BIND 10 #922: Callback for foreign module configuration

[BIND 10 Development]

#922: Callback for foreign module configuration
-------------------------------------+-------------------------------------
                   Reporter:         |                 Owner:  jelte
  vorner                             |                Status:  reviewing
                       Type:  task   |             Milestone:
                   Priority:  major  |  Sprint-20110517
                  Component:         |            Resolution:
  configuration                      |             Sensitive:  0
                   Keywords:         |           Sub-Project:  Core
            Defect Severity:  N/A    |  Estimated Difficulty:  0.0
Feature Depending on Ticket:         |           Total Hours:  0
        Add Hours to Ticket:  0      |
                  Internal?:  0      |
-------------------------------------+-------------------------------------
Changes (by vorner):

 * owner:  vorner => jelte


Comment:

 Hello

 Replying to [comment:3 jelte]:
 > Think this was already a problem, but easy to remedy now; we should move
 the subscribe() call down until after the last possible exception (or
 unsubscribe() before we throw) in the add_remote() call, so that we don't
 end up with 'loose' subscription if something goes wrong in the setup.

 OK, done.

 > Do we have/want a strict requirement that module names don't contain
 dots? They can't contain / due to another implementation limitation right
 now. If we do not have or want such a requirement, perhaps we should have
 two separate methods for adding them either with a module name or a file
 name. (Also, not necessarily a problem, but we should be aware of it:
 using a module name instead of a specfile means that said module has to be
 running (so that it is known for config manager) right now. Hopefully we
 can make this problem disappear soonish :)))

 I added a bool parameter. It is less duplicate code and the change was
 smaller.

 > Can you add a bit of text to the documentation that the handler should
 never throw? (or, alternatively, catch any exception when calling it, so
 that update and init do not bail out should the handler throw an
 exception)

 OK

 > There is a separate python implementation for ccsession etc. we should
 add this callback functionality there too (can be a separate ticket so as
 not to block the other tickets that depend on this one).

 I didn't need it right now, this is taken as a part of something else
 O:-). But OK, I'll create a ticket.

 > This was already a potential problem, and solving it will probably
 require some internal protocol change, but if an updated configuration is
 rejected by a module, I don't think the modules that are peeking in on its
 config values are notified of this. Same as for previous point, as this
 was already a problem, we should probably make this a ticket.

 Well, I wrote some notes about this into some ticket already. I'd like to
 move all checking of configuration into the python plugins, which would
 solve two things:
  * The problem with offline configuration.
  * The check is called now first and no notification is done if there's an
 error. We could even do some kind of transaction ‒ first check everything
 and then merge and notify. If one fails, all fail.

 Anyway, could you have a look at the new changes?

 Thanks

-- 
Ticket URL: <http://bind10.isc.org/ticket/922#comment:4>
BIND 10 Development <http://bind10.isc.org>
BIND 10 Development