COSMOS SDD Minutes 14JAN08

Logistics

• Date: 14JAN08
• Time: 1PM EST

Agenda

1) Use case identification -- two areas of focus

• Install/uninstall simple application
• Install/uninstall complex solution

Attendees

SAS: Jason Losh, Merrie Jensen, Jeff Hamm, Mark McCraw, Josh Hester, Robert DeMason
IBM: Chris Mildebrandt, Mark Weitzel, Charlie Halloran
CA: Jimmy Mohsin, Jack Devine

Minutes

Working on logistics. Iteration 10 is March/end June. Need to identify any dependencies among the project.
Q: What is scope of Use cases? Just installation of COSMOS components?
A: Scope is package authoring tooling & runtime of CL1 of the SDD spec plus examples. Maybe including installing pieces of Eclipse depending on how we line up with P2. This needs to be extensible so it can be expanded to add CL2 support when resources are available.
Overriding assumption: Maximum of one each of IU, LU, CU, and package descriptor per SDD. SDD means both, otherwise explicitly state it.

Tooling Use Cases
1) Use case: normal development creates artifacts which reside somewhere. Need to create SDDs for these artifacts based on the build information. Build information can be metadata somewhere or introspected from the artifact itself.
2) Separate use case: just introspect the artifact and create as much of the SDD as possible. This is useful especially in the case of a 3rd party component, where the build information is not available.
3) Support for build systems beyond the popular rpn, msi, etc. so that metadata from those systems could be brought in.
4) Need a set of rules for assembling the metadata and resolving conflicts (when the same set of data occurs with different values)
5) As the SDD is being created, validate that what has been created is syntactically correct. Validation may be turned on or off. A plug-in rules engine could also be useful to validate user extensions that are allowed by the validation process.
6) In addition to SDD creation, there is a use case for creating the actual package(Package Generation): SDD and artifact onto "media" that would make it consumable. May need to include SDD runtime and repository code in the packaging.
7) Tool must be able to be configured (where is data source, what rules are created, etc): basic administration of the tool parameters. For example, when moving the tool for use in another department or product. These things will be configurable:

• artifact inputs
• location of SDD outputs
• custom rules for creation and validation
• SDD content constraints (per IU, CU, LU)

8) Need a GUI front end to create and edit the SDD that has been initially created, or create one from scratch. GUI tool could also need the options to validate, package and deploy. Such a front end would need the same kind of location definitions and constraint administration to allow or disallow parts of the SDD that are "editable".

Deployment
Background: There are two categories of applications: Non-Federated and Federated (whether it is exposed to a federated inventory kind of database or not). For Federated Applications: need Web Services port open, take CMDBf graph queries and give replies about the MDR.
Assumptions: No way to know if a federated CMDB is accessible or not. But if it is, need to query it and write to it. Using the federated CMDB is the extra step in the below use cases for the federated case. Which repository to use should be transparent to the install program as much as possible, it should be handled by the runtime based on a configuration setting or documented install parm (e.g., CMDBf = True, CMDBServer=ipaddr).
The install target can be one of: Test/Development, Production, Hardened.
Requirement to run from media is supported by OASIS use cases 14, 15, 192.
What format is required for the repository? SML? Single MDR for the non-federated case that represents all software on the system? Store is not as important as being able to get the right info in and out of it.

Deployment Use cases:
1) Installing a Non-Federated application: include a runtime to interpret the package and perfrom these tasks:

• bootstrap the runtime if not already there
• install the artifacts
• register the existence of the resource in a repository (create/initialize if not already there)
• completion actions (reboot, start/stop service, cleanup, etc.).
  

2) Uninstalling a Non-Federated application: perform these tasks

• bootstrap the runtime if not already there
• de-register the resource from the repository
• remove the application
• completion actions (reboot, start/stop service, cleanup, etc.)
  

3) Configuring a Non-Federated application: perform these tasks

• bootstrap the runtime if not already there
• configure the artifacts
• register the configuration change in a repository (create/initialize if not already there)
• completion actions (reboot, start/stop service, cleanup, etc.).
  

4) Undo a Non-Federated application: perform these tasks

• bootstrap the runtime if not already there
• remove the last level of update to the application
• change the registration as needed. Undo only applies to an update to a resource, not a configuration.
  
• completion actions (reboot, start/stop service, cleanup, etc.)
  

5) Updating a Non-Federated application: perform these tasks

• bootstrap the runtime if not already there
• update the application
• change the registration as needed.
• completion actions (reboot, start/stop service, cleanup, etc.)
  

6) Be able to register in federated CMDB without requiring a re-install (for updating an already-installed application into the CMDB).

Action Items

• How to contribute code from both SAS and IBM? Mark/Charlie need to work with IBM IP Legal.
• To be a reference implementation there are some rules to abide by and we should document how much of CL1 we support within the reference implementation.
• Can a configuration action be rolled back, and how is that done? Merrie to investigate.