DDraceNetwork - bugs

DDraceNetwork

Development / bugs

Report client & server bugs, especially but not limited to release candidates.

Between 2024-11-19 00:00 and 2024-11-20 00:00

deen

I found a problem, we were counting the playtime double for a few days, fixing (edited)

I think the fix has broken something else, now there is no Playtime displayed at all for the last three weeks

I'm still fixing it, it will take a while

07:50

This is surprisingly expensive to recalculate

07:50

The server will be busy all day today and then we should have it back

07:51

Proof of busy server

Image attachment

rad font

07:54

60gigs of memory woaw

Best font: https://terminus-font.sourceforge.net/

07:56

That's just the website server btw, we have another 64 GB RAM server with more CPU for the database

deen

Proof of busy server

teletext

deen

That's just the website server btw, we have another 64 GB RAM server with more CPU for the database

thats alot of server

07:59

is this just for the website?

also lots of random stuff, database is heavily used by the game servers when you do /rank etc

deen

I'm still fixing it, it will take a while

oh ok, sorry!

deen

also lots of random stuff, database is heavily used by the game servers when you do /rank etc

i dont know the scaling rules for servers n stuff. why is /rank a heavy operation? you can store them sorted?

08:54

petition to add $totaltime because i dont remember

08:55

o its gone

Solly

i dont know the scaling rules for servers n stuff. why is /rank a heavy operation? you can store them sorted?

It needs to deduplicate

Tater

It needs to deduplicate

wdym (i guess thats y ur typing, oh its not) (edited)

Which i think ruins the sorting because the index is lost after the window

whats deduplication

You're really gonna do this to me on my phone

sorry

09:01

you can link spam me

09:01

oh i gues thas also bad

It's on google

is the thing which does /rank monolithic?

Yeah technically

then why deduplication

09:02

oh wait why deduplication if it was spread across multiple servers

Otherwise 1 person can have all top5 ranks

but that happens on map complete

It stores all your finishes in a normalized way

what do you mean normalized

That's a lot of typing sorry

who did it, end time?

Normalized database form

09:05

It never deletes anything or stores duplicates

09:05

So it has to calculate the top5 every time

linked list of references to finishes where you calculate rank when you need to?

09:06

you only ahve to search for additions then

You don't get to choose the container type in a sql database

09:06

It's always a balanced tree

no ): but can you not do linked lists

09:07

(crummy ones with uids)

It would be very very very slow

well slow is better than sorting every /rank5

It's not

i dont tend to touch databases but they cant be that bad at them

Also that would be un normalized

09:08

There's better ways

table 1: uid, next uid, prev uid, reference to finish(uid2) table 2: reference to finish (uid2), who, time, number of completions (edited)

If you want to unnormalize it's enough to store the finish table without duplicates with an index on time. And will query instantly

index on time?

Database index

but youd still have to sort?

It sorts things for you

or

09:10

okay

Kinda

09:10

It's more nuanced

Solly

table 1: uid, next uid, prev uid, reference to finish(uid2) table 2: reference to finish (uid2), who, time, number of completions (edited)

i dont see how this breaks normal form

The linked list itself breaks normal form because you can calculate it without storing it

i only know up to 3rd normal form

09:12

how many are there

6?

09:12

That's not important

i dont understand " you can calculate it without storing it" as a rule

09:13

oh are you saying its "excess" data so shouldnt be in a database

If storing the data is optional then it's not normalized

yeah oke

09:14

well you dont need the first table, i was just saying if you were forced to, you can just have a seperate linked list program (edited)

09:14

also if you were forced to it doesnt really matter that it breaks normal form if its for a good reason (that is faster /top5)

There are solutions which are very obvious deen just chooses to not use them to simplify everything (i assume) (edited)

deen

I found a problem, we were counting the playtime double for a few days, fixing (edited)

you delete this?

Image attachment

deen

Proof of busy server

idk what the core speed on these are, but thats alot of prcoessing happening

Deen does databases as a job so I assume he knows the most of anyone

i thought the db was relativly simple ~~not as i said it~~ table 1: per map: time, completion table 2: who, points (oh no my normal form), completions[], teamcompletions[] table 3/4: map, (team), best time, completions (edited)

Tater

Deen does databases as a job so I assume he knows the most of anyone

yeah i dont touch em

۔🌸𝕿𝖄4𝕶𝕬🌷۔

you delete this?

Yes, WIP

i just wana learn

(<- book)

map momento mori, block cordinates 1129.50 32.50

Image attachment

Image attachment

Tater

If you want to unnormalize it's enough to store the finish table without duplicates with an index on time. And will query instantly

For finishes I've looked into timeseries (TimescaleDB)

09:37

It doesnt fix the problem of /top 5, but you can use materialized views for that

Avo always has a service for everything of course. lol

Tater

Avo always has a service for everything of course. lol

KoG is a big playground

Avolicious

It doesnt fix the problem of /top 5, but you can use materialized views for that

Yes that's one of the "obvious solutions"

09:39

But adds maintenance

Wdym?

