Chaos Engineering & Resilience Testing

<default_to_action> When testing system resilience or injecting failures:

• DEFINE steady state (normal metrics: error rate, latency, throughput)

• HYPOTHESIZE system continues in steady state during failure

• INJECT real-world failures (network, instance, disk, CPU)

• OBSERVE and measure deviation from steady state

• FIX weaknesses discovered, document runbooks, repeat

Quick Chaos Steps:

• Start small: Dev → Staging → 1% prod → gradual rollout

• Define clear rollback triggers (error_rate > 5%)

• Measure blast radius, never exceed planned scope

• Document findings → runbooks → improved resilience

Critical Success Factors:

• Controlled experiments with automatic rollback

• Steady state must be measurable

• Start in non-production, graduate to production </default_to_action>

Quick Reference Card

When to Use

• Distributed systems validation

• Disaster recovery testing

• Building confidence in fault tolerance

• Pre-production resilience verification

Failure Types to Inject

Category Failures Tools

Network Latency, packet loss, partition tc, toxiproxy

Infrastructure Instance kill, disk failure, CPU Chaos Monkey

Application Exceptions, slow responses, leaks Gremlin, LitmusChaos

Dependencies Service outage, timeout WireMock

Blast Radius Progression

Dev (safe) → Staging → 1% prod → 10% → 50% → 100% ↓ ↓ ↓ ↓ Learn Validate Careful Full confidence

Steady State Metrics

Metric Normal Alert Threshold

Error rate < 0.1%

1%

p99 latency < 200ms

500ms

Throughput baseline -20%

Chaos Experiment Structure

// Chaos experiment definition const experiment = { name: 'Database latency injection', hypothesis: 'System handles 500ms DB latency gracefully', steadyState: { errorRate: '< 0.1%', p99Latency: '< 300ms' }, method: { type: 'network-latency', target: 'database', delay: '500ms', duration: '5m' }, rollback: { automatic: true, trigger: 'errorRate > 5%' } };

Agent-Driven Chaos

// qe-chaos-engineer runs controlled experiments await Task("Chaos Experiment", { target: 'payment-service', failure: 'terminate-random-instance', blastRadius: '10%', duration: '5m', steadyStateHypothesis: { metric: 'success-rate', threshold: 0.99 }, autoRollback: true }, "qe-chaos-engineer");

// Validates: // - System recovers automatically // - Error rate stays within threshold // - No data loss // - Alerts triggered appropriately

Agent Coordination Hints

Memory Namespace

aqe/chaos-engineering/ ├── experiments/* - Experiment definitions & results ├── steady-states/* - Baseline measurements ├── runbooks/* - Generated recovery procedures └── blast-radius/* - Impact analysis

Fleet Coordination

const chaosFleet = await FleetManager.coordinate({ strategy: 'chaos-engineering', agents: [ 'qe-chaos-engineer', // Experiment execution 'qe-performance-tester', // Baseline metrics 'qe-production-intelligence' // Production monitoring ], topology: 'sequential' });

Related Skills

• shift-right-testing - Production testing

• performance-testing - Load testing

• test-environment-management - Environment stability

Remember

Break things on purpose to prevent unplanned outages. Find weaknesses before users do. Define steady state, inject failures, measure impact, fix weaknesses, create runbooks. Start small, increase blast radius gradually.

With Agents: qe-chaos-engineer automates chaos experiments with blast radius control, automatic rollback, and comprehensive resilience validation. Generates runbooks from experiment results.