Rainer J. H. Brandt writes:
> > outside the zone can't 'see' the change.  So, why isn't this just an
> > exercise in proper application design, and why are zones necessarily
> > special here?
> They aren't.  I think I should have changed the subject line.
> (I've done that now.)  And probably switched to another mail list.
> (SMF? I hesitated because I hate cross-posting.)

Yes, this does look like a topic that's more appropriate for the SMF
list.  Though it's of interest to Zones users, it seems applicable to
arbitrary networking environments.

> I wasn't just talking about boot order.  That probably wasn't clear.
> That's just one of the many problems that could be solved by what I
> suggested.  And that was SMF across machines, which is what a cluster
> or a Service Oriented Architecture is about, and which is what we
> almost have with SMF.  A real thing instead of all this marketing buzz
> around SOA.
> I mean this:
>     <dependency name='my-great-database' type='service'
>         grouping='require_all' restart_on='none'>
>       <service_fmri value='svc://dbhost/application/postgresql' />
>     </dependency>
> Wouldn't that be nice?

I'm not so sure.

> Do you see that attribute "restart_on"?
> No, I won't ask the question that you mentioned up where I added the
> [***] comment.  I already know how to to that with SMF.

In general, I would not want my networking applications restarted
merely because a peer went haywire.  For a really simple example,
consider a web server.  The server likely depends on name services for
normal operation.  Should it go down if name services are temporarily
unavailable?  Should it restart?  Is networking unavailability or
partitioning equivalent to service failure?

Yes, it's possible to construct cases (such as a web server that's
wholly non-functional if a co-located database is missing) where such
behavior makes some sense.

I'm dubious, though, that this is the right way to design networking
applications.  It doesn't correspond to the multiple decades of design
practice in this area, so, before I'd endorse it, I think I'd have to
see some really compelling results, as well as, in the IETF tradition,
examples of problems that _can't_ be solved in the usual way.

Things get stranger still if you have a heterogeneous environment,
which I think is a fairly common situation.  SMF exists only on
Solaris.  So, how do you complete dependencies across platforms?

> So, should we move this to the SMF list?  Or forget it?
> (As I said, there are most likely people who have thought about this.)

The SMF list seems much more appropriate to me.  The mechanisms
involved would have to be in the SMF infrastructure itself -- making
the restarter and dependency tree aware of distributed applications --
and the usage of the new feature would be independent of (though
perhaps useful for) Zones.

James Carlson, Solaris Networking              <[EMAIL PROTECTED]>
Sun Microsystems / 1 Network Drive         71.232W   Vox +1 781 442 2084
MS UBUR02-212 / Burlington MA 01803-2757   42.496N   Fax +1 781 442 1677
zones-discuss mailing list

Reply via email to