Reforking?

[retorted]

What do you think about Pitchfork's reforking strategy to maximise shared memory between workers, and do you think Itsi could work with reforking?

On a related note; it seems like Itsi has the best throughput in the benchmarks, but I'm curious what the relative memory footprint of each server looks like, what do you think about adding some sort of memory profiling benchmark to the comparison?

[wouterken]

I think reforking is an interesting concept, but it’s not broadly applicable. Whether a warmed process settles into a stable memory footprint after N requests or a fixed period is very use-case dependent, and even when tuned well, it’s brittle (easy to break assumptions as apps evolve and require retuning). It makes sense at very large scale, where saving even a small percentage of memory across a fleet adds up, but it’s not something I’m aiming to specialize in.

Itsi’s goal is to be a solid and consistently good performer across a wide range of use-cases, from small hobby projects to large horizontally scaled production apps.