[tor-bugs] #16923 [Torflow]: Torflow Measured=N value does not update for overloaded relays returning HTTP 503 (was: Bandwidth / consensus weight adjustment slow to recognize rate-limit reduction)

[Tor Bug Tracker & Wiki]

#16923: Torflow Measured=N value does not update for overloaded relays returning
HTTP 503
------------------------------+----------------------
     Reporter:  someone_else  |      Owner:
         Type:  defect        |     Status:  assigned
     Priority:  major         |  Milestone:
    Component:  Torflow       |    Version:
   Resolution:                |   Keywords:
Actual Points:                |  Parent ID:
       Points:                |
------------------------------+----------------------
Changes (by hdqdak8v32aor):

 * priority:  normal => major


Comment:

 IMMEDIATE ATTENTION REQUIRED!

 Based on this plus an earlier bandwidth reduction event, have identified
 the probable cause of this issue.

 When a large bandwidth limit reduction is applied to a busy relay, the
 relay becomes saturated. Saturated relays cease processing DIR port
 requests and instead respond to all queries with

 {{{
 HTTP/1.0 503 Directory busy, try again later
 }}}

 It appears that Torflow passes over relays responding this way and leaves
 the measurement unchanged for as long as such relays reliably answer with
 the 503 error.

 The relay currently exhibiting this issue is ranked at about number six in
 the consensus per Vidalia and number three on Blutmagie.  Is rated
 Bandwidth=157000 and is set with the limit BandwidthRate=15000k (where
 k=1000).

 Torflow maintainer, please post to this ticket whether it is desired for
 the relay to remain in this state for diagnosis or if it should be taken
 offline briefly to exit the deadlock.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/16923#comment:6>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online