, the system is not completely dead, it is possible
that the software-watchdog will still be running too and the machine will
not reboot.
So, since the machine still reacts to ICMP echo request/reply, what do you
think about the hack to create a new ICMP subtype: "icmp reboot". if the
network layer detects such a packet (from a particular IP and/or ether-
address), it will simply call "hard_reset_now" ...
comments ?
regards, herbert rosmanith
herp@wildsau.idv.uni-linz.ac.at
rosmanith@edvz.uni-linz.ac.at