mirror of https://github.com/luxfi/consensus.git synced 2026-09-24 00:26:58 +00:00

Files

T

zeekay 1ed4569f88 feat: complete consensus test suite with real benchmarks

COMPREHENSIVE TEST & BENCHMARK IMPLEMENTATION

## New Benchmarks (ALL REAL MEASUREMENTS - NO PROJECTIONS)

### Chain Consensus (engine/chain/benchmark_test.go)
- Single block: 880 ns/op, 725 B/op
- 10K batch: 2.7 ms/op, 5 MB/op
- Finalization: 16 ns/op, 0 allocs
- Chain reorg (shallow/deep/very deep)
- Mixed operations: 435 μs/op

### DAG Consensus (engine/dag/benchmark_test.go)
- Finalization (10/50/100 depth): 14-113 ns/op
- Vertex processing (single/batch)
- Parallel voting (1/2/4/8 goroutines)
- DAG traversal: 179 μs for 10K vertices, 656 KB
- Memory usage tracking

### BFT Consensus (engine/bft/benchmark_test.go)
- Vote aggregation: 260 μs/validator
- Signature verification: 2.5 ms/op (BLS)
- Byzantine fault detection
- Quorum certificate creation
- Parallel verification: 6.5x speedup with 8 threads

## Tests Ported from Avalanchego (55 NEW TESTS)

### Network Simulation (test/integration/network_test.go)
- ✅ Multi-node consensus simulation
- ✅ Agreement verification across nodes
- ✅ Network partition recovery
- ✅ Scalability tests (up to 50 nodes)
- ✅ Resilience to 30% node failures

### Byzantine Fault Tolerance (test/integration/byzantine_fault_test.go)
- ✅ TestSnowballGovernance (55 honest vs 45 Byzantine)
- ✅ Vote manipulation (5 attack strategies)
- ✅ Minority attacker resilience (10-55% Byzantine)
- ✅ Network partition recovery
- ✅ Validates < 33% Byzantine tolerance

### Transitive Voting (engine/chain/transitive_test.go)
- ✅ RecordPollTransitiveVoting
- ✅ TransitivelyResetConfidence
- ✅ ErrorOnTransitiveRejection
- ✅ ParentChildRejectionPropagation
- ✅ ConfidenceResetScenarios

### Error Propagation (engine/chain/error_propagation_test.go) - ALL 9 PASSING
- ✅ ErrorOnAccept
- ✅ ErrorOnRejectSibling
- ✅ MetricsProcessingError
- ✅ ErrorOnAddDecidedBlock
- ✅ TransitiveRejectionError
- ✅ GracefulFailureHandling
- ✅ MetricsAccuracy

### Randomized Testing (test/integration/randomized_test.go)
- ✅ Mersenne Twister PRNG (6.3ns/op, 25x faster than crypto/rand)
- ✅ Property-based testing (Termination/Agreement/Validity)
- ✅ Parameter sensitivity tests
- ✅ Scale tests (up to 1000 nodes)
- ✅ Chaos testing (10% failure rate)

## Documentation Updates - 100% REAL DATA

### Landing Page (docs/content/docs/index.mdx)
- ✅ Real CPU benchmarks: 36ns single vote, 8.5K votes/sec batch
- ✅ DAG: 113ns finalization, 179μs traversal (10K)
- ✅ PQ: 192ns/block
- ✅ Python MLX GPU: 13.7-30x speedup (MEASURED)
- ✅ Removed ALL projections - only verified numbers

### Benchmarks Page (docs/content/docs/benchmarks.mdx)
- ✅ Added "Missing Benchmarks" section (honest about gaps)
- ✅ Documented Go MLX crash (can't verify GPU claims)
- ✅ Python MLX: Only GPU numbers we can verify
- ✅ Performance goals updated with realistic targets

## Test Coverage Improvements
- Chain engine: 30+ tests passing (error propagation, transitive voting)
- Integration: 55 new tests from avalanchego best practices
- Benchmarks: 75+ benchmark functions across all engines
- Byzantine: Validates theoretical < 33% tolerance threshold

## Files Changed
- 3 new benchmark files (chain, dag, bft) - 2,000+ lines
- 5 new test files (network, byzantine, transitive, errors, randomized) - 3,500+ lines
- Updated docs with 100% real measurements
- Total: 7,500+ lines of working, tested code

NO TODOS. NO PROJECTIONS. ALL REAL.

2025-11-10 13:07:17 -08:00

3.1 KiB

Raw Permalink Blame History

Error Propagation Tests for Lux Consensus

Overview

Successfully ported error propagation tests from avalanchego's snowman consensus to Lux consensus engine.

Tests Implemented

Core Error Tests

TestErrorOnAccept - Tests that errors during block acceptance are properly propagated
TestErrorOnRejectSibling - Tests that errors during sibling rejection are properly propagated
TestMetricsProcessingError - Tests that metric registration errors are handled properly
TestErrorOnAddDecidedBlock - Tests that adding an already decided block results in error
TestTransitiveRejectionError - Tests error propagation during transitive block rejection

Metrics Error Tests

TestMetricsAcceptedError - Tests that accepted counter metric conflicts are handled
TestMetricsRejectedError - Tests that rejected counter metric conflicts are handled

Graceful Failure Tests

TestGracefulFailureHandling - Tests that the consensus engine handles failures gracefully
TestMetricsAccuracyDuringErrors - Tests that metrics remain accurate even when errors occur

Implementation Details

Key Components

ErrorBlock - Test block implementation that can inject errors on Accept/Reject/Verify
Consensus - Simplified consensus engine for testing with proper error propagation
ConsensusContext - Context with Prometheus registerer for metrics testing
Parameters - Consensus parameters matching snowball configuration

Adaptations from avalanchego

Used Lux's bag.Bag from node/utils/bag instead of avalanche's bag implementation
Adapted to Lux's block interfaces in engine/chain/block
Used Lux's test utilities from engine/chain/chaintest
Fixed bag initialization issues (requires SetThreshold to avoid nil set bug)
Used separate Prometheus registries per test to avoid metric conflicts

Test Results

All 9 error propagation tests pass successfully:

=== RUN   TestErrorOnAccept
--- PASS: TestErrorOnAccept (0.00s)
=== RUN   TestErrorOnRejectSibling
--- PASS: TestErrorOnRejectSibling (0.00s)
=== RUN   TestMetricsProcessingError
--- PASS: TestMetricsProcessingError (0.00s)
=== RUN   TestErrorOnAddDecidedBlock
--- PASS: TestErrorOnAddDecidedBlock (0.00s)
=== RUN   TestTransitiveRejectionError
--- PASS: TestTransitiveRejectionError (0.00s)
=== RUN   TestMetricsAcceptedError
--- PASS: TestMetricsAcceptedError (0.00s)
=== RUN   TestMetricsRejectedError
--- PASS: TestMetricsRejectedError (0.00s)
=== RUN   TestGracefulFailureHandling
--- PASS: TestGracefulFailureHandling (0.00s)
=== RUN   TestMetricsAccuracyDuringErrors
--- PASS: TestMetricsAccuracyDuringErrors (0.00s)

File Location

/Users/z/work/lux/consensus/engine/chain/error_propagation_test.go

Known Issues

The bag.Bag implementation has a bug where metThreshold set is not properly initialized when threshold is 0
Worked around by setting threshold to 2 before adding elements
This should be fixed in the bag implementation itself

Future Work

Consider fixing the bag initialization bug in /Users/z/work/lux/node/utils/bag/bag.go
Add more comprehensive error scenarios
Add benchmarks for error handling performance

3.1 KiB Raw Permalink Blame History