[Search for users]
[Overall Top Noters]
[List of all Conferences]
[Download this site]
Title: | DECmcc user notes file. Does not replace IPMT. |
Notice: | Use IPMT for problems. Newsletter location in note 6187 |
Moderator: | TAEC::BEROUD |
|
Created: | Mon Aug 21 1989 |
Last Modified: | Wed Jun 04 1997 |
Last Successful Update: | Fri Jun 06 1997 |
Number of topics: | 6497 |
Total number of notes: | 27359 |
835.0. "Upward compatible mgmt models" by COOKIE::KITTELL (Richard - Architected Info Mgmt) Fri Mar 22 1991 15:52
Help me understand how to deal with upward compatibility of
management specifications. You've probably got this all
figured out for the network stuff, but it eludes me. As usual,
I have a scenario. Say V3.0 of my management spec has an
Flavor attribute defined as this enumeration:
TYPE Flavors = nn (Vanilla = 1, Chocolate = 2);
We ship V3, it's a hit so we ship a bunch, make bags 'o bucks
and bring DEC stock up $120. For V3.1 we want to add Strawberry:
TYPE Flavors = nn (Vanilla = 1, Chocolate = 2, Strawberry = 3);
So we update the management spec and the version returned for the
implementation of the object. We ship V3.1 and customers start
installing it.
Assuming for the moment that only the director (PM/AM) and the
entity know the details, and that they are both driven from
the MCC dictionary, we have four possible combinations. Two
aren't interesting, the ones where both the director and the
entity are from the same version kit. Of the other two, the
easy one is where the director is at V3.1 and the entity is at
V3.0. The V3.0 entity only sends a 1 or a 2 for the Flavor
attribute value, so the V3.1 director has no trouble with
that.
But we've put up the V3.1 kit on node A, which is where the
entity is running. On node B, which is still running V3.0, we
fire up MCC and Show the entity. The entity has had its Flavor
attribute set to Strawberry, so it returns a 3 for the
attribute integer value. What happens?
If the AM just passes through the 3 as the attribute value,
the MCC V1.1 FCL and Iconic PMs will display "%MCC-E-NOENTITY,
no corresponding entity instance exists", and not display any
of the other values, the Show fails. I know that it is on the
list for MCC V1.2 to make that message more meaningful. Have
any plans been made as to what it will do? For enumerations,
at least, I'd sure like it to display something like
Implementation = ({Component = Creamery,
Version = V3.1.0})
Flavor = Enumeration (3) not in dictionary
That would assure that a V3.0 director could still display
info from a V3.1 entity, even if it couldn't translate all the
enumerations. In addition, we could have the V3.1 release
notes indicate that the value 3 means Strawberry.
For MCC V1.1, an option that occurs to me, at least for those
enumerations I might be able to anticipate extending in the
future, would be to define the enumerations for V3.0 as
TYPE Flavors = nn (Vanilla = 1, Chocolate = 2,
Unknown Flavor = 99999999);
The AM would then "sniff" the values coming back from the
entity, and force anything it didn't recognize to 99999999,
thus giving the PM something reasonable to display:
Implementation = ({Component = Creamery,
Version = V3.1.0})
Flavor = Unknown Flavor
The difference being that the integer value isn't displayed, so
there isn't a way to tell which of the new flavors it might
be.
Another alternative, of course, is for the V3.0 AM to sample
the entity version and refuse to do business with an entity
newer than itself.
Or have the entity support both management models. Every
request from the AM would have to include the AM version so
the entity could decide what to return. But it isn't clear
what a V3.1 entity whose flavor was Strawberry could tell a
V3.0 AM without lying to it.
Which tricks am I missing?
T.R | Title | User | Personal Name | Date | Lines |
---|
835.1 | print the 3 | TOOK::CALLANDER | | Sat Mar 23 1991 13:48 | 8 |
| a bunch of worthwhile ideas...at this time the mcc-noentity error is
seen as just that an error, it has been qared. I don't know quite yet
what we are going to do about it yet. The most reqeusted change is to
simply do a hex dump of the value. Therefore in your strawberry example
if it isn't a known flavor we would print the number 3.
jill
|
835.2 | Yup, 3 | CAPN::SYLOR | Architect = Buzzword Generator | Sun Mar 24 1991 21:10 | 2 |
| Fix the PM. Print it as 3. That's what the NCL spec says to do in this
case. It's the only reasonable choice of the ones Richard listed.
|
835.3 | agreed | TOOK::CALLANDER | | Tue Mar 26 1991 10:13 | 0 |
835.4 | see QAR 556 in MCC_INTERNAL | TOOK::DITMARS | Pete | Fri Mar 29 1991 09:53 | 0
|