feat: add segment-based streaming ZIP writer, ZIP64 modes; replace legacy writer

sdk/internal/archive/benchmark_test.go

@@ -0,0 +1,371 @@
+package archive
+
+import (
+	"math/rand"
+	"testing"
+)
+
+// BenchmarkSegmentWriter_CRC32ContiguousProcessing benchmarks CRC32-related performance under
+// different segment write ordering patterns. The current implementation uses CRC32-combine over
+// per-segment CRCs and sizes and does not retain payload bytes between calls.
+//
+// Test patterns:
+//   - sequential: Optimal case where segments arrive in order (enables immediate processing)
+//   - reverse: Worst case where all segments must be buffered until the end
+//   - random: Pseudo-random order using deterministic pattern for reproducible results
+//   - interleaved: Moderate out-of-order (even indices first, then odd)
+//   - worst_case: Middle-out pattern that maximizes memory buffering requirements
+//
+// Measures: CRC32 calculation speed, memory allocation patterns, contiguous processing effectiveness
+func BenchmarkSegmentWriter_CRC32ContiguousProcessing(b *testing.B) {
+	testCases := []struct {
+		name         string
+		segmentCount int
+		segmentSize  int
+		writeOrder   string
+	}{
+		{"sequential_100x1KB", 100, 1024, "sequential"},
+		{"reverse_100x1KB", 100, 1024, "reverse"},
+		{"random_100x1KB", 100, 1024, "random"},
+		{"interleaved_100x1KB", 100, 1024, "interleaved"},
+		{"worst_case_100x1KB", 100, 1024, "worst_case"},
+		{"sequential_1000x1KB", 1000, 1024, "sequential"},
+		{"reverse_1000x1KB", 1000, 1024, "reverse"},
+		{"worst_case_1000x1KB", 1000, 1024, "worst_case"},
+	}
+
+	for _, tc := range testCases {
+		b.Run(tc.name, func(b *testing.B) {
+			// Generate write order based on pattern
+			writeOrder := generateWriteOrder(tc.segmentCount, tc.writeOrder)
+
+			b.ResetTimer()
+			b.ReportAllocs()
+
+			for i := 0; i < b.N; i++ {
+				writer := NewSegmentTDFWriter(tc.segmentCount)
+				ctx := b.Context()
+
+				// Create test segment data
+				segmentData := make([]byte, tc.segmentSize)
+				for j := range segmentData {
+					segmentData[j] = byte(j % 256)
+				}
+
+				// Write segments in specified order
+				for _, segIdx := range writeOrder {
+					_, err := writer.WriteSegment(ctx, segIdx, segmentData)
+					if err != nil {
+						b.Fatal(err)
+					}
+				}
+
+				// Finalize to trigger final CRC32 calculation
+				manifest := []byte(`{"test": "benchmark"}`)
+				_, err := writer.Finalize(ctx, manifest)
+				if err != nil {
+					b.Fatal(err)
+				}
+
+				writer.Close()
+			}
+		})
+	}
+}
+
+// BenchmarkSegmentWriter_VariableSegmentSizes benchmarks performance impact of variable segment sizes
+// on memory allocation and CRC32 processing efficiency.
+//
+// This benchmark tests how the segment writer handles dynamic memory allocation when segments have
+// unpredictable sizes. Variable sizes can impact both memory allocation patterns and CRC32 processing
+// efficiency, as larger segments require more memory and processing time.
+//
+// Test patterns:
+//   - uniform_1KB: Baseline with consistent 1KB segments for comparison
+//   - doubling: Exponentially increasing sizes (512B → 8KB) to test scaling
+//   - extreme_variance: Mixed small/large segments to stress memory allocator
+//   - fibonacci_like: Fibonacci-inspired progression for gradual size increases
+//   - large_mixed: Various large segments to test high memory usage patterns
+//
+// Measures: Memory allocation efficiency, CRC32 processing with varying data volumes, GC impact
+func BenchmarkSegmentWriter_VariableSegmentSizes(b *testing.B) {
+	testCases := []struct {
+		name  string
+		sizes []int
+	}{
+		{"uniform_1KB", []int{1024, 1024, 1024, 1024, 1024}},
+		{"doubling", []int{512, 1024, 2048, 4096, 8192}},
+		{"extreme_variance", []int{100, 10240, 200, 20480, 300}},
+		{"fibonacci_like", []int{256, 512, 768, 1280, 2048}},
+		{"large_mixed", []int{1024, 16384, 4096, 32768, 8192}},
+	}
+
+	for _, tc := range testCases {
+		b.Run(tc.name, func(b *testing.B) {
+			b.ResetTimer()
+			b.ReportAllocs()
+
+			for i := 0; i < b.N; i++ {
+				writer := NewSegmentTDFWriter(len(tc.sizes))
+				ctx := b.Context()
+
+				// Write segments with variable sizes
+				for segIdx, size := range tc.sizes {
+					segmentData := make([]byte, size)
+					for j := range segmentData {
+						segmentData[j] = byte((segIdx * j) % 256)
+					}
+
+					_, err := writer.WriteSegment(ctx, segIdx, segmentData)
+					if err != nil {
+						b.Fatal(err)
+					}
+				}
+
+				// Finalize
+				manifest := []byte(`{"variable_sizes": true}`)
+				_, err := writer.Finalize(ctx, manifest)
+				if err != nil {
+					b.Fatal(err)
+				}
+
+				writer.Close()
+			}
+		})
+	}
+}
+
+// BenchmarkSegmentWriter_MemoryPressure benchmarks memory allocation patterns and buffering efficiency
+// under various segment count and size combinations.
+//
+// This benchmark specifically targets memory allocation behavior to identify potential memory leaks,
+// inefficient buffering strategies, and garbage collection impact. It uses WithMaxSegments(count*2)
+// to allow extra buffering capacity and tests different buffer policies.
+//
+// Test scenarios:
+//   - small_segments: High segment count (1000) with minimal individual memory (512B each)
+//   - large_segments: Fewer segments (100) with larger memory footprint (8KB each)
+//   - mixed_sizes: Dynamic sizes from 512B to 4KB based on segment index modulo
+//
+// Write patterns test memory behavior:
+//   - sequential: Minimal buffering, immediate processing and cleanup
+//   - reverse: Maximum buffering until all segments received
+//   - interleaved: Moderate buffering with periodic cleanup opportunities
+//   - worst_case: Scattered pattern maximizing memory retention
+//
+// Measures: Peak memory usage, allocation patterns, buffer cleanup efficiency, GC pressure
+func BenchmarkSegmentWriter_MemoryPressure(b *testing.B) {
+	testCases := []struct {
+		name         string
+		segmentCount int
+		segmentSize  int
+		bufferPolicy string
+	}{
+		{"small_segments_sequential", 1000, 512, "sequential"},
+		{"small_segments_reverse", 1000, 512, "reverse"},
+		{"small_segments_worst_case", 1000, 512, "worst_case"},
+		{"large_segments_sequential", 100, 8192, "sequential"},
+		{"large_segments_reverse", 100, 8192, "reverse"},
+		{"large_segments_interleaved", 100, 8192, "interleaved"},
+		{"mixed_sizes_random", 500, 0, "mixed"}, // 0 = variable sizes
+	}
+
+	for _, tc := range testCases {
+		b.Run(tc.name, func(b *testing.B) {
+			writeOrder := generateWriteOrder(tc.segmentCount, tc.bufferPolicy)
+
+			b.ResetTimer()
+			b.ReportAllocs()
+
+			for i := 0; i < b.N; i++ {
+				writer := NewSegmentTDFWriter(tc.segmentCount, WithMaxSegments(tc.segmentCount*2))
+				ctx := b.Context()
+
+				// Write segments with focus on memory allocation patterns
+				for orderIdx, segIdx := range writeOrder {
+					var segmentData []byte
+
+					if tc.segmentSize == 0 { // Mixed sizes mode
+						size := 512 + (segIdx%8)*512 // Sizes from 512 to 4096
+						segmentData = make([]byte, size)
+					} else {
+						segmentData = make([]byte, tc.segmentSize)
+					}
+
+					// Fill with deterministic test data
+					for j := range segmentData {
+						segmentData[j] = byte((orderIdx * j) % 256)
+					}
+
+					_, err := writer.WriteSegment(ctx, segIdx, segmentData)
+					if err != nil {
+						b.Fatal(err)
+					}
+				}
+
+				// Finalize
+				manifest := []byte(`{"memory_test": true}`)
+				_, err := writer.Finalize(ctx, manifest)
+				if err != nil {
+					b.Fatal(err)
+				}
+
+				writer.Close()
+			}
+		})
+	}
+}
+
+// BenchmarkSegmentWriter_ZIPGeneration benchmarks ZIP archive structure generation performance,
+// focusing on the finalization process where the complete ZIP structure is assembled.
+//
+// This benchmark measures the overhead of generating ZIP format structures including local file headers,
+// central directory records, and data descriptors. It compares ZIP32 vs ZIP64 performance and tests
+// the final assembly process during Finalize() calls.
