[Spice-devel] gtk client side timeouts

[Jeremy White]

>> I looked for past discussions on spice-devel; I only readily
>> found discussion of the keepalive parameters on the server side,
>> and it looks like we take the idle timer down to 10 minutes now,
>> from 2 hours.  I didn't find discussion of something similar
>> client side.
>> 
>> Have we considered tuning them on the spice-gtk client side?
>> 
>> The attached patch applies a roughly 75 second timeout for me; is
>> this something that would be viable to add to the gtk client,
>> even if only as an option?
> 
> Ah, I've coincidentally been looking at this this week too, not
> with the same goal in mind though, my problem is SPICE connection
> sometimes getting dropped if idle for too long. But the current
> answer is also tweaking TCP keepalives. Definitely no objection
> having some form of keepalive client sides, though your proposed
> change is not going to work for the Windows client.

The other issue I had with my proposed change was that it was pretty
chatty; a packet every 30 seconds.  I guess that's not a lot in the
scheme of things, but it still troubles me.

In the interest of full disclosure, I also explored putting a tcp user
timeout on the socket.  That catches the case where we're trying to
transmit data, but there is no answer.  (So user wiggles mouse, hits a
key, some time later we time out).

That sort of worked; I got the cursor channels to close.  But having a
cursor channel close, it turns out, is not fatal to the client as a
whole :-/.

So another approach would be to add the user timeout, and have the
loss of a cursor channel become a fatal condition. That would avoid
the chatty packets, but would require a more invasive overall change.

But that would not solve your issue, so perhaps it's not the right choice.

Another question is - if we add a keepalive, does it need to be
configurable?

> 
> I've made some good progress on patches to add both Windows/linux 
> support to glib, but need to finish them up ;)

:-).  I'll see if I can work up an interim patch that works for both
Windows and Linux.

Cheers,

Jeremy