filer: improve FoundationDB performance by disabling batch by default (#7770)

* filer: improve FoundationDB performance by disabling batch by default This PR addresses a performance issue where FoundationDB filer was achieving only ~757 ops/sec with 12 concurrent S3 clients, despite FDB being capable of 17,000+ ops/sec. Root cause: The write batcher was waiting up to 5ms for each operation to batch, even though S3 semantics require waiting for durability confirmation. This added artificial latency that defeated the purpose of batching. Changes: - Disable write batching by default (batch_enabled = false) - Each write now commits immediately in its own transaction - Reduce batch interval from 5ms to 1ms when batching is enabled - Add batch_enabled config option to toggle behavior - Improve batcher to collect available ops without blocking - Add benchmarks comparing batch vs no-batch performance Benchmark results (16 concurrent goroutines): - With batch: 2,924 ops/sec (342,032 ns/op) - Without batch: 4,625 ops/sec (216,219 ns/op) - Improvement: +58% faster Configuration: - Default: batch_enabled = false (optimal for S3 PUT latency) - For bulk ingestion: set batch_enabled = true Also fixes ARM64 Docker test setup (shell compatibility, fdbserver path). * fix: address review comments - use atomic counter and remove duplicate batcher - Use sync/atomic.Uint64 for unique filenames in concurrent benchmarks - Remove duplicate batcher creation in createBenchmarkStoreWithBatching (initialize() already creates batcher when batchEnabled=true) * fix: add realistic default values to benchmark store helper Set directoryPrefix, timeout, and maxRetryDelay to reasonable defaults for more realistic benchmark conditions.
2025-12-15 13:03:34 -08:00
parent 44beb42eb9
commit 5a03b5538f
11 changed files with 321 additions and 36 deletions
--- a/weed/command/scaffold/filer.toml
+++ b/weed/command/scaffold/filer.toml
@@ -415,6 +415,16 @@ key_path=""
 # The name list used to verify the cn name
 verify_cn=""

+[foundationdb]
+# FoundationDB provides ACID transactions and horizontal scalability.
+# Requires: go build -tags foundationdb
+enabled = false
+cluster_file = "/etc/foundationdb/fdb.cluster"
+# api_version = 740
+# timeout = "5s"
+# directory_prefix = "seaweedfs"
+# For bulk ingestion, enable batching: batch_enabled = true
+
 [tarantool]
 address = "localhost:3301"
 user = "guest"
--- a/weed/filer/foundationdb/CONFIGURATION.md
+++ b/weed/filer/foundationdb/CONFIGURATION.md
@@ -27,6 +27,10 @@ export WEED_FOUNDATIONDB_API_VERSION=740
 export WEED_FOUNDATIONDB_TIMEOUT=5s
 export WEED_FOUNDATIONDB_MAX_RETRY_DELAY=1s
 export WEED_FOUNDATIONDB_DIRECTORY_PREFIX=seaweedfs
+# Write batching (disabled by default)
+export WEED_FOUNDATIONDB_BATCH_ENABLED=false
+export WEED_FOUNDATIONDB_BATCH_SIZE=100
+export WEED_FOUNDATIONDB_BATCH_INTERVAL=1ms
 ```

 ### 3. Command Line Arguments
@@ -56,6 +60,18 @@ While not directly supported, configuration can be specified via config files pa
 |--------|------|---------|-------------|
 | `directory_prefix` | string | `seaweedfs` | Directory prefix for key organization |

+### Write Batching Options
+
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `batch_enabled` | boolean | `false` | Enable write batching. Disabled by default for optimal S3 PUT latency. |
+| `batch_size` | integer | `100` | Maximum number of operations per batch (when batching is enabled) |
+| `batch_interval` | duration | `1ms` | Maximum time to wait before flushing a batch (when batching is enabled) |
+
+**Note:** Write batching is **disabled by default**. Each write commits immediately in its own 
+transaction, providing optimal latency for S3 PUT operations. Enable batching only for 
+high-throughput bulk ingestion workloads where you can tolerate slightly higher per-operation latency.
+
 ## Configuration Examples

 ### Development Environment
@@ -82,7 +98,25 @@ max_retry_delay = "5s"
 directory_prefix = "seaweedfs_prod"
 ```

