Testing Guide - ML Defender

Test Suite Overview

ML Defender employs a multi-layered testing strategy validated with real-world datasets and stress testing.

Unit Tests

25+ tests covering core algorithms

Integration Tests

End-to-end pipeline validation

Stress Tests

36K+ events, 17-hour stability

Test Structure

Unit Tests

Located in */tests/ directories:

sniffer/tests/
├── test_payload_analyzer.cpp      # Shannon entropy, PE detection
├── test_fast_detector.cpp         # Layer 1 heuristics
├── test_ransomware_processor.cpp  # Layer 2 features
└── test_integration_simple.cpp    # End-to-end flow

ml-detector/tests/
├── test_onnx_inference.cpp        # Model loading
├── test_feature_extraction.cpp    # 83-feature pipeline
└── test_zmq_consumer.cpp          # Message handling

firewall-acl-agent/tests/
├── test_crypto_decrypt.cpp        # ChaCha20-Poly1305
├── test_ipset_manager.cpp         # IPSet operations
└── test_batch_processor.cpp       # Queue management

Integration Tests

scripts/verify_rag_ecosystem.sh    # Full pipeline
scripts/verify_encryption.sh       # Crypto validation
scripts/verify_firewall_complete.sh # Firewall integration

Running Tests

Quick Unit Tests

# Build with tests
make PROFILE=debug sniffer

# Run all tests
cd sniffer/build-debug
ctest --output-on-failure

# Run specific test
./test_payload_analyzer

All Components

cd sniffer/build-debug
ctest --output-on-failure

# Individual tests:
./test_payload_analyzer
./test_fast_detector
./test_ransomware_feature_extractor
./test_integration_simple_event

With Valgrind (Leak Detection)

valgrind --leak-check=full \
         --show-leak-kinds=all \
         --track-origins=yes \
         ./test_payload_analyzer

Stress Testing

8-Hour Stability Test

Validated with 36,000 events across 4 progressive tests (source/stress_test_8h.sh).

# Full 8-hour test
./stress_test_8h.sh

# Monitor progress
tail -f /vagrant/stress_test_*/logs/sniffer.log

Test Phases (from source/stress_test_8h.sh:575-613):

Warm-up (30 min): Low load, gradual increase
Normal Load (2 hours): Mixed protocols (HTTP/HTTPS/DNS)
Stress Testing (1.5 hours): High bursts (50/s)
Ransomware Simulation (1 hour): Suspicious patterns
Sustained Load (3 hours): Continuous moderate traffic
Cool Down (30 min): Gradual reduction

Stress Test Results (Day 52)

From source/README.md:179-203:

Test	Events	Rate	CPU	Result
1	1,000	42.6/sec	N/A	✅ PASS
2	5,000	94.9/sec	N/A	✅ PASS
3	10,000	176.1/sec	41-45%	✅ PASS
4	20,000	364.9/sec	49-54%	✅ PASS

Totals (36K events):

crypto_errors: 0              ← Perfect crypto pipeline
decompression_errors: 0       ← Perfect LZ4 pipeline
protobuf_parse_errors: 0      ← Perfect message parsing
ipset_successes: 118          ← First ~1000 blocked
max_queue_depth: 16,690       ← Backpressure handled

Synthetic Traffic Generation

Tools Available

From source/tools/:

# Generate ML detector events
cd tools/build-debug
./synthetic_ml_output_injector 1000 50  # 1000 events, 50/sec

# Generate sniffer events
./synthetic_sniffer_injector --count 5000 --malicious-ratio 0.20

# Generate full event pipeline
./generate_synthetic_events 100 0.20  # 100 events, 20% malicious

Traffic Profiles

Hospital Benchmark (source/scripts/day11_hospital_benchmark/):

# Electronic Health Records load
scripts/day11_hospital_benchmark/traffic_profiles/ehr_load.sh

# Emergency department burst
scripts/day11_hospital_benchmark/traffic_profiles/emergency_test.sh

# PACS imaging traffic
scripts/day11_hospital_benchmark/traffic_profiles/pacs_burst.sh

Validation Scripts

Crypto Pipeline Validation

scripts/verify_encryption.sh

Verifies:

✅ ChaCha20-Poly1305 encryption/decryption
✅ LZ4 compression/decompression
✅ Protobuf serialization
✅ Zero errors at 36K events

Firewall Integration

scripts/verify_firewall_complete.sh

Validates:

✅ IPSet creation and rules
✅ Event reception from ml-detector
✅ Decryption and decompression
✅ IP blocking via iptables

Full Ecosystem

scripts/verify_rag_ecosystem.sh

End-to-end test:

Start etcd-server
Start sniffer (eBPF capture)
Start ml-detector (inference)
Start firewall-acl-agent (blocking)
Generate synthetic traffic
Verify all components healthy

Test Datasets

CTU-13 Neris Botnet

Used for ransomware detection validation (source/README.md:289-292).

