Re: Server-Sent Events contradiction

On Wed, 2 Jan 2013, Glenn Maynard wrote:
> On Wed, Jan 2, 2013 at 4:23 PM, Ian Hickson <ian@hixie.ch> wrote:
> > 
> > Since none of the browsers I could test reconnect for 500s currently 
> > as far as I can tell, I've changed the spec to not make 5xxs 
> > reconnect. The server load issue seems like a pretty big deal.
> 
> Ordinary exponential backoff deals with the server load issue.

Exponential backoff is a pretty horrible user experience. It basically 
doesn't work in UI, especially for transient problems.


> It doesn't make sense for a transient error to cause the connection to 
> permanently fail.

We're not really talking about a transient error, though. We're talking 
about a situation where you connect and immediately have a problem, or the 
connection drops, you try to reconnect, and _then_ you immediately have a 
problem.


> All this does is make web developers do it themselves, which we 
> shouldn't need to deal with.

I would rather make Web devs who want this to reconnect have their error 
case for the EventSource URL a 200 OK with the right type that then drops 
the connection right away, than hammer the server so hard that the Web dev 
can't log in to fix the problem in the first place. (Or, even better, they 
can simply not open the incoming connection until the problem is resolved 
-- that will also cause attempts to reconnect.)

-- 
Ian Hickson               U+1047E                )\._.,--....,'``.    fL
http://ln.hixie.ch/       U+263A                /,   _.. \   _\  ;`._ ,.
Things that are impossible just take longer.   `._.-(,_..'--(,_..'`-.;.'

Received on Thursday, 3 January 2013 02:16:28 UTC