test: add load tests (#93)

* setup env for k6

* test: load tests to package.json

* test: add tests

* test: add load test

* test: granualize metrics

* test: have a standard workload

* test: load 50 users 1 request/sec/user

* test: log the response failures

* test: randomize the iteration interval

* test: adjust the load scenario

* test: doc update

* test: move globals to shared esconfig

.devcontainer/devcontainer.json

@@ -20,7 +20,8 @@
       "version": "latest",
       "installBicep": true
     },
-    "ghcr.io/azure/azure-dev/azd:latest": {}
+    "ghcr.io/azure/azure-dev/azd:latest": {},
+    "ghcr.io/devcontainers-contrib/features/k6:1": {}
   },
 
   // Configure tool-specific properties.

package.json

@@ -13,6 +13,7 @@
     "start:indexer": "npm run dev --workspace=indexer",
     "test": "npm run test -ws --if-present",
     "test:playwright": "npx playwright test",
+    "test:load": "k6 run tests/load/index.js",
     "build": "npm run build -ws --if-present",
     "clean": "npm run clean -ws --if-present",
     "docker:build": "npm run docker:build -ws --if-present",
@@ -33,6 +34,7 @@
     "@playwright/test": "^1.39.0",
     "@tapjs/nock": "^3.1.13",
     "@types/node": "^18.15.3",
+    "@types/k6": "^0.47.1",
     "concurrently": "^8.2.1",
     "eslint-config-shared": "^1.0.0",
     "lint-staged": "^14.0.1",

packages/eslint-config/index.js

@@ -1,4 +1,7 @@
 module.exports = {
+  globals: {
+    __ENV: 'readonly',
+  },
   parserOptions: {
     ecmaVersion: 'latest',
     sourceType: 'module',

tests/load/README.md

@@ -0,0 +1,22 @@
+The tests use [k6](https://k6.io/) to perform load testing.
+
+# Install k6
+
+k6 is already included in the dev container, so no further installation is required.
+
+For manual installation, refer to [k6 installation docs](https://k6.io/docs/get-started/installation/).
+
+# To run the test
+
+Set the following environment variables to point to the deployment.
+
+```
+export WEBAPP_URI=<webapp_uri>
+export SEARCH_API_URI=<search_api_uri>
+```
+
+Once set, you can now run load tests using the following command:
+
+```
+npm run test:load
+```

tests/load/chat.js

@@ -0,0 +1,66 @@
+import http from 'k6/http';
+import { Trend } from 'k6/metrics';
+import { group, sleep } from 'k6';
+
+const chatStreamLatency = new Trend('chat_stream_duration');
+const chatNoStreamLatency = new Trend('chat_nostream_duration');
+
+function between(min, max) {
+  min = Math.ceil(min);
+  max = Math.floor(max);
+  return Math.floor(Math.random() * (max - min) + min); // The maximum is exclusive and the minimum is inclusive
+}
+
+function choose(list) {
+  return list[between(0, list.length)];
+}
+
+export function chat(baseUrl, stream = true) {
+  group('Chat flow', function () {
+    const defaultPrompts = [
+      'How to search and book rentals?',
+      'What is the refund policy?',
+      'How to contact a representative?',
+    ];
+
+    const payload = JSON.stringify({
+      messages: [{ content: choose(defaultPrompts), role: 'user' }],
+      context: {
+        retrieval_mode: 'hybrid',
+        semantic_ranker: true,
+        semantic_captions: false,
+        suggest_followup_questions: true,
+        retrievalMode: 'hybrid',
+        top: 3,
+        useSemanticRanker: true,
+        useSemanticCaptions: false,
+        excludeCategory: '',
+        promptTemplate: '',
+        promptTemplatePrefix: '',
+        promptTemplateSuffix: '',
+        suggestFollowupQuestions: true,
+        approach: 'rrr',
+      },
+      stream,
+    });
+
+    const parameters = {
+      headers: {
+        'Content-Type': 'application/json',
+      },
+      tags: { type: 'API' },
+    };
+
+    const response = http.post(`${baseUrl}/chat`, payload, parameters);
+
+    if (response.status !== 200) {
+      console.log(`Response: ${response.status} ${response.body}`);
+    }
+
+    // add duration property to metric
+    const latencyMetric = stream ? chatStreamLatency : chatNoStreamLatency;
+    latencyMetric.add(response.timings.duration, { type: 'API' });
+
+    sleep(between(5, 20)); // wait between 5 and 20 seconds between each user iteration
+  });
+}

tests/load/config.js

@@ -0,0 +1,16 @@
+export const thresholdsSettings = {
+  'http_req_failed{type:API}': [{ threshold: 'rate<0.01' }], // less than 1% failed requests
+  'http_req_failed{type:content}': [{ threshold: 'rate<0.01' }], // less than 1% failed requests
+  'http_req_duration{type:API}': ['p(90)<40000'], // 90% of the API requests must complete below 40s
+  'http_req_duration{type:content}': ['p(99)<200'], // 99% of the content requests must complete below 200ms
+};
+
+// 5.00 iterations/s for 1m0s (maxVUs: 100-200, gracefulStop: 30s)
+export const standardWorkload = {
+  executor: 'constant-arrival-rate',
+  rate: 5,
+  timeUnit: '1s',
+  duration: '1m',
+  preAllocatedVUs: 100,
+  maxVUs: 200,
+};

tests/load/index.js

@@ -0,0 +1,19 @@
+import { mainpage } from './mainpage.js';
+import { chat } from './chat.js';
+import { thresholdsSettings, standardWorkload } from './config.js';
+
+export const options = {
+  scenarios: {
+    staged: standardWorkload,
+  },
+  thresholds: thresholdsSettings,
+};
+
+const webappUrl = __ENV.WEBAPP_URI;
+const searchUrl = __ENV.SEARCH_API_URI;
+
+export default function () {
+  mainpage(webappUrl);
+  chat(searchUrl, true);
+  //chat(searchUrl, false);
+}

tests/load/mainpage.js

@@ -0,0 +1,15 @@
+import http from 'k6/http';
+import { Trend } from 'k6/metrics';
+import { group, sleep } from 'k6';
+
+const mainpageLatency = new Trend('mainpage_duration');
+
+export function mainpage(baseUrl) {
+  group('Mainpage', function () {
+    // save response as variable
+    const response = http.get(`${baseUrl}`, { tags: { type: 'content' } });
+    // add duration property to metric
+    mainpageLatency.add(response.timings.duration, { type: 'content' });
+    sleep(1);
+  });
+}