IRCaBot 2.1.0
GPLv3 © acetone, 2021-2022
#ls2
/2022/02/07
@eyedeekay
+R4SAS
+RN
+RN_
+T3s|4
+Xeha
+not_bob
+orignal
FreeRider
Irc2PGuest15271
Onn4l7h
Onn4|7h
T3s|4_
aargh3
acetone_
anon4
cancername
eyedeekay_bnc
profetikla
shiver_1
u5657
weko_
x74a6
zzz successfully split() with matching keys on both sides
orignal great
orignal do you publish SSU2 address already?
zzz no, this is mostly test code
zzz standalone test, as I did with NTCP2
zzz data phase success:
zzz Got I2NP block: [UnknownI2NPMessage:
zzz Type: 107
zzz Length: 424]
zzz Got PADDING block, len: 48 in frame len: 487
orignal so I can't try to connect to you yet
zzz oh gosh no
zzz I said 6 weeks about 2 weeks ago :)
orignal you didn't get my idea ))
zzz also, as I said, the header encryption isn't right, needs to be reworked
orignal since you have Bob side already
zzz what idea was that?
orignal I could start with Alice only
zzz no way, I'm not testing your (untested) Alice for you. Do both sides and test it yourself ))
zzz I've fixed 100 problems testing both sides
orignal too lazy to implement both at the time ))
orignal anyway will do porbably this week
zzz yeah but we know from experience, testing with somebody else is slowwwwwwwww and painful.
zzz also, right now, I have a lot of things just stubbed out, doesn't match the spec
zzz the good news is, the data phase is really easy
orignal will check
zzz ok got both data directions working
zzz 0) Hi
zzz woo, full house
zzz what's on the list for today?
orignal R4SAS is missing
orignal introducers for SSU2
zzz ok that's 1)
zzz I'll add 2) SSU bug post mortem
zzz 3) SSU2 coding status
eyedeekay1 Are there any release details to discuss?
zzz 4) release schedule
zzz ok that's a good list for now
zzz 1) introducers for SSU2
orignal specs don't meantion how they should be published
orignal I guess it must include s and i
zzz yeah, the proposal is incomplete about introducers and peer testing, we decided that was "phase 2"
zzz but yes, to do it properly, it would need s and i
orignal let me explain
R4SAS hi, I'm here, but bit busy with android stuff problems
zzz the relay and peer test blocks also are not properly specified yet
orignal I started with refactoring
orignal I want to implement common NTCP2/SSU data structures
orignal so when I parse a SSU2 address I have to understand how I parse introducers
zzz I understand
orignal and data structures for it
zzz the problem is we haven't even done a full security analysis of relay and peer testing
orignal I have SSU1's interiducers for now and ofc it's not right
zzz so I'm not even sure _how_ it's going to work
zzz like, can we do relay and peer test "out of session" or do we need a full session handshake first
orignal same way as SSU1, no?
zzz I don't think so
zzz I don't think it's secure
orignal I think we should do it with "N" Noise
orignal like we do in ratchers
orignal e.g. no handshake just one time encryption
zzz maybe
zzz but with UDP address spoofing, we need a handshake
orignal another question if we can introduce SSU2 routers through SSU1 and vice-versa
zzz the problem is that adds one or more round trips to the process
zzz I don't have any answers for you right now
zzz good question
orignal that's fine
orignal but also SSU2 wothout introducers is useless
orignal because it's not better than NTCP2
zzz remember we decided to stop working on the phase 2 proposal and start coding phase 1...
orignal agree
orignal so for now I assume SSU2 address doesn't include introducers
orignal if it does it's error
zzz yeah, until we figure it out.
orignal fine. will implement tghis way
zzz but I think you're right, it would need both s and i
orignal is and ii ))
zzz yup
zzz because I may do SSU and SSU2 on one address, like I did for NTCP2. I'm not sure yet and probably won't know for another 2 months
orignal ofc I will do SSU2 address only like I did for NTCP2
zzz all I'm coding now is test code. integrating it into the existing code will be much later
zzz good stuff, glad you're thinking about introducers
orignal so is SSU addres contains v it supports SSU2
zzz yeah I think so
zzz anything else on 1) ?
orignal how do you differentiat SSU and SSU2 request then?
zzz I don't know
zzz next month ))
orignal then start use SSU2 addresses only
zzz my test code publishes both
zzz but I can change it
orignal probably no worth
zzz anything else on 1) ?
orignal just separate SSU
orignal yes, one more question
zzz go ahead, don't mean to rush you
orignal I'm Bob I receive incoming SSU2 session
orignal and then Alice's RI
orignal how do I verify her s?
orignal SSU2-capable address and transport?
zzz same as NTCP2. the noise static key must match the s in the RI
orignal when we had NTCP2 it was only on s
orignal now we also have SSU2
orignal do you think it should be the same s and NTCP2 or might be different?