-### High-Performance Setup
+### High-Performance Setup (Low Latency)
+
+For S3 workloads requiring low latency per operation:
+
+```toml
+[foundationdb]
+enabled = true
+cluster_file = "/etc/foundationdb/fdb.cluster"
+api_version = 740
+timeout = "5s"
+max_retry_delay = "1s"
+directory_prefix = "sw"  # Shorter prefix for efficiency
+# Batching disabled (default) for optimal per-operation latency
+batch_enabled = false
+```
+
+### High-Throughput Bulk Ingestion
+
+For bulk data loading where throughput matters more than per-operation latency:

 ```toml
 [foundationdb]
@@ -91,7 +125,11 @@ cluster_file = "/etc/foundationdb/fdb.cluster"
 api_version = 740
 timeout = "60s"
 max_retry_delay = "10s"
-directory_prefix = "sw"  # Shorter prefix for efficiency
+directory_prefix = "sw"
+# Enable batching for higher throughput
+batch_enabled = true
+batch_size = 100
+batch_interval = "1ms"
 ```

 ### Path-Specific Configuration
--- a/weed/filer/foundationdb/README.md
+++ b/weed/filer/foundationdb/README.md
@@ -51,6 +51,9 @@ directory_prefix = "seaweedfs"
 | `timeout` | Operation timeout duration | `5s` | No |
 | `max_retry_delay` | Maximum retry delay | `1s` | No |
 | `directory_prefix` | Directory prefix for organization | `seaweedfs` | No |
+| `batch_enabled` | Enable write batching (see Performance section) | `false` | No |
+| `batch_size` | Max operations per batch | `100` | No |
+| `batch_interval` | Max time before batch flush | `1ms` | No |

 ### Path-Specific Configuration

@@ -109,12 +112,38 @@ make setup

 ## Performance Considerations

+### Write Batching Configuration
+
+By default, write batching is **disabled** (`batch_enabled = false`). Each write commits 
+immediately in its own transaction. This provides optimal latency for S3 PUT operations.
+
+**When to enable batching:**
+- High-throughput bulk ingestion workloads
+- Scenarios where you can tolerate slightly higher per-operation latency
+- Workloads with many concurrent small writes
+
+**Batching configuration options:**
+
+```toml
+[foundationdb]
+# Enable write batching (disabled by default for optimal S3 latency)
+batch_enabled = true
+# Maximum operations per batch
+batch_size = 100
+# Maximum time to wait before flushing a batch
+batch_interval = "1ms"
+```
+
+**Performance comparison:**
+- **Batching disabled**: Each S3 PUT commits immediately (~1-5ms per op depending on FDB latency)
+- **Batching enabled**: Operations are grouped, reducing total commits but adding batch interval latency
+
 ### Optimal Configuration

 - **API Version**: Use the latest stable API version (720+)
 - **Directory Structure**: Use logical directory prefixes to isolate different SeaweedFS instances
 - **Transaction Size**: Keep transactions under 10MB (FDB limit)
- **Batch Operations**: Use transactions for multiple related operations
+- **Concurrency**: Use multiple client connections for parallel operations

 ### Monitoring

--- a/weed/filer/foundationdb/foundationdb_store.go
+++ b/weed/filer/foundationdb/foundationdb_store.go
@@ -45,8 +45,12 @@ const (
 	MAX_DIRECTORY_LIST_LIMIT = 1000

 	// Write batching defaults
+	// Note: Batching is disabled by default because S3 semantics require waiting
+	// for durability, and the batch timer adds latency to each operation.
+	// Enable batching only for workloads that can tolerate potential latency.
 	DEFAULT_BATCH_SIZE     = 100
-	DEFAULT_BATCH_INTERVAL = 5 * time.Millisecond
+	DEFAULT_BATCH_INTERVAL = 1 * time.Millisecond
+	DEFAULT_BATCH_ENABLED  = false
 )

 func init() {
@@ -129,11 +133,33 @@ func (b *writeBatcher) run() {
 		timer.Reset(b.interval)
 	}

