Why We Rewrote Our HTTP Server in Zig

Our built-in parser was four times slower than it needed to be. So we wrote our own, from scratch, in a language most JavaScript developers have never touched.

Last January, I was running benchmark comparisons against three other JavaScript runtimes and the numbers did not add up. Our throughput on simple GET requests was stuck at 48,000 requests per second on an 8-core machine. Volt's built-in HTTP module was pulling 52,000 on identical hardware.

The Breaking Point

We profiled the request lifecycle and found the culprit in five minutes: our HTTP parser was spending 40 percent of its time copying memory between intermediate buffers. Every incoming connection triggered three allocations before application code ran. The fix was not incremental. We needed a zero-copy parser that could interpret raw bytes straight from the socket buffer.