discordjs/ws big bot memes (old) | discord.js - imagine an app | Page 2

stable hatch Mar 28, 2023, 5:49 PM

#

Wait what exactly are you doing

#

What are clusters

sullen snow Mar 29, 2023, 12:43 AM

#

clusters are basically a process where it runs a single d.js client
threads is basically discordjs/ws spawned threads

#

basically imagine d.js sharder but with clustering method

#

similar to kurasuta if you are familiar with it

stable hatch Mar 29, 2023, 1:49 PM

#

so you mixed djs/ws with djs itself?

sullen snow Mar 29, 2023, 3:12 PM

#

its basically what the pr that ports djs/ws

#

but my own implementation

#

do not be confused with it doing major overwrites other than the identify thing in the ws package

#

its not mixed but rather a port to make djs/ws compatible with d.js

#

our sharder is https://github.com/Deivu/Indomitable which supports v14 stock just fine

GitHub

GitHub - Deivu/Indomitable: A lightweight (the actual ship is heavy...

A lightweight (the actual ship is heavy though), performant, powerful & no dependency sharder for Discord.JS - GitHub - Deivu/Indomitable: A lightweight (the actual ship is heavy though), ...

#

sharder spawns clusters, which has discord.js client, which has ported discord.js/ws on it

#

we say clusters because each clusters handle multiple websockets (in this case, threads that is made by discord.js/ws)

#

each thread handles one discord.js/ws websocket connection that is mostly untouched as well

#

did that clear the confusion here

dim oracle Mar 30, 2023, 8:01 PM

#

Surpassed 3 days of no issues now

dusty dove Mar 31, 2023, 5:01 AM

#

pogging

dim oracle Mar 31, 2023, 1:09 PM

#

and just when I say that a shard dies

#

let me get the debug logs

#

seems to be the same as here

dusty dove Mar 31, 2023, 1:22 PM

#

mmmh

#

dunno, wait for us to merge stuff

dim oracle Mar 31, 2023, 1:23 PM

#

this is on 0.8.0-dev.1679832251-b8b852e.0 though

#

shard is never seen again after this

📎 logs.txt

dusty dove Mar 31, 2023, 2:06 PM

#

dusty dove dunno, wait for us to merge stuff

okay, but

#

there's a rather large pending PR

#

that refactors the waitforevent logic

#

though the shard never being seen again is unrelated, it's mostly that's a hard error throw so the worker just died

#

and we don't have any auto-restarting logic in place

#

probably should in hindsight

stable hatch Mar 31, 2023, 2:08 PM

#

Deadge

dusty dove Mar 31, 2023, 2:12 PM

#

though idk how that should work like

#

it throws cus invalid token

#

and i just... restart anyway? like that is

#

not great

#

i guess the alternative would be to handle the handshake timeout different

#

but i think i just can't

#

the stack trace is fully internal as you can see

stable hatch Mar 31, 2023, 2:14 PM

#

oh its the bubbled hello wait error

#

wont that be fixed with your pr

dusty dove Mar 31, 2023, 2:15 PM

#

genuinely don't know

#

i cant tell if ws is just throwing internally

#

or if they actually fire an error event on handshake timeout

#

either way i do not see an easy way to make it retry if that happens

#

which is lovely

stable hatch Mar 31, 2023, 2:22 PM

#

dusty dove either way i do not see an easy way to make it retry if that happens

and i'm gonna ask again

#

didn't your PR changing waitForEvent fix it

dusty dove Mar 31, 2023, 2:22 PM

#

no dude

#

again

#

this is the handshake timeout

#

its a number I pass into the WS constructor

#

this is not a waitForEvent call at all for me

#

its literally WS internally waiting for the TCP connection being opened

#

before I start getting any events from the connection

stable hatch Mar 31, 2023, 2:23 PM

#

I thought waitForEvent listened for error too

dusty dove Mar 31, 2023, 2:23 PM

#

yes, but again

#

I am not calling waitForEvent at all at this point, I don't think

#

I'm not quite certain what WS is doing internally here

stable hatch Mar 31, 2023, 2:24 PM

#

girl you dont even know your own code at this point

#

[DEBUG[ [Fri,03/31/23,10:19:46] (Cluster Process [ID: 3]): [[EventHandler](Discord.JS): [WS => Shard 79 => Worker] Waiting for event hello for 60000ms[

#

literally the log before the handshake error

#

so you DO call wait for event

dusty dove Mar 31, 2023, 2:24 PM

#

shrug then

stable hatch Mar 31, 2023, 2:24 PM

#

so you CAN handle it

dusty dove Mar 31, 2023, 2:24 PM

#

or not

#

I still don't know if they emit an error event

stable hatch Mar 31, 2023, 2:24 PM

#

Deadge

dusty dove Mar 31, 2023, 2:24 PM

#

but yes, we'll see if the PR helps

#

if not we need to find something else

stable hatch Mar 31, 2023, 2:25 PM

#

aka Promise.race(once(), once(error), timeout)

dusty dove Mar 31, 2023, 2:25 PM

#

oh you know what

stable hatch Mar 31, 2023, 2:25 PM

#

OMEGALUL

dusty dove Mar 31, 2023, 2:25 PM

#

why do we pass a handshake timeout at all

#

if the hello one will just fire anyway

#

lol

#

i had a review to add it in the original /ws pr because it "helped solve mainlib issues"

stable hatch Mar 31, 2023, 2:25 PM

#

handshake timeouts are also node only

dusty dove Mar 31, 2023, 2:25 PM

#

and i just did

#

but i never checked up on why its there

stable hatch Mar 31, 2023, 2:26 PM

#

believe me idek why half the code in current ws exists either

dusty dove Mar 31, 2023, 2:26 PM

#

u can ask

#

and ill have an answer

stable hatch Mar 31, 2023, 2:26 PM

#

current ws not /ws

dim oracle Mar 31, 2023, 2:26 PM

#

side note, any word back about the big bot/currency bucketing

dusty dove Mar 31, 2023, 2:27 PM

#

oh lol

stable hatch Mar 31, 2023, 2:27 PM

#

dim oracle side note, any word back about the big bot/currency bucketing

the sweet sweet sounds of silence

dim oracle Mar 31, 2023, 2:27 PM

#

the classic

stable hatch Mar 31, 2023, 2:27 PM

#

eh, i cant blame em

#

I also didn't ping them, I'll ping ian 1 later

dim oracle Apr 3, 2023, 10:46 AM

#

Can we actually continue this topic? Hopefully the explanation here (#1086244458900750367 message) cleared some confusion as to why we call them clusters

dusty dove Apr 3, 2023, 11:04 AM

#

its still unclear to me what your shardsPerWorker is

dim oracle Apr 3, 2023, 11:15 AM

#

@sullen snow I guess

dim oracle Apr 3, 2023, 11:16 AM

#

dusty dove its still unclear to me what your shardsPerWorker is

but I guess that was answered already

dusty dove Apr 3, 2023, 12:52 PM

#

u dont seem v sure abt it

sullen snow Apr 3, 2023, 1:27 PM

#

dusty dove u dont seem v sure abt it

im sure about it

#

I coded the whole thing so I know it's 1

dusty dove Apr 3, 2023, 1:27 PM

#

well, i cannot reproduce

#

so you're gonna have to get me a reproduceable code sample that I can easily run myself

sullen snow Apr 3, 2023, 1:28 PM

#

did you try running it with lots of shards?

#

on my 100k bot this doesnt seem to happen

dusty dove Apr 3, 2023, 1:28 PM

#

where does the warning actually come from

#

is it one of the threads or the main process

#

scrap that, it's def the threads

#

since the stack trace includes the bootstrapper

sullen snow Apr 3, 2023, 1:29 PM

#

 MaxListenersExceededWarning: Possible EventTarget memory leak detected. 11 message listeners added to [MessagePort [EventTarget]]. Use events.setMaxListeners() to increase limit
    at [kNewListener] (node:internal/event_target:514:17)
    at eventEmitter.<computed> (node:internal/worker/io:307:12)
    at MessagePort.addEventListener (node:internal/event_target:623:23)
    at MessagePort.on (node:internal/event_target:873:10)
    at VanguardBootstrap.setupThreadEvents (/main/node_modules/@discordjs/ws/src/utils/WorkerBootstrapper.ts:83:5)
    at VanguardBootstrap.bootstrap (/main/node_modules/vanguard/src/worker/VanguardBootstrap.ts:51:14)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)

dusty dove Apr 3, 2023, 1:30 PM

#

sullen snow on my 100k bot this doesnt seem to happen

how many shards is that

#

w/e

#

ill run 100

sullen snow Apr 3, 2023, 1:31 PM

#

1440

dusty dove Apr 3, 2023, 1:31 PM

#

you what

#

you run ~70 guilds per shard?

sullen snow Apr 3, 2023, 1:31 PM

#

no this is on chip

#

1440 shards
24 shards per cluster

dusty dove Apr 3, 2023, 1:32 PM

#

well

#

hmm

#

Error: Not enough sessions remaining to spawn 1440 shards; only 986 remaining; resets at 2023-04-04T13:30:54.853Z lol

sullen snow Apr 3, 2023, 1:32 PM

#

or rather 24 "threads" per cluster

dusty dove Apr 3, 2023, 1:33 PM

#

@stable hatch i'll use our magic bot token

sullen snow Apr 3, 2023, 1:33 PM

#

its kinda a weird bug since I'd seen this happen on my original bot, but didnt care about it

#

and this is on older 0.6.x master ws

#

its there but I'm unsure how it even emits or how it even emits

#

and it doesnt also seem to happen again after the initial emit anyways

dusty dove Apr 3, 2023, 1:34 PM

#

yeah I see it too at 1440 shards

#

yeah its directly tied to the shard count

#

at 9 shards it doesnt happen

#

at 10 it does

sullen snow Apr 3, 2023, 1:36 PM

#

at 1 shard per thread?

