[sldev] Preferred CAPS timeouts and retries?

[John Hurliman]

Ryan Williams wrote:
> John Hurliman wrote:
>   
>> Ryan Williams wrote:
>>     
>>> I have a few more questions:
>>>
>>>   -is this happening on socket open, or is it refusing to deliver data
>>> over the open socket, or both?
>>>   - what are the logs from the official client that you're reading to
>>> determine that it's occasionaly problematic there?
>>>   - have you noticed any pattern for which regions have the problem?
>>>
>>> Thanks for bringing this to our attention, BTW, we appreciate it.
>>>
>>> -RYaN
>>>   
>>>       
>> I don't have any of the old log files from the official viewer so I'm
>> going to start looking for it again and save the log when it happens so
>> I can post. Translating .NET callbacks in libsecondlife to what is
>> actually happening at the lower layer, lets see here:
>>
>> BeginGetRequestStream succeeds
>> EndGetRequestStream succeeds
>> BeginGetResponse hangs (indefinitely?)
>>
>> So it looks like the socket opens, the client sends a request, but the
>> server never sends a response. I only go to three or four sims in SL
>> usually, but out of those I've most often noticed it in Ahern on a busy
>> night. The sim turns in to what I call a "blackhole sim" because you can
>> teleport in just fine, but since CAPS is unavailable in that sim there
>> is no way to walk or TP out, and you get trapped in the sim.
>>     
>
> Hi, sorry for the long delay on this, but we think we've identified one
> issue that might cause this, and deployed the fix to the beta grid.
>
> The problem that we identified was that the caps server was not
> responding when an invalid cap was invoked, it should have been
> returning a 404.  Now it returns a 404 instead of keeping the connection
> open forever.
>
> Would you let me know if you still see the issues with connecting to the
> caps server on the beta grid?
>
> -RYaN
>
>   

I've been monitoring this and haven't seen any more 404s (only get them 
in a semi-rare timing issue where the sim disconnects and the CAPS 
thread tries to reconnect before being told to shut down), but the 
initial CAPS connections appear to be much more reliable then 2-3 months 
ago.

John