[tor-bugs] #21394 [Core Tor/Tor]: connection timeouts are affecting Tor Browser usability

[Tor Bug Tracker & Wiki]

#21394: connection timeouts are affecting Tor Browser usability
-------------------------------------------------+-------------------------
 Reporter:  arthuredelstein                      |          Owner:  (none)
     Type:  defect                               |         Status:
                                                 |  needs_review
 Priority:  Very High                            |      Milestone:  Tor:
                                                 |  0.3.1.x-final
Component:  Core Tor/Tor                         |        Version:
 Severity:  Normal                               |     Resolution:
 Keywords:  029-backport 030-backport            |  Actual Points:
  031-backport tbb-performance, tbb-usability,   |
  performance, tbb-needs                         |
Parent ID:                                       |         Points:
 Reviewer:                                       |        Sponsor:
-------------------------------------------------+-------------------------

Comment (by Dhalgren):

 Replying to [comment:81 Sebastian]:
 > I am not sure that retrying in the case of named is actually beneficial,
 which is why I didn't include it in the patch. Even if we succeed on the
 later retry it's unlikely that the client will actually benefit from that,
 because it will also give up on the request.

 I'm not following the the line of thought.  If evdns.c (in libevent) re-
 attempts the request at five seconds and succeeds, the client knows
 nothing about it.  All the client knows is that it required six-seven-
 whatever seconds for the circuit connection to complete.  In the retried-
 circuit scenario, it could take twelve, thirteen, on the outside fourteen
 seconds for success.

 Admittedly this will rarely matter (my Unbound histograms say about 0.5%
 of requests), but if attempts=1 timeout=5 is set, then the exit will
 literally throw away responses arriving after five seconds and the client
 will twiddle it's thumbs for the remaining five or ten seconds of its
 timeout interval with no chance of success.  My understanding is that
 evdns all-retries-exhausted events are _not_ relayed back to the client.

 > I had no intention to slight you in any way, I felt that I pushed very
 hard to actually get your work here recognized and a proper fix merged
 after your contributions had unfortunately been ignored for a very long
 time. I am just a volunteer myself and spending this time because I care
 about having a good outcome for the network.

 Very sorry for any misunderstanding!  I was not referring to you or anyone
 on this ticket above.  I was expressing my trepidation at spending time
 contributing in general because several times in the past I submitted
 carefully thought-out code changes (to Tor and other projects) that were
 summarily hacked on for no obvious reason, and without involving me.  Took
 much of the enjoyment out of it.

 In regard to this issue I am rather liking the experience.  I spent about
 six weeks understanding the problem, working on the fix, following up the
 results and documenting what I found on the wiki, and I it's a pleasure to
 have the work recognized and for it to effect a critical improvement to
 the network--though my original motive was simply to tame a single berserk
 relay.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/21394#comment:82>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online