You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
520 lines
16 KiB
520 lines
16 KiB
package main |
|
|
|
import ( |
|
"context" |
|
"fmt" |
|
"sort" |
|
"sync" |
|
"time" |
|
|
|
"git.mleku.dev/mleku/nostr/encoders/event" |
|
"git.mleku.dev/mleku/nostr/encoders/filter" |
|
"git.mleku.dev/mleku/nostr/encoders/hex" |
|
"git.mleku.dev/mleku/nostr/encoders/kind" |
|
"git.mleku.dev/mleku/nostr/encoders/tag" |
|
"git.mleku.dev/mleku/nostr/interfaces/signer/p8k" |
|
"lukechampine.com/frand" |
|
"next.orly.dev/pkg/database" |
|
) |
|
|
|
const ( |
|
// GraphBenchNumPubkeys is the number of pubkeys to generate for graph benchmark |
|
GraphBenchNumPubkeys = 100000 |
|
// GraphBenchMinFollows is the minimum number of follows per pubkey |
|
GraphBenchMinFollows = 1 |
|
// GraphBenchMaxFollows is the maximum number of follows per pubkey |
|
GraphBenchMaxFollows = 1000 |
|
// GraphBenchSeed is the deterministic seed for frand PRNG (fits in uint64) |
|
GraphBenchSeed uint64 = 0x4E6F737472 // "Nostr" in hex |
|
// GraphBenchTraversalDepth is the depth of graph traversal (3 = third degree) |
|
GraphBenchTraversalDepth = 3 |
|
) |
|
|
|
// GraphTraversalBenchmark benchmarks graph traversal using NIP-01 style queries |
|
type GraphTraversalBenchmark struct { |
|
config *BenchmarkConfig |
|
db *database.D |
|
results []*BenchmarkResult |
|
mu sync.RWMutex |
|
|
|
// Cached data for the benchmark |
|
pubkeys [][]byte // 100k pubkeys as 32-byte arrays |
|
signers []*p8k.Signer // signers for each pubkey |
|
follows [][]int // follows[i] = list of indices that pubkey[i] follows |
|
rng *frand.RNG // deterministic PRNG |
|
} |
|
|
|
// NewGraphTraversalBenchmark creates a new graph traversal benchmark |
|
func NewGraphTraversalBenchmark(config *BenchmarkConfig, db *database.D) *GraphTraversalBenchmark { |
|
return &GraphTraversalBenchmark{ |
|
config: config, |
|
db: db, |
|
results: make([]*BenchmarkResult, 0), |
|
rng: frand.NewCustom(make([]byte, 32), 1024, 12), // ChaCha12 with seed buffer |
|
} |
|
} |
|
|
|
// initializeDeterministicRNG initializes the PRNG with deterministic seed |
|
func (g *GraphTraversalBenchmark) initializeDeterministicRNG() { |
|
// Create seed buffer from GraphBenchSeed (uint64 spread across 8 bytes) |
|
seedBuf := make([]byte, 32) |
|
seed := GraphBenchSeed |
|
seedBuf[0] = byte(seed >> 56) |
|
seedBuf[1] = byte(seed >> 48) |
|
seedBuf[2] = byte(seed >> 40) |
|
seedBuf[3] = byte(seed >> 32) |
|
seedBuf[4] = byte(seed >> 24) |
|
seedBuf[5] = byte(seed >> 16) |
|
seedBuf[6] = byte(seed >> 8) |
|
seedBuf[7] = byte(seed) |
|
g.rng = frand.NewCustom(seedBuf, 1024, 12) |
|
} |
|
|
|
// generatePubkeys generates deterministic pubkeys using frand |
|
func (g *GraphTraversalBenchmark) generatePubkeys() { |
|
fmt.Printf("Generating %d deterministic pubkeys...\n", GraphBenchNumPubkeys) |
|
start := time.Now() |
|
|
|
