[tahoe-dev] Issue with Unplugging One Node from private grid

[Kevin Reid]

On Nov 29, 2010, at 15:57, Brian Warner wrote:

> he first problem (which isn't the most important one) is how to detect
> non-explicit connection loss The best you can do here is a  
> heuristic, by
> sending a keepalive packet every once in a while, and abandoning any
> connection that has been quiet for too long.

On Mac OS X, at least, you can monitor whether there is an active  
network connection using the SCReachability API. I don't know whether  
other platforms have anything similar.

http://developer.apple.com/library/mac/#documentation/Networking/ 
Conceptual/SystemConfigFrameworks/SC_ReachConnect/SC_ReachConnect.html

It's stupid to drop connections preemptively, of course, but you can  
do better (at least heuristically) than waiting for a TCP timeout on  
an existing connection. For example, if the network status goes from  
'down' to 'up', do one or both of these things:

   - send an explicit keepalive packet and assume the connection is dead
     if there is not a prompt response

   - open a second connection while keeping the first and drop the first
     if the second succeeds

The latter can be thought of as selectively discarding TCP's ordering
guarantees: you want new messages to go through even if they arrive
before old messages do (or can be assumed to have failed to arrive).

(Disclaimer: I have not actually implemented any of the above anywhere.)

-- 
Kevin Reid                                  <http://switchb.org/kpreid/>