17:01:10 <ahf> #startmeeting network team meeting, 22nd november 2021
17:01:10 <MeetBot> Meeting started Mon Nov 22 17:01:10 2021 UTC.  The chair is ahf. Information about MeetBot at http://wiki.debian.org/MeetBot.
17:01:10 <MeetBot> Useful Commands: #action #agreed #help #info #idea #link #topic.
17:01:21 * ahf just got out of another meeting
17:01:30 <ahf> our pad is at https://pad.riseup.net/p/tor-netteam-2021.1-keep
17:02:01 <ahf> how are people's boards looking? https://gitlab.torproject.org/groups/tpo/core/-/boards
17:02:02 <mikeperry> o/
17:02:38 <nickm> decent over here
17:02:38 <eta> o/
17:02:52 <ahf> i'll sync with cohosh later today about the s30 (and friends) tickets we have and figure out what needs to be prioritized there
17:03:31 * eta moves some things aroung
17:03:33 <eta> d*
17:03:50 <ahf> eta: you have gotten your head around how all the board stuff works in the team?
17:04:10 <eta> think so (?) — I mean I was just moving things out of doing that were actually done, etc.
17:04:19 <ahf> ya, that is pretty much it XD
17:04:26 <ahf> awesome
17:04:49 <ahf> dgoulet: anything we need to talk about regarding releases here?
17:05:10 <dgoulet> nope
17:05:23 <ahf> i am gonna try to get those warnings done this week that weasel reported
17:05:32 <ahf> some of them are funny, some of them are weird
17:05:37 <ahf> ok
17:06:01 <nickm> eta: for arti#238, if you're not on that this week, mind if I take a look?
17:06:11 <ahf> dgoulet: you have everything under control with NHT from what i can tell, i have not everything under control with "for anti-censorship team" label right now, but we sync about that later today to figure out what should be next on the plate there
17:06:26 <ahf> we have no reminders, no announcements
17:06:45 <ahf> thanks all for updating the pad with stuff! it looks good already
17:06:54 <ahf> mikeperry: i think you can takeover now for s61
17:07:26 <mikeperry> Ok last week a lot happened in the congestion control piece!
17:07:27 <eta> nickm, sure, go ahead
17:07:44 <mikeperry> Last week, I finished running the Vegas queue parameter sims in Shadow. Queue parameter values that match the outbuf queue plus 1 TLS record of cells in the circuit queues, per hop, are performing the best.
17:07:50 <mikeperry> I documented this in the experiments doc, and also reorganized it a bit to separate out the outbuf tuning.
17:08:00 <mikeperry> dgoulet: could read over the new outbuf section 3 for sanity and let me know of any thoughts you have wrt extra things we might want to log?
17:08:10 <mikeperry> https://gitlab.torproject.org/mikeperry/tor/-/blob/cc_shadow_experiments_v2/SHADOW_EXPERIMENTS.txt#L302
17:08:22 <dgoulet> sure
17:08:38 <mikeperry> great, thanks
17:08:52 <mikeperry> jnewsome and hiro have been working hard on splitting the simulation and live perf measurement sources by country. They are running some final plotting tests on this in toy sims.
17:09:01 <mikeperry> (This will be very helpful for further tuning and comparison to baseline, as it helps us separate the latency effects of location from the latency effects of changes we make. Yay!)
17:09:12 <ahf> nice
17:09:34 <mikeperry> I also updated the simulation doc with the next tuning stage, which is "slow start". I have a new list of named paramter sets, along with their git commit, so we can easily use gitlab to go through those:
17:09:41 <mikeperry> https://gitlab.torproject.org/mikeperry/tor/-/blob/cc_shadow_experiments_v2/SHADOW_EXPERIMENTS.txt#L261
17:09:48 <mikeperry> (I will be off starting Weds through the weekend, but I am hoping to keep the sim box warm with those during that time!)
17:10:14 <ahf> ohhh, is this the holiday week in the US?