zzz you look through the RI for the SSU2 address, pull out the s, and check it
zzz might be different. probably should be different
orignal and SSU2 or SSU2+transport?
orignal e.g. SSU2 ipv4 and SSU2 ipv6
orignal can they have different s?
zzz right, that's why the keys should be the same for all SSU2 addresses.
zzz same rules as NTCP2
zzz I guess you could look for the v4 or v6 address
orignal for NTCP2 I look by NTCP2 + transports
orignal alrhough I publish the same s
zzz what do you mean "+transports" ?
orignal but assume they might be different per address
orignal ipv4/ipv6/ygg
zzz oh, ok
zzz anything else on 1) ?
zzz 2) SSU bug post mortem
zzz I don't care how the bug happened
orignal tell me
zzz but I want to talk about how we can find and fix things like this faster
orignal I know how
orignal my question is different
orignal how it has so big impact
orignal especially since "dropping is good"
zzz you estimated it was dropping 5%... zlatinb estimated 40-50%
zzz no protocol can handle 50% drops
zlatinb 5% * 6 hops = more
orignal but not every how was SSU
zzz we try to keep the same peers for client tunnels. If we have a bad one, or the same bad one in all tunnels, it's really bad
zzz I saw it the most when I had the same peer as first hop in all outbound tunnels
zzz so if that connection was SSU...
orignal it's obvious
orignal because you prefer SSU
orignal while first i2pd peer prefers NTCP2
zzz it's a mix. what we prefer depends on our connection count and conn limits for both transports, and other factors
orignal how to resolve it faster?
orignal you must tell me the problem
orignal because I was not aware for a long time
zzz yes. I worked on it for two months, and last week I gave up, then zlatinb started doing testnet
orignal nobody complained
zzz I complained two months ago, I said it was i2pd
zzz then a few weeks later I changed my mind
orignal then you said it was not
orignal you compained about leasesets
orignal that they are not published
zzz it took me a long time because it wasn't my bug :)
zzz yes, we publish LS through client tunnels
zlatinb m0ar testnet seems like an obvious answer, no?
orignal the problem was you didn't identify the problem properly
orignal zlatinb no quite
orignal you can use testnet when you know what kind of tests you need to run
orignal and you must indetify a problem first
zzz I didn't understand the problem
orignal you did
orignal when you said that data didn't go through a tunnel
zzz one thing for sure, the release was Nov. 29, and network reliability started getting worse almost immediately
zzz by early December, people were complaining, and IRC disconnects were a lot more often
orignal I started noticing these disconnect
zzz we all saw it
zlatinb I thought RosKomNadzor was blackholing us, so I looked at .ru routers. That was because they had just blocked tor, all in all lucky strike
orignal after postman went down
zlatinb what are we trying to solve with this discussion? It is impossible to come up with a bulletproof way to avoid all future problems
zzz I just felt like I was the only one investigating for two months. Everybody else said 'not my problem'
zzz not trying to avoid, trying to find and fix faster
orignal wrong
orignal you didn't explain details
zlatinb but if we decide that every transport change needs to run through testnet we have some level of confidence
orignal like "I build a tunnel trought routes A,B,C and tunnel worked wrong"
orignal so my point is
orignal you should shared more info about a problem
orignal with tehcnical details
orignal and reproducable cases
zzz I don't have any tests for the live net that "proves" a router is bad
zzz I don't think it's possible
zlatinb I found a bad router manually
orignal you have given me the list of suspicious routers
zlatinb single tunnel pool, force that router to be in it
orignal that's kinda info I expect next time
zzz most of those suggestions only came a week ago
orignal but for LS problem for example you could tell what exatly did you di
orignal FF, tunnels, etc.
orignal if could help to narrow down the problem
zzz is there anything you can do in i2pd to make sure SSU gets more testing? this isn't the first SSU problem. Remember when you were dropping after 5 seconds?
zzz you should use outbound SSU for some % of connections to make sure it's always being tested
orignal maybe
zzz how are we going to ever be sure i2pd SSU2 works if it's not ever "preferred" ?
orignal but you always know that SSU in i2pd is shitty and slow
orignal SSU2 yes
zzz use it anyway, so it gets better
orignal I'm going to implement properly
orignal while fixing SSU in i2pd is worse than rewrite it completely
orignal also I do use SSU
orignal 1. after peertest I have some SSU connections
orignal and reuse them
orignal 2. Incoming connection if firewalled
orignal an IB comes through SSU ofc
orignal one sec
zzz that's all I have for 2)
zlatinb I have more
orignal I give some data
zzz go ahead
zlatinb we need to come up with some preventative masures for the future, and I think testnet suite is it
zlatinb like decide on minimum number of tests before a new release
zlatinb I'm happy to do them
orignal SSU ( 824 )
zlatinb like building tunnels, no tunnel test failures, simple file transfer, etc.
