DDraceNetwork - developer

DDraceNetwork

Development / developer

Development discussion. Logged to https://ddnet.tw/irclogs/ Connected with DDNet's IRC channel, Matrix room and GitHub repositories — IRC: #ddnet on Quakenet | Matrix: #ddnet-developer:matrix.org GitHub: https://github.com/ddnet

Between 2021-10-03 00:00:00Z and 2021-10-04 00:00:00Z

[ddnet/ddnet-libs] Pull request closed: #15 Add libpng for Jupeyy (untested)

07:40

[ddnet/ddnet-libs] Pull request closed: #17 Update libs, add libpng 1.6.37

07:40

[ddnet-libs:master] 2 new commits

66baca4 Update libs, add libpng 1.6.37 - def- ca5394e Merge pull request #17 from ddnet/pr-libs - def-

Learath2

it's nice to finally have someone that knows what they are doing with xdp and has access to some decent hardware

Lesson for today: If you're going to allocate a bpf map entry for each unverified packet, make sure it's atleast limited to a certain size so it won't fill up the entire RAM, rendering the entire logic for said filter unusable

07:50

Easy solution is BPF_MAP_TYPE_LRU_HASH

Or avoid any allocation for unverified packets if at all possible

We allocate the entries to verify challenge

07:54

if(check_ddnet_handshake_header(udp_bytes, data_end)){
uint32_t challenge = bpf_get_prandom_u32();
bpf_map_update_elem(&ddnet_hashes_list, &challenge, &val BPF_ANY);

07:55

then once we receive the challenge reply, it checks that map to match ip addr & challenge, then deletes the hash from the list

07:55

I've simply changed the type to LRU HASH, that way if the map fills up, it won't cause huge issues

07:55

and instead "recycle" old entries

07:56

Speaking about xdp

07:56

$ddos

flag_unk

server| +- | ▲ pps | ▼ pps

MAIN | up | 570| 565

MASTER| up | 95| 145 flag_unk

flag_unk

FRA | up | 2.0k| 2.31k

GER1 | up | 123| 96

GER2 | up | 428| 508

POL2 | up | 223| 241

RUS4 | up | 2.45k| 2.79k

TUR2 | up | 701| 706

IRN | up | 363| 197

CHL2 |ddos| 1.44k| 298

BRA3 | up | 1.12k| 1.36k

USA | up | 8| 7

USA2 | up | 383| 494

CAN2 | up | 37| 36

CHN1 | up | 8.68k| 12.98k

CHN2 | up | 1.96k| 2.47k

CHN3 | up | 2.96k| 3.78k

CHN4 | up | 3.18k| 3.94k

CHN5 | up | 3.51k| 4.49k

CHN6 | up | 3.44k| 4.26k

CHN8 | up | 3.05k| 3.73k

JAP2 | up | 2.17k| 2.6k

KOR2 | up | 2.35k| 2.67k

SGP2 | up | 247| 340

ZAF2 | up | 134| 112

._. FRA is most certainly under attack right now

Nope, it's not

Had trouble connecting & staying on

07:58

but I guess it got mitigated

No, there is no mitigation active right now and I saw no attack

07:58

Just judging by the fact I couldn't connect at all really

I am currently hacking around with freeze textures because i tend to confuse frozen tees with ninjajetpack or normal ninja does the client know if other tees are frozen?

08:07

got it https://github.com/ddnet/ddnet/blob/4ee5aa585b6bd11f4e12522ca9d6572db460ce05/src/game/client/gameclient.h#L353

ddnet/gameclient.h at 4ee5aa585b6bd11f4e12522ca9d6572db460ce05 · dd...

DDraceNetwork, a cooperative racing mod of Teeworlds - ddnet/gameclient.h at 4ee5aa585b6bd11f4e12522ca9d6572db460ce05 · ddnet/ddnet

Is this server-sync'ed too?

i think the client started at some point to reimplement server logic for better prediction

Just asking as otherwise players could just compile the client, and force isFrozen to false, no?

I would be very suprised if the client could sync up to the server and tell it about state

08:08

no :D

Good good

this is not minecraft haha

Just wondering as i've actually seen cheat clients for this

08:09

(mainly people complaining about it in #reports) (edited)

i doubt that

08:09

what did you send?

08:09

image.png

Image attachment

# reports

08:10

Anyway, time for work.

the tw protocol is pretty safe regarding cheats by passing physics

iirc, the protocol is mostly just the client sending keyboard input to the server and the server returning a gamestate

ye

08:34

how it should be in any sane game

chillerdragon: uwu

OwO

08:37

screenshot_2021-10-03_10-36-55.png

Image attachment

08:37

send halp itube they bully me while coding .-.

