Is this performance expected of Elysia?

[winstxnhdw]

I have 3 replica pods with 1 vCPU and 1 GB memory hosting an Elysia server. The server performs a trivial task as shown below.

import swagger from '@elysiajs/swagger'
import { Elysia, t } from 'elysia'

new Elysia()
  .use(swagger())
  .get(
    '/health',
    ({ request, query }) => {
      request.headers.delete('host')

      return {
        endpoint: query.endpoint,
        proxy: query.proxy,
        disable_proxy: query.disable_proxy,
      }
    },
    {
      query: t.Object({
        endpoint: t.String({
          error: 'The `endpoint` query parameter is missing!',
        }),
        proxy: t.Optional(
          t.String({
            default: Bun.env['HTTP_PROXY'] || Bun.env['HTTPS_PROXY'],
          }),
        ),
        disable_proxy: t.Optional(t.Boolean()),
      }),
    },
  )

Running wrk on my server produces the following stats.

  8 threads and 400 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    91.35ms   94.45ms   1.40s    91.11%
    Req/Sec   632.72    136.92     5.12k    85.27%
  Latency Distribution
     50%   95.00ms
     75%  102.10ms
     90%  141.46ms
     99%  466.25ms
  151580 requests in 30.07s, 41.69MB read
  Non-2xx or 3xx responses: 16
Requests/sec:   5040.83
Transfer/sec:      1.39MB

~5000 RPS and a P99 of 466 ms for such a trivial task seems to be bad. For reference, my Granian server has a P99 of 100 ms to serve its entire Swagger documentation (which is arguably more work). Do let me know if this is what you are seeing on your side as well.

[re-thc]

for such a trivial task seems to be bad

What are you benchmarking here? The underlying Bun server vs a Rust HTTP server? Well it's obvious that Rust will likely be faster.

For trivial tasks you're barely triggering any Javascript or Python code. It won't give you an indication of what the real world usage looks like.

For reference, my Granian server has a P99 of 100 ms to serve its entire Swagger documentation (which is arguably more work)

i.e. make some assumptions instead of actually benchmarking and comparing?

At the end of the day, P90 looks fine. It's hard to tell without more information (e.g. metrics). It could be a bug in Bun, Elysia or something else.

[winstxnhdw]

Hey, I am writing a really simple forward proxy so barely any JavaScript or Python code will be executed anyways. In general, I just want to know if this is roughly the expected performance observed by the community, which you answered, so thank you.