Twitter Space 4/25/2022 -- Fail Whaling

Some great stories about Twitter scalability were told on the Oxide and Friends podcast, featuring core Solaris engineers and the co-founder of Twitter's then cloud host, Joyent. There is one story about Twitter taking 100ms of CPU to serve a request, because Rails was using an exception to exit from a loop that was in code executing 434 stack frames deep, and the profiling showed huge amounts of time spent in bcopy, grabbing all those stack frames as part of exception handling to simply... exit a loop. Whole thing is gold but skip to 25:20 for that specific story: link to this video

Oxide and Friends hosted Jason Hoffmann, who used to provide hosting for Twitter in the early days, and he had some classic war stories of how early Twitter got to learn some hard lessons about scaling, and why Ruby exceptions weren't a great control structure choice for hot code paths, especially combined with a deep call stack... link to this video