[nanomsg] Re: The nanoconfig preview
- From: Paul Colomiets <paul@xxxxxxxxxxxxxx>
- To: "nanomsg@xxxxxxxxxxxxx" <nanomsg@xxxxxxxxxxxxx>
- Date: Fri, 4 Oct 2013 19:22:33 +0300
Hi Martin,
On Fri, Oct 4, 2013 at 6:50 PM, Martin Sustrik <sustrik@xxxxxxxxxx> wrote:
> On 2013-10-04 17:39, Paul Colomiets wrote:
>
>> My opinion is quite biased: I hate synchronous stuff. But anyway:
>>
>> 1. Application startup should not be blocked by network lag
>>
>> 2. Address resolving for multiple sockets might be parallel
>>
>> 3. Nanoconfig rechecks addresses every 5 min, and when update message
>> arrives
>
>
> I was thinking about the case when the topology doesn't exist. Shouldn't the
> fact be reported to the user?
>
Should. But that will not make lib fully synchronous, so nc_close will
be async anyway (this what discussion started from). How to deliver
initial error is an interesting question though (right now all errors
are just printed to stderr, so it's easier to debug).
Actually, for development asynchronous stderr should be ok. For
production, the failure will be seen in monitoring anyway. Also
non-existent topology request will be logged in configuration service
I think.
--
Paul
Other related posts:
