Bug#768168: watchdog does not start at boot

[Michael Meskes]

On Wed, Nov 12, 2014 at 02:44:53PM +0100, Michael Biebl wrote:
> Yeah, it shows that watchdog.service is involved in dependency cycle and
> systemd therefore removes it from the transaction.

Why doesn't it show the same reaction on my system then? Uwe's system doesn't
seem to have any changes out of the ordinary. Any idea?

> > After=runlevel1.target runlevel2.target runlevel3.target runlevel4.target runlevel5.target
> 
> I don't think this does and please don't use that in a service file. We
> don't really want to drag that legacy concept of SysV runlevels into
> native .service files.

I'm fine with any other solution. Problem is that this piece of systemd doesn't
seem to be very well documented or I just didn't find anything.

> Couldn't you just run "ExecStartPre=-/sbin/modprobe $watchdog_module"?
> The "-" would ignore any errors

Should do, just never occured to me.

> > ExecStartPre=/bin/systemctl stop wd_keepalive
> 
> Calling systemctl from service files is highly discouraged, I'd go as
> far as say forbidden unless you know exactly what you're doing.
> Can you elaborate why this is necessary?

Sure, wd_keepalive is supposed to run whenever watchdog is not. But they cannot
run at the same time, so wd_keepalive has to be stopped for watchdog to be
started.

> > ExecStart=/usr/sbin/watchdog $watchdog_options
> > ExecStopPost=/bin/systemctl start wd_keepalive
> 
> Same here. Does that mean, on shutdown, when all services are about to
> be stopped, you start wd_keepalive? Why?

Yes, indeed that is what we need. Watchdog triggers a special device to tell
the hardware watchdog that everything is well. When it is stopped it closes the
device and obviously stops triggering it. Normally that means the hardware
watchdog is disabled. However, there is a kernel setting called
CONFIG_WATCHDOG_NOWAYOUT or somesuch that makes the kernel not accept the
closing, meaning the hardware watchdog stays on alert and triggers a reset
after a while. With watchdog having to be stopped early (it could be monitoring
other services) the system might run into a reset by the watchdog before its
shutdown has been finished. Therefore we need a small program doing nothing but
triggering the device, which is wd_keepalive.

Hopefully this very brief explanation makes sense.

Michael
-- 
Michael Meskes
Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org)
Michael at BorussiaFan dot De, Meskes at (Debian|Postgresql) dot Org
Jabber: michael.meskes at gmail dot com
VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL