[tor-bugs] #25705 [Core Tor/Tor]: Refactor circuit_build_failed to separate build vs path failures

[Tor Bug Tracker & Wiki]

#25705: Refactor circuit_build_failed to separate build vs path failures
--------------------------+------------------------------
 Reporter:  mikeperry     |          Owner:  (none)
     Type:  defect        |         Status:  needs_review
 Priority:  Medium        |      Milestone:
Component:  Core Tor/Tor  |        Version:
 Severity:  Normal        |     Resolution:
 Keywords:                |  Actual Points:
Parent ID:  #25546        |         Points:
 Reviewer:                |        Sponsor:  SponsorV-can
--------------------------+------------------------------

Comment (by asn):

 Posting a discussion with armadev from yesterday which might be relevant
 to this ticket. I dont have enough time to dig into this right now, so
 I'll just post the raw logs:
 {{{
 16:01 <+armadev> yeah, i am expecting that when i read 25705, i will want
 to remind people that right now we *do* want to call some of that code for
 circuits where we
                  failed to pick the first hop,
 16:01 <+armadev> because that is how we rate limit the amount of thrashing
 that tor does when it goes offline and stays there for a while
 16:02 <+asn> hmm
 16:02 <+asn> what you mean?
 16:02 <+asn> i have not considered that
 16:02 <+armadev> that is, if you're offline and you've marked all your
 tors down, you wake up every so often, and try some circuits, and they all
 fail because you don't
                  think anything is up, and you sleep again for a bit
 16:02 <+armadev> it is possible that prop#271 changed that behavior
 16:02 -zwiebelbot:#tor-dev- Prop#271: Another algorithm for guard
 selection [CLOSED]
 16:03 <+asn> we sleep for a bit?
 16:03 <+asn> do we?
 16:03 <+asn> i dont think we do? either with prop271 or before
 16:03 <+asn> do we?
 16:04 <+armadev> check out did_circs_fail_last_period
 16:04 <+armadev> if we have MAX_CIRCUIT_FAILURES failures in this period,
 16:04 <+armadev> and also we had MAX_CIRCUIT_FAILURES in the last period
 16:04 <+armadev> then we abort all circuits
 16:05 <+armadev> that's part of why it is important to not have too many
 failures that aren't really failures
 16:06 <+armadev> because if you have too many, you'll trigger the "woah
 stop for a bit" feature
 16:06 <+asn> hmmm
 16:06 <+armadev> which is good if they're really failures
 16:06 <+asn> i was not aware of this
 16:06 <+asn> feature
 16:07 <+armadev> now you are! :)
 16:07 <+armadev> it is mostly for tors that are actually offline right now
 16:07 <+armadev> so they thrash less
 16:07 <+armadev> and in that respect it's probably related to the "use
 less cpu when not in active use" tickets
 }}}

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/25705#comment:4>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online