with a worst-case production-looking load the p99 jumped to just under 1ms (but definitely still quite good) and throughput of around 44k req/sec @ ~650% CPU which seems like the practical limits of a single instance of the service
just finished up queueing updates during initial load etc.
Comments
today has been experimenting with jemalloc since under heavy load i've been getting memory fragmentation as well as simulating worst-case traffic
just finished up queueing updates during initial load etc.
Anxious
Sad
Confused
Disappointed
….