A web server that JIT-compiles your Caddyfile to machine code claims 3x throughput — the thread asked: 35k qps for whom? (su3.io)
zeroserve's new Caddy-compat mode JIT-compiles a subset of Caddyfile directives to eBPF, then to native machine code, running it in an io_uring event loop for 3x throughput and 70% lower latency over Caddy.
The thread didn't dispute the numbers — it disputed the point. 'Caddy compatible minus everything that matters,' one wrote, since ACME and plugins are gone. Another did the math: at 35k qps on two threads, you're optimizing a metric that was never the bottleneck.
Sign in to save