+	// Collect available ops without blocking
+	collectAvailable := func() {
+		for {
+			select {
+			case op := <-b.ops:
+				batch = append(batch, op)
+				batchBytes += op.size()
+				if len(batch) >= b.size || batchBytes >= FDB_BATCH_SIZE_LIMIT {
+					return
+				}
+			default:
+				return
+			}
+		}
+	}
+
 	for {
 		select {
 		case op := <-b.ops:
 			batch = append(batch, op)
 			batchBytes += op.size()
+
+			// Optimization: When an operation arrives, try to collect more
+			// available operations without blocking. This improves throughput
+			// when multiple goroutines are submitting concurrently.
+			collectAvailable()
+
 			// Flush when batch count or size limit is reached
 			if len(batch) >= b.size || batchBytes >= FDB_BATCH_SIZE_LIMIT {
 				flush()
@@ -185,8 +211,10 @@ type FoundationDBStore struct {
 	directoryPrefix string
 	timeout         time.Duration
 	maxRetryDelay   time.Duration
-	// Write batching
+	// Write batching - disabled by default for optimal S3 latency
+	// Enable for high-throughput bulk ingestion workloads
 	batcher       *writeBatcher
+	batchEnabled  bool
 	batchSize     int
 	batchInterval time.Duration
 }
@@ -225,6 +253,10 @@ func (store *FoundationDBStore) Initialize(configuration util.Configuration, pre
 	configuration.SetDefault(prefix+"timeout", "5s")
 	configuration.SetDefault(prefix+"max_retry_delay", "1s")
 	configuration.SetDefault(prefix+"directory_prefix", "seaweedfs")
+	// Batching is disabled by default - each write commits immediately.
+	// This provides optimal latency for S3 PUT operations.
+	// Enable batching for high-throughput bulk ingestion workloads.
+	configuration.SetDefault(prefix+"batch_enabled", DEFAULT_BATCH_ENABLED)
 	configuration.SetDefault(prefix+"batch_size", DEFAULT_BATCH_SIZE)
 	configuration.SetDefault(prefix+"batch_interval", DEFAULT_BATCH_INTERVAL.String())

@@ -247,6 +279,7 @@ func (store *FoundationDBStore) Initialize(configuration util.Configuration, pre
 	}

 	// Parse batch configuration
+	store.batchEnabled = configuration.GetBool(prefix + "batch_enabled")
 	store.batchSize = configuration.GetInt(prefix + "batch_size")
 	if store.batchSize <= 0 {
 		store.batchSize = DEFAULT_BATCH_SIZE
@@ -288,10 +321,16 @@ func (store *FoundationDBStore) initialize(clusterFile string, apiVersion int) e
 		return fmt.Errorf("failed to create/open kv directory: %w", err)
 	}

-	// Start write batcher for improved throughput
-	store.batcher = newWriteBatcher(store, store.batchSize, store.batchInterval)
-	glog.V(0).Infof("FoundationDB: write batching enabled (batch_size=%d, batch_interval=%v)",
-		store.batchSize, store.batchInterval)
+	// Conditionally start write batcher
+	// Batching is disabled by default for optimal S3 latency.
+	// When disabled, each write commits immediately in its own transaction.
+	if store.batchEnabled {
+		store.batcher = newWriteBatcher(store, store.batchSize, store.batchInterval)
+		glog.V(0).Infof("FoundationDB: write batching enabled (batch_size=%d, batch_interval=%v)",
+			store.batchSize, store.batchInterval)
+	} else {
+		glog.V(0).Infof("FoundationDB: write batching disabled (direct commit mode for optimal latency)")
+	}

 	glog.V(0).Infof("FoundationDB store initialized successfully with directory prefix: %s", store.directoryPrefix)
 	return nil
--- a/weed/filer/foundationdb/foundationdb_store_test.go
+++ b/weed/filer/foundationdb/foundationdb_store_test.go
@@ -9,6 +9,7 @@ import (
 	"fmt"
 	"os"
 	"strings"
+	"sync/atomic"
 	"testing"
 	"time"

