[tor-project] Network team meeting notes, 27 Nov 2017
nickm at freehaven.net
Mon Nov 27 17:44:54 UTC 2017
You can see a transcript of our latest meeting at
Below is a copy of our pad.
Network team meeting pad, 27 November 2017
Welcome to our meeting! Every Monday at 1700 UTC on #tor-meeting on
OFTC. (This channel is logged while meetings are in progress.)
Want to participate? Awesome! Here's what to do:
1. If you have updates, enter them below, under your name.
2. If you see anything you want to talk about in your updates, put
them in boldface!
3. Show up to the IRC meeting and say hi!
Note the meeting location: #tor-meeting on OFTC!
This of course is the way to talk to dragons, if you don't want
to reveal your proper name (which is wise), and don't want to
infuriate them by a flat refusal (which is also very wise). No
dragon can resist the fascination of riddling talk and of wasting
time trying to understand it.
-- J.R.R. Tolkien, _The Hobbit_
Meeting notes from last week:
- Sponsor M is no longer billable: please let me or isabela know
if you need to do more work on M stuff, and we can work something out.
- On the roadmap spreadsheet: Please start taking december/january
tasks. (If somebody else has already taken something you want, please
talk to them and/or add yourself too.)
- Team rotation for December please:
- Please review stuff in review-group-26:
- Remember: many people take time off in December! If you're
going to do something in December that requires another person, please
confirm that they will be around.
- Soon it will be time to triage 0.3.3.x tickets. Please assign
tickets in that milestone that you will do to yourself.
- Aiming for many releases next Tuesday (5 Dec)
- meeting time -- 1800 UTC is yes. Not sure whether to track DST
yet. Going to say "yes", but we can change our mind between now and
the next DST transition. Will flip a coin to decide whose DST.
- Last week: (actually two weeks, sorry it's so long!)
- Revised prop#283, which moves relay IPv6 ORPorts from
microdescs to the microdesc consensus
- Started designing relay IPv6 ORPort reachability checks,
turns out they need IPv6 relay extends
- consensus health got a "ReachableIPv6" pseudo-flag, thanks
- Relay Search (Atlas) got IPv6 ORPort and IPv6 Exit
"additional flags", thanks irl! (#10401)
- Merged code that adds rend point IPv6 addresses to HSv3
client intro cells, thanks neelc! (#23577)
(There's still more work to be done for HSv3 IPv6, see #23493.)
- Patched metrics onion service graphs to show which ones
measure v2 and v3 services (#24048)
- Wrote and rewrote parts of the new "PrivCount in Tor using
Shamir Secret Sharing" proposal
- Fixed a nasty little bug where Tor checks cached bridge
descriptors, rather than confirmed
running bridges (#24392)
- Opened a ticket about pruning our list of supported consensus methods
- Why is a zstd-compressed microdesc consensus larger than a
Am I doing it wrong?
[nick: I tried to answer, but found even more questions! See
- Tried to help out asn with the microdesc fetch bugs
- Kept implementing PrivCount features for the next release
- This week:
- PrivCount Experimental features, bug fixes, and release
- Submitting #23881 (rust logging) for review (hopefully Tuesday)
- sent out reminder to testers for Nov test
- worked on dec 1st deadlines
- met with asn, geko and antonela on HS UX tasks - elected a
couple for December
- focus on my deadlines
- organize hs states on google doc for antonela to add icons
- checking in on moat server <- question for isis
last week (2017-W47):
- investigated Tor Browser error reporting oddities (#24367,
now specifically in #24428; thanks brade!)
- investigated client-in-future clock skew errors
* with an expired consensus that's not reasonably live
(+27.5h) it looks like stuff hangs far earlier than with the expired
yet reasonably live (+3.5h) case
* looks like there are situations where we might hang indefinitely but
don't because enough relays are unreachable that we hit the warning
* it also looks like +27.5h gets a "trusted" clock skew indication
from a dirauth sooner than +3.5h does?
this week (2017-W48):
- bug triage
- start sketching out larger-scale bootstrap progress/error
- roof monsters are eating my roof so i'll be sporadically unavailable
- investigate sporadic oniongit CI pipeline failures (low disk space?)
- starting to think that a heuristic for inferring client
clock skew from non-dirauth relays might start looking like EWMA (with
incremental variance too); maybe we can refactor the circuitmux EWMA
code to help with this?
- Released 0.3.2.5-alpha
- Reviewed open proposals wrt privcount; commented.