g.initializeDeterministicRNG() |
|
g.pubkeys = make([][]byte, GraphBenchNumPubkeys) |
|
g.signers = make([]*p8k.Signer, GraphBenchNumPubkeys) |
|
|
|
for i := 0; i < GraphBenchNumPubkeys; i++ { |
|
// Generate deterministic 32-byte secret key from PRNG |
|
secretKey := make([]byte, 32) |
|
g.rng.Read(secretKey) |
|
|
|
// Create signer from secret key |
|
signer := p8k.MustNew() |
|
if err := signer.InitSec(secretKey); err != nil { |
|
panic(fmt.Sprintf("failed to init signer %d: %v", i, err)) |
|
} |
|
|
|
g.signers[i] = signer |
|
g.pubkeys[i] = make([]byte, 32) |
|
copy(g.pubkeys[i], signer.Pub()) |
|
|
|
if (i+1)%10000 == 0 { |
|
fmt.Printf(" Generated %d/%d pubkeys...\n", i+1, GraphBenchNumPubkeys) |
|
} |
|
} |
|
|
|
fmt.Printf("Generated %d pubkeys in %v\n", GraphBenchNumPubkeys, time.Since(start)) |
|
} |
|
|
|
// generateFollowGraph generates the random follow graph with deterministic PRNG |
|
func (g *GraphTraversalBenchmark) generateFollowGraph() { |
|
fmt.Printf("Generating follow graph (1-%d follows per pubkey)...\n", GraphBenchMaxFollows) |
|
start := time.Now() |
|
|
|
// Reset RNG to ensure deterministic follow graph |
|
g.initializeDeterministicRNG() |
|
// Skip the bytes used for pubkey generation |
|
skipBuf := make([]byte, 32*GraphBenchNumPubkeys) |
|
g.rng.Read(skipBuf) |
|
|
|
g.follows = make([][]int, GraphBenchNumPubkeys) |
|
|
|
totalFollows := 0 |
|
for i := 0; i < GraphBenchNumPubkeys; i++ { |
|
// Determine number of follows for this pubkey (1 to 1000) |
|
numFollows := int(g.rng.Uint64n(uint64(GraphBenchMaxFollows-GraphBenchMinFollows+1))) + GraphBenchMinFollows |
|
|
|
// Generate random follow indices (excluding self) |
|
followSet := make(map[int]struct{}) |
|
for len(followSet) < numFollows { |
|
followIdx := int(g.rng.Uint64n(uint64(GraphBenchNumPubkeys))) |
|
if followIdx != i { |
|
followSet[followIdx] = struct{}{} |
|
} |
|
} |
|
|
|
// Convert to slice |
|
g.follows[i] = make([]int, 0, numFollows) |
|
for idx := range followSet { |
|
g.follows[i] = append(g.follows[i], idx) |
|
} |
|
totalFollows += numFollows |
|
|
|
if (i+1)%10000 == 0 { |
|
fmt.Printf(" Generated follow lists for %d/%d pubkeys...\n", i+1, GraphBenchNumPubkeys) |
|
} |
|
} |
|
|
|
avgFollows := float64(totalFollows) / float64(GraphBenchNumPubkeys) |
|
fmt.Printf("Generated follow graph in %v (avg %.1f follows/pubkey, total %d follows)\n", |
|
time.Since(start), avgFollows, totalFollows) |
|
} |
|
|
|
// createFollowListEvents creates kind 3 follow list events in the database |
|
func (g *GraphTraversalBenchmark) createFollowListEvents() { |
|
fmt.Println("Creating follow list events in database...") |
|
start := time.Now() |
|
|
|
ctx := context.Background() |
|
baseTime := time.Now().Unix() |
|
|
|
var mu sync.Mutex |
|
var wg sync.WaitGroup |
|
var successCount, errorCount int64 |
|
latencies := make([]time.Duration, 0, GraphBenchNumPubkeys) |
