Files
beads/internal/storage/sqlite/cycle_bench_test.go
Peter Chanthamynavong 3342dc2fe8 perf(sqlite): replace O(2^n) cycle detection with O(V+E) DFS
Replace the recursive SQL CTE in DetectCycles with Go-layer DFS using
shared visited set. The previous implementation enumerated all paths
through the dependency graph, causing exponential blowup with diamond
patterns (multiple issues depending on the same target).

Changes:
- Add loadDependencyGraph() to load deps as adjacency list in one query
- Implement DFS cycle detection with recStack for back-edge detection
- Add normalizeCycle() for consistent cycle deduplication
- Add DetectCycles-specific benchmarks (Linear, Dense, Tree graphs)
- Use direct SQL INSERT in benchmarks to bypass AddDependency overhead

Performance improvement on dense graph (500 nodes, 2500 edges):
- Before: >120s timeout
- After: 1.6ms

Benchmarks:
- DetectCycles_Linear_1000: 0.84ms (1000 nodes, 999 edges)
- DetectCycles_Dense_500: 1.59ms (500 nodes, ~2500 edges)
- DetectCycles_Tree_1000: 0.85ms (1000 nodes, 999 edges)
2025-12-28 06:04:05 -08:00

391 lines
12 KiB
Go

//go:build bench
package sqlite
import (
"context"
"fmt"
"testing"
"github.com/steveyegge/beads/internal/types"
)
// BenchmarkCycleDetection benchmarks the cycle detection performance
// on various graph sizes and structures
//
// Benchmark Results (Apple M4 Max, 2025-10-16):
//
// Linear chains (sparse):
// 100 issues: ~3.4ms per AddDependency (with cycle check)
// 1000 issues: ~3.7ms per AddDependency (with cycle check)
//
// Tree structure (branching factor 3):
// 100 issues: ~3.3ms per AddDependency
// 1000 issues: ~3.5ms per AddDependency
//
// Dense graphs (each issue depends on 3-5 previous):
// 100 issues: Times out (>120s for setup + benchmarking)
// 1000 issues: Times out
//
// Conclusion:
// - Cycle detection adds ~3-4ms overhead per AddDependency call
// - Performance is acceptable for typical use cases (linear chains, trees)
// - Dense graphs with many dependencies can be slow, but are rare in practice
// - No optimization needed for normal workflows
// BenchmarkCycleDetection_Linear_100 tests linear chain (sparse): bd-1 → bd-2 → bd-3 ... → bd-100
func BenchmarkCycleDetection_Linear_100(b *testing.B) {
benchmarkCycleDetectionLinear(b, 100)
}
// BenchmarkCycleDetection_Linear_1000 tests linear chain (sparse): bd-1 → bd-2 → ... → bd-1000
func BenchmarkCycleDetection_Linear_1000(b *testing.B) {
benchmarkCycleDetectionLinear(b, 1000)
}
// BenchmarkCycleDetection_Linear_5000 tests linear chain (sparse): bd-1 → bd-2 → ... → bd-5000
func BenchmarkCycleDetection_Linear_5000(b *testing.B) {
benchmarkCycleDetectionLinear(b, 5000)
}
// BenchmarkCycleDetection_Dense_100 tests dense graph: each issue depends on 3-5 previous issues
func BenchmarkCycleDetection_Dense_100(b *testing.B) {
b.Skip("Dense graph setup slow (creates 5*n deps). AddDependency CTE is O(n), not affected by DetectCycles fix.")
benchmarkCycleDetectionDense(b, 100)
}
// BenchmarkCycleDetection_Dense_1000 tests dense graph with 1000 issues
func BenchmarkCycleDetection_Dense_1000(b *testing.B) {
b.Skip("Dense graph setup slow (creates 5*n deps). AddDependency CTE is O(n), not affected by DetectCycles fix.")
benchmarkCycleDetectionDense(b, 1000)
}
// BenchmarkCycleDetection_Tree_100 tests tree structure (branching factor 3)
func BenchmarkCycleDetection_Tree_100(b *testing.B) {
benchmarkCycleDetectionTree(b, 100)
}
// BenchmarkCycleDetection_Tree_1000 tests tree structure with 1000 issues
func BenchmarkCycleDetection_Tree_1000(b *testing.B) {
benchmarkCycleDetectionTree(b, 1000)
}
// Helper: Create linear dependency chain
func benchmarkCycleDetectionLinear(b *testing.B, n int) {
store, cleanup := setupBenchDB(b)
defer cleanup()
ctx := context.Background()
// Create n issues
issues := make([]*types.Issue, n)
for i := 0; i < n; i++ {
issue := &types.Issue{
Title: fmt.Sprintf("Issue %d", i),
Status: types.StatusOpen,
Priority: 2,
IssueType: types.TypeTask,
}
if err := store.CreateIssue(ctx, issue, "benchmark"); err != nil {
