Dave,
Thanks for the review. See my response in-line.
Dave Miner wrote:
> Sundar Yamunachari wrote:
>> Hi,
>>
>> The proposal to simplify AI manifest by removing criteria manifest
>> is updated and available at
>> http://www.opensolaris.org/os/project/caiman/auto_install/manifest_simplification_proposal_v2.
>>
>> The document has more information and user scenarios. Please review
>> the proposal and provide your feedback.
>>
>
> I think the proposal should address the following points:
>
> - Relationship between services and manifests needs better definition
> (perhaps this is just a reference to one of the other specs written
> recently)
I will add the information.
>
> - How does this relate to the proposed derived manifest functionality?
> It feels somewhat redundant, so I think we need much clearer
> understanding of the cases in which one would use one or the other.
okay.
>
> - The potential complexity of the command line required to define
> multiple ranges needs more fleshing out to conclude whether this is in
> fact an improvement, usability-wise. A syntax that can both
> accommodate complex criteria sets and remain memorable enough to not
> require constant reference to a man page seems non-trivial to solve.
That is one of my concerns too. One compromise could be that the user
can use command line for simple criteria and use an input file for
complex criteria sets.
>
> In a more detailed sense, I'm not sure how it really improves the
> handling of the problems in section 2:
>
> 2.1 - The problem of changing criteria could be simplified in the
> current scheme by direct replacement, rather than a remove/republish.
Yes, We could do that. If you separate criteria specification from the
manifest file, we could update criteria sets, remove a criteria and
replace the criteria with a new criteria set. So this is much more
flexible than simple replacement.
>
> 2.2 - There still seem to be scenarios (for example, 4.2) in which the
> duality issue here still exists.
It is possible to use MAC address in create-client and add-manifest (as
a criteria) but it is not necessary to setup both. In the current
scheme, you may have to setup in both the places (create-client and
criteria as part of the manifest) to make it work
>
> 2.3 - Retrieval of criteria seems like it could be accomplished in the
> current setup without requiring this change. This seems more like a
> matter of not accounting for the requirement in the implementation
> than a fundamental design problem.
Agreed. But removing the criteria specification from the manifest makes
it easy for users with small set of machines. By providing a way to
manipulate criteria (add, remove or replace), it will be more flexible
to change the criteria rather than providing the whole manifest as the
input for changing the criteria.
Further by separating criteria specification from the manifest, we could
support loading several manifests with each one having its own criteria
set using a single command. For example, I could provide a file with the
following format:
Manifest 1
criteria 1
criteria 2
criteria 3
Manifest 2
criteria 4
criteria 5
.....
This will be really useful when we are setting up a new service with
existing manifests and criteria.
>
> All of the above is not to say that I am in favor of the status quo,
> but merely that I don't yet see this proposal as providing an obvious
> improvement.
Conceptually I think separating criteria from the manifest is a good
idea. We need to come up with the right set of interfaces to make it
easy to use and easy to maintain.
Thanks,
Sundar
>
> Dave
>
