[Standards-JIG] Closing idle streams

Chris Mullins chris.mullins at coversant.net
Fri Jun 2 17:56:15 UTC 2006


Peter Saint-Andre Wrote:

> I keep waffling on this -- back and forth between including 
> a stream error or not. Really I don't think there *is* an 
> error, just a normal condition that the stream isn't being 
> used. So I lean toward just closing the stream.

The area where we encountered (and continue to encounter) issues with
this is uni-directional s2s streams. 

Many of the implementations seem to have issues when one side of the
communication channel goes down. We often end up seeing, "I can send you
a message, but you can't send me a message" and this continues for quite
some time.

I would tend towards a solution that says, "If this channel goes down,
take down the corresponding channel going the other way". I realize not
all s2s connections are symmetric (especially when you get into
clusters), but I think such a rule would increase the trend towards more
reliable s2s communications in the general case. 

The other option is to force the acceleration of the switch to
bi-directional SASL for s2s, but that's way outside the scope of this
particular discussion.

-- 
Chris Mullins




More information about the Standards mailing list