[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Possible Agenda - July 12
Possible
agenda… -
Educational ppt approval -
Main spec updates (see below) -
Symptoms/Prescriptions for Cloud profile (see way below) -
CatalogueSource -> CatalogueEditor (or
something that denotes the more fine grained control)
Jeff – makes sense. -
Case Manager -> remove? We have no real use for a
Case Manager in the spec and certainly not in our implementation. I think this
is redundant unless we consider adding "Case" as well as a concept
–which might complicate things. Jeff – Case
Manager is an Orchestrator, and seems to be implementation specific. -
Add TypeRegistry –maybe with a better name? Also, this
by no means mandates the actual existence of a registry in the system, but
simply a point where users can find existing types. How this is to be
implemented is a detail out of the spec (e.g. could be implemented as an actual
registry/store, or as a query mechanism that collects this information from the
various places in the framework). Jeff – Why
wouldn’t the Catalog be a good place to store the Type information? Answering
my own question: Architecturally, it is probably best to separate
Practitioner/SymptomEmitter registered information from CatalogEditor supplied
information. In an implementation, a single data store could be used for
both (simplifying administration, maintenance, etc). Does that sound
about right? -
I have some issues with the interfaces as set out in
the spec. "Interfaces" imply specific interaction semantics, and the
way they are currently spelled out makes me uncomfortable. In a sense, what we
have there are not interfaces, but Roles! But we already have the term role for
the active roles in the framework, so this makes things tricky (imagine:
"The role of SymptomHandler can be taken by the role of
Diagnostician" :-\ ) And these roles have some operations which we should
spell out not in the common format ("getPrescriptionTypes" etc) but
in a more loose way ("get/list prescription types"). Then in the
relevant appendix we make it more concrete with specificities, e.g. REST. Jeff – I think the first column in our Interface
table is just the ‘name’ of the Interface, not a role. For example,
SymptomHandler is not a role, but just the name of an interface. Agree
that the operations should be defined in a loose way. ------------------------------------------------------------------------------------ Below is a
list of possible Symptom types for Cloud management. Two
concerns were mentioned on the call today: 1. Lack of taxonomy – Is there an
existing taxonomy that could be used for the symptom type (and/or prescription
type)? 2. Symptom boundary – How do we decide
what should be a symptom and what shouldn’t be a symptom? For example,
resource performance notifications (ie: cpu1 on server xyz is 39% utilized)
doesn’t seem like a good candidate for a symptom. Can we articulate a
‘rule’ for what should be a symptom? Please
offer your thoughts.
|
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]