[ddnet/ddnet] Pull request opened: #4189 Build macOS build

Undefined symbols for architecture x86_64: "_SCDynamicStoreCopyProxies", referenced from: _Curl_resolv in libcurl.a(libcurl_la-hostip.o) ld: symbol(s) not found for architecture x86_64

Checklist

[ ] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test if it works standalone, system.c especially
[ ] Considered possible null pointers and out o...

[236/236] Linking CXX executable DDNet real 0m13.971s user 1m55.663s sys 0m7.239s

08:53

compile time on r5 5600x

08:53

with clang

congrats on the update

thanks

Release build?

cmake .. -GNinja -DMYSQL=ON -DCMAKE_BUILD_TYPE:STRING=RelWithDebInfo \ -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DVIDEORECORDER=ON -DDOWNLOAD_GTEST=ON \ -D_CMAKE_TOOLCHAIN_PREFIX=llvm-

08:54

ye

08:54

i think my old cpu was bottlenecking the gpu

08:54

on dota i got nearly 50 more fps

08:55

well i also got new ram at 3200mhz

08:55

from 2933

08:55

but i doubt it matters much

Twice as fast as my i7 6700k (28s)

:o

08:56

with clang?

yes, same command as you

ill try with gcc

08:56

well i had this too export CXX=/usr/bin/clang++

ah, llvm- is not enough

ye

20 s then

17.7s for me with gcc

08:57

so gcc is slower

08:58

which means clang is even better for development

08:58

faster iterations

chillerdragon

send halp itube they bully me while coding .-.

I come in 1 hour

[ddnet/ddnet] Pull request opened: #4190 Add notes for settings only available

with specific compile options or on specific builds Also applies to https://ddnet.tw/settingscommands/

Checklist

[ ] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test if it works standalone, system.c especially
[ ] Considered possible null pointers and out of bounds array indexing
[ ] Changed no physics that affect existing maps
[ ] Tested ...

10:25

[ddnet-web:master] 1 new commit

9618e35 Update settings & commands with https://github.com/ddnet/ddnet/pull/4190 - def-

@Learath2 weird behaviour on GER1 server

13:20

everyone got connection problems

13:21

without attacks :D after i rejoined it was also fully okay again

should i think about endianness while using std streams?

You should almost never think about endianness

justatest

1

okeey

A teeworlds data file will have a magic at the top. From which you can determine whether you need to flip or not

13:34

(idk why that approach was taken instead of always saving datafiles little endian and swapping on big endian or vice versa)

troll

1

fokkonaut

without attacks :D after i rejoined it was also fully okay again

Might be caused by handshake timeout(?)

yeah, no sane in this thing

meh, 130 clients handshaked

chillerdragon

screenshot_2021-10-03_10-36-55.png

chillerdragon: send ur client version

ger1 is so laggy rn

14:02

@Fän

dont think i caught it on time

14:15

got it

why is it laggy?

1 moment

wow

14:25

connection problems again

14:26

wtf

that's me recompiling the code

DUDE

14:26

D:

14:26

omg

14:26

i tohught saves are gone

well, if people wouldnt throw attacks using protocols from 1998

14:26

towards it (edited)

14:26

this wouldnt happen

14:26

anyway, there we go

xd

just dropping any unknown protocol now

:D

ok, good to go now

you probably fucked a lot of runs with this btw

and saved a lot of them in the future

14:28

Rather have it taken care of now than never tbh (edited)

14:30

I'm surprised how protocols from literally 1998 are being used to attack stuff nowadays

who attacked, do you know that?

nope, doesnt matter anyway

is there an attack running rn?

nope

but was?

any attack that would affect the server should show up @ $ddos

14:31

and yes, until I patched that flaw (edited)

oh okay

Code in germany is really far behind as it's quite outdated

14:33

and can't be updated to up2date until the new router comes in (edited)

still lagging

no attack here

well

14:34

its lagging :D

Might be you?

