Code review comment for lp://staging/~jimbaker/pyjuju/juju-status-changes-spec

On 2012/03/16 02:53:42, niemeyer wrote:

https://codereview.appspot.com/5837051/diff/1/source/drafts/juju-status-changes.rst
> File source/drafts/juju-status-changes.rst (right):

https://codereview.appspot.com/5837051/diff/1/source/drafts/juju-status-changes.rst#newcode24
> source/drafts/juju-status-changes.rst:24: - {name: db, service: blog2,
state:
> down}
> On 2012/03/16 01:30:59, hazmat wrote:
> > Its useful i think to explicitly state the instance of a particular
relation
> is
> > down rather than relying on a user inferring against the absence of
a value
> > that there is a problem.

> If the goal is to make things clear and obvious, having a massive
amount of
> information for each relation won't help.

> What about splitting up, then:

> relations:
> db: [blog1]
> relations-pending:
> db: [blog2]

that sounds reasonable.. it would be relations-down.. pending isn't
valid from the unit's perspective, it either knows about and its up or
down, i mean we could infer, but its a very transient state, effectively
the time between the add-relation and a watch firing.

also i wonder if it makes sense to only use list format for relations
that are server oriented, for client relations its only a singleton by
name. it does add minor ambiguity to the parsing as interface role isn't
available in status. we should probably run this by the list before it
lands, as its going to break the extant status parsers.

https://codereview.appspot.com/5837051/