orignal SSUv6 ( 146 )
orignal that's FF
orignal number of sessions
orignal so SSU it not so rare in i2pd
orignal NTCP2 ( 1955 )
orignal olle 2 times less
zlatinb without tests like this the problem can happen again
orignal and I also have 5) for today
orignal number of routers in netdb
zzz ok then let's wrap up 2)
zzz zlatinb, that's a great suggestion to do mixed testnet testing
orignal so what's the problem
orignal yes I agree
orignal before I had around 5K routers on FF
orignal and 8K now
zzz we're still on 2)
zzz anything else on 2) ?
zlatinb I'll come up with a list of test cases
zzz sounds good zlatinb
zlatinb and put on wiki somewhere
zlatinb eot on 2)
zzz 3) SSU2 coding status
zzz as I said, I got a basic handshake and data phase working
zzz not all according to the spec, some things are stubbed out
zzz test code only
zzz orignal, what's your status?
orignal I have started with Noise
orignal than with RouterInfo and data structures for addresses
orignal now implementing SessionRequest send and receive
orignal that's all
orignal hope to finish handshake by the end of this week
zzz as I said, the header encryption and header protection need to change. For now, you can follow what's in the spec, or just don't do it at all
zzz I hope to fix it up in the next week
zzz and hope to be ready for testing with you in 3-4 weeks, as promised
zzz I'm also making fixes to the proposal almost every day
zzz anything else on 3) ?
zzz 4) release schedule
zzz we plan to release in two weeks, feb. 21
zzz what about i2pd?
orignal we can do it
zzz the earlier the better to get the fix out there
zzz eyedeekay1, what did you have for this topic?
R4SAS fine, but depending on situation maybe I'll want to start it on 20th
orignal 20th is also fine ))
zzz I'd be happy if you released today, but I know you have your own processes and schedules
zlatinb I want to run some basic tests in a mixed testnet after checkin deadlines, testing final code for both i2p and i2pd
zlatinb I want at least one day for that
orignal I don't want to mark it as 0.9.53
R4SAS 2 weeks ahead, so you won't know what will happen))
eyedeekay1 I didn't have anything specific to the topic that applies to everybody
eyedeekay1 But I was wondering if I should reach into the easy-install bundle router.config files and set router.disableTunnelTesting=false with the installer
zzz zlatinb, our checkin deadline will be Feb. 18, probably around noon eastern
eyedeekay1 No, right? Since the default has been changed?
zzz I do plan to leave tunnel testing enabled by default for the release, even though I do have code to not put i2pd .52 in tunnels
zzz I'm a little concerned about what it will do for conn limits though
zzz tunnel testing definitely isn't free
orignal don't do it all time
orignal do it once
eyedeekay1 Also it's kind of nasty to try and enable on Android without us having any UI for it, dev builds for Android are awesome right now but release builds are not
orignal after a tunnel was built
zzz 'doing it once' wouldn't catch a peer that only fails 5% of the time
orignal tunnel either works of not
orignal if not build a new tunnel instead
zzz so you're saying the bug is 100% of the time?
zzz I'm confused
orignal peer fails but not session
zzz we don't ever fail a tunnel after one failed test
orignal I mean if session fails it fails for good until recreated
zzz I've enhanced a lot of the testing to find bad peers, and made netdb lookups and stores more flexible also
zzz so hopefully the network will get better quickly after the release
orignal my point is even a single peer test is enough
zzz I'm talking tunnel tests
zzz that go through two tunnels
zzz that's 6 routers to blame
zzz anyway, we're far off the topic of 4)
zzz anything else on 4) ?
zzz 5) number of routers in netdb
orignal 5k before and 8K now
orignal I think it's nothing but waste of memory
orignal we should change the algorithm
zzz what you keep and what you don't is implementation-dependent, up to you
orignal I use your cleanup algorithm
orignal and I think you see the same picture
zzz I have 3500 on a non-ff and 5200 on a ff right now
zzz you can use the same algorithm but tweak the parameters
zzz or change the algorithm. or cache them on disk and flush out of memory
zzz whatever you want
zzz "5K before" before what?
orignal before recently
orignal before Tor in Russia
orignal literally
orignal FF doesn't have an algorithm
orignal just 1 hour
zzz I haven't seen such a big increase here
orignal strange
zzz so make it half an hour
orignal maybe it's might be my problem only
zzz also, don't do exploration on ffs
zzz no need
orignal Routers: 4936
orignal not-FF
orignal good point
orignal will change the code to not explore on FF
zzz another thing that helps is sending netdb responses out exploratory tunnels, so you don't have a million connections to other people's IBGWs
zzz that you have to look up
zzz and also rerouting responses to a different tunnel
zzz getting late, anything else on 5) ?
orignal and I have to run
orignal talk later
zzz ok, good meeting, later
zzz thanks zlatinb will take a look tomorrow