tests/stress_tests/chat_session_with_file.js

// This file aims to test a scenario where 80 users send 3 messages to ONE model (pessimistic).

import http from 'k6/http';
import { sleep, check } from 'k6';
import { SharedArray } from 'k6/data';

const file_name = __ENV.FILE;

const message_examples = new SharedArray('chat messages', function () {
  const f = open('./message_examples.txt');
  // .split('\n') creates an array where each line is one element
  // .filter removes empty lines at the end of the file
  return f.split('\n').map(line => line.trim()).filter(line => line.length > 0);
});


const testFile = open(`./${file_name}`, 'b'); 

export const options = {
  scenarios: {
    my_spike_test: {
      executor: 'per-vu-iterations',
      vus: 80,              // 80 total users
      iterations: 1,        // Each user runs the function exactly once.
    },
  },
};

export default function () {
  // Each VU must wait a random time period to prevent them from
  // sending their messages at the exact same time.
  sleep(Math.random() * 10);

  const url = __ENV.URL

  // Each VU first sends 1 file
  const data = {
    file: http.file(testFile, file_name, 'application/pdf'), // The file
    session_id: `VU${__VU}`,
  };

  const res = http.put(url + "/file", data);

  check(res, { 'status is 200': (r) => r.status === 200 });

  // Each VU sends 3 messages
  for (let i = 0; i < 3; i++) {
    const messageIndex = ((__VU - 1) * 3 + i) % message_examples.length;
    const userMessage = message_examples[messageIndex];

    const payload = {
      user_id: `VU${__VU}`,
      session_id: `VU${__VU}`,
      conversation_id: `VU${__VU}`,
      human_message: userMessage,
      model_type: __ENV.MODEL_TYPE,
      consent: true,
      age_group: "0-18",
      gender: "M",
      roles: ["other"],
      participant_id: `VU${__VU}`,
      lang: "en"
    };
    const params = { headers: { 'Content-Type': 'application/json' } };

    const res = http.post(url + "/chat", JSON.stringify(payload), params);

    check(res, {'status is 200': (r) => r.status === 200})

    let reply = '';
    if (res.status === 200) {
      // k6 does not support streaming response bodies. It waits for the entire response until
      // the stream is 'done'. Therefore, we do not need to read the chunks one by one.
      let data = "";
      try {
        data = res.json();
      } catch (error) {
        // However, if the response contains streamed data, it is not in JSON format.
        // We would have to read the body to access that data.
        data = res.body;
      }
      reply = data.reply || 'no_reply';
    } else {
      console.error(res.status);
      console.error(res.body);
    }

    // Simulating reading time, thinking time and writing time.
    // Simulate reading speed: ~200ms per word in the reply + 2s thinking time
    const readingTime = (reply.split(' ').length * 0.2) + 2;
    // Cap it so it doesn't wait forever, but add some randomness (jitter)
    const finalSleep = Math.min(readingTime, 15) + (Math.random() * 3);
    
    sleep(finalSleep);  
  }
}

// TEST RESULT ANALYSIS
// CHAMP (GROK) - T4 small - 80VUs - with a small file (15 Ko)
//   █ TOTAL RESULTS

//     checks_total.......: 320     3.761939/s
//     checks_succeeded...: 100.00% 320 out of 320
//     checks_failed......: 0.00%   0 out of 320

//     ✓ status is 200

//     HTTP
//     http_req_duration..............: avg=5.45s min=77.65ms med=5.58s max=13.39s p(90)=10.38s p(95)=11.18s
//       { expected_response:true }...: avg=5.45s min=77.65ms med=5.58s max=13.39s p(90)=10.38s p(95)=11.18s
//     http_req_failed................: 0.00%  0 out of 320
//     http_reqs......................: 320    3.761939/s

//     EXECUTION
//     iteration_duration.............: avg=1m11s min=42.01s  med=1m11s max=1m25s  p(90)=1m21s  p(95)=1m23s
//     iterations.....................: 80     0.940485/s
//     vus............................: 1      min=1        max=80
//     vus_max........................: 80     min=80       max=80

//     NETWORK
//     data_received..................: 571 kB 6.7 kB/s
//     data_sent......................: 1.4 MB 17 kB/s

// running (01m25.1s), 00/80 VUs, 80 complete and 0 interrupted iterations
// my_spike_test ✓ [======================================] 80 VUs  01m25.1s/10m0s  80/80 iters, 1 per VU