#

or per worker

dusty dove Apr 3, 2023, 1:36 PM

#

yes

#

1 per worker/thread

#

10 total

#

im p sure i know why now though

#

not a bug as i thought

sullen snow Apr 3, 2023, 1:37 PM

#

yeah it doesnt really emit

#

after the initial boot

stable hatch Apr 3, 2023, 1:37 PM

#

dusty dove `Error: Not enough sessions remaining to spawn 1440 shards; only 986 remaining; ...

you can always causae the token to reset

#

(dont)

sullen snow Apr 3, 2023, 1:37 PM

#

so my guess is its not really an issue

#

but then again

#

kyoso hates seeing those errors

dusty dove Apr 3, 2023, 1:37 PM

#

stable hatch you can always causae the token to reset <:mm:272107213911228416>

that was my own bot, not ours

sullen snow Apr 3, 2023, 1:37 PM

#

so he's been knocking on me about it

dusty dove Apr 3, 2023, 1:37 PM

#

yeah so

#

I think node does something crazy

stable hatch Apr 3, 2023, 1:38 PM

#

those are not errors

dusty dove Apr 3, 2023, 1:38 PM

#

and the messagePort object is exactly the same

#

in every worker thread

stable hatch Apr 3, 2023, 1:38 PM

#

and are also non issues

dusty dove Apr 3, 2023, 1:38 PM

#

which is why its saying it leaks

#

well it could've been an actual leak vlad

#

but really its every thread assigning 1 listener

#

but there's just a bunch of threads

#

and the object is just shared i guess

#

i have no idea how that works

sullen snow Apr 3, 2023, 1:38 PM

#

can't we just try to increase the event emitter based on how many shards per worker we have

dusty dove Apr 3, 2023, 1:38 PM

#

i dont have any other explanation otherwise

sullen snow Apr 3, 2023, 1:38 PM

#

so we can just have this error silenced

dusty dove Apr 3, 2023, 1:38 PM

#

its kinda tricky i think

sullen snow Apr 3, 2023, 1:39 PM

#

oh well I guess I'll just let him know

#

but yeah thanks for confirming it as well that its not really an issue

dusty dove Apr 3, 2023, 1:41 PM

#

nevermind

#

im insane

#

I was running shardsPerWorker: 'all' on accident

stable hatch Apr 3, 2023, 1:41 PM

#

dusty dove im insane

we know

dusty dove Apr 3, 2023, 1:41 PM

#

where it does make sense to happen

#

yeah cant repro with shardsPerWorker 1

sullen snow Apr 3, 2023, 1:41 PM

#

thats also our default setting

#

right now

#

oh wait a second

#

let me check real quick

#

I think Im missing something crucial here

dusty dove Apr 3, 2023, 1:42 PM

#

no wonder it was starting 1440 threads so easily

#

➜ node index.js

<--- Last few GCs --->


<--- JS stacktrace --->


#
# Fatal javascript OOM in GC during deserialization
#


<--- Last few GCs --->


<--- JS stacktrace --->

FATAL ERROR: Committing semi space failed. Allocation failed - JavaScript heap out of memory```

lmao

#

i just crashed trying to start 1440

sullen snow Apr 3, 2023, 1:43 PM

#

https://safe.saya.moe/tf0hh7y944xg.png

#

drakeLUL

dusty dove Apr 3, 2023, 1:43 PM

#

PepeLaugh

sullen snow Apr 3, 2023, 1:43 PM

#

it was all all along

dusty dove Apr 3, 2023, 1:43 PM

#

"are you sure you're doing shardsPerWorker 1?" "yes we're sure"

sullen snow Apr 3, 2023, 1:43 PM

#

why did I forgot that

#

~~its not like my memory is good forgive me~~

dusty dove Apr 3, 2023, 1:43 PM

#

neither is mine but at least check properly PepegaHands

sullen snow Apr 3, 2023, 1:44 PM

#

hey at least we know there is an issue on shards per worker all

#

a "non issue"

dusty dove Apr 3, 2023, 1:44 PM

#

when r u going over my pr @stable hatch

sullen snow Apr 3, 2023, 1:44 PM

#

LUL

dusty dove Apr 3, 2023, 1:44 PM

#

yeah actually u can fix that urself

#

if you provide your own uh

#

workerFile

#

just do parentPort.setMaxListeners() somewhere in your script

sullen snow Apr 3, 2023, 1:44 PM

#

it was more on a confusion on my part

#

I thought "all" means 1 ws per thread

dusty dove Apr 3, 2023, 1:44 PM

#

lol no it means all of them in one thread

sullen snow Apr 3, 2023, 1:45 PM

#

thats why I was telling it was 1

dusty dove Apr 3, 2023, 1:45 PM

#

tho if u were running that in prod w no issues

#

that is impressive

sullen snow Apr 3, 2023, 1:45 PM

#

yeah I think so because the thread only runs ws

dusty dove Apr 3, 2023, 1:45 PM

#

means my stuff is goated

#

well yeah of course

sullen snow Apr 3, 2023, 1:45 PM

#

its not like its handling d.js insane class conversions LUL

dusty dove Apr 3, 2023, 1:45 PM

#

a big part of my stuff is that basically everything is async

#

so there's like no thread-blockers

#

other than zlib

stable hatch Apr 3, 2023, 1:45 PM

#

dusty dove when r u going over my pr <@139836912335716352>

if lucky, this week

sullen snow Apr 3, 2023, 1:46 PM

#

and why I opted to do my own impl. of porting it to v14 using your ws when there are choices crabRave

dusty dove Apr 3, 2023, 1:46 PM

#

your biggest bottleneck is literally always JSON.parse

#

or erlpack

#

unironically

stable hatch Apr 3, 2023, 1:46 PM

#

zlib w

sullen snow Apr 3, 2023, 1:46 PM

#

erlpack is actually less bottleneck than json.parse

#

but then again

dusty dove Apr 3, 2023, 1:46 PM

#

i'd buy that

#

but you end up sending more data over the wire

#

in most cases

#

w/o zlib

stable hatch Apr 3, 2023, 1:46 PM

#

receiving*

dusty dove Apr 3, 2023, 1:46 PM

#

with, erlpack is much easier for zlib to compress

sullen snow Apr 3, 2023, 1:46 PM

#

we used erlpack so thats maybe the reason why we are able to run 24 shards on one thread of your ws

stable hatch Apr 3, 2023, 1:46 PM

#

you dont send zlib data

dusty dove Apr 3, 2023, 1:46 PM

#

yeah ty, receiving

sullen snow Apr 3, 2023, 1:47 PM

#

with stock d.js ws handling

#

you can run about 16-20 shards

#

with erlpack in single process

dusty dove Apr 3, 2023, 1:47 PM

#

fwiw we're starting work on TSR soon-ish

#

lots of convos abt cache n stuff rn

sullen snow Apr 3, 2023, 1:47 PM

#

if you can remove the channels bottleneck

dusty dove Apr 3, 2023, 1:47 PM

#

so soon you should be able to ditch all that ancient jank garbage

stable hatch Apr 3, 2023, 1:47 PM

#

leaking chats

#

Banned

dusty dove Apr 3, 2023, 1:47 PM

#

true

sullen snow Apr 3, 2023, 1:47 PM

#

well tbh the main issue or main issue with d.js for me rn is

#

its insane class conversions and caching this class

stable hatch Apr 3, 2023, 1:47 PM

#

not to mention thats a stretch dd

dusty dove Apr 3, 2023, 1:47 PM

#

yeah soon is a stretch lol

stable hatch Apr 3, 2023, 1:48 PM

#

we're having discussions, wont mean we even start work on it till those discussions yield a conclusion

dusty dove Apr 3, 2023, 1:48 PM

#

its still soon in terms of like

#

how long their bot's been around for tho realistically

stable hatch Apr 3, 2023, 1:48 PM

#

tldr dont listen to dd unless its ws stuff

dusty dove Apr 3, 2023, 1:48 PM

#

mmlol

#

just have to put it right

sullen snow Apr 3, 2023, 1:48 PM

#

main reason I just dont move to eris or other libs there is because d.js is just maintained and been using this since like v11 LUL

#

no point on moving to other libs if I can just make it work I guess

#

kek

#

I get timely updates on parts I dont want to maintain, and just maintain the parts I want to maintain kinda thing

sullen snow Apr 11, 2023, 7:56 AM

#

any ideas on what is c lose code 4200 could be @dusty dove

#

cant find any documents on it so you might know what that is?

dusty dove Apr 11, 2023, 7:56 AM

#

its just the code we specify when we resume

#

since it needs to be 4xxx

sullen snow Apr 11, 2023, 7:57 AM

#

oh so basically the sharder tries to reconnect then?

dusty dove Apr 11, 2023, 7:57 AM

#

resume specifically, yes

sullen snow Apr 11, 2023, 7:57 AM

#

and if it fails, what happens

#

tries to reident or resume again?

dusty dove Apr 11, 2023, 7:57 AM

#

tries to fresh reconnect

sullen snow Apr 11, 2023, 7:58 AM

#

ok, just wondering since theres some kind of outage last night

#

and some shards didn't connect

#

same on my side but my logs is 1gb so LUL

dim oracle Apr 11, 2023, 7:59 AM

#

Probably related to previous issues, is the big™️ PR merged already?

dusty dove Apr 11, 2023, 7:59 AM

#

kyra approved now

dim oracle Apr 11, 2023, 7:59 AM

#

Doesn't help we got 10k+ 520's from Discord in like 40 minutes lol

dusty dove Apr 11, 2023, 7:59 AM

#

just waiting on space

#

should be sometime today

dim oracle Apr 11, 2023, 8:01 AM

#

Alright

stable hatch Apr 11, 2023, 11:47 AM

#

dim oracle Doesn't help we got 10k+ 520's from Discord in like 40 minutes lol

i k r

dim oracle Apr 11, 2023, 11:50 AM

#

We love 3500 zombie shard disconnects in ~20 minutes

sullen snow Apr 11, 2023, 12:50 PM

#

zombies

#

The Last of Us: Discord.JS edition

stable hatch Apr 11, 2023, 9:47 PM

#

Just heard news about ya boy, max_concurrency, sounds like we're handling it correctly in djs/ws pog

dim oracle Apr 11, 2023, 11:28 PM

#

stable hatch Just heard news about ya boy, max_concurrency, sounds like we're handling it cor...

