add-unit relation question

[William Reade]

On Mon, Jun 3, 2013 at 9:34 PM, David Britton
<david.britton at canonical.com>wrote:

> I tried changing that, and it is still racy.  I don't think there is any
> policy in juju that *all* joined hooks are run before *all* changed hooks.
>  Sure, they are run serially on the same unit, but across units if joined
> takes too long, joining units can start running changed before joined is
> finished.  I think I've verified this with a simple test charm.
>

For a reasonably detailed look at what's guaranteed wrt hook execution,
http://bazaar.launchpad.net/~juju/juju-core/trunk/view/head:/doc/charms-in-action.txt
should
tell you almost all you ever wanted to know, and possibly more. It's not an
exceptionally fun read, but it should be accurate: it's not very recent,
but I don't think it's rotted much since I wrote it. "Relations in depth"
and subsequent sections, starting at :246, are probably most relevant, but
you probably want to skim the foregoing for context.

In this particular scenario, the postgres charm doesn't write any
remote-unit-specific data to the relation, so there is indeed currently no
way to tell when pg_hba.conf has been written with the information required
for a particular client. I'd recommend option (3) in the original mail to
resolve this situation; client charms should just exit -relation-changed
without error if their name is not present in the list of acknowledged
clients, and just relax until their -relation-changed is run against valid
data before trying to connect. (FWIW, I'm a little surprised this hasn't
come up before... surely *some* other charms exist that handle analogous
situations?)

Cheers
William
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/juju/attachments/20130603/4da664f1/attachment.html>