17:10:23 <mikeperry> yeah
17:10:24 <nickm> yup. Thu and Fri are company holidays
17:10:37 <ahf> ahhhh, ok, that is very good to know. more CI time for the europeans!
17:10:43 <GeKo> heh
17:10:54 <nickm> remember to take other balancing holidays!
17:11:03 <ahf> ok, that is good to know. i think that menas we also wont be doing our thursday sync this week
17:11:49 <mikeperry> I also finished up the last implementation pieces of Exit-side negotiation, and rebased dgoulet's work on onion service negotiation on top of that: https://gitlab.torproject.org/mikeperry/tor/-/commits/cc_negotiation_v4+svc
17:11:55 * eta will probably just take the US holidays off to keep things simpler
17:12:06 <mikeperry> The Exit-side implementation is now in a new merge request (without dgoulet's work): https://gitlab.torproject.org/tpo/core/tor/-/merge_requests/495
17:12:17 <mikeperry> It needs unit tests, but is otherwise ready for review.
17:12:26 <mikeperry> For the unit tests, it seems fairly straightforward to add tests for the CPU worker queues, as well as onion_skin_server_handshake and onion_skin_client_handshake to cover negotiation. I will add these soon.
17:12:49 <ahf> sweet
17:13:01 <mikeperry> However, it seems like there's no existing test coverage of protover, extendinfo -> handshake choice, circuit completion, or the actualy cpuworker execution itself...
17:13:08 * nickm volunteers as Second Reviewer, or Third, depending on whatever you need
17:13:10 <mikeperry> FWIW, I exercised it in Chutney, and we will be exercising it in Shadow once we get to the mixed client sims, in Round 6:
17:13:13 <mikeperry> https://gitlab.torproject.org/mikeperry/tor/-/blob/cc_shadow_experiments_v2/SHADOW_EXPERIMENTS.txt#L430
17:13:22 <mikeperry> Those sims will give us a lot of exercise of the code, and I also added a hidden torrc option to let us exercise it on an exit with our own clients regardless of consensus param.
17:13:41 <mikeperry> nickm: yeah. I'm wondering if you could take a look at that MR and see if there's any additional tests you could easily add for those currently uncovered areas in tor, if it seems worth it?
17:14:24 <nickm> sure; add a link to the coverage report and a list of the stuff I should look at?
17:15:24 <mikeperry> I have a list of notes of stuff that I can translate into something more legible. does CI generate the coverage report, or do I need to do sth extra?
17:15:31 <ahf> it does not
17:15:35 <ahf> sadly
17:16:09 <mikeperry> how do we get one? do I have to make one and upload it somewhere?
17:16:30 <ahf> there is in doc/ somewhere a file describing how you get the gcov files out
17:16:34 <nickm> ah, braino.  I can make my own coverage report if you need
17:16:43 <nickm> not thinking my best today :/
17:17:22 <ahf> hm, we could store main branch gcov reports in a gitlab page..
17:17:40 <ahf> next to the doxygen/rustdoc stuff
17:17:52 <mikeperry> as far as reviewers, I think it def makes sense for dgoulet and nickm to reveiw this, in case triagebot disagrees. dgoulet needs to use some of it for onion svc negotiation anyway (that's why I did the rebase for him)
17:18:11 <nickm> yes
17:19:02 <ahf> you can override triage-bot by just setting a reviewer manually
17:19:10 <ahf> it only handles tickets with no reviewer on
17:19:18 <mikeperry> the onion service code does also test protover, but only for its descriptor version
17:19:23 <mikeperry> I ran into a test failure there
17:19:36 <mikeperry> so perhaps some of that can be repurposed or generalized?
17:22:01 <mikeperry> moving on, at some point we should integrate the overload checking into shadow. does the latest tor alpha have the version of that that lets us tune the DNS rate?
