Code review comment for lp:~cmars/juju-core/no-default-series

Revision history for this message
John A Meinel (jameinel) wrote :

On Thu, Mar 20, 2014 at 6:51 PM, Casey Marshall
<email address hidden> wrote:
...
>
> When a client deploys or upgrades a charm, it is responsible for
> resolving the URL with the charm repository before making the API call.
> As far as I know, these are the only entry points for new charm URLs
> from the user into the API server. If this is naive, let's iterate on
> improving the defensiveness & I'm open to suggestion.
>
> I'd like the state server to never have to resolve a charm URL series,
> and reject all unresolved URLs.

I'd like to understand this a bit better. Given the API server already
supports "deploy cs:precise/foo" which it will then go to the charm
store and download, it seems perfectly reasonable that the client
could connect to *1* remote machine (the API Server) rather than also
needing to know who the Charm Store is and how it is configured, and
how it can resolve information there.
Why do you feel it is better to push this work into needing Clients to
be smart? (One argument, I guess, is that Juju GUI is already listing
the charm store in order to present anything to clients, so the only
other client today is juju-core.)
But given that the API Server already needs to know how to talk to the
charm store (to install cs:precise/foo) it doesn't seem that bad to
have it also resolve URLs.

John
=:->

« Back to merge proposal