|
| 1 | +package testcases |
| 2 | + |
| 3 | +import ( |
| 4 | + "bytes" |
| 5 | + "context" |
| 6 | + "encoding/json" |
| 7 | + "fmt" |
| 8 | + "io" |
| 9 | + "net/http" |
| 10 | + "strings" |
| 11 | + "time" |
| 12 | + |
| 13 | + pkgtestcases "github.com/vllm-project/semantic-router/e2e/pkg/testcases" |
| 14 | + "k8s.io/client-go/kubernetes" |
| 15 | +) |
| 16 | + |
| 17 | +func init() { |
| 18 | + pkgtestcases.Register("chat-completions-stress-request", pkgtestcases.TestCase{ |
| 19 | + Description: "Send 1000 sequential requests and measure success rate", |
| 20 | + Tags: []string{"llm", "stress", "reliability"}, |
| 21 | + Fn: testStressTest, |
| 22 | + }) |
| 23 | +} |
| 24 | + |
| 25 | +// StressTestResult tracks the result of a single request |
| 26 | +type StressTestResult struct { |
| 27 | + RequestID int |
| 28 | + Success bool |
| 29 | + StatusCode int |
| 30 | + Duration time.Duration |
| 31 | + ErrorMessage string |
| 32 | +} |
| 33 | + |
| 34 | +func testStressTest(ctx context.Context, client *kubernetes.Clientset, opts pkgtestcases.TestCaseOptions) error { |
| 35 | + if opts.Verbose { |
| 36 | + fmt.Println("[Test] Starting stress test: 1000 sequential requests") |
| 37 | + } |
| 38 | + |
| 39 | + // Setup service connection and get local port |
| 40 | + localPort, stopPortForward, err := setupServiceConnection(ctx, client, opts) |
| 41 | + if err != nil { |
| 42 | + return err |
| 43 | + } |
| 44 | + defer stopPortForward() // Ensure port forwarding is stopped when test completes |
| 45 | + |
| 46 | + const totalRequests = 1000 |
| 47 | + var results []StressTestResult |
| 48 | + successCount := 0 |
| 49 | + totalDuration := time.Duration(0) |
| 50 | + |
| 51 | + // Send 1000 requests sequentially |
| 52 | + for i := 1; i <= totalRequests; i++ { |
| 53 | + result := sendSingleRequest(ctx, i, localPort, opts.Verbose) |
| 54 | + results = append(results, result) |
| 55 | + |
| 56 | + if result.Success { |
| 57 | + successCount++ |
| 58 | + } |
| 59 | + totalDuration += result.Duration |
| 60 | + |
| 61 | + // Print progress every 100 requests |
| 62 | + if opts.Verbose && i%100 == 0 { |
| 63 | + currentSuccessRate := float64(successCount) / float64(i) * 100 |
| 64 | + fmt.Printf("[Test] Progress: %d/%d requests completed (%.2f%% success rate)\n", |
| 65 | + i, totalRequests, currentSuccessRate) |
| 66 | + } |
| 67 | + } |
| 68 | + |
| 69 | + // Calculate statistics |
| 70 | + successRate := float64(successCount) / float64(totalRequests) * 100 |
| 71 | + failureCount := totalRequests - successCount |
| 72 | + avgDuration := totalDuration / time.Duration(totalRequests) |
| 73 | + |
| 74 | + // Set details for reporting |
| 75 | + if opts.SetDetails != nil { |
| 76 | + opts.SetDetails(map[string]interface{}{ |
| 77 | + "total_requests": totalRequests, |
| 78 | + "successful": successCount, |
| 79 | + "failed": failureCount, |
| 80 | + "success_rate": fmt.Sprintf("%.2f%%", successRate), |
| 81 | + "avg_duration_ms": avgDuration.Milliseconds(), |
| 82 | + }) |
| 83 | + } |
| 84 | + |
| 85 | + // Print summary |
| 86 | + printStressTestResults(results, totalRequests, successCount, successRate, avgDuration) |
| 87 | + |
| 88 | + if opts.Verbose { |
| 89 | + fmt.Printf("[Test] Stress test completed: %d/%d successful (%.2f%% success rate)\n", |
| 90 | + successCount, totalRequests, successRate) |
| 91 | + } |
| 92 | + |
| 93 | + return nil |
| 94 | +} |
| 95 | + |
| 96 | +func sendSingleRequest(ctx context.Context, requestID int, localPort string, verbose bool) StressTestResult { |
| 97 | + result := StressTestResult{ |
