1. Why Load Testing Is Non-Negotiable in Fintech Microservices
In financial systems, an untested endpoint that runs fast on a local developer machine can trigger catastrophic cascading failures under production concurrent load. When thousands of mobile banking users simultaneously initiate balance inquiries or flash transactions, microservices face:
- Connection Pool Exhaustion: Inefficient HTTP client connection reuse causing TCP socket exhaustion (`TIME_WAIT` saturation).
- JSON Serialization CPU Bottlenecks: High CPU overhead marshaling heavy JSON payloads on REST endpoints under 10k+ requests per second.
- Garbage Collection (GC) Stalls: Excessive heap allocations during request decoding causing microsecond-level GC pauses that compound across distributed call chains.
ghz is a high-performance command-line load testing tool and library specifically engineered for gRPC microservices, capable of generating sustained concurrent traffic and reporting sub-millisecond latency distribution histograms.
2. Architectural Benchmark: gRPC (HTTP/2) vs REST (HTTP/1.1)
We conducted standardized load tests across our Go microservices handling identical wallet transaction balance lookup payloads under 50,000 total requests with 200 concurrent workers:
| Performance Metric | REST API (JSON / HTTP 1.1) | gRPC (Protobuf / HTTP 2) | Efficiency Gain |
|---|---|---|---|
| Throughput (Queries Per Second) | 4,820 QPS | 18,450 QPS | 3.8x Higher |
| Median Latency (p50) | 18.4 ms | 1.8 ms | 10x Faster |
| 99th Percentile Latency (p99) | 92.1 ms | 3.1 ms | 96.6% Lower |
| Network Bandwidth (Payload Size) | 420 bytes / req | 88 bytes / req | 79% Compression |
| CPU Utilization at 10k QPS | 84% CPU | 22% CPU | 73% Less Load |
3. Automated ghz Load Testing Script in CI/CD
To prevent performance regressions before merging feature pull requests, we run automated `ghz` benchmark sweeps targeting gRPC staging servers:
#!/usr/bin/env bash
set -euo pipefail
TARGET_HOST="127.0.0.1:50051"
PROTO_PATH="./proto/wallet.proto"
CALL="wallet.WalletService.GetBalance"
echo "⚡ Executing 50,000 gRPC load requests (200 concurrency)..."
ghz --insecure \
--proto="${PROTO_PATH}" \
--call="${CALL}" \
-d '{"user_id": "usr_998877", "currency": "IDR"}' \
-c 200 \
-n 50000 \
--duration=30s \
--format=summary \
"${TARGET_HOST}"4. Topic Cluster: Complete Engineering Case Study Series
Explore all technical deep dives in our Go, Fintech, & Distributed Systems Cluster:
📚 Explore the Complete Engineering Series
Inter-connected production case studies authored by Nur Awwabin Yoga Saputra: