Juju is still too hard

[Mark Shuttleworth]

On 22/09/14 15:06, Mark Ramm-Christensen (Canonical.com) wrote:
> I think we need to make sure that we do the best error reporting we can, so
> if Juju isn't working because of Azure issues, we should find some way to
> let users know that so that they can try another cloud, contact microsoft,
> or otherwise find another way forward.
>
> --Mark Ramm
>

We have to take responsibility for the experience of the user.

That means it's a bug in Juju if cloud-level failures are obscured into
Juju errors that are hard to debug.

The measure of quality in software is "how it deals with the
unexpected", that means anticipating and handling errors in a way which
is appropriate:

 * retry a few times if that might help (if it's transient the user
experiences delays but not a failure)
 * fail gracefully, meaning:
   * don't leave unclean bits that the user has to tidy up
   * provide a clear guide as to where the problem is and how to fix it

Please, review code landings from that perspective. Look at implicit
assumptions that a call-out (especially external call-outs to things
like cloud services) worked, and insist on appropriate handling, to the
standard above, of all the cases where it actually failed, including:

 * explicit failures (500 etc)
 * responses that are unexpected (OK with data that doesn't validate to
expectations)
 * hangs and timeouts (no response at all)
 * crashes / exceptions (for local calls)

Mark
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/juju/attachments/20140923/ad68eec8/attachment.html>