[tor-bugs] #21969 [Core Tor/Tor]: We're missing descriptors for some of our primary entry guards

[Tor Bug Tracker & Wiki]

#21969: We're missing descriptors for some of our primary entry guards
-------------------------------------------------+-------------------------
 Reporter:  asn                                  |          Owner:  asn
     Type:  defect                               |         Status:  new
 Priority:  Medium                               |      Milestone:  Tor:
                                                 |  unspecified
Component:  Core Tor/Tor                         |        Version:  Tor:
                                                 |  0.3.0.6
 Severity:  Normal                               |     Resolution:
 Keywords:  needs-proposal, tor-guard, tor-      |  Actual Points:
  bridge, tor-client, tbb-needs,                 |
  034-triage-20180328, 034-removed-20180328      |
Parent ID:  #30817                               |         Points:  1.5
 Reviewer:                                       |        Sponsor:
-------------------------------------------------+-------------------------

Comment (by s7r):

 Replying to [comment:99 teor]:
 > Thanks for the bug reports, but they aren't really helping us make
 progress on this issue.
 >
 > We know that tor needs some design changes to resolve this issue:
 >
 > Replying to [comment:93 teor]:
 > > #16844 and #21969 have conflicting goals, so we need to write a
 proposal that balances these goals. See #30817.
 >
 > In particular, tor needs to try harder to download descriptors if it has
 a good link, but backoff if it has a bad link, so that at least some
 descriptors can download. We don't know which case applies to your tor
 instance, because you only sent us 1 second worth of logs.
 >

 I see - however regardless the case if Tor states it is missing just 1/3
 descriptors of our primary entry guards, it should still be able to build
 circuits with the available 2/3 descriptors that apparently we are not
 missing. That was the only thing that raised my attention.

 I sent 1 second worth of logs because those are the only lines that were
 interesting and needed attention. Just before them were the regular
 heartbeat messages (I have sent / received / our onion service received
 etc. etc.) and that was everything in the log file as it was recently
 truncated by logrotate (and in the previous log file before rotation there
 was nothing besides the heartbeat messages as well). So, even if I would
 have sent complete log still only this 1 second worth of log would make
 reference to the issue.

 > If you'd like, we can make a ticket that delays those annoying logs
 until tor has been missing descriptors for an hour or so.
 > That is a simple change, it will stop annoying relay operators, and it
 might get us better diagnostics.

 1 hour seams too much. Might get a new consensus by that time. Maybe like
 20 minutes if we really make this change? But I don't see a huge benefit,
 the logs are not annoying as they are, and we can see this way how often
 they occur in the wild and how many users are reporting them. What raises
 attention are not the logs itself but only the part where it complains
 about missing JUST 1/3 descriptors and stating it is not able to build
 circuits (why? still has 2/3 apparently). If this is expected, we should
 document and rephrase. Also a more interesting thing is that this instance
 was running with `NumPrimaryGuards = 6` yet it was complaining about
 missing descriptors for 1/3 of primary guards but I opened this as a
 separate issue in #31644 because it is slightly different and this ticket
 is already very large.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/21969#comment:100>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online