Idk, I assume deen would rather pay the high server bill than deal with a materialized view or similar. That's his choice

But you create the materialized view once & then use incremental refresh

09:41

You an do a full refresh, but this is expensive in terms of CPU

I don't run the servers pepeH

pepeH

deen runs databases out of pocket? (edited)

Yeah gotcha

I guess I can open source my database insights

Solly

deen runs databases out of pocket? (edited)

Donation based

then it matters sorta if it could be less intesnive

Solly

deen runs databases out of pocket? (edited)

It's donation funded but he makes up the difference if there is any, usually none recently

1

o

KoG is also donation based smiles

merge and become koddnget

09:43

^ wip name

Avolicious

But you create the materialized view once & then use incremental refresh

Who supports incremental refresh, that sounds interesting

Tater

Who supports incremental refresh, that sounds interesting

Timescale does: https://docs.timescale.com/use-timescale/latest/continuous-aggregates/

Timescale Documentation | Continuous aggregates

Continuous aggregates make queries run faster on very large datasets

Ddnet is on mariadb tho

Yep

09:46

MariaDB has views as well, but not incremential refresh I assume

Avolicious

Timescale does: https://docs.timescale.com/use-timescale/latest/continuous-aggregates/

What this mentions postgresql, is it a wrapper for postgres?

We tried Materialize for ddnets records table. The amount of ram required was absurd, so we gave up on that idea

Tater

What this mentions postgresql, is it a wrapper for postgres?

Timescale is a plugin, yes

09:47

It is based on PostgreSQL

Oh plugin

Tater

Oh plugin

But it works perfectly fine

If you want to work on the top5 issue, I think it's settled that a separate table with only top records of people would help the most, and you can maintain one using triggers

Yeah

Learath2

If you want to work on the top5 issue, I think it's settled that a separate table with only top records of people would help the most, and you can maintain one using triggers

This is better than materialized view?

Tater

This is better than materialized view?

It's not cheap to refresh it at all

No, its just a different way to do

I guess the trigger is pretty light

09:50

In this case

But the data aggregation is heavy

Materialize aims to solve the incremental update issue, but as I said it required an absurd amount of ram. I didn't try Timescale

09:52

I thought Timescale was more for time series data, our data isn't very dense in time per map or per player. It is however kinda dense in time in general without filtering. So maybe it might work decently

Finishes are time-based

09:52

You cant finish 10 years ago right now (edited)

But our queries are almost never about the most recent of all finishes. It's usually never about the most recent but rather the shortest

09:53

Anyway, idk what sorts of query Timescale is optimized for, so maybe it would work/help

That would be a big migration

09:54

The more tables idea sounds easier

09:58

Also team finishes are a thing, there's non fixed team size and the website needs a link to everyone's profile so you can't just put it in a string I think

I would definitely try the table with triggers thing before any other drastic action

I had a single idea for a clever index trick that I was going to try but not having the actual indexes was an imaginary blocker I made, I'll try it in a couple days.

do 0.7 players always appear like this?

Image attachment

justatest

1

is this a bug? how do i get here? im on team 0 no practice btw

Image attachment

444

is this a bug? how do i get here? im on team 0 no practice btw

wats a bug?

Solly

wats a bug?

is possible to get there?

are the seats supposed to be solid

ig tee hitbox is bigger than that lil pixel

are the white squares solid?

ye

17:01

think so

there shouldnt.. be a way to get there?

what map is this @444 ?

Solly

there shouldnt.. be a way to get there?

Image attachment

teleport lazer/grenade/gun/etc! (edited)

meloƞ

what map is this @444 ?

Justix

444

Click to see attachment 🖼️

now we are 2

you can get through the gap with downspeed, it works similar to going through the these freeze tiles i dew but its not about position but speed (edited)

Image attachment

lol

problem is, you can't get out kekw

kekw

meloƞ

problem is, you can't get out kekw

kekw

jhwahjwahjahj ye

17:07

thx 4 help guys

guess how many people are in here

Image attachment

Image attachment

17:08

now we are 10

the more time i play this game, the funnier i find it

brownbear

2

cyberfighter 2

do 0.7 players always appear like this?

yeah

17:33

i guess its intentional ?

shouldnt they appear as default skin

Seems like the server doesn't send any skin info for those clients (edited)

むぎ

relased btw

@deen

pls read further, this has been mentioned a few times now, no need to ping :p

i looked it up tho

22:26

only saw one mention

22:26

what am i supposed to search for that is not "relased"

22:28

but ur right it probably wouldve been mentioned a few times

22:28

but actually he should be pinged every day its not fixed so

22:28

im kinda right anyways

baby frog

but actually he should be pinged every day its not fixed so

ah yes, lets annoy a random individual online every single day because there's a typo

u ignored the other half of my argument though

I don't see how this is a point to argue about.

i agree, and i think we should stop arguing

Exported 172 message(s)

Timezone: UTC+0