[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

Re: [tor-bugs] #1772 [Tor - Tor client]: Counting circuit timeouts when we can't establish any TLS connection

#1772: Counting circuit timeouts when we can't establish any TLS connection
------------------------------+---------------------------------------------
 Reporter:  arma              |       Owner:  mikeperry         
     Type:  defect            |      Status:  assigned          
 Priority:  normal            |   Milestone:  Tor: 0.2.2.x-final
Component:  Tor - Tor client  |     Version:                    
 Keywords:                    |      Parent:                    
------------------------------+---------------------------------------------

Comment(by mikeperry):

 arma -

 A). You're right. I think I am going to revert my fix for #1741 to
 explicitly check for 3 hop paths and only count those.

 B). The logic to determine if the network is dead is present in
 circuit_build_times_network_close() and
 circuit_build_times_network_is_live(). cbt_network_is_live() is called
 every time we get a cell, and resets the internal counter
 cbt->liveness.nonlive_timeouts to 0. Every time you get to the 95th
 percentile on the CBT curve, circuit_build_times_count_close() is called.
 It calls the cbt_network_close(), which increments
 cbt->liveness.nonlive_timeouts. circuit_build_times_network_check_live()
 is the call that checks this counter, and if it gets above 3, we stop
 counting timeouts. If it gets above 6, we throw away the 3 previous
 timeouts (which were recorded, but presumably happened when the network
 was believed to be down).

 If there is a problem in how we decide to stop counting circuits when your
 network goes away, it's in that pile of madness. Please have a look at it,
 and/or grep your logs for "Network is no longer live" and "Network is
 flaky" and other associated logs from those functions.

 C). Yes, that's the idea. The purpose should first be non-13 in the first
 call to expire. It gets switched to 13, and then is allowed to build until
 the timeout value for the 95th percentile on the CBT curve, at which point
 circuit_expire_building decides to really kill it.

-- 
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/1772#comment:4>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online
_______________________________________________
tor-bugs mailing list
tor-bugs@xxxxxxxxxxxxxxxxxxxx
https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-bugs