+//
+// Test scenarios:
+//   - zip32_small/large: Standard ZIP format (supports files <4GB) with varying segment counts
+//   - zip64_small/large: ZIP64 format (handles >4GB files) with extended headers
+//   - zip64_huge_segments: Large 64KB segments that require ZIP64 format
+//
+// The benchmark focuses on finalization overhead including:
+//   - Data descriptor generation for streaming entries
+//   - Central directory assembly with file metadata
+//   - ZIP64 extended information extra fields when needed
+//   - Final ZIP structure validation and writing
+//
+// Measures: ZIP structure generation speed, ZIP32 vs ZIP64 overhead, finalization efficiency
+func BenchmarkSegmentWriter_ZIPGeneration(b *testing.B) {
+	testCases := []struct {
+		name         string
+		segmentCount int
+		segmentSize  int
+		zip64Mode    Zip64Mode
+	}{
+		{"zip32_small", 10, 1024, Zip64Never},
+		{"zip32_large", 100, 1024, Zip64Never},
+		{"zip64_small", 10, 1024, Zip64Always},
+		{"zip64_large", 100, 1024, Zip64Always},
+		{"zip64_huge_segments", 5, 65536, Zip64Auto}, // Auto triggers ZIP64 by size
+	}
+
+	for _, tc := range testCases {
+		b.Run(tc.name, func(b *testing.B) {
+			options := []Option{WithZip64Mode(tc.zip64Mode)}
+
+			b.ResetTimer()
+			b.ReportAllocs()
+
+			for i := 0; i < b.N; i++ {
+				writer := NewSegmentTDFWriter(tc.segmentCount, options...)
+				ctx := b.Context()
+
+				// Create test segment data
+				segmentData := make([]byte, tc.segmentSize)
+				for j := range segmentData {
+					segmentData[j] = byte(j % 256)
+				}
+
+				// Write all segments
+				for segIdx := 0; segIdx < tc.segmentCount; segIdx++ {
+					_, err := writer.WriteSegment(ctx, segIdx, segmentData)
+					if err != nil {
+						b.Fatal(err)
+					}
+				}
+
+				// Focus benchmark on finalization (ZIP generation)
+				manifest := []byte(`{"zip_generation_test": true}`)
+				_, err := writer.Finalize(ctx, manifest)
+				if err != nil {
+					b.Fatal(err)
+				}
+
+				writer.Close()
+			}
+		})
+	}
+}
+
+// generateWriteOrder creates deterministic segment write orders for consistent benchmark testing.
+//
+// This function generates various write patterns to test different aspects of the segment writer's
+// performance characteristics. All patterns are deterministic to ensure reproducible benchmark results.
+//
+// Supported patterns:
+//   - "sequential": Natural order (0,1,2,3...)
+//   - "reverse": Backward order (...3,2,1,0)
+//   - "interleaved": Even indices first (0,2,4...), then odd (1,3,5...) - moderate out-of-order
+//   - "worst_case": Middle-out pattern starting from center, alternating left/right - maximizes buffering
+//   - "random"/"mixed": Deterministic shuffle using rand.Shuffle with a fixed seed
+//
+// The patterns are designed to stress different aspects:
+//   - Memory usage patterns
+//   - Processing efficiency
+//   - Cache locality (how segments are accessed in memory)
+func generateWriteOrder(count int, pattern string) []int {
+	order := make([]int, count)
+
+	switch pattern {
+	case "sequential":
+		for i := 0; i < count; i++ {
+			order[i] = i
+		}
+	case "reverse":
+		for i := 0; i < count; i++ {
+			order[i] = count - 1 - i
+		}
+	case "interleaved":
+		// Interleaved pattern: 0,2,4,6...1,3,5,7... (moderate out-of-order)
+		idx := 0
+		// First pass: even indices
+		for i := 0; i < count; i += 2 {
+			if idx < count {
+				order[idx] = i
+				idx++
+			}
+		}
+		// Second pass: odd indices
+		for i := 1; i < count; i += 2 {
+			if idx < count {
+				order[idx] = i
+				idx++
+			}
+		}
+	case "worst_case":
+		// A scattered pattern that stresses segment bookkeeping
+		mid := count / 2
+		order[0] = mid
+		left, right := mid-1, mid+1
+		idx := 1
+
+		// Alternate left and right from middle
+		for left >= 0 || right < count {
+			if left >= 0 && idx < count {
+				order[idx] = left
+				idx++
+				left--
+			}
+			if right < count && idx < count {
+				order[idx] = right
+				idx++
+				right++
+			}
+		}
+	case "random", "mixed":
+		// Deterministic shuffle using a fixed seed for reproducible benchmarks
+		for i := 0; i < count; i++ {
+			order[i] = i
+		}
+		r := rand.New(rand.NewSource(42))
+		r.Shuffle(len(order), func(i, j int) { order[i], order[j] = order[j], order[i] })
+		return order
+	default:
+		// Default to sequential
+		for i := 0; i < count; i++ {
+			order[i] = i
+		}
+	}
+
+	return order
+}