Quote:
Originally Posted by NiLuJe
@makue: I'd first take a look at a *full* log dumped via showlog and not ;dm, because Amazon's logging facilities sometime do funny things when coalescing logs...
(i.e., here, I'm only seeing the replay of dmesg's backlog done once the system logger boots up, which doesn't tell us much about when exactly it hung, besides "before syslog-ng appeared to have been brought up").
|
Could we automate that?
Consider; the first (the only) program the kernel runs after it finishes start-up is "init".
Perhaps jam a "dmesg" and/or "showlog" call very early into the list of things "init" starts?
If such a modification never gets called, then we know the system crashed before reaching "init".
The root fs is mounted when "init" is started ...
But if we are having file system errors, then writing our extra log output could be a "bad idea" ...
But if we set-up a ram disk to write (or re-direct) our special logging too...
Also, if required, we could net-cat it off of the machine somewhat later.
Whatever, it is going to be hard to debug.
(Note: I think debugfs is mounted, but I am not sure when in the process it gets mounted. Early, with /proc and /sys ??)