14:34

like i said, no attack running

14:34

$ddos

flag_unk

server| +- | ▲ pps | ▼ pps

MAIN | up | 659| 652

MASTER| up | 105| 153 flag_unk

flag_unk

FRA |down| |

GER1 | up | 2.02k| 2.52k

GER2 | up | 3.01k| 3.61k

POL2 | up | 475| 529

RUS4 | up | 4.37k| 4.88k

TUR2 | up | 1.93k| 2.12k

IRN | up | 232| 47

CHL2 |ddos| 1.41k| 199

BRA3 | up | 2.03k| 2.53k

USA | up | 33| 35

USA2 | up | 234| 269

CAN2 | up | 44| 63

CHN1 | up | 8.03k| 9.41k

CHN2 | up | 900| 1.2k

CHN3 | up | 2.03k| 2.56k

CHN4 | up | 2.98k| 3.63k

CHN5 | up | 2.84k| 3.48k

CHN6 | up | 5.07k| 6.37k

CHN8 | up | 2.62k| 3.33k

JAP2 | up | 2.16k| 2.56k

KOR2 | up | 2.18k| 2.53k

SGP2 | up | 310| 452

ZAF2 | up | 222| 188

^

more likely the server

14:34

^^

14:34

i am not lagging

14:36

then it has to be the CPu

You mean the CPU running at 10%?

or some other hardware

14:37

idk

14:37

its definitely the server

https://ddnet.tw/status/

14:38

considering that most ppl are 19-40ms, atleast as of right now, I don't think anything major is going on

have you ever joined and tested yourself?

I am in it right now.

its not unplayable but there are stutters

jumping around with no lags whatsoever.

xD

You're also at 38ms

14:39

14:39

Have you tried reconnecting?

wont help lol

Check.

no im in the run

I'm jumping around with no lags whatsoever

14:40

Only thing i could think of is that it somehow messed up packet sync or whatever during the time I had to recompile / restart filtering

14:41

which i've seen happening in source games a lot (edited)

zooming out a lot makes the game lag

14:54

only on ger1

14:55

(not the game, but it looks like the traffic doesnt work correctly, still > 500fps)

Fän

that's me recompiling the code

Making changes in production just hits differently

(edited)

@Fän seems like ur firewall is not working correctly yet

fokkonaut

@Fän seems like ur firewall is not working correctly yet

Maybe its anti fokkonaut

u

SPYRES

Making changes in production just hits differently

(edited)

Didn't affect much in this case, but we usually only do this during "emergencies"

14:56

or well, in this case, getting things to run ok

Fän

Didn't affect much in this case, but we usually only do this during "emergencies"

Yeah ik. I'm just messing with yah. I do it as well on our firewall. Waste time in CAB.

fokkonaut

(not the game, but it looks like the traffic doesnt work correctly, still > 500fps)

idk how that would be related to game traffic / filtering. Once you are connected, you're whitelisted to send / receive traffic. There's no "traffic blocking" at that point

14:57

Like I said, try rejoining. Chances are something messed up during the reconnect.

When you zoom out, you tell the server your zoom level

14:58

maybe it's because you receive too many objects then?

Ok, like I said, you're either whitelisted to send traffic, or not. There's no in-between.

Reconnecting doesnt help

14:58

Maybe try it yourself ^^

I did a while ago

Again?

sure

If I could I'd let deen host at where I work. 20GB links and we own a /16 on the internet. Pitty I dont own the organisation

On my own server with way more objects it doesnt lag at all

SPYRES

If I could I'd let deen host at where I work. 20GB links and we own a /16 on the internet. Pitty I dont own the organisation

Even with "20gbps" links, if your server can't handle the l7 load, there's not much you can do

14:59

from deen's blog post, the server literally dies at < 100mbps

@Fän just tried it again, it really is only on your server

fokkonaut

@Fän just tried it again, it really is only on your server

hm, maybe fragment-related?

15:00

i just started dropping fragmented responses as this would bypass the filtering

15:00

but i can just let whitelisted clients send fragments, shouldnt be an issue

it seems to be only on jao shooter

Any other GER1 server works ok?

15:01

that's odd, maybe map-related or something?

i guess, but this is also the one with most players

other players have no impact on your network performance

15:01

or mitigation

15:01

or anything really (edited)

15:01

We've kept that in mind

Fän

other players have no impact on your network performance

of course, more objects to be sent, i dont know how your firewall is constructed

We do not perform any rate limiting.

Fän

Even with "20gbps" links, if your server can't handle the l7 load, there's not much you can do

True but we got ddos mitigation upstream as well by our ISP before it comes to us over the 20Gbps links. I dont see why it wouldnt not handle it. We got a proper VMware infrastructure. But things like XDP and reading cloudflare blogs is more interesting.

like i said, once you're validated, you're ok to throw as much traffic as you want

then it seems like i am a psycho (edited)

15:02

and there is no lag

maybe a bug?