|
|
|
// Use worker pool for parallel event creation |
|
numWorkers := g.config.ConcurrentWorkers |
|
if numWorkers < 1 { |
|
numWorkers = 4 |
|
} |
|
|
|
workChan := make(chan int, numWorkers*2) |
|
|
|
// Rate limiter: cap at 20,000 events/second |
|
perWorkerRate := 20000.0 / float64(numWorkers) |
|
|
|
for w := 0; w < numWorkers; w++ { |
|
wg.Add(1) |
|
go func() { |
|
defer wg.Done() |
|
|
|
workerLimiter := NewRateLimiter(perWorkerRate) |
|
|
|
for i := range workChan { |
|
workerLimiter.Wait() |
|
|
|
ev := event.New() |
|
ev.Kind = kind.FollowList.K |
|
ev.CreatedAt = baseTime + int64(i) |
|
ev.Content = []byte("") |
|
ev.Tags = tag.NewS() |
|
|
|
// Add p tags for all follows |
|
for _, followIdx := range g.follows[i] { |
|
pubkeyHex := hex.Enc(g.pubkeys[followIdx]) |
|
ev.Tags.Append(tag.NewFromAny("p", pubkeyHex)) |
|
} |
|
|
|
// Sign the event |
|
if err := ev.Sign(g.signers[i]); err != nil { |
|
mu.Lock() |
|
errorCount++ |
|
mu.Unlock() |
|
ev.Free() |
|
continue |
|
} |
|
|
|
// Save to database |
|
eventStart := time.Now() |
|
_, err := g.db.SaveEvent(ctx, ev) |
|
latency := time.Since(eventStart) |
|
|
|
mu.Lock() |
|
if err != nil { |
|
errorCount++ |
|
} else { |
|
successCount++ |
|
latencies = append(latencies, latency) |
|
} |
|
mu.Unlock() |
|
|
|
ev.Free() |
|
} |
|
}() |
|
} |
|
|
|
// Send work |
|
for i := 0; i < GraphBenchNumPubkeys; i++ { |
|
workChan <- i |
|
if (i+1)%10000 == 0 { |
|
fmt.Printf(" Queued %d/%d follow list events...\n", i+1, GraphBenchNumPubkeys) |
|
} |
|
} |
|
close(workChan) |
|
wg.Wait() |
|
|
|
duration := time.Since(start) |
|
eventsPerSec := float64(successCount) / duration.Seconds() |
|
|
|
// Calculate latency stats |
|
var avgLatency, p95Latency, p99Latency time.Duration |
|
if len(latencies) > 0 { |
|
sort.Slice(latencies, func(i, j int) bool { return latencies[i] < latencies[j] }) |
|
avgLatency = calculateAvgLatency(latencies) |
|
p95Latency = calculatePercentileLatency(latencies, 0.95) |
|
p99Latency = calculatePercentileLatency(latencies, 0.99) |
|
} |
|
|
|
fmt.Printf("Created %d follow list events in %v (%.2f events/sec, errors: %d)\n", |
|
successCount, duration, eventsPerSec, errorCount) |
|
fmt.Printf(" Avg latency: %v, P95: %v, P99: %v\n", avgLatency, p95Latency, p99Latency) |
|
|
|
// Record result for event creation phase |
|
result := &BenchmarkResult{ |
|
TestName: "Graph Setup (Follow Lists)", |
|
Duration: duration, |
|
TotalEvents: int(successCount), |
|
EventsPerSecond: eventsPerSec, |
|
AvgLatency: avgLatency, |
|
P95Latency: p95Latency, |
|
P99Latency: p99Latency, |
|
ConcurrentWorkers: numWorkers, |
|
MemoryUsed: getMemUsage(), |
|
SuccessRate: float64(successCount) / float64(GraphBenchNumPubkeys) * 100, |
|
} |
|
|
|
g.mu.Lock() |
|
g.results = append(g.results, result) |
|
g.mu.Unlock() |
|
} |
|
|
|