b.Fatalf("Failed to create issue: %v", err)
}
issues[i] = issue
}
// Create linear chain: each issue depends on the previous one
for i := 1; i < n; i++ {
dep := &types.Dependency{
IssueID: issues[i].ID,
DependsOnID: issues[i-1].ID,
Type: types.DepBlocks,
}
if err := store.AddDependency(ctx, dep, "benchmark"); err != nil {
b.Fatalf("Failed to add dependency: %v", err)
}
}
// Now benchmark adding a dependency that would NOT create a cycle
// (from the last issue to a new unconnected issue)
newIssue := &types.Issue{
Title: "New issue",
Status: types.StatusOpen,
Priority: 2,
IssueType: types.TypeTask,
}
if err := store.CreateIssue(ctx, newIssue, "benchmark"); err != nil {
b.Fatalf("Failed to create new issue: %v", err)
}
b.ResetTimer()
for i := 0; i < b.N; i++ {
// Add dependency from first issue to new issue (safe, no cycle)
dep := &types.Dependency{
IssueID: issues[0].ID,
DependsOnID: newIssue.ID,
Type: types.DepBlocks,
}
// This will run cycle detection on a chain of length n
_ = store.AddDependency(ctx, dep, "benchmark")
// Clean up for next iteration
_ = store.RemoveDependency(ctx, issues[0].ID, newIssue.ID, "benchmark")
}
}
// Helper: Create dense dependency graph
func benchmarkCycleDetectionDense(b *testing.B, n int) {
store, cleanup := setupBenchDB(b)
defer cleanup()
ctx := context.Background()
// Create n issues
issues := make([]*types.Issue, n)
for i := 0; i < n; i++ {
issue := &types.Issue{
Title: fmt.Sprintf("Issue %d", i),
Status: types.StatusOpen,
Priority: 2,
IssueType: types.TypeTask,
}
if err := store.CreateIssue(ctx, issue, "benchmark"); err != nil {
b.Fatalf("Failed to create issue: %v", err)
}
issues[i] = issue
}
// Create dense graph: each issue (after 5) depends on 3-5 previous issues
for i := 5; i < n; i++ {
for j := 1; j <= 5 && i-j >= 0; j++ {
dep := &types.Dependency{
IssueID: issues[i].ID,
DependsOnID: issues[i-j].ID,
Type: types.DepBlocks,
}
if err := store.AddDependency(ctx, dep, "benchmark"); err != nil {
b.Fatalf("Failed to add dependency: %v", err)
}
}
}
// Benchmark adding a dependency
newIssue := &types.Issue{
Title: "New issue",
Status: types.StatusOpen,
Priority: 2,
IssueType: types.TypeTask,
}
if err := store.CreateIssue(ctx, newIssue, "benchmark"); err != nil {
b.Fatalf("Failed to create new issue: %v", err)
}
b.ResetTimer()
for i := 0; i < b.N; i++ {
dep := &types.Dependency{
IssueID: issues[n/2].ID, // Middle issue
DependsOnID: newIssue.ID,
Type: types.DepBlocks,
}
_ = store.AddDependency(ctx, dep, "benchmark")
_ = store.RemoveDependency(ctx, issues[n/2].ID, newIssue.ID, "benchmark")
}
}
// Helper: Create tree structure (branching)
func benchmarkCycleDetectionTree(b *testing.B, n int) {
store, cleanup := setupBenchDB(b)
defer cleanup()
ctx := context.Background()
// Create n issues
issues := make([]*types.Issue, n)
for i := 0; i < n; i++ {
issue := &types.Issue{
Title: fmt.Sprintf("Issue %d", i),
Status: types.StatusOpen,
Priority: 2,
IssueType: types.TypeTask,
}
if err := store.CreateIssue(ctx, issue, "benchmark"); err != nil {
b.Fatalf("Failed to create issue: %v", err)
}
issues[i] = issue
}
// Create tree: each issue (after root) depends on parent (branching factor ~3)
for i := 1; i < n; i++ {
parent := (i - 1) / 3
dep := &types.Dependency{
IssueID: issues[i].ID,
DependsOnID: issues[parent].ID,
Type: types.DepBlocks,
}
if err := store.AddDependency(ctx, dep, "benchmark"); err != nil {
b.Fatalf("Failed to add dependency: %v", err)
}
}
// Benchmark adding a dependency
newIssue := &types.Issue{
Title: "New issue",
Status: types.StatusOpen,
Priority: 2,
IssueType: types.TypeTask,
}
if err := store.CreateIssue(ctx, newIssue, "benchmark"); err != nil {
b.Fatalf("Failed to create new issue: %v", err)
}
b.ResetTimer()
for i := 0; i < b.N; i++ {
dep := &types.Dependency{
IssueID: issues[n-1].ID, // Leaf node
DependsOnID: newIssue.ID,
Type: types.DepBlocks,
}
_ = store.AddDependency(ctx, dep, "benchmark")
_ = store.RemoveDependency(ctx, issues[n-1].ID, newIssue.ID, "benchmark")
}
}
// ============================================================================
