a distributed discord sharder.

this guy makes shards to connect to discord. distributes the messages across workers.

there's a whitepaper about this design, it's pretty interesting and you should read it some time.

scenario: my discord bot has lots of big image commands. i use python because it has great image manipulation libraries and a great discord library. but my bot uses so much memory and the CPU is constantly pinned, so it runs really slow.

problem: you are getting bottlenecked by your own code, but because your bot is coupled so tightly to discord, you can't really do anything about it. more shards = more complexity and more points of failure.

answer: use a super fast language to talk to discord. keep a couple of these guys answering dispatch events. drop the ones you don't care about, put the ones you do care about in a big queue. cache the data you want in an in-memory database, get rid of the rest of it because it's fat and wastes memory. spawn a million python workers to answer discord events, bring them up and down as you choose, have basically zero downtime.

this design is owed to zeyla and wolke who each have their own implementations of this system.

this one was written for fun, mostly.

no where, yet.

when someone asks.

Licensed under ISC. LICENSE