- Started implementing some privcount backend stuff in Rust
- Worked on a faster monotime_coarse_absolute() for 32-bit
- Worked on free-macro rewrite (#24337)
- Opened review-group 26.
- Reviewed and merged various fixes
- Fixed bugs in consensus diff corner cases.
- Analyzed hashring storage failure probabilities (#23170) and
workarounds (#24425, #24426)
- Wrote trivial C backend for Rust to call our logging functions
- Analyzed and wrote fixes for the bugs GeKo found with STACK (#24423)
- Try android performance tools (sponsor 8.)
- Inform packagers and users of upcoming releases on/around December 5.
- Backport more tickets to earlier stable series, as appropriate.
- Write/revise TROVE advisories and revise patches (as needed).
- (Also get CVEs)
- Prototype and publish revised privcount/shamir proposal
- Review some stuff in review-group-26
- Publish other pending proposals (if any)
- Clarify and revise our security policy on bug severities (#22962)
- Review December roadmap; convince everybody to take on tickets there.
- Other TBD.
- Please grab me to talk about design or code whenever you want
- Meet with a namecoin person who's in town.
- Last week:
* Worked on TROVE-2017-12. Needs review/merge on the security list.
* Worked on #24313 HS bug. A patch exists.
* Finalized #23709 that is removing in/out channel's queue. That branch is
in need_review for 033 and has bumped the code coverage of channels to
84%. I've been running that code on both my relays with running with
valgrind. So far so good.
* After the above, I've made a "cell tracing" branch that tracks cells from
inbuf to outbuf. Idea is to have a way to measure performance of the cell
fast path at every steps and try to find possible bugs with cells stuck
for too long or try to answer questions like "why 2.78 seconds on
average?" (not a "confirmed number").
* Tried to find the bug on #24346 but we need to add more logging to hunt
* Review couple tickets in review-group-25.
- This week:
* Re-re-revalidate the cell tracing branch, clean it up and run it more
thoroughly to try to identify where and why we might have cell latencies
in order to identity if the scheduler is at fault.
* Continue ticket review/patches for 032 and 033.
* If I receive my Android device this week, I'll maybe try with ahf to
setup the dev. environement to profile HS and scheduler.
- Got #23817 merged. This should help a lot with #21969.
- Attended UX+onion+TorBrowser meeting. tl;dr is that we should focus on
#23247 in the short-term, and we also discussed #21952. We have another
meeting this Wednesday.
- Catching up with last week's backlog. Please let me know if I should pay
attention to something that I'm not mentioning below.
- Continue review of #23100 and #23114.
- Plan development on md/guard issues (#24113, #23863)
- Figure out next most important guard/prop224 stuff that I should be doing.
- Began testing #23101 (pre-building HS-specific circuits for
vanguards/pinned middles). Still needs more testing, unit tests,
- Cleaning up #13837 (pinned middles) and #23101. Let them run
for a while with prop247 simulator,
monitor paths built and timeout rate, write tests, etc.
- Maybe write a simple patch for #24228 (to slow down CBT
circuit building by 3X or so)?
- Landed `simpleperf` notes in tor.git
- Optimization meeting with Nick
- Looked into n8fr8's new split orbot work.
- Do multiple builds for different ABI's for testing to
get 64-bit information.
- Put Tor's bench+tests on device via the Orbot APK, but still no
way of automatically running them without using adb/shell.
- Started looking into the event loop code in Tor for
- Looked into HC's (from Guardian Project) work with
- Get event loop instrumentation into tor.git
- Update sponsor 8 wiki documentation on n8fr8's split orbot changes as
that will probably simplify things for people.
- Get `simpleperf` results for via-onion/via-exit/bootstrap tests for
both 32-bit and 64-bit for comparison.
- Help others (David and Nick) with getting up and running on Android.
- Look into #24368 [compression, zlib, zstd] and maybe see if
we should tune our parameters there
(as per Nick's comment).
- Tor specification meeting with Mozilla.
- Look into David's tracing code in Tor and see if it can be made to
work with the Android NDK tracing features.
- revisited prop#249 prop#269 and prop#270 and began revising
- looked at my beginning implementation of prop#269 from last year
- re-read a couple lattice papers and reviewed a reference
implementation of one of the NIST submissions
- took the rest of the week off
- re-raise prop#249 discussion on the ML and hopefully put
that one into "accepted" state
- re-write prop#269 and prop#270
- track down and fix some bug in the moat server
- work with dcf on getting the meek tunnel working?
More information about the tor-project