[brandz-discuss] Behavior when init dies

[David Bustos]

Quoth Nils Nieuwejaar on Tue, Aug 22, 2006 at 05:28:19PM -0400:
> On Tue 08/22/06 at 14:02 PM, David.Bustos at sun.com wrote:
> > I hear that if init dies, the zone will be rebooted.  Is this true?  If
> > so, it seems heavy-handed.  Why not just log an error?
> 
> The Solaris kernel and init(1M) work together to provide a clean warm
> restart when it dies.  So when the Solaris init dies, the kernel simply
> restarts it and everything else in the zone keeps purring along.
> 
> When the Linux init restarts, it goes through its entire boot sequence
> again.  This leads to oddities like having duplicate crond, syslogd, and
> other services.  In a perfect world, each of these services would be smart
> enough to detect that it already has an instance running, and would simply
> go away.  In this world, the failure mode is completely undefined, and
> varies from service to service.  We decided that a clean restart was
> preferable to leaving the zone in an unknown, untested, unexpected, and
> generally insane state.

Sure, but I didn't say it should be restarted.  I just said an error
should be logged.  Why should all of my processes be killed just because
zombies won't be reaped, and some services may not be restarted?  At
a minimum, it seems that this should be an administrative preference.


David