// DetectCycles Benchmarks
// These benchmark the DetectCycles function directly (not AddDependency).
// The Go DFS fix changed DetectCycles from O(2^n) to O(V+E).
// ============================================================================
// BenchmarkDetectCycles_Linear_1000 benchmarks DetectCycles on a linear chain
func BenchmarkDetectCycles_Linear_1000(b *testing.B) {
store, cleanup := setupBenchDB(b)
defer cleanup()
ctx := context.Background()
createLinearGraph(b, store, ctx, 1000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
_, _ = store.DetectCycles(ctx)
}
}
// BenchmarkDetectCycles_Dense_500 benchmarks DetectCycles on dense graph
// This was O(2^n) before the fix, now O(V+E)
func BenchmarkDetectCycles_Dense_500(b *testing.B) {
store, cleanup := setupBenchDB(b)
defer cleanup()
ctx := context.Background()
createDenseGraphDirect(b, store, ctx, 500)
b.ResetTimer()
for i := 0; i < b.N; i++ {
_, _ = store.DetectCycles(ctx)
}
}
// BenchmarkDetectCycles_Tree_1000 benchmarks DetectCycles on tree structure
func BenchmarkDetectCycles_Tree_1000(b *testing.B) {
store, cleanup := setupBenchDB(b)
defer cleanup()
ctx := context.Background()
createTreeGraph(b, store, ctx, 1000)
b.ResetTimer()
for i := 0; i < b.N; i++ {
_, _ = store.DetectCycles(ctx)
}
}
// createLinearGraph creates n issues with linear chain dependencies
func createLinearGraph(b *testing.B, store *SQLiteStorage, ctx context.Context, n int) []*types.Issue {
issues := make([]*types.Issue, n)
for i := 0; i < n; i++ {
issue := &types.Issue{
Title: fmt.Sprintf("Issue %d", i),
Status: types.StatusOpen,
Priority: 2,
IssueType: types.TypeTask,
}
if err := store.CreateIssue(ctx, issue, "benchmark"); err != nil {
b.Fatalf("Failed to create issue: %v", err)
}
issues[i] = issue
}
// Create linear chain using direct SQL (faster than AddDependency)
for i := 1; i < n; i++ {
_, err := store.db.ExecContext(ctx, `
INSERT INTO dependencies (issue_id, depends_on_id, type, created_at, created_by)
VALUES (?, ?, 'blocks', datetime('now'), 'bench')
`, issues[i].ID, issues[i-1].ID)
if err != nil {
b.Fatalf("Failed to add dependency: %v", err)
}
}
return issues
}
// createDenseGraphDirect creates n issues with dense deps using direct SQL
// Each issue (after 5) depends on the 5 previous issues
// Uses direct SQL to bypass AddDependency's cycle check (O(n) vs O(n²) setup)
func createDenseGraphDirect(b *testing.B, store *SQLiteStorage, ctx context.Context, n int) []*types.Issue {
issues := make([]*types.Issue, n)
for i := 0; i < n; i++ {
issue := &types.Issue{
Title: fmt.Sprintf("Issue %d", i),
Status: types.StatusOpen,
Priority: 2,
IssueType: types.TypeTask,
}
if err := store.CreateIssue(ctx, issue, "benchmark"); err != nil {
b.Fatalf("Failed to create issue: %v", err)
}
issues[i] = issue
}
// Create dense graph using direct SQL (bypasses cycle check during setup)
for i := 5; i < n; i++ {
for j := 1; j <= 5 && i-j >= 0; j++ {
_, err := store.db.ExecContext(ctx, `
INSERT INTO dependencies (issue_id, depends_on_id, type, created_at, created_by)
VALUES (?, ?, 'blocks', datetime('now'), 'bench')
`, issues[i].ID, issues[i-j].ID)
if err != nil {
b.Fatalf("Failed to add dependency: %v", err)
}
}
}
return issues
}
// createTreeGraph creates n issues in tree structure (branching factor 3)
func createTreeGraph(b *testing.B, store *SQLiteStorage, ctx context.Context, n int) []*types.Issue {
issues := make([]*types.Issue, n)
for i := 0; i < n; i++ {
issue := &types.Issue{
Title: fmt.Sprintf("Issue %d", i),
Status: types.StatusOpen,
Priority: 2,
IssueType: types.TypeTask,
}
if err := store.CreateIssue(ctx, issue, "benchmark"); err != nil {
b.Fatalf("Failed to create issue: %v", err)
}
issues[i] = issue
}
// Create tree using direct SQL
for i := 1; i < n; i++ {
parent := (i - 1) / 3
_, err := store.db.ExecContext(ctx, `
INSERT INTO dependencies (issue_id, depends_on_id, type, created_at, created_by)
VALUES (?, ?, 'blocks', datetime('now'), 'bench')
`, issues[i].ID, issues[parent].ID)
if err != nil {
b.Fatalf("Failed to add dependency: %v", err)
}
}
return issues
}