start/stop hook guarantees
gustavo.niemeyer at canonical.com
Tue Dec 6 19:48:33 UTC 2011
> Sadly, it's started to make much less sense to me as I delve deeper into
> the state restoration work. As I understand it, the intent is to allow
There are several intentions involved. I'd say that the most critical
goal is to attempt to put the service to run whenever we don't know
better, and I think we can consistently run "start" on most of the
cases you describe, with "install_error" and "configuration_error"
being indeed exceptions.
That said, there's some uncomfortable confusion going on here that is
an eye-opener. With all the background you have, if you can't clearly
and comfortably understand and define the behavior of "start" and
"stop", I don't have many hopes that anyone else will. People are also
unhappy with the whole "non-idempotent" start hook, and while I don't
think it's reasonable to expect people to write idempotent hooks, I
can correlate with the overall problem they're trying to solve.
I think we have a very good chance to revamp that mechanism a bit, and
solve all of these problems. I already have some ideas in mind, but
please give me a day to think about this problem. I'll try to write
something in the plane and come back to you tomorrow here in the list
(will be in San Francisco, so maybe too late for you).
-- I'm not absolutely sure of anything.
More information about the Juju