// runThirdDegreeTraversal runs the third-degree graph traversal benchmark |
|
func (g *GraphTraversalBenchmark) runThirdDegreeTraversal() { |
|
fmt.Printf("\n=== Third-Degree Graph Traversal Benchmark ===\n") |
|
fmt.Printf("Traversing 3 degrees of follows for each of %d pubkeys...\n", GraphBenchNumPubkeys) |
|
|
|
start := time.Now() |
|
ctx := context.Background() |
|
|
|
var mu sync.Mutex |
|
var wg sync.WaitGroup |
|
var totalQueries int64 |
|
var totalPubkeysFound int64 |
|
queryLatencies := make([]time.Duration, 0, GraphBenchNumPubkeys*3) |
|
traversalLatencies := make([]time.Duration, 0, GraphBenchNumPubkeys) |
|
|
|
// Sample a subset for detailed traversal (full 100k would take too long) |
|
sampleSize := 1000 |
|
if sampleSize > GraphBenchNumPubkeys { |
|
sampleSize = GraphBenchNumPubkeys |
|
} |
|
|
|
// Deterministic sampling |
|
g.initializeDeterministicRNG() |
|
sampleIndices := make([]int, sampleSize) |
|
for i := 0; i < sampleSize; i++ { |
|
sampleIndices[i] = int(g.rng.Uint64n(uint64(GraphBenchNumPubkeys))) |
|
} |
|
|
|
fmt.Printf("Sampling %d pubkeys for traversal...\n", sampleSize) |
|
|
|
numWorkers := g.config.ConcurrentWorkers |
|
if numWorkers < 1 { |
|
numWorkers = 4 |
|
} |
|
|
|
workChan := make(chan int, numWorkers*2) |
|
|
|
for w := 0; w < numWorkers; w++ { |
|
wg.Add(1) |
|
go func() { |
|
defer wg.Done() |
|
|
|
for startIdx := range workChan { |
|
traversalStart := time.Now() |
|
foundPubkeys := make(map[string]struct{}) |
|
|
|
// Start with the initial pubkey |
|
currentLevel := [][]byte{g.pubkeys[startIdx]} |
|
startPubkeyHex := hex.Enc(g.pubkeys[startIdx]) |
|
foundPubkeys[startPubkeyHex] = struct{}{} |
|
|
|
// Traverse 3 degrees |
|
for depth := 0; depth < GraphBenchTraversalDepth; depth++ { |
|
if len(currentLevel) == 0 { |
|
break |
|
} |
|
|
|
nextLevel := make([][]byte, 0) |
|
|
|
// Query follow lists for all pubkeys at current level |
|
// Batch queries for efficiency |
|
batchSize := 100 |
|
for batchStart := 0; batchStart < len(currentLevel); batchStart += batchSize { |
|
batchEnd := batchStart + batchSize |
|
if batchEnd > len(currentLevel) { |
|
batchEnd = len(currentLevel) |
|
} |
|
|
|
batch := currentLevel[batchStart:batchEnd] |
|
|
|
// Build filter for kind 3 events from these pubkeys |
|
f := filter.New() |
|
f.Kinds = kind.NewS(kind.FollowList) |
|
f.Authors = tag.NewWithCap(len(batch)) |
|
for _, pk := range batch { |
|
// Authors.T expects raw byte slices (pubkeys) |
|
f.Authors.T = append(f.Authors.T, pk) |
|
} |
|
|
|
queryStart := time.Now() |
|
events, err := g.db.QueryEvents(ctx, f) |
|
queryLatency := time.Since(queryStart) |
|
|
|
mu.Lock() |
|
totalQueries++ |
|
queryLatencies = append(queryLatencies, queryLatency) |
|
mu.Unlock() |
|
|
|
if err != nil { |
|
continue |
|
} |
|
|
|
// Extract followed pubkeys from p tags |
|
for _, ev := range events { |
|
for _, t := range *ev.Tags { |
