[plug] TCP connections refused

[Craig Ringer]

>> Anybody have any idea what it could be doing based on the basic info
>> I've given?  It is a remote machine, and I can't connect to it any time
>> soon to see what is going on.

>> I'm guessing some dodgy RAM or something, has anyone seen this behaviour
>> before and could put a different slant on it?  There are no IDS things
>> running on it.

If something kills userland, but the kernel keeps running OK, you can 
get behaviour like that.

> I had a machine here do a similar thing where the kernel "died" with an 
> oops. This stopped ALL userland programs from running, and thus 
> accepting connections, but the in-kernel stuff like NAT and normal 
> routing continued along quite happily.

Yep. Kernel oops-es but doesn't do a full panic causing a system halt. 
The kernel keeps on NATing packets etc but userland doesn't respond at 
at all. However, the fact that it still responds to UDP queries is 
interesting, as it suggests that /something/ is alive in there.

I've also had disk failures do this - things like apache and sshd that 
look at the disk die, but anything running in RAM only just keeps on 
plodding, including xinetd, bind, etc. Could it be that your HDD is 
going dogdy?

I don't suppose that its located somewhere where a serial console can be 
used to capture the panic, if any? If not, consider editing 
/etc/syslog.conf to do something like

*.*				@some-remote-host

where the remote host starts syslog as "syslog -r". Then at least you'll 
be able to watch the logs remotely.

All my machines here at work log to each other, and its very handy for 
figuring out what's going on if something goes wrong. Very soon the two 
main linux servers will have null-modem lines to each other and will be 
replicating the console on them, too. The new xeon box can even give me 
BIOS access via the serial port :-)

Craig