15:03

I have no idea, but i'm on there just fine

15:03

surely odd though

zoom out

15:03

max

15:03

with opengl 3 on, there should be a limit in zooming out

15:03

and that state is still playable with 500 fps at least

15:03

for me, but the network is lagging, and thats only on this server

SPYRES

True but we got ddos mitigation upstream as well by our ISP before it comes to us over the 20Gbps links. I dont see why it wouldnt not handle it. We got a proper VMware infrastructure. But things like XDP and reading cloudflare blogs is more interesting.

"ddos mitigation" which is based on best-effort. You will still see >1Gbps leaks every now and then. There's not just packet size to consider, but also the rate of those coming in.

15:04

From what i can see, ddnet can handle about 300k pps with a decent CPU

fokkonaut

for me, but the network is lagging, and thats only on this server

tried any other ger1 server though?

15:05

I'll check, my mac won't let me zoom out though, so it might take a while

i'll check if thats the map, because it has an unusual amount of draggers and @deen changed the networkclipping to always send all objects in your area zoom distance (which on my server again works fine with wwaaaay more objects (so many that the snap fills up to max size when zooming out completely))

Maybe it is fragment-related after all.

15:06

We'll see in a moment

its a bit too on another server with this map, but the current server seems to be the worst (probably due to all the players)

check again?

still

Doesn't seem filter-related then.

Fän

"ddos mitigation" which is based on best-effort. You will still see >1Gbps leaks every now and then. There's not just packet size to consider, but also the rate of those coming in.

My palo alto seems to be doing a decent enough job. No complains yet. We dont go as in depth as XDP at where I work. We pay for things to work out the box and let someone else handle development! :p

client restart also didnt help, must be the full snap then

hmm actually

15:11

what about now?

still

yeah sorry, i can't really tell what it is then

15:12

1 mom

15:12

oh dang

15:12

might've fat fingered there

15:13

-_-

yea seems like it did you crash the server?=

no

the fuck is happening then

give it a second

15:13

I fat fingered

connection problems, out of game back to back

ok, give it another try now

same

can't really tell then

im pretty sure its all the draggers when zooming out

I'll check this out when on pc

the client i use includes bots :-. so im not sharing it (@shiro)

chillerdragon: send ur client version

16:23

@ddnet staff pls no ban

chillerdragon

the client i use includes bots :-. so im not sharing it (@shiro)

its ok, bot is nice

chillerbot confirmed

nice, you can now generate mostly redundant comments automatically for your school projects greenthing

greenthing

Image attachment

troll

3

looks like the comments are incorrect

[ddnet-web:master] 1 new commit

db46795 fix typo - def-

22:29

[ddnet/ddnet] Issue closed: #3797 Would it be possible to ship SDL for the Steam binary with Wayland support?

22:31

[ddnet-libs:master] 1 new commit

5287ccd Add zlib1.dll for curl, add pnglibconf.h - def-

22:32

[ddnet/ddnet] Pull request opened: #4191 Add zlib1.dll for curl, add pnglibconf.h

As reported by Jupeyy in https://github.com/ddnet/ddnet-libs/pull/17#issuecomment-933034244

Checklist

[ ] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test if it works standalone, system.c especially
[ ] Considered possible null pointers and out of bounds array indexing
[ ] Changed no physics that affect existing maps
[ ] Tested the change wit...

[ddnet/ddnet] Pull request closed: #4191 Add zlib1.dll for curl, add pnglibconf.h

22:45

[ddnet:master] 2 new commits

b733f36 Add zlib1.dll for curl, add pnglibconf.h - def- deaec27 Merge #4191 - bors[bot]

is the SDL lib, the 2.0.16 release, or current master?

[ddnet:master] 2 new commits

e8bc6fc Add notes for settings only available - def- e7252b5 Merge #4190 - bors[bot]

22:56

[ddnet/ddnet] Pull request closed: #4190 Add notes for settings only available

2.0.16 release with the 2 patches

alright

22:59

well if somebody with wayland wants to test it, would be nice @nori or someone

22:59

i dunno, SDL merges like 200 wayland patches per second, so no idea when it works and when not xd

[ddnet:master] 2 new commits

e143555 Build macOS build - def- 1e8fd0d Merge #4189 - bors[bot]

23:10

[ddnet/ddnet] Pull request closed: #4189 Fix macOS build

Exported 259 message(s)