idk if it is alright, will it break?
https://discord.com/developers/docs/events/gateway#sharding-max-concurrency
#weird ass algorithm with zig
keen owl
spare spade
I see an alloc and a nested loop. I don't really know what the question is (was) asking
keen owl
oh just this
spare spade
are you asking us to debug your code for you, without even knowing if there's a bug?
spare spade
maybe someone familiar with the algorithm will have more useful feedback. I just don't really get the specific question and problem you're running into. it seems a bit too abstract. like. yeah, that's some init code.
keen owl
okay it's like dividing an array into chunks
[a,b,c,d,e,f,g] with a concurrency of 3, [[a,b,c],[d,e,f],[g]]
then it has some logic to create the ID of the shard
spare spade
can you post the text instead of a screenshot? that way I don't have to retype lines
and there's a line I think needs correction
keen owl
hold on
glossy forum
[a,b,c,d,e,f,g] with a concurrency of 3, [[a,b,c],[d,e,f],[g]]
I believe the keyword is "partition".
keen owl
/// spawn buckets in order
/// https://discord.com/developers/docs/events/gateway#sharding-max-concurrency
fn spawnBuckets(self: *Self) [][]Shard {
const concurrency = self.options.info.session_start_limit.?.max_concurrency;
const range = self.shard_start - self.shard_end;
const bucket_count = (range + concurrency - 1) / concurrency;
self.logif("#0 preparing buckets", .{});
const buckets = try self.allocator.alloc([][]Shard, bucket_count);
for (buckets, 0..) |bucket, index| {
const bucket_size = if ((index + 1) * concurrency > range) range - (index * concurrency) else concurrency;
bucket.* = try self.allocator.alloc([]Shard, bucket_size);
for (bucket, 0..) |shard, i| {
shard.* = try self.create(self.shard_start + index * concurrency + i);
}
}
self.logif("{d} buckets created", .{bucket_count});
return buckets;
}
thx
glossy forum
Kinda smells like the "gather/scatter" approach
spare spade
I am wondering if this line:
const buckets = try self.allocator.alloc([][]Shard, bucket_count);
is allocating [][][]Shard and not [][]Shard?
Is that what you're intending?
keen owl
I think that's wrong
spare spade
cause allocating a string is try allocator.alloc(u8, string_len);, not try allocator.alloc([]u8, string_len);
keen owl
cause allocating a string is `try allocator.alloc(u8, string_len);`, not `try al...
oh crap
OK what about now
😭
glossy forum
You'll want |*bucket| in order to get a pointer
spare spade
that seems more like what I'd expect. but I'm not digging deep into the API or expected data set, so maybe I'm wrong
glossy forum
No that sounds right
keen owl
You'll want `|*bucket|` in order to get a pointer
like thiz
keen owl
depends on the guild count of the bot
I believe every shard may handle like 1000 guilds
it's really not that exact
glossy forum
Not sure what sort of concurrency you're doing here, but -- Generally, you want to have fairly large partitions, and then give each one their own entirely distinct blob of data which they work on (by which I mean, a separate allocation, or more than 64 bytes in memory away from another thread's blob), only to then recombine at the end once the thread is done with it.
That way you minimize contention, which is the single most important thing with getting all the perf from them.
Allocating each 'row' separately as you're doing there works fine, but is inefficient if the bucket size is small, or you have a lot of buckets.
keen owl
it is everything explained here https://discord.com/developers/docs/events/gateway#sharding-max-concurrency
glossy forum
(If the bucket size is small because it means you'll have many more things being handed between threads; i.e contention)
(And if you have a lot of buckets, because you have to free each one individually - or use an arena)