Bug#353053: watchdog: this needs to be resolved for xen when many domains need to be shut down

2007-02-12 Thread Michael Meskes
On Fri, Feb 09, 2007 at 09:42:26PM +1100, James Harper wrote:
 I thought of this as well... one at K01 and one at K99. K01 would create
 /etc/watchdog/shutdown, which watchdog would notice and not do any
 further network or process checks. K99 would actually stop the process.
 The startup script would automatically clear the shutdown file. Does
 that action of the rc files go against any Debian policies?

Could you please try 5.2.6-7? I did not go this route but instead
enabled the wd_keepalive binary which is a watchdog with just the
keepalive routines.

Michael
-- 
Michael Meskes
Email: Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org)
ICQ: 179140304, AIM/Yahoo: michaelmeskes, Jabber: [EMAIL PROTECTED]
Go SF 49ers! Go Rhein Fire! Use Debian GNU/Linux! Use PostgreSQL!


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]



Bug#353053: watchdog: this needs to be resolved for xen when many domains need to be shut down

2007-02-09 Thread Michael Meskes
On Fri, Feb 09, 2007 at 03:03:33PM +1100, James Harper wrote:
 Would it be possible for watchdog to detect a change to runlevel 0 or
 6, and simply stop doing the checks which become unreliable once a
 shutdown has started? This behaviour could be controlled by a config
 option if necessary.

And then make watchdog become the last daemon stopped? Yes, that might
work. Hmm, thinking about it, it might even be better to do this with
two programs. 

Thanks or this idea.

Michael

-- 
Michael Meskes
Email: Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org)
ICQ: 179140304, AIM/Yahoo: michaelmeskes, Jabber: [EMAIL PROTECTED]
Go SF 49ers! Go Rhein Fire! Use Debian GNU/Linux! Use PostgreSQL!


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]



Bug#353053: watchdog: this needs to be resolved for xen when many domains need to be shut down

2007-02-09 Thread James Harper
 On Fri, Feb 09, 2007 at 03:03:33PM +1100, James Harper wrote:
  Would it be possible for watchdog to detect a change to runlevel 0
or
  6, and simply stop doing the checks which become unreliable once a
  shutdown has started? This behaviour could be controlled by a config
  option if necessary.
 
 And then make watchdog become the last daemon stopped? Yes, that might
 work. Hmm, thinking about it, it might even be better to do this with
 two programs.

I thought of this as well... one at K01 and one at K99. K01 would create
/etc/watchdog/shutdown, which watchdog would notice and not do any
further network or process checks. K99 would actually stop the process.
The startup script would automatically clear the shutdown file. Does
that action of the rc files go against any Debian policies?

You'd also have to consider the possibility of races between watchdog
noticing that a process has stopped, and noticing that
/etc/watchdog/shutdown has been created. The only way around this I can
think of is that you somehow make sure that watchdog has noticed that
the file has been created before you let K01 finish. Maybe by having it
in turn create a /etc/watchdog/shutdown_ack file.

Checking the runlevel frequently in watchdog could be 'expensive' in
terms of CPU (I think it needs to parse the utmp file), and even more
'expensive' in trying to prevent races.

James



Bug#353053: watchdog: this needs to be resolved for xen when many domains need to be shut down

2007-02-08 Thread James Harper
Package: watchdog
Version: 5.2.6-6
Followup-For: Bug #353053

I'm having a problem where xen needs an unknown amount of time to shut down the 
virtual machines before it can shut down itself, which means shutting down 
watchdog when nowayout has been set causes a premature reboot.

Would it be possible for watchdog to detect a change to runlevel 0 or 6, and 
simply stop doing the checks which become unreliable once a shutdown has 
started? This behaviour could be controlled by a config option if necessary.

Thanks

James


-- System Information:
Debian Release: 4.0
  APT prefers testing
  APT policy: (500, 'testing')
Architecture: i386 (i686)
Shell:  /bin/sh linked to /bin/bash
Kernel: Linux 2.6.18-3-xen-686
Locale: LANG=C, LC_CTYPE=C (charmap=ANSI_X3.4-1968)

Versions of packages watchdog depends on:
ii  debconf [debconf-2.0]1.5.11  Debian configuration management sy
ii  libc62.3.6.ds1-8 GNU C Library: Shared libraries
ii  makedev  2.3.1-83creates device files in /dev
ii  udev 0.103-2 /dev/ and hotplug management daemo

watchdog recommends no packages.

-- debconf information:
* watchdog/run: true
* watchdog/restart: false


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]