|
if len(t.T) >= 2 && string(t.T[0]) == "p" { |
|
pubkeyHex := string(t.ValueHex()) |
|
if _, exists := foundPubkeys[pubkeyHex]; !exists { |
|
foundPubkeys[pubkeyHex] = struct{}{} |
|
// Decode hex to bytes for next level |
|
if pkBytes, err := hex.Dec(pubkeyHex); err == nil { |
|
nextLevel = append(nextLevel, pkBytes) |
|
} |
|
} |
|
} |
|
} |
|
ev.Free() |
|
} |
|
} |
|
|
|
currentLevel = nextLevel |
|
} |
|
|
|
traversalLatency := time.Since(traversalStart) |
|
|
|
mu.Lock() |
|
totalPubkeysFound += int64(len(foundPubkeys)) |
|
traversalLatencies = append(traversalLatencies, traversalLatency) |
|
mu.Unlock() |
|
} |
|
}() |
|
} |
|
|
|
// Send work |
|
for _, idx := range sampleIndices { |
|
workChan <- idx |
|
} |
|
close(workChan) |
|
wg.Wait() |
|
|
|
duration := time.Since(start) |
|
|
|
// Calculate statistics |
|
var avgQueryLatency, p95QueryLatency, p99QueryLatency time.Duration |
|
if len(queryLatencies) > 0 { |
|
sort.Slice(queryLatencies, func(i, j int) bool { return queryLatencies[i] < queryLatencies[j] }) |
|
avgQueryLatency = calculateAvgLatency(queryLatencies) |
|
p95QueryLatency = calculatePercentileLatency(queryLatencies, 0.95) |
|
p99QueryLatency = calculatePercentileLatency(queryLatencies, 0.99) |
|
} |
|
|
|
var avgTraversalLatency, p95TraversalLatency, p99TraversalLatency time.Duration |
|
if len(traversalLatencies) > 0 { |
|
sort.Slice(traversalLatencies, func(i, j int) bool { return traversalLatencies[i] < traversalLatencies[j] }) |
|
avgTraversalLatency = calculateAvgLatency(traversalLatencies) |
|
p95TraversalLatency = calculatePercentileLatency(traversalLatencies, 0.95) |
|
p99TraversalLatency = calculatePercentileLatency(traversalLatencies, 0.99) |
|
} |
|
|
|
avgPubkeysPerTraversal := float64(totalPubkeysFound) / float64(sampleSize) |
|
traversalsPerSec := float64(sampleSize) / duration.Seconds() |
|
queriesPerSec := float64(totalQueries) / duration.Seconds() |
|
|
|
fmt.Printf("\n=== Graph Traversal Results ===\n") |
|
fmt.Printf("Traversals completed: %d\n", sampleSize) |
|
fmt.Printf("Total queries: %d (%.2f queries/sec)\n", totalQueries, queriesPerSec) |
|
fmt.Printf("Avg pubkeys found per traversal: %.1f\n", avgPubkeysPerTraversal) |
|
fmt.Printf("Total duration: %v\n", duration) |
|
fmt.Printf("\nQuery Latencies:\n") |
|
fmt.Printf(" Avg: %v, P95: %v, P99: %v\n", avgQueryLatency, p95QueryLatency, p99QueryLatency) |
|
fmt.Printf("\nFull Traversal Latencies (3 degrees):\n") |
|
fmt.Printf(" Avg: %v, P95: %v, P99: %v\n", avgTraversalLatency, p95TraversalLatency, p99TraversalLatency) |
|
fmt.Printf("Traversals/sec: %.2f\n", traversalsPerSec) |
|
|
|
// Record result for traversal phase |
|
result := &BenchmarkResult{ |
|
TestName: "Graph Traversal (3 Degrees)", |
|
Duration: duration, |
|
TotalEvents: int(totalQueries), |
|
EventsPerSecond: traversalsPerSec, |
|
