[tor-bugs] #29231 [Core Tor/Tor]: Relays vastly underreport write-total in padding-counts line in extrainfo descriptor

[Tor Bug Tracker & Wiki]

#29231: Relays vastly underreport write-total in padding-counts line in extrainfo
descriptor
---------------------------------------+-----------------------------------
 Reporter:  arma                       |          Owner:  (none)
     Type:  defect                     |         Status:  merge_ready
 Priority:  Very High                  |      Milestone:  Tor:
                                       |  0.4.1.x-final
Component:  Core Tor/Tor               |        Version:
 Severity:  Normal                     |     Resolution:
 Keywords:  tor-relay padding wtf-pad  |  Actual Points:
Parent ID:  #28634                     |         Points:
 Reviewer:  dgoulet                    |        Sponsor:
---------------------------------------+-----------------------------------

Comment (by mikeperry):

 Replying to [comment:8 dgoulet]:
 > lgtm;
 >
 > Just to clarify Mike, in `write_packed_cell()`, we count the cell only
 if the cell was successfully put in the connection buffer. Any error there
 will not count the cell.
 >
 > That it OK? As I recall some stats we want to count every cell, not just
 the ones that succeeded.

 Yes, this is what we want here -- we want to know the actual written cell
 counts and actual written padding counts, so we can measure what the
 overhead is on the wire and make sure the actual overhead is reasonable
 for the current network conditions.

 The patch also relocates the circuitpadding stat accounting to behave
 similarly -- if the #29204 fix causes us to not pad because the queue is
 full, we should not count those cells in our statistics.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/29231#comment:9>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online