How to use all my CPU cores with Elysia?

[stefanos82]

I have tried to find an example how to use all my CPU cores and couldn't find anything.

Can someone share their snippet if possible?

I'm interested in Elysia and would like to learn how to push its limits.

[loclv]

@stefanos82

I think this is common question for running Bun/Node.js on across all CPUs available, without any code modifications.

You can read about how to do it with pm2 - Cluster Mode.
And how to run Bun as a daemon with PM2.

Besides, you can use Bun - workers

[stefanos82]

Thank you @loclv 👍

[stefanos82]

It seems I did not need to use the Node.js cluster after all; Bun supports spawn which I took advantage of and produced the desired results I wanted from the very beginning.

Below is the code which you can place in a spawn.ts file and execute it via bun spawn.ts:

import os from 'node:os';

// @ts-ignore
const numCPUs = os.availableParallelism();
for (let i = 0; i < numCPUs; i++) {
    Bun.spawn(['bun', 'src/index.ts'], {
        stdio: ['inherit', 'inherit', 'inherit'],
        env: { ...process.env },
    });
}

With wrk -c 100 -t 4 --latency http://127.0.0.1:3000/ I've managed to get:

Running 10s test @ http://127.0.0.1:3000/
  4 threads and 100 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   697.66us    0.90ms  15.11ms   87.23%
    Req/Sec    52.88k     4.80k   64.45k    66.75%
  Latency Distribution
     50%  310.00us
     75%    0.95ms
     90%    1.83ms
     99%    4.08ms
  2108813 requests in 10.03s, 257.42MB read
Requests/sec: 210260.53
Transfer/sec:     25.67MB

What's interesting, without spawn the default single threaded bun run src/index.ts produces

Running 10s test @ http://127.0.0.1:3000/
  4 threads and 100 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     0.87ms  152.01us   7.14ms   92.55%
    Req/Sec    28.95k     2.05k   58.95k    98.26%
  Latency Distribution
     50%  839.00us
     75%    0.87ms
     90%    0.92ms
     99%    1.38ms
  1158176 requests in 10.10s, 141.38MB read
Requests/sec: 114671.46
Transfer/sec:     14.00MB

which does not match the node.js cluster behavior, that is single-threaded result times the number of CPU cores available, in my case 4.

Is this an expected behavior?

[loclv]

@stefanos82
Thank you for sharing! I appreciate this way a lot!