(This email is being sent in preparation for putting together
 a fast track to go together with these CRs.)

In attending to 6236881, we're currently looking at turning
svc:/network/ipfilter up into 6 services in order to get the
correct level of access via "refresh" and "start"/"stop" methods
to managing related data.  The new list of services is currently
planned to be:

svc:/network/ipfilter (milestone)
svc:/network/ipfilter/ipf
svc:/network/ipfilter/ipmon
svc:/network/ipfilter/ippool
svc:/network/ipfilter/ipfnat
svc:/network/ipfilter/ipfconf

An unavoidable part of this change will be that administrators
will be expected to change their behaviour from:

svcadm enable ipfilter

to

svcadm enable -r ipfilter

but I cannot see any other way around this if the service
is to be broken up in a meaningful manner.

Because this will change the nature of svc:/network/ipfilter
from a service to a milestone, we will need to propogate the
"enable"d-ness of this service through to all of the others
(above) to correctly reflect previous in the new environment
with those settings prior to BFU and/or upgrades.  To handle
this, it appears necessary to use the file
/var/svc/profile/upgrade and establish a contract with
PSARC/2002/547, correct?


On a seperate note...

In putting together the dependency graph, what I would have
liked would be to have a link between services that was heavier
than "optional_all" but lighter than "require_all".  What I
was looking for was to be able to say:
- start X if Y has started or start X if Y is disabled
- wait for Y to finish starting and then start X

In this, "svcadm enable X" if Y is disabled does not enable
Y nor stop X from reaching online.  However, if Y is in
maintainence, then "svcadm enable X" should wait until Y
is online.

Should this be an RFE for SMF or does this expression of
a dependency relationship between services raise a red
flag to others?

As it is, we've opted to use "optional_all" here.


For those internal to Sun, if you want to see, in detail,
what the current manifest files look like, point your
browser at:

http://greatwall.prc/~zf203873/cr6236881/

Darren


Reply via email to