AvgLatency: avgTraversalLatency, |
|
P90Latency: calculatePercentileLatency(traversalLatencies, 0.90), |
|
P95Latency: p95TraversalLatency, |
|
P99Latency: p99TraversalLatency, |
|
Bottom10Avg: calculateBottom10Avg(traversalLatencies), |
|
ConcurrentWorkers: numWorkers, |
|
MemoryUsed: getMemUsage(), |
|
SuccessRate: 100.0, |
|
} |
|
|
|
g.mu.Lock() |
|
g.results = append(g.results, result) |
|
g.mu.Unlock() |
|
|
|
// Also record query performance separately |
|
queryResult := &BenchmarkResult{ |
|
TestName: "Graph Queries (Follow Lists)", |
|
Duration: duration, |
|
TotalEvents: int(totalQueries), |
|
EventsPerSecond: queriesPerSec, |
|
AvgLatency: avgQueryLatency, |
|
P90Latency: calculatePercentileLatency(queryLatencies, 0.90), |
|
P95Latency: p95QueryLatency, |
|
P99Latency: p99QueryLatency, |
|
Bottom10Avg: calculateBottom10Avg(queryLatencies), |
|
ConcurrentWorkers: numWorkers, |
|
MemoryUsed: getMemUsage(), |
|
SuccessRate: 100.0, |
|
} |
|
|
|
g.mu.Lock() |
|
g.results = append(g.results, queryResult) |
|
g.mu.Unlock() |
|
} |
|
|
|
// RunSuite runs the complete graph traversal benchmark suite |
|
func (g *GraphTraversalBenchmark) RunSuite() { |
|
fmt.Println("\n╔════════════════════════════════════════════════════════╗") |
|
fmt.Println("║ GRAPH TRAVERSAL BENCHMARK (100k Pubkeys) ║") |
|
fmt.Println("╚════════════════════════════════════════════════════════╝") |
|
|
|
// Step 1: Generate pubkeys |
|
g.generatePubkeys() |
|
|
|
// Step 2: Generate follow graph |
|
g.generateFollowGraph() |
|
|
|
// Step 3: Create follow list events in database |
|
g.createFollowListEvents() |
|
|
|
// Step 4: Run third-degree traversal benchmark |
|
g.runThirdDegreeTraversal() |
|
|
|
fmt.Printf("\n=== Graph Traversal Benchmark Complete ===\n\n") |
|
} |
|
|
|
// GetResults returns the benchmark results |
|
func (g *GraphTraversalBenchmark) GetResults() []*BenchmarkResult { |
|
g.mu.RLock() |
|
defer g.mu.RUnlock() |
|
return g.results |
|
} |
|
|
|
// PrintResults prints the benchmark results |
|
func (g *GraphTraversalBenchmark) PrintResults() { |
|
g.mu.RLock() |
|
defer g.mu.RUnlock() |
|
|
|
for _, result := range g.results { |
|
fmt.Printf("\nTest: %s\n", result.TestName) |
|
fmt.Printf("Duration: %v\n", result.Duration) |
|
fmt.Printf("Total Events/Queries: %d\n", result.TotalEvents) |
|
fmt.Printf("Events/sec: %.2f\n", result.EventsPerSecond) |
|
fmt.Printf("Success Rate: %.1f%%\n", result.SuccessRate) |
|
fmt.Printf("Concurrent Workers: %d\n", result.ConcurrentWorkers) |
|
fmt.Printf("Memory Used: %d MB\n", result.MemoryUsed/(1024*1024)) |
|
fmt.Printf("Avg Latency: %v\n", result.AvgLatency) |
|
fmt.Printf("P90 Latency: %v\n", result.P90Latency) |
|
fmt.Printf("P95 Latency: %v\n", result.P95Latency) |
|
fmt.Printf("P99 Latency: %v\n", result.P99Latency) |
|
fmt.Printf("Bottom 10%% Avg Latency: %v\n", result.Bottom10Avg) |
|
} |
|
}
|
|
|