16:58:40 <ahf> #startmeeting Network team meeting, 31st october 2022
16:58:40 <MeetBot> Meeting started Mon Oct 31 16:58:40 2022 UTC.  The chair is ahf. Information about MeetBot at http://wiki.debian.org/MeetBot.
16:58:40 <MeetBot> Useful Commands: #action #agreed #help #info #idea #link #topic.
16:58:42 <ahf> hello hello
16:58:47 <Diziet> o/
16:58:56 <ahf> o/
16:58:59 <juga> o/
16:59:05 <dgoulet> o/
16:59:05 <ahf> pad is at https://pad.riseup.net/p/tor-netteam-2022.1-keep
16:59:32 <nickm> hihi
16:59:44 <jnewsome> o/
17:00:09 <ahf> ok, let's go
17:00:21 <ahf> how are folks doing with their boards: https://gitlab.torproject.org/groups/tpo/core/-/boards
17:00:29 <ahf> things have been cleaned up quite a bit here in the last two weeks
17:00:39 <nickm> my board is semi-relevant till 1.1 is out, but I think i'm doing ok...
17:00:51 <Diziet> What nickm said
17:00:57 <ahf> ya, let's not think much further than 1.1 for arti folks :D
17:01:16 <ahf> i don't see anything off either
17:01:35 <ahf> dgoulet: anything on the tor release front?
17:02:53 <dgoulet> I'm hoping this week! Trying to finalize loose ends
17:02:58 <dgoulet> but should be "soon"
17:03:01 <dgoulet> (tm)
17:03:10 <ahf> cool
17:03:18 <nickm> if there's anything i can do to help, just ask
17:03:27 <dgoulet> sure thanks
17:03:38 <nickm> I'm busy with arti 1.1, but there are frequently times when a distraction would help my brain
17:03:46 <ahf> we don't have anything incoming from other teams other than the faravahar ticket, but dgoulet is on that
17:04:27 <ahf> eta, Diziet, nickm: are y'all up for an irc sync on where we are with 1.1.0 tomorrow at 14 UTC?
17:04:58 <eta> sure
17:05:21 <Diziet> SGTM
17:05:25 <nickm> +1
17:05:55 <ahf> ok, perfect
17:06:11 <ahf> we have no announcement or discussion topics
17:06:21 <ahf> i think non-s61 people can go do something else now and then mikeperry can take over?
17:06:45 <mikeperry> ok
17:07:33 <mikeperry> so I am working on sims to test david's latest fixes and changes, and test the ability to lower the outbound queue limit at relays
17:07:56 <mikeperry> that involves tuning onion service congestion control params, and doing exit upload sims
17:08:27 <mikeperry> the end result should be less memory pressure at guards, and possibly some ancillary mitigation of exit connect floods
17:09:26 <mikeperry> geko also looked at the sbws ratios for Artikel10. it looks like their situation was vastly improved by their surgeprotector tool
17:10:24 <ahf> nice
17:10:25 <mikeperry> geko,juga: how is sbws going otherwise? I remain very distracted :/
17:10:46 <juga> mikeperry: going good, deploying tomorrow sbws!150
17:10:53 <juga> to try the uploads
17:11:11 <juga> (in longclaw's sbws)
17:11:39 <mikeperry> juga: is that weird issue with SS flag toggling between 0 and 1 still happening?
17:11:43 <juga> nope
17:11:50 <juga> need to reply to that still
17:12:15 <juga> but after debugging more, it isn't happeing, i was just confused
17:12:50 <mikeperry> ok. and did you get a chance to see if the connection failures/transfer termination were correlated with CIRC CLOSED/FAILED REASON=DESTROY control port lines?
17:13:06 <juga> i only got REASON=REQUESTED
17:13:10 <juga> so i guess not
17:13:32 <juga> i wanted to do more tests on that
17:13:45 <mikeperry> hrmm ok lmk
17:14:42 <juga> ok
17:14:44 <ahf> what was the thought here if the reason was DESTROY?