17:23:20 <ahf> no, it does not
17:23:23 <mikeperry> are we satisfied with that generally, geko/dgoulet?
17:23:49 <GeKo> with what?
17:23:56 <mikeperry> the overload lines
17:24:00 <dgoulet> the new system is working as in as expected but the 1% right now is too low as far as we can tell
17:24:09 <dgoulet> and we are investigating this DNS problem with our own Exit now
17:24:36 <GeKo> https://metrics.torproject.org/rs.html#details/50485E03CA39D393BD54D315CEBA65E6DD0FDDB9 it is
17:24:51 <GeKo> and teh overload we see is solely due to port 8074 open :)
17:25:19 <mikeperry> what us 8074?
17:25:55 <GeKo> gadu-gadu
17:26:01 <GeKo> (seriously)
17:26:06 <mikeperry> or rather, why does that nmatter? it is high bandwidth?
17:26:06 <nickm>17:26:16 <ahf> o.O
17:26:31 <ahf> wth is gadu-gadu doing here? :o
17:26:33 <GeKo> it's because almost all those dns requests over that port time out
17:26:40 <GeKo> i have no clue
17:26:51 <GeKo> it's weird
17:27:03 <ahf> sounds like there is a dysfunctional botnet out there or a very fun day in ghidra with a gadu-gadu client there
17:27:05 <nickm> GeKo: you mean, almost all DNS requests for connections to that port, right?
17:27:12 <GeKo> yes
17:27:19 <nickm> whew!
17:27:21 <GeKo> sorry
17:27:33 * nickm was having horrible visions of DNS-over-gadu-gadu
17:27:41 <GeKo> :)
17:27:45 <GeKo> anyway
17:28:03 <GeKo> we need to dig into that and figure out a solution that is suable at the end
17:28:06 <GeKo> *usable
17:28:24 <GeKo> i think right now some exit operators turns their exits into non-exits
17:28:25 <ahf> fun project lol, i was not expecting to see gadu gadu being mentioned in a tor meeting
17:28:33 <GeKo> because of the stupid overload
17:28:37 <GeKo> which is not optimal
17:28:48 <ahf> can we tell them to disable gadu gadu for now? or is that not enough?
17:28:56 <GeKo> not enough
17:29:02 <GeKo> there are more pieces to the puzzle
17:29:08 <ahf> ok
17:30:41 <mikeperry> juga: I still have this sbws graph open to look at.. the CDF one.. sorry for that delay
17:30:55 <mikeperry> how is sbws world going?
17:31:24 <GeKo> well, i added an item on the pad
17:31:40 <juga> mikeperry: good, not much going on last week
17:31:43 <GeKo> i think we need to return debugging the issue we are seeing on gabelmoo
17:31:50 <GeKo> juga: oh, okay
17:31:56 <GeKo> you are here
17:31:58 <juga> i still have to update the graph anyway, still waiting for data to be processed
17:32:10 <GeKo> didn't mean to jump ahead and give the summary ;)
17:32:31 <juga> GeKo: re. gabelmoo, i think we're fine, i'm going to update the issue
17:32:35 <juga> GeKo: np
17:32:41 <juga> that's all from my side
17:32:59 <GeKo> great
17:33:36 <ahf> anything else for today?
17:34:06 <mikeperry> that's all from me, unless geko has more
17:34:18 <GeKo> nope, i am good, thanks
17:35:00 <ahf> let's call it then. hope folks enjoy the holiday here
17:35:10 <ahf> monday next week is off for US folks too then, right?
17:35:25 <ahf> or is that a normal workday
17:35:41 <gaba> holidays is only thursday and friday
17:35:43 <gaba> o/
17:35:46 <ahf> ah
17:35:46 <ahf> ok
17:35:52 <nickm> yup
17:35:52 <ahf> perfect, see you all for netteam meeting next monday then
17:35:55 <ahf> o/
17:35:57 <nickm> see y'all then!
17:35:58 <ahf> #endmeeting