Stress Tests

Comprehensive stress testing infrastructure for Neumann targeting 1M entity scale with extensive coverage of concurrency, data volume, and sustained load.

Quick Start

# Run all stress tests (45+ min total)
cargo test --release -p stress_tests -- --ignored --nocapture

# Run specific test suite
cargo test --release -p stress_tests --test hnsw_stress -- --ignored --nocapture

# Run with custom duration (30s instead of default)
STRESS_DURATION=30 cargo test --release -p stress_tests -- --ignored --nocapture

Test Suites

Key Performance Findings

TensorStore (DashMap)

• 7.5M writes/sec at 1M entities
• Sub-microsecond median latency
• Handles 16:1 contention ratio with 2.5M ops/sec

HNSW Index

• 3,372 vectors/sec insert rate at 1M scale
• 0.11ms search latency (p50)
• 99.8% recall@10 under concurrent load

BloomFilter

• 0.88% FP rate at 1M keys (target 1%)
• 15M+ ops/sec bit-level operations
• Thread-safe with AtomicU64

Mixed Workloads

• All engines can operate concurrently
• Graph operations (5us p50) and vector ops (< 1us p50) are fastest
• Relational engine adds ~12ms p50 overhead due to schema operations

Configuration

Environment Variables

Config Presets

#![allow(unused)]
fn main() {
// Quick mode (CI): 100K entities, 4 threads, 30s
let config = quick_config();

// Full mode (local): 1M entities, 16 threads, 10min
let config = full_config();

// Endurance mode: 500K entities, 8 threads, 1 hour
let config = endurance_config();
}

Latency Metrics

All tests report percentile latencies using HdrHistogram:

Running in CI

For CI pipelines, use quick_config with limited duration:

- name: Run stress tests
  run: |
    STRESS_DURATION=30 cargo test --release -p stress_tests -- --ignored --nocapture
  timeout-minutes: 15