| 98 | + RequestID: requestID, |
| 99 | + Success: false, |
| 100 | + } |
| 101 | + |
| 102 | + start := time.Now() |
| 103 | + |
| 104 | + // Prepare request body |
| 105 | + requestBody := map[string]interface{}{ |
| 106 | + "model": "MoM", |
| 107 | + "messages": []map[string]string{ |
| 108 | + { |
| 109 | + "role": "user", |
| 110 | + "content": fmt.Sprintf("Request #%d: What is 2+2?", requestID), |
| 111 | + }, |
| 112 | + }, |
| 113 | + } |
| 114 | + |
| 115 | + jsonData, err := json.Marshal(requestBody) |
| 116 | + if err != nil { |
| 117 | + result.ErrorMessage = fmt.Sprintf("marshal error: %v", err) |
| 118 | + result.Duration = time.Since(start) |
| 119 | + return result |
| 120 | + } |
| 121 | + |
| 122 | + url := fmt.Sprintf("http://localhost:%s/v1/chat/completions", localPort) |
| 123 | + req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewBuffer(jsonData)) |
| 124 | + if err != nil { |
| 125 | + result.ErrorMessage = fmt.Sprintf("create request error: %v", err) |
| 126 | + result.Duration = time.Since(start) |
| 127 | + return result |
| 128 | + } |
| 129 | + |
| 130 | + req.Header.Set("Content-Type", "application/json") |
| 131 | + |
| 132 | + httpClient := &http.Client{ |
| 133 | + Timeout: 30 * time.Second, |
| 134 | + } |
| 135 | + |
| 136 | + resp, err := httpClient.Do(req) |
| 137 | + if err != nil { |
| 138 | + result.ErrorMessage = fmt.Sprintf("send request error: %v", err) |
| 139 | + result.Duration = time.Since(start) |
| 140 | + return result |
| 141 | + } |
| 142 | + defer resp.Body.Close() |
| 143 | + |
| 144 | + body, err := io.ReadAll(resp.Body) |
| 145 | + if err != nil { |
| 146 | + result.ErrorMessage = fmt.Sprintf("read response error: %v", err) |
| 147 | + result.Duration = time.Since(start) |
| 148 | + result.StatusCode = resp.StatusCode |
| 149 | + return result |
| 150 | + } |
| 151 | + |
| 152 | + result.Duration = time.Since(start) |
| 153 | + result.StatusCode = resp.StatusCode |
| 154 | + |
| 155 | + if resp.StatusCode == http.StatusOK { |
| 156 | + result.Success = true |
| 157 | + } else { |
| 158 | + result.ErrorMessage = fmt.Sprintf("status %d: %s", resp.StatusCode, string(body)) |
| 159 | + } |
| 160 | + |
| 161 | + return result |
| 162 | +} |
| 163 | + |
| 164 | +func printStressTestResults(results []StressTestResult, totalRequests, successCount int, successRate float64, avgDuration time.Duration) { |
| 165 | + separator := strings.Repeat("=", 80) |
| 166 | + fmt.Println("\n" + separator) |
| 167 | + fmt.Println("Stress Test Results") |
| 168 | + fmt.Println(separator) |
| 169 | + fmt.Printf("Total Requests: %d\n", totalRequests) |
| 170 | + fmt.Printf("Successful: %d\n", successCount) |
| 171 | + fmt.Printf("Failed: %d\n", totalRequests-successCount) |
| 172 | + fmt.Printf("Success Rate: %.2f%%\n", successRate) |
| 173 | + fmt.Printf("Average Duration: %v\n", avgDuration) |
| 174 | + fmt.Println(separator) |
| 175 | + |
| 176 | + // Show first 10 failures if any |
| 177 | + failureCount := 0 |
| 178 | + fmt.Println("\nFirst 10 Failures (if any):") |
| 179 | + for _, result := range results { |
| 180 | + if !result.Success && failureCount < 10 { |
| 181 | + failureCount++ |
| 182 | + fmt.Printf(" Request #%d: %s (duration: %v)\n", |
| 183 | + result.RequestID, result.ErrorMessage, result.Duration) |
| 184 | + } |
| 185 | + if failureCount >= 10 { |
| 186 | + break |
| 187 | + } |
| 188 | + } |
| 189 | + |
| 190 | + if failureCount == 0 { |
| 191 | + fmt.Println(" No failures! 🎉") |
| 192 | + } |
| 193 | + fmt.Println() |
| 194 | +} |
0 commit comments