# Replay CTU-13 dataset
make test-replay-neris

# Manual replay
sudo tcpreplay -i eth1 --mbps=10 \
  datasets/ctu13/botnet-capture-20110810-neris.pcap

Expected Results:

492K events processed
97.6% ransomware detection accuracy
0 crashes, 0 memory leaks

Dataset Structure

datasets/
├── ctu13/
│   ├── smallFlows.pcap          # Quick test (1K flows)
│   ├── botnet-capture-neris.pcap # Ransomware (492K events)
│   └── bigFlows.pcap            # Stress test (10M+ flows)
└── synthetic/
    ├── normal_traffic.pcap
    └── ddos_simulation.pcap

Performance Benchmarks

17-Hour Stability Test Results

From source/TESTING.md:164-182:

Total Runtime:              17h 2m 10s (61,343 seconds)
Total Packets Processed:    2,080,549
Payloads Analyzed:          1,550,375 (74.5%)
Peak Throughput:            82.35 events/second
Average Throughput:         33.92 events/second
Memory Footprint:           4.5 MB (STABLE)
CPU Usage (load):           5-10%
Kernel Panics:              0
Memory Leaks:               0

Status: ✅ PRODUCTION-READY

Component Latency

From source/TESTING.md:282-296:

Component	Latency	Notes
eBPF capture	<1 μs	Kernel space
Ring buffer	<1 μs	Zero-copy
PayloadAnalyzer (fast)	1.01 μs	Normal traffic
PayloadAnalyzer (slow)	149.3 μs	Suspicious (entropy ≥ 7.0)
FastDetector	<1 μs	O(1) heuristics
RansomwareProcessor	Async	Every 30s batch
Protobuf serialize	~10 μs	Per event
ZMQ PUSH	~50 μs	Network I/O

End-to-end latency:

Normal path: ~64 μs
Suspicious path: ~212 μs

Coverage and CI/CD

Test Coverage Goals

Unit tests: >80% code coverage
Integration tests: All critical paths
Stress tests: 24h+ continuous operation

Current Coverage

From source/TESTING.md:353-361:

Test Suite	Tests	Status	Coverage
PayloadAnalyzer	8	✅ All pass	Entropy, PE, patterns
FastDetector	5	✅ All pass	Heuristics, windows
RansomwareProcessor	7	✅ All pass	Features, aggregation
Integration	5	✅ All pass	End-to-end flow
Total	25	✅ 100%	Comprehensive

CI/CD Pipeline (Planned)

.github/workflows/test.yml

name: Test Suite

on: [push, pull_request]

jobs:
  unit-tests:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v2
      - name: Build
        run: make PROFILE=debug all
      - name: Run Tests
        run: |
          cd sniffer/build-debug && ctest --output-on-failure
          cd ml-detector/build-debug && ctest --output-on-failure

  stress-test:
    runs-on: ubuntu-latest
    steps:
      - name: 1-hour stress test
        run: ./stress_test_8h.sh
        timeout-minutes: 70

Test-Driven Development

Writing New Tests

Example: Adding a new feature to PayloadAnalyzer

sniffer/tests/test_payload_analyzer.cpp

#include <gtest/gtest.h>
#include "payload_analyzer.hpp"

TEST(PayloadAnalyzer, DetectsCryptoMiningStratum) {
    PayloadAnalyzer analyzer;
    std::string payload = "{\"id\":1,\"method\":\"mining.subscribe\"}";
    
    auto result = analyzer.analyze(
        reinterpret_cast<const uint8_t*>(payload.data()),
        payload.size()
    );
    
    EXPECT_TRUE(result.is_suspicious);
    EXPECT_GT(result.suspicious_strings, 0);
    EXPECT_TRUE(result.has_crypto_pattern);
}

Test Naming Convention

test_<component>_<feature>.cpp
TEST(<Component>, <BehaviorDescription>)

Debugging Failed Tests

Common Issues

Test fails intermittently:

# Run test 100 times
for i in {1..100}; do 
  ./test_payload_analyzer || echo "FAILED at iteration $i"
done

Memory errors:

# TSAN build
make PROFILE=tsan sniffer
cd sniffer/build-tsan
TSAN_OPTIONS="history_size=7" ./test_payload_analyzer

# ASAN build
make PROFILE=asan sniffer
cd sniffer/build-asan
./test_payload_analyzer

Performance regression:

# Compare before/after
hyperfine './build-debug/test_payload_analyzer' \
          './build-production/test_payload_analyzer'

Next Steps

Build System

Understand CMake and Makefile configuration

Stress Testing

Deep dive into stress test methodology

eBPF/XDP

Learn eBPF packet capture internals

Performance

Optimize and benchmark components

Contributing

Advanced

Documentation Index

​Test Suite Overview

Unit Tests

Integration Tests

Stress Tests

​Test Structure

​Unit Tests

​Integration Tests

​Running Tests

​Quick Unit Tests

​All Components

​With Valgrind (Leak Detection)

​Stress Testing

​8-Hour Stability Test

​Stress Test Results (Day 52)

​Synthetic Traffic Generation

​Tools Available

​Traffic Profiles

​Validation Scripts

​Crypto Pipeline Validation

​Firewall Integration

​Full Ecosystem

​Test Datasets

​CTU-13 Neris Botnet

​Dataset Structure

​Performance Benchmarks

​17-Hour Stability Test Results

​Component Latency

​Coverage and CI/CD

​Test Coverage Goals

​Current Coverage

​CI/CD Pipeline (Planned)

​Test-Driven Development

​Writing New Tests

​Test Naming Convention

​Debugging Failed Tests

​Common Issues

​Next Steps

Build System

Stress Testing

eBPF/XDP

Performance

Build docs developers (and LLMs) love

Test Suite Overview

Test Structure

Unit Tests

Integration Tests

Running Tests

Quick Unit Tests

All Components

With Valgrind (Leak Detection)

Stress Testing

8-Hour Stability Test

Stress Test Results (Day 52)

Synthetic Traffic Generation

Tools Available

Traffic Profiles

Validation Scripts

Crypto Pipeline Validation

Firewall Integration

Full Ecosystem

Test Datasets

CTU-13 Neris Botnet

Dataset Structure

Performance Benchmarks

17-Hour Stability Test Results

Component Latency

Coverage and CI/CD

Test Coverage Goals

Current Coverage

CI/CD Pipeline (Planned)

Test-Driven Development

Writing New Tests

Test Naming Convention

Debugging Failed Tests

Common Issues

Next Steps