So buckets aren't a thing at all?

stable hatch Apr 12, 2023, 8:35 AM

#

dim oracle So buckets aren't a thing at all?

You'll need to explain that buckets means to you

dim oracle Apr 12, 2023, 11:03 AM

#

stable hatch You'll need to explain that buckets means to you

You need 5s between identifies for shards in the same bucket, where bucket = shard_id % max_concurrency

#

other than that you can do whatever order you want

stable hatch Apr 12, 2023, 11:03 AM

#

dim oracle other than that you can do whatever order you want

not what the docs say

#

also yeah we're doing it correct shrug

#

oh wait

dim oracle Apr 12, 2023, 11:04 AM

#

stable hatch Just heard news about ya boy, max_concurrency, sounds like we're handling it cor...

but what did they confirm here, that what I said is wrong?

stable hatch Apr 12, 2023, 11:04 AM

#

ugh your phrasing is so confusing

#

its not a bucket

#

its a ratelimit key

#

the formula is correct

#

and yes, you cannot do shard 0 and 16 at the same time for example

#

you need to wait the 5s between them

dim oracle Apr 12, 2023, 11:05 AM

#

But for example, I can start 0..7 and 40..47 at the same time

stable hatch Apr 12, 2023, 11:06 AM

#

well

#

technically you can

#

we just go sequentially

dim oracle Apr 12, 2023, 11:06 AM

#

Yeah yeah, but just hypthetically

#

So what I said was right and the docs are wrong?

#

Or what

stable hatch Apr 12, 2023, 11:07 AM

#

yes, hypothetically you COULD spawn 0..7 and 40..47 at the same time

#

because they have different ratelimit keys

#

(which is what discord calls them, not buckets)

sullen snow Apr 12, 2023, 11:07 AM

#

wait what

stable hatch Apr 12, 2023, 11:07 AM

#

someone hasn't read the docs page for big sharding

sullen snow Apr 12, 2023, 11:07 AM

#

what we do is we spawn shards at the same "bucket"

stable hatch Apr 12, 2023, 11:08 AM

#

https://discord.com/developers/docs/topics/gateway#sharding-max-concurrency

stable hatch Apr 12, 2023, 11:08 AM

#

sullen snow what we do is we spawn shards at the same "bucket"

........?

#

so you spawn

sullen snow Apr 12, 2023, 11:08 AM

#

that wasn't there before

stable hatch Apr 12, 2023, 11:08 AM

#

that has been there for months

#

lol

#

Dare I even say years

sullen snow Apr 12, 2023, 11:09 AM

#

is it? it was never even noticed there

stable hatch Apr 12, 2023, 11:09 AM

#

it has been there for a LONG while,yeah

sullen snow Apr 12, 2023, 11:09 AM

#

wouldn't be asking questions on some big bot managers guild if it was there all the time

#

LUL

stable hatch Apr 12, 2023, 11:09 AM

#

sullen snow what we do is we spawn shards at the same "bucket"

so let me get this straight

sullen snow Apr 12, 2023, 11:09 AM

#

here I'll show you the actual stuff

stable hatch Apr 12, 2023, 11:10 AM

#

you're spawning shards 0, 16, 32, ... in one process?

sullen snow Apr 12, 2023, 11:10 AM

#

            const bucket = shardId % concurrency;
            const cooldown = BASE + OFFSET;
            let unlocked = await cache.redis.set(`current:lock:${bucket}`, 1, 'PX', cooldown, 'NX');
            while(!unlocked) {
                unlocked = await cache.redis.set(`current:lock:${bucket}`, 1, 'PX', cooldown, 'NX');
                await WaitAdmiral(128);
            }
            return message.success(true);

#

basically once you set that bucket on redis

stable hatch Apr 12, 2023, 11:10 AM

#

I'm confused what you're doing

sullen snow Apr 12, 2023, 11:11 AM

#

it will connect allow all the shards on that bucket to connect

stable hatch Apr 12, 2023, 11:12 AM

#

ok i'm confused why you're doing it

#

when instead you could, and should, spawn your clusters in batches of 16 (or w/e max_concurrency is)

#

and wait for them to ready up

sullen snow Apr 12, 2023, 11:12 AM

#

we just spawn our clusters regardless, and keep them in "waiting" mode until its their ws's turn to connect

stable hatch Apr 12, 2023, 11:12 AM

#

but we handle it for you

#

sullen snow Apr 12, 2023, 11:13 AM

#

nope

#

in some sense yes

stable hatch Apr 12, 2023, 11:13 AM

#

in all sense, yes

sullen snow Apr 12, 2023, 11:13 AM

#

like for example, if you have 16 shards

#

in that cluster

#

it will handle that

#

but we don't care about the number of the shards in the cluster

dim oracle Apr 12, 2023, 11:13 AM

#

I think there's still confusion between what a cluster is here

stable hatch Apr 12, 2023, 11:14 AM

#

~~bc y'all have 50 layers of namings~~

sullen snow Apr 12, 2023, 11:14 AM

#

i mean

#

node named it clusters

#

so we call it clusters

#

KEKW

#

ok let me clear this up

dim oracle Apr 12, 2023, 11:14 AM

#

just show them a code sample maybe? dunno what would help to clear things up

sullen snow Apr 12, 2023, 11:15 AM

#

d.js doesnt handle max_concurrency in a way that all clusters are in sync

stable hatch Apr 12, 2023, 11:15 AM

#

because we expect shards to be linear, as you should to speed up boots as much as possible

#

instead of spreading them around so hard

sullen snow Apr 12, 2023, 11:15 AM

#

what you mean by spreading them around?

#

if you mean I spawn a lot of clusters with little shards on it, we don't do that

#

in fact we cram as much as shards as possible in a single process

stable hatch Apr 12, 2023, 11:16 AM

#

well it sounds like your clusters are keyed by the ratelimit_key

#

instead of spawning a cluster with 0..15, one with 16..31, and so on

dusty dove Apr 12, 2023, 11:17 AM

#

oh i did not see that portion in the docs

#

like ever

stable hatch Apr 12, 2023, 11:17 AM

#

yall dead

sullen snow Apr 12, 2023, 11:17 AM

#

see dd also related

stable hatch Apr 12, 2023, 11:17 AM

#

do you not read docs OMEGALUL

dusty dove Apr 12, 2023, 11:17 AM

#

i do i just never found this somehow

#

i think i looked for it on another page

sullen snow Apr 12, 2023, 11:17 AM

#

i wont be confused about max_concurrency if it was there all along

dim oracle Apr 12, 2023, 11:17 AM

#

can we easily find in the github repo when that part was added

#

just to be sure

stable hatch Apr 12, 2023, 11:17 AM

#

dusty dove i think i looked for it on another page

on what other page other than gateway would you have looked

sullen snow Apr 12, 2023, 11:17 AM

#

anyways, our system doesnt spawn shards in a cluster based on max_concurrency

dusty dove Apr 12, 2023, 11:17 AM

#

idk

#

dont remember

#

doesnt matter

#

point is

sullen snow Apr 12, 2023, 11:18 AM

#

that means, you can keep the clusters equal to your # of threads of your cpu

dusty dove Apr 12, 2023, 11:18 AM

#

in theory i think they're right if they do stuff over the network

sullen snow Apr 12, 2023, 11:18 AM

#

so that means our shard count can reach 30 shards per cluster

stable hatch Apr 12, 2023, 11:18 AM

#

dim oracle can we easily find in the github repo when that part was added

Google_Chrome_Dev_-_2023-04-12_at_14.18.292x.png

#

2 years ago

dusty dove Apr 12, 2023, 11:18 AM

#

if they don't guarantee their clusters start sequentially

#

this will break yes

sullen snow Apr 12, 2023, 11:18 AM

#

now on boot, we just start everything we dont care about it

#

this is why we have a redis lock

#

where in shards will take turns on connecting

dusty dove Apr 12, 2023, 11:19 AM

#

right now with our current built in strategies things work though

dim oracle Apr 12, 2023, 11:19 AM

#

stable hatch

welp, lmao

dusty dove Apr 12, 2023, 11:19 AM

#

if you just run them just like that

stable hatch Apr 12, 2023, 11:19 AM

#

also

#

ngl

#

you could probably make all shards in one process with worker strategy and 16 per worker

sullen snow Apr 12, 2023, 11:19 AM

#

like with this setup, how fast our boot up time again @dim oracle

dusty dove Apr 12, 2023, 11:19 AM

#

ok yeah sec

#

let me test this but not do 1440 workers

sullen snow Apr 12, 2023, 11:19 AM

#

stable hatch you could probably make all shards in one process with worker strategy and 16 pe...

but is not efficient as I designed the code to be able to even handle 50k + guilds in single process

dusty dove Apr 12, 2023, 11:20 AM

#

since that just crashed node last time

dim oracle Apr 12, 2023, 11:20 AM

#

in theory 1440 shards can be up in 7,5 minutes (1440 / 16 *5)

sullen snow Apr 12, 2023, 11:20 AM

#

thats why we need the redis to get them all sync up with the ratelimited identify

stable hatch Apr 12, 2023, 11:20 AM

#

sullen snow but is not efficient as I designed the code to be able to even handle 50k + guil...

if you spawn shards at discords recommendations you'd have 16k guilds

sullen snow Apr 12, 2023, 11:20 AM

#

which is what our version of max_concurrency does

dusty dove Apr 12, 2023, 11:21 AM

#

yeah tbh

#

i have no idea what is going on anymore

sullen snow Apr 12, 2023, 11:21 AM

#

we spawn shards on discord's recommendations

dusty dove Apr 12, 2023, 11:21 AM

#

im doing too many things at once

sullen snow Apr 12, 2023, 11:21 AM

#

we just control how many clusters we spawn

#

aka process

#

so each shard still has avg of 1000 guilds

stable hatch Apr 12, 2023, 11:21 AM

#

you know how discord is saying this ratelimit key exists?

#

uh

#

i might have debunked that

dim oracle Apr 12, 2023, 11:21 AM

#

Full boot this morning to the latest dev release took from 13:01:23 to 13:10:03 which is 8 minutes 40 seconds

stable hatch Apr 12, 2023, 11:22 AM

#

let me check it again

sullen snow Apr 12, 2023, 11:22 AM

#

DEBUG [Wed,04/12/23,11:10:34] ([Cluster 3 (2218313)]): Discord.JS; [WS => Proxy] [Info] Spawn settings
        Shards: [ 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111 ]
        Shard Count: 28
        Total Shards: 112

like eg I did a full restart today

#

this is a process

#

it contains 28 shards

#

max_concurrency is set to 1

#

so it means even I spawn this clusters all at the same time

#

I don't need to worry about exceeding the ratelimit key

#

as they are synched process wide

dusty dove Apr 12, 2023, 11:23 AM

#

im p sure discord is literally not enforcing those ratelimit keys

#

lol

stable hatch Apr 12, 2023, 11:23 AM

#

they are

#

but kinda

dusty dove Apr 12, 2023, 11:23 AM

#

idk, our impl doesn't seem right now that i see this docs page

stable hatch Apr 12, 2023, 11:23 AM

#

tried with ```
shardCount: 64,
shardIds: [0, 16, 32],

dusty dove Apr 12, 2023, 11:24 AM

#

yeah that should throw right?

stable hatch Apr 12, 2023, 11:24 AM

#

📎 message.txt

#

it didn't

#

but also kinda did?

dusty dove Apr 12, 2023, 11:24 AM

#

yeah so its not enforced

#

oh

stable hatch Apr 12, 2023, 11:24 AM

#

out of the 3, shard 16 got invalid session

dusty dove Apr 12, 2023, 11:24 AM

#

Invalid session; will attempt to resume: false
def did

#

yeah odd that its not both that got it

stable hatch Apr 12, 2023, 11:24 AM

#

even tho realistically both 32 and 16 should've invalid session'd

dusty dove Apr 12, 2023, 11:24 AM

#

hmm

#

max conc 16 right?

stable hatch Apr 12, 2023, 11:24 AM

#

ye

#

its our ws tester

dusty dove Apr 12, 2023, 11:25 AM

#

no wait

#

you're wrong

#

just 16 should've

#

0 % 16 = 0
32 % 16 = 0

#

0 and 32 share

#

oh wait

#

off by one errors

stable hatch Apr 12, 2023, 11:25 AM

#

girl

dusty dove Apr 12, 2023, 11:25 AM

#

classic programmer mistsake

stable hatch Apr 12, 2023, 11:25 AM

#

your math high

#

0, 16 and 32, all modulo 16 is 0

dusty dove Apr 12, 2023, 11:25 AM

#

yeah

#

lol

stable hatch Apr 12, 2023, 11:26 AM

#

which means 2 of them should've invalid session'd

#

but they didn't

dusty dove Apr 12, 2023, 11:26 AM

#

no

#

wait

#

oh my god

stable hatch Apr 12, 2023, 11:26 AM

#

now i tried it again and now it was 32 that did

#

so

#

Discord's enforcement of this is dead

#

I dont

dusty dove Apr 12, 2023, 11:27 AM

#

okay, I follow now

#

yes so our impl is incomplete

stable hatch Apr 12, 2023, 11:27 AM

#

yeah but also discord is not properly handling it

dusty dove Apr 12, 2023, 11:27 AM

#

like many other things

#

either way

#

waitForIdentify should give a shardId param to support allowing this properly for their cluster setup/over the network stuff

sullen snow Apr 12, 2023, 11:27 AM

#

yes

dusty dove Apr 12, 2023, 11:27 AM

#

and the identifythrottler needs to account for it too for our builtin strategies

sullen snow Apr 12, 2023, 11:27 AM

#

finally master dd

#

but yes, this is why we rarely get invalid sessions even our cluster shard count is wack

#

or do we even get those

#

LUL

stable hatch Apr 12, 2023, 11:28 AM

#

dusty dove `waitForIdentify` should give a shardId param to support allowing this properly ...

KINDA?

#

i mean

#

it still wont help

dusty dove Apr 12, 2023, 11:28 AM

#

wut

sullen snow Apr 12, 2023, 11:28 AM

#

but im fine with just passing a custom identifyThrottler

dusty dove Apr 12, 2023, 11:28 AM

#

we just need to also account what ratelimit key each shard has

stable hatch Apr 12, 2023, 11:28 AM

#

I mean ok, how will you account max concurrency for the shard bucket?

sullen snow Apr 12, 2023, 11:28 AM

#

the devs should be the one who implement it over the network

stable hatch Apr 12, 2023, 11:28 AM

#

ok dd but they spawn the same bucket in a process

#

so in the end there is no max_concurrency to handle, only 5s wait between identifies

#

it needs to be done but gl handling that, and not opening the conn before identify is available

dusty dove Apr 12, 2023, 11:29 AM

#

🤔

#

i dont think its that complicated

stable hatch Apr 12, 2023, 11:30 AM

#

ok let me give you these two test cases then

#

on our ws tester

dusty dove Apr 12, 2023, 11:30 AM

#

the only tricky case if you like

#

never want one of the shards AT ALL

sullen snow Apr 12, 2023, 11:30 AM

#

not opening conn before identify is available?

stable hatch Apr 12, 2023, 11:30 AM

#

given shardCount: 64
shards 0..15 should all start and identify at once
shard 0, 16, 32, 48 should each identify and wait 5s between each other

sullen snow Apr 12, 2023, 11:30 AM

#

right now, ws opens con even ident is not available it just not identifies

stable hatch Apr 12, 2023, 11:31 AM

#

yeah which might have its own side effects

dusty dove Apr 12, 2023, 11:31 AM

#

yeah vlad but thats what i mean

#

or like

#

okay wait

stable hatch Apr 12, 2023, 11:31 AM

#

stable hatch given shardCount: 64 shards 0..15 should all start and identify at once shard 0,...

no i just gave you params to test your changes

#

idc what you do to make it work OMEGALUL

dusty dove Apr 12, 2023, 11:31 AM

#

no but clarify

#

so just shardCount: 64?

#

no other params?

stable hatch Apr 12, 2023, 11:31 AM

#

shardCount: 64
shardIds: the numbers there

dusty dove Apr 12, 2023, 11:32 AM

#

yeah see but this makes no sense as an isolated test case with our built in strategies

#

thats what i was saying

stable hatch Apr 12, 2023, 11:32 AM

#

it doesnt need to make sense

dusty dove Apr 12, 2023, 11:32 AM

#

this is the tricky one

stable hatch Apr 12, 2023, 11:32 AM

#

it needs to work

dusty dove Apr 12, 2023, 11:32 AM

#

why

#

what use case are you covering

#

other than "I never want shard X to exist"

stable hatch Apr 12, 2023, 11:32 AM

#

what

#

Deadge

#

the point is testing the identify throttler

#

not

#

"i dont want shard x"

dusty dove Apr 12, 2023, 11:32 AM

#

yes, but this test makes no sense with how it should work

stable hatch Apr 12, 2023, 11:33 AM

#

How so?

dusty dove Apr 12, 2023, 11:33 AM

#

well the way I imagine it is

sullen snow Apr 12, 2023, 11:33 AM

#

well if you want to test this

stable hatch Apr 12, 2023, 11:33 AM

#

It tests both ranges that wouldn't have any ratelimit_keys clashing (0..15, each element modulo 16 is a diff number), as well as spawning shards that share the same ratelimit_key (thus each id modulo 16 is 0 for example)

#

idk if you're missing something or i suck at explaining lmao

sullen snow Apr 12, 2023, 11:35 AM

#

I recommend to just test this without exact max_concurrency values on each cluster shard count
lets say you have 16 max_concurrency.

have first process have 10 shards, second 10 shards, then remaining on last process

then spawn all the clusters at the same time.

now 10 shards 1st cluster and 6 shards on 2nd cluster should spawn at the same time
wait 5 seconds

now last shards on 2nd cluster and 3rd cluster should spawn

dusty dove Apr 12, 2023, 11:35 AM

#

no we're just thinking of diff ways to implement i think

#

let me pseudocode

sullen snow Apr 12, 2023, 11:35 AM

#

thats what the redis impl. solves on our case

stable hatch Apr 12, 2023, 11:36 AM

#

shardCount: 64
shardIds: {start: 0, end: 16 (or 15, idfk how our range works LMAO)}

