Scott Lawrence wrote:
> On Tue, 2008-09-09 at 00:19 -0400, Damian Krzeminski wrote:
>> Kevin Thorley wrote:
>>> On Mon, 2008-09-08 at 20:46 +0000, Scott Lawrence wrote:
>>> [...]
>>>> Other than a general objection to polling, I guess not.  What frequency
>>>> were you thinking of using?
>>>>
>>> I have a general objection to polling as well, but this seems to be the
>>> easiest way to squeeze this in to sipXconfig during this sprint.  Damian
>>> had mentioned a 30 second polling interval.  I think we could safely go
>>> up to 2 minutes and still maintain a reasonable amount of "liveness" in
>>> the system.  We could also make this a configurable param (either
>>> through sipXconfig or a config file) if it were to cause problems in
>>> certain network environments.
>>>
>>> Kevin
>>>
>>>
>> sipXconfig is only going to poll when admin adds a new server which is not
>> configured yet, right?
>> It's not going to poll all the servers, all the time (at least not for that
>> reason).
>>
>> And if that's the case we can get down to about 10 seconds: the goal is to
>> get the new server noticed and configure it as soon as it starts.
>>
>> How about the Raymond's question though: how does sipXconfig is going to
>> contact supervisor for the first time?
>>
>> Looks like the initial exchange was to be done through HTTP PUT of the
>> gzipped file. See:
>>
>> http://track.sipfoundry.org/secure/attachment/15317/ClusterManagement.html#config.if
>>
>> Which would suggest that the polling could be implemented as HTTP GET on
>> the same URL. Bad idea?
> 
> We need to review the initial setup scenarios in more detail, but this
> is _not_ just to support that.
> 
> Suppose that any configuration change is made, and the push of that
> change to the distributed system fails because that system is down?
> When the distributed system comes back, the checkin provides the
> stimulus to tell sipXconfig that it can now successfully send the
> update.
> 
> On the other hand, that doesn't help if the problem was a network
> problem but both systems were up, so I guess you have to poll for that
> one anyway.
> 
> 

Yes - one of the sides has to actively poll if we want to make sure that
config changes are properly replicated. I guess what I was trying to
propose is the phase based approach to implement the current spec.

The initial phase - something that we would do in sipXconfig as a part of
this sprint - would be just adding a "Send Profile" button to server page.
That would be in line with what we are doing for phones, gateways and other
device and if I do not miss anything would allow admin to configure remote
system from sipXconfig UI.

Then we can add servers polling - more frequent for newly added servers,
less frequent for the ones that have been configured. That would allow us
to push

And even later we can implement check-in interface: that would let new and
restarted servers to register with sipXconfig and allow it to react very
quickly when server is up.

I am hoping this might help us to concentrate on the most important problem
at hand: how to initially configure remote server and will leave the open
door for the improvements.

Damian

_______________________________________________
sipx-dev mailing list
[email protected]
List Archive: http://list.sipfoundry.org/archive/sipx-dev
Unsubscribe: http://list.sipfoundry.org/mailman/listinfo/sipx-dev

Reply via email to