[tor-bugs] #34031 [Metrics/Onionperf]: Figure out warning about unknown error type when exporting .tpf file

[Tor Bug Tracker & Wiki]

#34031: Figure out warning about unknown error type when exporting .tpf file
---------------------------------------+--------------------------------
 Reporter:  karsten                    |          Owner:  acute
     Type:  defect                     |         Status:  needs_review
 Priority:  Medium                     |      Milestone:
Component:  Metrics/Onionperf          |        Version:
 Severity:  Normal                     |     Resolution:
 Keywords:  metrics-team-roadmap-2020  |  Actual Points:  0.2
Parent ID:                             |         Points:  2
 Reviewer:  karsten                    |        Sponsor:  Sponsor59-must
---------------------------------------+--------------------------------

Comment (by karsten):

 Thanks for the very detailed analysis!

 Regarding the three unknown failure types, I think we should rather stop
 generating .tpf files (#34141) than add these three cases to the existing
 dictionary. However, that doesn't solve the problem entirely:
  - We have similar code in metrics-lib that converts OnionPerf's analysis
 files to Torperf results. Right now, that code contains a similar
 map/dictionary with known failure types, and if it doesn't find a key, it
 simply includes the combined error code, as in: `PROXY_END_INTERNAL`. I
 was somehow under the impression that OnionPerf would do the same, but I
 was wrong.
  - Maybe we should avoid keeping a map/dictionary with known failure types
 in metrics-lib and simply put together the error code in the format we
 want, which would be `TOR/END/INTERNAL` in this case. I'll add a new
 ticket for that. Should be an easy change.

 Regarding the additional cases, I strongly believe that these are the
 result of OnionPerf's approach to match TGen transfers to streams and
 circuits, which is somewhat broken. The approach there is to match them by
 source port, but that fails as soon as the same source port is being used
 more than once per day. In metrics-lib I'm doing a somewhat more complex
 approach that also looks at start/end times of transfers and streams. That
 avoids these cases, so I'd say that there's nothing else to do here,
 assuming that we do #34141.

 If you agree, I'll leave a comment on #34141 that we need to fix this
 ticket, if we don't do that one. And I'll open a new ticket to tweak
 metrics-lib's code for including yet unknown error codes in Torperf
 results. And then I'll close this ticket. Makes sense?

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/34031#comment:14>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online