All shards should identify at once and none should invalid session

shardCount: 64
shardIds: [0, 16, 32, 48]

Shard should identify and wait 5s before the next one identifies, none should invalid session

sullen snow Apr 12, 2023, 11:36 AM

#

thats what I learned on ~~some python dev that has a big bot on what he does~~

stable hatch Apr 12, 2023, 11:36 AM

#

thats what I want you to test

#

how you make it work is up to you bc you're smart

dusty dove Apr 12, 2023, 11:36 AM

#

waitForIdentify(shardId: number): Promise<void>

const key = formulaHere;
const shardsInKey = this.whatever.get(key); // whatever is a Map<key, number[]>
const promise = new Promise((resolve) => this.promises.set(shardId, resolve));

shardsInKey.push(shardId);

if (shardsInKey.length === 16) {
  for (const shard of shardsInKey) {
    this.promises.get(shard).resolve();
    this.promises.delete(shard);
  }
}

return promise;```

stable hatch Apr 12, 2023, 11:36 AM

#

but those two test cases need to work sweats

dusty dove Apr 12, 2023, 11:36 AM

#

you get the idea?

stable hatch Apr 12, 2023, 11:36 AM

#

idc about the code behind it dd

#

I really couldn't care less

#

KEKW

dusty dove Apr 12, 2023, 11:36 AM

#

no but what I'm saying is

stable hatch Apr 12, 2023, 11:36 AM

#

thats on you

dusty dove Apr 12, 2023, 11:37 AM

#

god you won't listen at all

stable hatch Apr 12, 2023, 11:37 AM

#

and you're ignore me

dusty dove Apr 12, 2023, 11:37 AM

#

I've read everything u sent, im just trying to tell you that

#

shardCount: 64 shardIds: [0, 16, 32, 48]

will literally never resolve with how I want to do it with our current strategies

stable hatch Apr 12, 2023, 11:37 AM

#

our strategies have nothing to do with the identify throttler

dusty dove Apr 12, 2023, 11:38 AM

#

because to make this work we need to make the IdentifyThrottler wait for all max_concurrency shards to attempt their Identify call before we send off the batch

#

they literally do though

stable hatch Apr 12, 2023, 11:38 AM

#

well sounds like you need to redo it then shrug

dusty dove Apr 12, 2023, 11:38 AM

#

no, there's no other way

#

lol

stable hatch Apr 12, 2023, 11:38 AM

#

stable hatch shardCount: 64 shardIds: {start: 0, end: 16 (or 15, idfk how our range works LMA...

these two tests need to pass

#

thats all I'll say

sullen snow Apr 12, 2023, 11:38 AM

#

and i assume

#

you cant use redis here?

dusty dove Apr 12, 2023, 11:38 AM

#

you can

#

thats what im trying to tell him

#

but lol

#

donowall

sullen snow Apr 12, 2023, 11:38 AM

#

I think you should just have

#

2 in built strategies

#

then 3rd = custom to the 3rd party dev

dusty dove Apr 12, 2023, 11:39 AM

#

like

sullen snow Apr 12, 2023, 11:39 AM

#

1 will use redis
1 will use not
1 will use whatever they like or make

#

I dont think we can cover your tests in a single strategy here

dusty dove Apr 12, 2023, 11:39 AM

#

shardIds: [0, 16, 32, 48]
this does not make sense, if we want to benefit from max_concurrency then the throttling implementation needs to wait for shards 1..15 to have the full 0..16 pack before it sends off the batch

stable hatch Apr 12, 2023, 11:39 AM

#

dusty dove <a:donowall:741699151544451092>

and I gave you test cases that should pass but you're worse than a brick wall

dusty dove Apr 12, 2023, 11:39 AM

#

it makes sense if you had more managers on different processes/places, yes

sullen snow Apr 12, 2023, 11:40 AM

#

which is what we do ^

dusty dove Apr 12, 2023, 11:40 AM

#

and you'd need a custom strategy to make it work

#

but as an isolated test case

stable hatch Apr 12, 2023, 11:40 AM

#

No you wouldn't

dusty dove Apr 12, 2023, 11:40 AM

#

I literally cannot test that

stable hatch Apr 12, 2023, 11:40 AM

#

you'd need the identify throttler to handle it

#

yes you can

#

Deadge

#

you give it those ids

#

you make it connect

dusty dove Apr 12, 2023, 11:40 AM

#

??????????? dude what

stable hatch Apr 12, 2023, 11:40 AM

#

you see it doesnt invalid session

#

you profit

#

whats so hard

#

You literally build a map of ratelimit_key -> remainingConcurrency

#

or well

dusty dove Apr 12, 2023, 11:40 AM

#

yes

sullen snow Apr 12, 2023, 11:40 AM

#

oh well, unless you have access to the "main process" of whatever sharder they use

dusty dove Apr 12, 2023, 11:40 AM

#

and you're telling me

stable hatch Apr 12, 2023, 11:41 AM

#

no, ratelimit_key -> 1

dusty dove Apr 12, 2023, 11:41 AM

#

to connect [0, 16, 32, 48]

stable hatch Apr 12, 2023, 11:41 AM

#

if its not 1, you wait 5s

dusty dove Apr 12, 2023, 11:41 AM

#

so, when am I supposed to decide when to connect just 0

#

because 1..15 aren't coming through

#

which are in the same bucket

stable hatch Apr 12, 2023, 11:41 AM

#

they're NOT

dusty dove Apr 12, 2023, 11:41 AM

#

that is what im telling you makes no sense here

stable hatch Apr 12, 2023, 11:41 AM

#

omg

#

they are not

dusty dove Apr 12, 2023, 11:41 AM

#

"bucket"

#

terminology

stable hatch Apr 12, 2023, 11:41 AM

#

ratelimit_key = shard_id % concurrency

dusty dove Apr 12, 2023, 11:41 AM

#

This puts your shards into "buckets" of max_concurrency size. When you start your bot, you may start up to max_concurrency shards at a time, and you must start them by "bucket" in order. To explain another way, let's say you have 16 shards, and your max_concurrency is 16:

stable hatch Apr 12, 2023, 11:41 AM

#

0 -> 0
1 -> 1
...
16 -> 0

dusty dove Apr 12, 2023, 11:41 AM

#

they do use the term bucket on the docs

#

okay, so let's be clear

#

when I say bucket

#

I refer to shards that need to identify at once, yes?

#

agreed?

stable hatch Apr 12, 2023, 11:42 AM

#

no

dusty dove Apr 12, 2023, 11:42 AM

#

that is literally what it says man

#

oh my god

#

do you want to argue over the english language

#

or semantics

stable hatch Apr 12, 2023, 11:42 AM

#

do you not read the whole fucking codeblock

#

where they explain what a bucket is

#

lmao

#

zoomers i s2g

dusty dove Apr 12, 2023, 11:42 AM

#

k

#

you know what

#

implement it yourself

#

im not dealing with it if this is the attitude i get

stable hatch Apr 12, 2023, 11:43 AM

#

shard_id: 0, rate limit key (0 % 16): 0
shard_id: 1, rate limit key (1 % 16): 1
shard_id: 2, rate limit key (2 % 16): 2
shard_id: 3, rate limit key (3 % 16): 3
shard_id: 4, rate limit key (4 % 16): 4
shard_id: 5, rate limit key (5 % 16): 5
shard_id: 6, rate limit key (6 % 16): 6
shard_id: 7, rate limit key (7 % 16): 7
shard_id: 8, rate limit key (8 % 16): 8
shard_id: 9, rate limit key (9 % 16): 9
shard_id: 10, rate limit key (10 % 16): 10
shard_id: 11, rate limit key (11 % 16): 11
shard_id: 12, rate limit key (12 % 16): 12
shard_id: 13, rate limit key (13 % 16): 13
shard_id: 14, rate limit key (14 % 16): 14
shard_id: 15, rate limit key (15 % 16): 15
shard_id: 16, rate limit key (16 % 16): 0
shard_id: 17, rate limit key (17 % 16): 1
shard_id: 18, rate limit key (18 % 16): 2
shard_id: 19, rate limit key (19 % 16): 3
shard_id: 20, rate limit key (20 % 16): 4
shard_id: 21, rate limit key (21 % 16): 5
shard_id: 22, rate limit key (22 % 16): 6
shard_id: 23, rate limit key (23 % 16): 7
shard_id: 24, rate limit key (24 % 16): 8
shard_id: 25, rate limit key (25 % 16): 9
shard_id: 26, rate limit key (26 % 16): 10
shard_id: 27, rate limit key (27 % 16): 11
shard_id: 28, rate limit key (28 % 16): 12
shard_id: 29, rate limit key (29 % 16): 13
shard_id: 30, rate limit key (30 % 16): 14
shard_id: 31, rate limit key (31 % 16): 15

> 
> In this case, you must start the shard buckets **in "order"**. That means that you can start shard 0 -> shard 15 concurrently, and then you can start shard 16 -> shard 31.

dusty dove Apr 12, 2023, 11:43 AM

#

or get someone actually willing to have a conversation without snarky remarks every 10 messages

