> Firecracker didn't have a very compelling I/O story
Can you provide any sources for this claim? We're running Firecracker in production over at blacksmith dot sh and haven't been able to reproduce any perf regressions in Firecracker over CH in our internal benchmarking.