Completely rewritten design attached.
On 04/28/2010 07:57 AM, Stephen Gallagher wrote:
On 04/28/2010 06:04 AM, Sumit Bose wrote:
> On Tue, Apr 27, 2010 at 03:37:02PM -0400, Stephen Gallagher wrote:
>> Allow backends to set a callback in the be_ctx that should be
>> invoked when the ID provider goes online.
>>
>> This can be used to perform regular maintenance tasks that are
>> valid only when going online.
>>
>> Specifically, we can use this to perform a deferred kinit on behalf
>> of the user when we go online. We can also use this to trigger a
>> dynamic DNS update for the IPA provider.
>>
>> --
>> Stephen Gallagher
>> RHCE 804006346421761
>>
>> Delivering value year after year.
>> Red Hat ranks #1 in value among software vendors.
>>
http://www.redhat.com/promo/vendor/
>
>> }
>>
>> + /* Reconnection succeeded
>> + * Run any post-connection routines
>> + */
>> + if (state->be->online_cb_list) {
>> + DLIST_FOR_EACH(callback, state->be->online_cb_list) {
>> + ret = callback->cb(callback->pvt);
>> + if (ret != EOK) {
>> + DEBUG(0, ("Post-connection callback returned
[%d][%s]",
>> + ret, strerror(ret)));
>> + tevent_req_error(req, ret);
>> + break;
>> + }
>> + }
>> + }
>> +
>
> I have several comments here:
> - I would suggest to move this loop into a separate subroutine like
> be_run_online_cb()
Yeah, that would make sense. Then we could expose that routine.
> - Maybe it would make sense to call the callbacks with a timer event to
> make sure that they do not block the current request or other
> callbacks
I was going back and forth on this. As you said, the risk is that it
could cause a block. So I think I'll do as you suggest and make it a
stepped series of timer events. The catch is that return values from the
callbacks won't affect the request they're attached to.
> - Currently the callbacks are run on every reconnect which might be
> intended, but is IMO different from running after the transition from
> offline to online. E.g. with GSSAPI we need to reconnect if the ticket
> becomes invalid although we haven't been offline. To catch only the
> offline-online transition I would suggest to add a run_online_cb flag
> which is set by be_mark_offline() and checked and unset by the
> proposed be_run_online_cb().
Good idea, I didn't think of that. I also wasn't thinking of the GSSAPI
case. Doing it this way would also allow me to make these callbacks
available to the proxy provider.
If the callbacks should be run after
> every successful connection it would be better to replace
> '*online_cb*' by something like '*connect_cb*'.
> - Is it intended that only the id provider is running the callbacks?
>
This was mostly an optimization. The ID provider must be online for any
of the other providers to actually work (since they will all verify the
user before proceeding). So it's not actually possible in our current
design for anything but the ID provider to cause the transition from
offline to online operation. (Obviously the reverse transition can occur
in any of them)
--
Stephen Gallagher
RHCE 804006346421761
Delivering value year after year.
Red Hat ranks #1 in value among software vendors.