#

maybe space or kyra

sullen snow Apr 12, 2023, 11:43 AM

#

well tbh

#

you cant implement this easily without something that could make it lock

#

I tried without using redis or smth and my mind came out blank

#

so thats just my 2 cents here

dusty dove Apr 12, 2023, 11:44 AM

#

of course, but making it work with redis is your responsibility

sullen snow Apr 12, 2023, 11:44 AM

#

thats why I never added support for max_concurrency on my sharder package

dusty dove Apr 12, 2023, 11:44 AM

#

considering you're using custom strategies

stable hatch Apr 12, 2023, 11:44 AM

#

I mean if you're gonna spawn random shards across processes then yeah there's no way for us to keep it in sync

sullen snow Apr 12, 2023, 11:44 AM

#

yes but if you can make that work with a "lock" that can substitute from redis it may work

#

the question is what

stable hatch Apr 12, 2023, 11:44 AM

#

..unless we pull a sessionInfo and make it a method you can use?

dusty dove Apr 12, 2023, 11:44 AM

#

if you want to make that work saya, you need a single dedicated process just to deal with that

stable hatch Apr 12, 2023, 11:45 AM

#

dd would that work?

dusty dove Apr 12, 2023, 11:45 AM

#

kinda like its own microservice

sullen snow Apr 12, 2023, 11:45 AM

#

i mean that calculation we do is offloaded on a thread

#

and we just do a await promise() to wait for shards to connect on where discordjsws is

dusty dove Apr 12, 2023, 11:45 AM

#

so the idea is

#

if you want to do it as efficiently as you can

#

you need a central place that aggregates this state

#

the way you should do it is like

#

imagine you have a message broker

#

and you send it a shard id

#

and you create a promise

#

and you resolve it when you get a reply through the broker

#

like, it sends you that shard id back

stable hatch Apr 12, 2023, 11:46 AM

#

stable hatch ..unless we pull a sessionInfo and make it a method you can use?

dd PepeHands

sullen snow Apr 12, 2023, 11:46 AM

#

yes thats how technically it works

dusty dove Apr 12, 2023, 11:46 AM

#

meanwhile in the broker end

sullen snow Apr 12, 2023, 11:46 AM

#

just in other layman terms on us

dusty dove Apr 12, 2023, 11:46 AM

#

you have one unique consumer for the queue

#

and once it has all the shards of a given batch together

#

like [0, 15]

#

it fires the messages for those 16 shards

#

makes sense?

dusty dove Apr 12, 2023, 11:47 AM

#

stable hatch ..unless we pull a sessionInfo and make it a method you can use?

what do you mean

#

waitForIdentify?

sullen snow Apr 12, 2023, 11:47 AM

#

yes that makes sense

stable hatch Apr 12, 2023, 11:47 AM

#

i mean

dusty dove Apr 12, 2023, 11:47 AM

#

no im asking

stable hatch Apr 12, 2023, 11:47 AM

#

we could make the throttler extensible

dusty dove Apr 12, 2023, 11:47 AM

#

because it already is a strategy method

#

exactly because of this

#

lol

#

is what i was telling you

sullen snow Apr 12, 2023, 11:48 AM

#

yes thats what I was asking since we made the thread

dusty dove Apr 12, 2023, 11:48 AM

#

export interface IContextFetchingStrategy {
    readonly options: FetchingStrategyOptions;
    retrieveSessionInfo(shardId: number): Awaitable<SessionInfo | null>;
    updateSessionInfo(shardId: number, sessionInfo: SessionInfo | null): Awaitable<void>;
    waitForIdentify(): Promise<void>;
}```

#

for the built in ones that just calls an identify throttler

#

well

#

simple calls one

stable hatch Apr 12, 2023, 11:48 AM

#

OK but what if you wanna use current strategies with a diff throttler

dusty dove Apr 12, 2023, 11:48 AM

#

worker sends a message to the parent thread

#

oh

#

yeah but why is my question

stable hatch Apr 12, 2023, 11:49 AM

#

i mean we have a use case right here that answers your why

dusty dove Apr 12, 2023, 11:49 AM

#

yeah but

#

I think we're thinking of different places to decouple the logic

sullen snow Apr 12, 2023, 11:49 AM

#

right now I "block" the original op on the worker strategy and foward it to my ws manager class which contains my custom throttler

dusty dove Apr 12, 2023, 11:49 AM

#

to me your waitForIdentify() should be responsible for throtting logic

#

our IdentifyThrottler is its own class because we use it in 2 places

#

its just a sane internal impl

stable hatch Apr 12, 2023, 11:49 AM

#

mostly sane, we need to handle the ratelimit_key too if someone does [0, 1, 2, 3, 15, 16, 17]

dusty dove Apr 12, 2023, 11:49 AM

#

you shouldnt really need a different one with our built in strategies

#

like

#

if they need to send their identify requests over redis

#

that already won't work with the default worker strategy

#

they need to extend it

sullen snow Apr 12, 2023, 11:50 AM

#

once waitForIdentify has the shardId, I can just extend the waitForIdentify function

#

without me doing any hacks

#

and give it to the sharding strategy

dusty dove Apr 12, 2023, 11:50 AM

#

ye

#

either way

#

back to the tests topic

#

let's start over

stable hatch Apr 12, 2023, 11:51 AM

#

Cat_Shake_Yes

dusty dove Apr 12, 2023, 11:51 AM

#

i'll use the term batches as to not irritate anyone

#

so you have numbers like

#

0..15

#

that is a batch

#

16..31

#

another batch

#

they go together at once

#

yea?

stable hatch Apr 12, 2023, 11:51 AM

#

no..?

dusty dove Apr 12, 2023, 11:51 AM

#

ded

#

y

stable hatch Apr 12, 2023, 11:51 AM

#

you wait for first batch to finish

#

then start the second one

dusty dove Apr 12, 2023, 11:51 AM

#

oh yes

#

thats what i mean

#

a batch is a group of shards that go at once

stable hatch Apr 12, 2023, 11:51 AM

#

dead

dusty dove Apr 12, 2023, 11:51 AM

#

mfw semantics

#

anyway

stable hatch Apr 12, 2023, 11:51 AM

#

pls english

dusty dove Apr 12, 2023, 11:52 AM

#

so

dim oracle Apr 12, 2023, 11:52 AM

#

wait to finish what, just wait 5s and start the next

dusty dove Apr 12, 2023, 11:52 AM

#

ye

#

so the thing is

#

if you create a manager with [0, 1, 2, 3, 15, 16, 17] and no other manager(s)

#

im never completing the 0..15 batch

#

and in my mind that isn't a usecase we should... support

#

there's 2 cases it doesn't get completed

#

faulty strategy - not our problem as long as its not the default ones

#

or 2, someone is trying to never connect X or Y shard - makes no sense to account for that

stable hatch Apr 12, 2023, 11:53 AM

#

dusty dove and in my mind that isn't a usecase we should... support

Its not a usecase we want to support but its a usecase we should gracefully handle

dusty dove Apr 12, 2023, 11:53 AM

#

but shoooooould we

stable hatch Apr 12, 2023, 11:54 AM

#

Bc right now 16 and 17 will just invalid session

#

Yes we should

dusty dove Apr 12, 2023, 11:54 AM

#

I feel like we might end up trading perf in some ways if we do

stable hatch Apr 12, 2023, 11:54 AM

#

Girl

dusty dove Apr 12, 2023, 11:54 AM

#

no no but what i mean is

#

what do we do if a batch is never completed

stable hatch Apr 12, 2023, 11:54 AM

#

dusty dove I feel like we might end up trading perf in some ways if we do

You don't call identify every second

dusty dove Apr 12, 2023, 11:54 AM

#

because i was thinking "the promise just never resolves" lol

#

like, that makes sense to me

stable hatch Apr 12, 2023, 11:54 AM

#

Adding on a few ms won't kill the world

dusty dove Apr 12, 2023, 11:54 AM

#

dusty dove what do we do if a batch is never completed

.

#

and how do we determine what "never" means

#

how long?

stable hatch Apr 12, 2023, 11:54 AM

#

Ok but like

dusty dove Apr 12, 2023, 11:54 AM

#

this complicates the logic so much

#

this is what i was saying

stable hatch Apr 12, 2023, 11:54 AM

#

One thing you're thinking differently than discord

#

Is in batches

#

Whereas discord thinks in ratelimit keys

#

Which is what we should also do

dusty dove Apr 12, 2023, 11:55 AM

#

idk

#

In this case, you must start the shard buckets in "order". That means that you can start shard 0 -> shard 15 concurrently, and then you can start shard 16 -> shard 31.

stable hatch Apr 12, 2023, 11:55 AM

#

You have 0..15 a batch

#

Each element in the batch has its own key

dusty dove Apr 12, 2023, 11:55 AM

#

nvm

#

I have a new idea

#

i got it

stable hatch Apr 12, 2023, 11:56 AM

#

Shoot away

#

Realistically we need to consider the concurrency and the ratelimit key

#

But I'll wait to see your idea

dusty dove Apr 12, 2023, 12:01 PM

#

it has come to me via divine revelation

stable hatch Apr 12, 2023, 12:01 PM

#

Bit early, easter is in a few days

dim oracle Apr 12, 2023, 12:03 PM

#

stable hatch Bit early, easter is in a few days

I'm not even going to question what you mean by that

stable hatch Apr 12, 2023, 12:03 PM

#

(divine revelation, our easter is this friday/monday)

#

its a bad joke

dim oracle Apr 12, 2023, 12:04 PM

#

Never knew easter was celebrated on different days across the world

#

TIL

stable hatch Apr 12, 2023, 12:05 PM

#

Different easters

#

Catholic and Christian easter

dim oracle Apr 12, 2023, 12:05 PM

#

apparently there's 5

stable hatch Apr 12, 2023, 12:09 PM

#

dusty dove it has come to me via divine revelation

Will you share with the class or

dusty dove Apr 12, 2023, 12:11 PM

#

no im having lunch

#

then ill write it

dusty dove Apr 12, 2023, 1:23 PM

#

i lied im working on something else

#

i might do it today still though

dusty dove Apr 13, 2023, 10:13 AM

#

@stable hatch yeah nvm this is unfeasible

#

if we could like

#

better guarantee the order of shards requesting identifies

#

it would work

#

otherwise this is insane

#

we bounce so much between incomplete batches

#

I'm like 99% sure its best we just do what I described where it waits for completed batches

stable hatch Apr 13, 2023, 10:17 AM

#

Is doing what I suggested that bad?

