It is important to node here that by "Server" they mean the Discord logical server concept which may more generically be called a "community", "group" or "workspace". It has nothing to do with the physical compute architecture.
So a title that would make sense to people not familiar with Discord terminology could be:
> How Discord Serves 15M Members of a Single Workspace
These type of articles are always fun, though I mistakenly thought they meant Discord was running on a single hardware server, that would be fun to read. Though considering they run on BEAM they can spread out the workload plenty. Still, I remember when the hard limit for a Discord used to be 1 million, and then along came MidJourney.
It's funny how in hindsight simple changes to major web applications have a major impact. I remember, Tumblr used to show you every statistic about your blog on the dashboard, that meant every single time someone hit the front-page they had to query the latest data to show you your follower count. Sometime after they removed it from the Dashboard, I noticed Tumblr went offline / down significantly less, I doubt that was the only optimization they did, but I presume they did loads of optimizations to their platform at the time.
Yeah, statistical things like that need to be pre-calculated and cached for best performance.
When you first get stuff working doing a query is easier and faster. At some point you remove it and move to counters that are updated via your data pipe.
So a title that would make sense to people not familiar with Discord terminology could be:
> How Discord Serves 15M Members of a Single Workspace