17:15:24 <mikeperry> DESTROY events could be caused by our queue limits closing sbws circuits
17:15:33 <ahf> ah
17:15:43 <mikeperry> but reliability could also be bad because of the exit dos. it is just unclear at this point
17:16:36 <ahf> yeah
17:16:43 <mikeperry> dgoulet: do you have those additional metricsport metrics for congestion control written down? did you want to go over those again this week, or you just gonna write up a patch?
17:16:57 <ahf> does sbws have something like metricsport that could be put into the shiny new castle that dgoulet is building?
17:17:01 <dgoulet> mikeperry: I'm just going to go for the patch. And yes, I basically copied the IRC log :)
17:17:22 <mikeperry> ok. I am happy to review that
17:17:25 <dgoulet> +1
17:17:57 <juga> ahf: nope, but an issue can be opened
17:18:03 <mikeperry> this might help with sbws, too. I can think about adding stuff for that if it seems obvious to me during review to add easy additional things there
17:18:27 <dgoulet> keep in mind that MetricsPort right now, except for HS, is solely relay side
17:18:42 <dgoulet> and I kind of want to keep it that way because our glorious arti :)
17:18:42 <ahf> juga: i can create one for sure, but it is not something you have to jump on just because i create it
17:18:44 <mikeperry> oh, so normal client things can't go into it?
17:18:45 <trinity-1686a> (I don't think meetbot was started, is that wanted?)
17:19:02 <dgoulet> mikeperry: yeah... no, client side stuff, not really
17:19:09 <ahf> i think it was, trinity-1686a ? a bit over 17 UTC
17:19:13 <dgoulet> I mean things work for client also sometimes but mostly it is for relay monitoring
17:21:53 <mikeperry> ok
17:22:18 <mikeperry> hiro: got your pm about torctl dropped events. did you want to talk about that now, or just bbb or sth later?
17:23:20 <trinity-1686a> ahf: my bad the matrix-irc bridge was restarted just at that time, so it probably missed a few messages
17:23:26 <ahf> ah :D
17:23:50 <mikeperry> otherwise, next monthly s61 sync is 18utc nov 7th. I will have the pad updated by Friday
17:24:05 <jnewsome> mikeperry: i take it the uploads in the shadow sims are working for you now?
17:25:03 <mikeperry> jnewsome: they seem to be. I might try to grep out the DESTROY events from client logs this week. I am gonna try some more sims first tho
17:25:32 <jnewsome> mikeperry: ok cool :)
17:25:42 <mikeperry> jnewsome: did you make other shadow changes that got folded in? I am seeing slightly lower queue overload in these sims generally
17:25:57 <jnewsome> mikeperry: hmmm, no
17:26:15 <mikeperry> it could be because doing bidirectional activity means that we're more symmetric and less likely to pile all traffic in the inbound queues
17:26:42 <jnewsome> well, when you enable 5 MB uploads, it'll do fewer of the other downloads; maybe that's it?
17:26:53 <mikeperry> yeah possibly
17:27:25 <jnewsome> we probably ought to bump the shadow version sometime, but yeah I've been avoiding it since I don't want to introduce changes that'd make results not comparable
17:27:41 <mikeperry> ok
17:27:51 <mikeperry> looks like this irc box is about to reboot. so might as well wrap it up
17:28:38 <hiro> mikeperry: I have opened this ticket https://gitlab.torproject.org/tpo/network-health/metrics/onionperf/-/issues/40055
17:29:13 <mikeperry> hiro: ok, will take a look. about to get booted off irc tho
17:29:14 <hiro> I have a bug when processing multiple days data for onionperf a clients with tor 0.4.7
17:29:28 <hiro> and I wonder if we are parsing the torctl logs correcly
17:30:33 <hiro> ok is that a netsplit or something?
17:30:34 <ahf> ugh.. mike is gone it seems
17:30:35 <ahf> lol
17:30:57 <ahf> ok, maybe we do the remaining sync when the network have returned to it senses
17:31:10 <ahf> i'll mark the meeting as done and then we can talk in #tor-dev when folks return
17:31:12 <ahf> thanks all <3
17:31:14 <ahf> #endmeeting