#

@dusty dove

#

👀

dusty dove Apr 13, 2023, 10:18 AM

#

stable hatch Is doing what I suggested that bad?

re-iterate

stable hatch Apr 13, 2023, 10:19 AM

#

map of ratelimit key and if it was used + when it resets
Get the ratelimit key from the shard id that wants to identify
If its used, wait, otherwise try sending it

#

Optimal? Not even by a long shot
But would it work? Probably

dusty dove Apr 13, 2023, 10:20 AM

#

i think i just mind blocked

#

since thats basically what i was trying to do w/o having even gone over what u said again

#

but this doesn't feel right

#

oh

#

yeah i see now

#

let me tweak that

#

export class IdentifyThrottler {
    private readonly states = new Collection<number, SessionState>();

    public constructor(private readonly manager: WebSocketManager) {}

    public async waitForIdentify(shardId: number): Promise<void> {
        const info = await this.manager.fetchGatewayInformation();
        const key = shardId % info.session_start_limit.max_concurrency;

        const state = this.states.ensure(key, () => ({
            queue: new AsyncQueue(),
            remaining: info.session_start_limit.max_concurrency,
            resetsAt: Date.now() + 5_000,
        }));

        await state.queue.wait();

        try {
            if (state.remaining <= 0) {
                const diff = state.resetsAt - Date.now();
                if (diff <= 5_000) {
                    // To account for the latency the IDENTIFY payload goes through, we add a bit more wait time
                    const time = diff + Math.random() * 1_500;
                    await sleep(time);
                }

                state.remaining = info.session_start_limit.max_concurrency;
                state.resetsAt = Date.now() + 5_000;
            }

            state.remaining--;
        } finally {
            state.queue.shift();
        }
    }
}``` @stable hatch 👀

#

test('basic case', async () => {
    fetchGatewayInformation.mockReturnValue({
        session_start_limit: {
            max_concurrency: 2,
            reset_after: 5_000,
        },
    });

    // Those shouldn't wait since they're in different keys

    await throttler.waitForIdentify(0);
    expect(sleep).not.toHaveBeenCalled();

    await throttler.waitForIdentify(1);
    expect(sleep).not.toHaveBeenCalled();

    // Those shouldn't wait either since they're in the same key and max_concurrency is 2

    await throttler.waitForIdentify(2);
    expect(sleep).not.toHaveBeenCalled();

    await throttler.waitForIdentify(3);
    expect(sleep).not.toHaveBeenCalled();

    // These should wait since the key is full

    await throttler.waitForIdentify(4);
    expect(sleep).toHaveBeenCalledTimes(1);

    await throttler.waitForIdentify(5);
    expect(sleep).toHaveBeenCalledTimes(2);
});

#

passes

stable hatch Apr 13, 2023, 10:34 AM

#

Smexy

#

I'll review fully when back home

dusty dove Apr 13, 2023, 10:36 AM

#

https://github.com/discordjs/discord.js/pull/9375

GitHub

refactor: properly support max_concurrency ratelimit keys by didine...

Please describe the changes this PR makes and why it should be merged:
Status and versioning classification:

Code changes have been tested against the Discord API, or there are no code changes
I k...

#

@dim oracle @sullen snow this also adds the new param

#

you'll still need custom waitForIdentify for your cluster setup though

#

the default identifythrottler obviously cant deal with cross process stuff

stable hatch Apr 13, 2023, 10:38 AM

#

@dusty dove can you hunor me and test the two cases I gave you live?

#

Pls

dusty dove Apr 13, 2023, 10:38 AM

#

uhh

#

let me find them again

stable hatch Apr 13, 2023, 10:39 AM

#

Shards 0..15 instant

#

Shard 0,16,32 wait

dusty dove Apr 13, 2023, 10:41 AM

#

stable hatch shardCount: 64 shardIds: {start: 0, end: 16 (or 15, idfk how our range works LMA...

#

➜ node max-concurrency-test-1.js
Hello 2
Hello 5
Hello 6
Hello 9
Hello 4
Hello 7
Hello 14
Hello 11
Hello 13
Hello 8
Hello 0
Hello 3
Hello 1
Hello 10
Hello 12
Hello 15
Ready 14
Ready 11
Ready 6
Ready 2
Ready 9
Ready 5
Ready 3
Ready 0
Ready 8
Ready 4
Ready 1
Ready 13
Ready 10
Ready 12
Ready 7
Ready 15```

#

first test seems fine

#

if thats all

#

aaa

#

no this is wrong

#

beeproll

#

asyncqueue is trolling me

#

wth

#

@stable hatch ok i fixed it

#

the 2nd case passes

stable hatch Apr 13, 2023, 10:55 AM

#

What about the first

dusty dove Apr 13, 2023, 10:55 AM

#

passes too

#

all run instantly

#

if i add an extra shard just that one waits

#

e.g. 0..16

➜ node max-concurrency-test-1.js
Hello 2
Hello 3
Hello 9
Hello 10
Hello 16
Hello 12
Hello 0
Sleeping 5875.688359300205 for shard 0
Hello 1
Hello 4
Hello 6
Hello 7
Hello 5
Hello 11
Hello 8
Hello 14
Hello 13
Ready 2
Ready 6
Ready 16
Ready 10
Ready 3
Ready 9
Ready 4
Ready 13
Ready 5
Ready 12
Ready 7
Ready 11
Ready 14
Ready 8
Ready 1
Hello 15
Ready 15
Ready 0```

#

@stable hatch hmmm

#

I have a question for u

#

do u think its fine if i refactored this

    public constructor(private readonly manager: WebSocketManager) {}

    public async waitForIdentify(shardId: number): Promise<void> {
        const info = await this.manager.fetchGatewayInformation();```

#

to like

#

public constructor(private readonly info: APIGatewayBotInfo) {}```

#

what's the worst that could happen

#

esp. considering how aggressively fetchGatewayInformation caches anyway

#

that would also make our class re-useable for kyoso and saya

stable hatch Apr 13, 2023, 11:07 AM

#

Hmm

#

You wouldn't have the cache expiry this way

dusty dove Apr 13, 2023, 11:07 AM

#

yeah

#

but how bad can that be

#

cus like

stable hatch Apr 13, 2023, 11:08 AM

#

Could make it instead take an interface that provides the fetching method for the gw info

#

Then we can pass in the manager, they can pass in w/e

dusty dove Apr 13, 2023, 11:08 AM

#

oh wait lol

stable hatch Apr 13, 2023, 11:08 AM

#

To fetch from redis or whatever

dusty dove Apr 13, 2023, 11:08 AM

#

i just looked at the code

#

we only use uh

#

https://sucks-to-b.eu/NWVKAW.png

#

one reference to it

#

and its for max concurrency

#

loool

#

can just take it in the constructor

#

that should be ok

stable hatch Apr 13, 2023, 11:08 AM

#

Oh yeah lol

dusty dove Apr 13, 2023, 11:09 AM

#

since ull always need to fully restart to change it anyway

stable hatch Apr 13, 2023, 11:09 AM

#

Yeah

dusty dove Apr 13, 2023, 11:09 AM

#

oh nooo

#

this also tries to emit a debug log

#

                this.manager.emit(WebSocketShardEvents.Debug, {
                    shardId,
                    message: `Waiting ${time}ms before IDENTIFYing`,
                });```

#

i had just added that one

#

PepeHands

#

to track sleep times

#

oh well

#

its not that important

sullen snow Apr 13, 2023, 11:11 AM

#

@dusty dove will check once I have time

stable hatch Apr 13, 2023, 11:11 AM

#

I'll also try the pr locally

#

Just to double check

sullen snow Apr 13, 2023, 11:11 AM

#

just clarifying something, this is extendible?

dusty dove Apr 13, 2023, 11:11 AM

#

yeah i need to fix some stuff up

dusty dove Apr 13, 2023, 11:11 AM

#

sullen snow just clarifying something, this is extendible?

what is

sullen snow Apr 13, 2023, 11:12 AM

#

the identify throttler

dusty dove Apr 13, 2023, 11:12 AM

#

you wont need to

#

but yes

#

sure

sullen snow Apr 13, 2023, 11:12 AM

#

why i wont need to?

