hi meem

Peter Memishian wrote:
>  > > But I personally think that temporary datalinks should also be 
>  > > represented as SMF instances, as they do provide datalink services.
>  > >
>  > yeah, i can see that there's a reasonable argument to do this. i think
>  > the key question is whether we expect the SMF representation
>  > to be complete regarding what is configured on the system,
>  > versus being complete about what is _persistently_ configured
>  > on the system. the latter is probably less intuitive to administrators,
>  > but i don't know that that is such a big problem - my assumption
>  > is they'll be using dladm to query link status rather than
>  > "svcs datalink/*". the problem i see with exhaustively
>  > representing temporary configuration is having to undo the
>  > temporary portions of the configuration explicitly (i.e. the
>  > instances that were created via temporary commands) on
>  > reboot.
>
> I thought SMF already had a way to mark e.g. property groups as being
> temporary?  Is that functionality insufficient?
>
>   
temporary property groups will work fine (they disappear
on reboot) but the problem is they need to be stored in instances,
which are always persistent across reboot. this is a problem because
such temporary instances would need to be torn down
explicitly on reboot as a result. while this is doable, it does
stray somewhat from the SMF model which is geared towards
storing configuration intent, and i'm concerned about the
knock-on effects of having SMF representations for such temporary
entities, e.g. the GUI would likely then have to determine
the nature of the object represented (temporary or persistent),
and likely not display such temporary instances.

since it appears that the temporary information we need to
store maps to data that dlmgmtd needs to retain (but not
across reboot) for restart purposes - i.e. the linkid-linkname
mappings - my suggestion is to store that data locally
within the datalink-management instance in a
temporary property group.

the motivation for this distinction is that the methods
for these persistent instances could potentially simply
instantiate the object they represent temporarily,
e.g. in the case of an aggregation instance start method,
we'd have: "dladm create-aggr -t  ...". the benefit of this
approach is that the existence of the aggregation is
predicated on the state of the instance, as opposed
to having to modify "dladm up-aggr" say to correlate
configuration data with instance state (i.e. the only
persistent aggregations that are created on boot are
those that have instances that are enabled).
> FWIW, I agree with Cathy that temporary datalinks should be represented as
> SMF instances, for the same reason that temporarily enabling a service
> still causes that service to show up in `svcs'.
>
>   
right, but consider the consequences of representing
temporary entities as SMF instances - the semantics
of enabling an instance imply that the enable action
is persistent, i.e. the service instance will be enabled
at next boot also. if we support an instance representation
for temporary entities (which can be manipulated by
svcadm), my fear is that it will set expectations in this
direction.

alan

Reply via email to