Evaluasi Performa API Berbasis Arsitektur Microservices pada Sistem Logistik Menggunakan k6 dan Apache JMeter

Abstract

The growth of Indonesia's e-commerce and logistics industry demands high reliability of microservices-based systems, making API performance a critical factor of service quality. This research evaluates logistics API performance under various load scenarios while comparing two load testing tools, k6 and Apache JMeter. Testing was conducted in a staging environment on the pickup request and shipment manifest endpoints through load, stress, and spike test scenarios with identical configurations, comprising 36 sessions (18 per tool), measuring response time (p95, p99), throughput, error rate, and the CPU and memory overhead of both tools. The results reveal two distinct degradation mechanisms: pickup request degrades with concurrency, its error rate rising from 6.68% (5 VU) to 31.48% (50 VU) and p95 from 2,569 ms to 3,365 ms, whereas manifest degrades with load duration, keeping a low error rate (0.00–0.14%) while p95 rises from 2,831 ms to 4,623 ms. Under normal load, the p95 of both endpoints (2,249–3,263 ms) remains far above the 1,000 ms service level threshold adopted in this study. Both tools cross-validate each other, with relative p95 differences of 2.6–4.0% and 4.5–13.5%; k6 is more resource-efficient (2.27% CPU, 35.6 MB) than Apache JMeter (3.90% CPU, 454.9 MB), whereas Apache JMeter offers greater data depth through its JTL format. The contribution of this research is the identification of two fundamentally different degradation mechanisms on microservices endpoints residing on the same infrastructure.