#

is there some kind of built in function now? or something?

dusty dove Apr 13, 2023, 11:13 AM

#

cus you cant change what throttler we use internally anyway in our strategies

#

im just telling you that if you go the redis/message broker approach i told you about

#

you can use the class on the receiving end now

#

without having to re-invent the logic

#

since it doesnt take a manager in the constructor anymore

sullen snow Apr 13, 2023, 11:13 AM

#

wait im confused, give me a quick rundown on how it works?

stable hatch Apr 13, 2023, 11:13 AM

#

Tbf @dusty dove they'd still need to make their own

#

Because they want to pull from redis

dusty dove Apr 13, 2023, 11:13 AM

#

yeah, but thats just the strategy

dusty dove Apr 13, 2023, 11:14 AM

#

sullen snow wait im confused, give me a quick rundown on how it works?

will do in a sec

#

like, the actual sleep() logic they can re-use from us

#

is what im saying vlad

#

they dont need to re-inevnt that anymore

stable hatch Apr 13, 2023, 11:14 AM

#

I still feel like we should make our existing strategies have the ability to pass in a custom identify throttler

dusty dove Apr 13, 2023, 11:14 AM

#

they just need to change where the wait requests go to

#

look thru the code

#

and ull see why that doesnt make a lot of sense

#

haha

stable hatch Apr 13, 2023, 11:14 AM

#

Yes it makes sense

#

We've been through this

dusty dove Apr 13, 2023, 11:15 AM

#

no ehehe

stable hatch Apr 13, 2023, 11:15 AM

#

Having to reinvent strategies to replace just this one piece that's standalone is not ideal

#

Yes DD.

dusty dove Apr 13, 2023, 11:16 AM

#

no

#

like, you're 100% wrong here

stable hatch Apr 13, 2023, 11:16 AM

#

???

dusty dove Apr 13, 2023, 11:16 AM

#

lol

stable hatch Apr 13, 2023, 11:16 AM

#

How?

#

And do explain

dusty dove Apr 13, 2023, 11:16 AM

#

because as things stand right now

#

the goal of Strategy#waitForIdentify and similar methods, like the session fetching ones

#

are to do effectively do messaging to get a reply back

#

for instance

#

    public async waitForIdentify(shardId: number): Promise<void> {
        const nonce = Math.random();
        const payload = {
            op: WorkerReceivePayloadOp.WaitForIdentify,
            nonce,
            shardId,
        } satisfies WorkerReceivePayload;
        // eslint-disable-next-line no-promise-executor-return
        const promise = new Promise<void>((resolve) => this.waitForIdentifyPromises.set(nonce, resolve));
        parentPort!.postMessage(payload);
        return promise;
    }```

#

and on the other end

#

            case WorkerReceivePayloadOp.WaitForIdentify: {
                await this.throttler.waitForIdentify(payload.shardId);
                const response: WorkerSendPayload = {
                    op: WorkerSendPayloadOp.ShardCanIdentify,
                    nonce: payload.nonce,
                };
                worker.postMessage(response);
                break;
            }```

#

you never need to overwrite the throttler's behavior, is what I was saying

#

the throttler is just responsible for being that central point that figures out how long to wait

#

like, being able to pass in your own throttler class does not help at all

#

because the throttler is not responsible for messaging

#

they need a custom strategy because they do cross-process stuff over the network

stable hatch Apr 13, 2023, 11:18 AM

#

But dd you're missing the point

#

The throttler sleeps

#

That's what hangs identifies

#

They want to base the info from redis

#

They can't do that without overriding the throttler

#

Which is what sleeps

dusty dove Apr 13, 2023, 11:19 AM

#

well, they have to override the strategy first and foremost

#

to send the data over redis

stable hatch Apr 13, 2023, 11:19 AM

#

They don't have to tho

#

The strategy works perfectly fine

dusty dove Apr 13, 2023, 11:19 AM

#

?

stable hatch Apr 13, 2023, 11:19 AM

#

The throttler needs to be changed

#

Because that's ALL that changes

dusty dove Apr 13, 2023, 11:19 AM

#

you want to move this abstraction to the throttler

#

but i just dont agree

stable hatch Apr 13, 2023, 11:19 AM

#

Is when the throttler says "ya you can identify now"

dusty dove Apr 13, 2023, 11:19 AM

#

since that makes it inconsistent with everything else

stable hatch Apr 13, 2023, 11:19 AM

#

You don't move anything dd

#

Omg

dusty dove Apr 13, 2023, 11:20 AM

#

you literally are though, like, what im saying is, the strategies are actually responsible for throttling

#

straight up

#

the identifythrottler is just a convenient helper

stable hatch Apr 13, 2023, 11:20 AM

#

Ok but you're missing the point

dusty dove Apr 13, 2023, 11:20 AM

#

since we implement it in two places

#

I'm not. I understand what you're proposing and I wholeheartedly disagree

stable hatch Apr 13, 2023, 11:21 AM

#

I don't think you are understanding, which is fine and I'm trying to explain you it

#

Lets go step by step

#

The strategy had a throttler, which waits to tell a shard it can identify, correct?

dusty dove Apr 13, 2023, 11:22 AM

#

the shard asks the strategy for an identify, which internally leverages the throttler to determine, yes

#

past the messaging layer

stable hatch Apr 13, 2023, 11:22 AM

#

We don't care about how the strategies communicate

#

But

#

The throttler in the end is what waits for n time before the strategy gets the go-ahead

#

And tells the shard that it can proceed

#

Correct?

dusty dove Apr 13, 2023, 11:23 AM

#

sure

stable hatch Apr 13, 2023, 11:24 AM

#

Ok now say I want to reuse the current worker strategy, but I want the locks for the shards to be stored in redis

#

Everything in the worker strategy is already perfectly wired up for it (messaging to / from worker to say it can proceed)

#

I just need to change the throttler to make it fetch from/set to redis

#

See what I'm hinting at now?

dusty dove Apr 13, 2023, 11:26 AM

#

yes

#

v much so from the start

stable hatch Apr 13, 2023, 11:26 AM

#

Then I fail to see the issue with making the throttler replaceable

#

We'd default to our impl and let people change it should they know what they're doing

dusty dove Apr 13, 2023, 11:27 AM

#

I just see major footguns with it

stable hatch Apr 13, 2023, 11:27 AM

#

That's on users side

dusty dove Apr 13, 2023, 11:27 AM

#

no but

stable hatch Apr 13, 2023, 11:27 AM

#

Just like how we allow custom strategies

dusty dove Apr 13, 2023, 11:27 AM

#

on one hand sure, that's the case with any indirection like this that you let the user potentially implement

#

but you kind of need a really good reason to allow it, or if the user has no other way to accomplish it

#

once you give them 2 ways to do it

#

they're likely to not figure out what is best for their case

#

and create some abomination

stable hatch Apr 13, 2023, 11:28 AM

#

I mean you do realize this is already doable by extending our current classes and overriding the class

#

I'm just asking we make it more friendly

dusty dove Apr 13, 2023, 11:29 AM

#

whatever ¯_(ツ)_/¯

#

im not fond of it

#

but i can do it

stable hatch Apr 13, 2023, 11:29 AM

#

I want to control as much of the flow as possible too

#

But we also want experienced users to build out their system if they know what they're doing

#

Remember that this is mega low level

#

You can decide if you want the manager to make the throttler or the strategy

dusty dove Apr 13, 2023, 11:31 AM

#

ugh i broke my tests

dim oracle Apr 13, 2023, 11:32 AM

#

PepePopcorn

dusty dove Apr 13, 2023, 11:32 AM

#

nvm

#

the unit test was just wrong

#

lol

#

the ones u had me do work fine still

#

what a mess

stable hatch Apr 13, 2023, 11:35 AM

#

Kek

dusty dove Apr 13, 2023, 12:27 PM

#

ok im done w everything now

#

so as per vlad's moaning @dim oracle @sullen snow you now have 2 ways to go about this once the PR is merged

if you don't need a custom strategy anywhere for any other reason, then you now have the option to implement IIdentifyThrottler and pass in a buildIdentifyThrottler to the manager with a custom one

#

though just make sure you do it right theCutest

dim oracle Apr 13, 2023, 12:36 PM

#

I'm guessing the double I in IIdentify is a typo?

dusty dove Apr 13, 2023, 12:36 PM

#

no

#

its an interface meant to be implemented

#

I prefix

#

just like IContextFetchingStrategy

#

just unfortunate it ends up being double

dim oracle Apr 13, 2023, 12:36 PM

#

I see

stable hatch Apr 13, 2023, 12:44 PM

#

~~or make an abstract class~~ okno

dusty dove Apr 13, 2023, 1:07 PM

#

i made this thread stop auto hiding after an hour

#

since it was annoying

stable hatch Apr 13, 2023, 1:11 PM

#

thank u

#

i keep forgetting we can do that

#

lol

#

also @dusty dove it looks perfect, I think the only thing that looks off is something for future PRs, which is that connections are open even if we cant identify

#

so maybe for future, we should start the conns after we have identify available, unless we have resume available

dusty dove Apr 13, 2023, 1:12 PM

#

i think that's ok

stable hatch Apr 13, 2023, 1:12 PM

#

(imagine if you start like 20 conns, discord will kill those that dont identify fast enough, and uhhh, that we haven't tested yet)

dusty dove Apr 13, 2023, 1:12 PM

#

hmmm

#

i think that takes like a good minute to happen, haha

stable hatch Apr 13, 2023, 1:12 PM

#

still

dusty dove Apr 13, 2023, 1:12 PM

#

maybe

#

will see if they have any trouble w it

stable hatch Apr 13, 2023, 1:13 PM

#

(its also a good ratelimit bc if you open 16 conns constantly discord tends to have some conns that never respond with hello)

dusty dove Apr 13, 2023, 1:13 PM

#

im curious actually

#

2440 shards here goes

#

POOGERS

#

never mind

#

Error: Not enough sessions remaining to spawn 2440 shards; only 1856 remaining; resets at 2023-04-14T11:35:33.799Z

#

we really abused our test bot today

#

lmao

stable hatch Apr 13, 2023, 1:13 PM

#

well

#

we can always make it reset the token

dusty dove Apr 13, 2023, 1:14 PM

#

how many do we even start with

dusty dove Apr 13, 2023, 1:14 PM

#

stable hatch (its also a good ratelimit bc if you open 16 conns constantly discord tends to h...

actually never ran into this

stable hatch Apr 13, 2023, 1:15 PM

#

i have while trying to help others with the ws

#discordjs/ws big bot memes (old)