@@ -325,6 +326,120 @@ func BenchmarkFoundationDBStore_KvOperations(b *testing.B) {
 	}
 }

+// BenchmarkFoundationDBStore_InsertEntry_NoBatch benchmarks insert performance
+// with batching disabled (direct commit mode - optimal for S3 PUT latency)
+func BenchmarkFoundationDBStore_InsertEntry_NoBatch(b *testing.B) {
+	store := createBenchmarkStoreWithBatching(b, false, 100, 1*time.Millisecond)
+	defer store.Shutdown()
+
+	ctx := context.Background()
+	entry := &filer.Entry{
+		FullPath: "/benchmark_nobatch/file.txt",
+		Attr: filer.Attr{
+			Mode:  0644,
+			Uid:   1000,
+			Gid:   1000,
+			Mtime: time.Now(),
+		},
+	}
+
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		entry.FullPath = util.NewFullPath("/benchmark_nobatch", fmt.Sprintf("%x", uint64(i))+".txt")
+		err := store.InsertEntry(ctx, entry)
+		if err != nil {
+			b.Fatalf("InsertEntry failed: %v", err)
+		}
+	}
+}
+
+// BenchmarkFoundationDBStore_InsertEntry_WithBatch benchmarks insert performance
+// with batching enabled (higher throughput for bulk ingestion)
+func BenchmarkFoundationDBStore_InsertEntry_WithBatch(b *testing.B) {
+	store := createBenchmarkStoreWithBatching(b, true, 100, 1*time.Millisecond)
+	defer store.Shutdown()
+
+	ctx := context.Background()
+	entry := &filer.Entry{
+		FullPath: "/benchmark_batch/file.txt",
+		Attr: filer.Attr{
+			Mode:  0644,
+			Uid:   1000,
+			Gid:   1000,
+			Mtime: time.Now(),
+		},
+	}
+
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		entry.FullPath = util.NewFullPath("/benchmark_batch", fmt.Sprintf("%x", uint64(i))+".txt")
+		err := store.InsertEntry(ctx, entry)
+		if err != nil {
+			b.Fatalf("InsertEntry failed: %v", err)
+		}
+	}
+}
+
+// BenchmarkFoundationDBStore_ConcurrentInsert_NoBatch benchmarks concurrent inserts
+// with batching disabled (simulates S3 PUT concurrency)
+func BenchmarkFoundationDBStore_ConcurrentInsert_NoBatch(b *testing.B) {
+	store := createBenchmarkStoreWithBatching(b, false, 100, 1*time.Millisecond)
+	defer store.Shutdown()
+
+	var counter atomic.Uint64
+
+	b.ResetTimer()
+	b.RunParallel(func(pb *testing.PB) {
+		ctx := context.Background()
+		for pb.Next() {
+			n := counter.Add(1)
+			entry := &filer.Entry{
+				FullPath: util.NewFullPath("/benchmark_concurrent_nobatch", fmt.Sprintf("%d.txt", n)),
+				Attr: filer.Attr{
+					Mode:  0644,
+					Uid:   1000,
+					Gid:   1000,
+					Mtime: time.Now(),
+				},
+			}
+			err := store.InsertEntry(ctx, entry)
+			if err != nil {
+				b.Fatalf("InsertEntry failed: %v", err)
+			}
+		}
+	})
+}
+
+// BenchmarkFoundationDBStore_ConcurrentInsert_WithBatch benchmarks concurrent inserts
+// with batching enabled (tests batch efficiency under concurrent load)
+func BenchmarkFoundationDBStore_ConcurrentInsert_WithBatch(b *testing.B) {
+	store := createBenchmarkStoreWithBatching(b, true, 100, 1*time.Millisecond)
+	defer store.Shutdown()
+
+	var counter atomic.Uint64
+
+	b.ResetTimer()
+	b.RunParallel(func(pb *testing.PB) {
+		ctx := context.Background()
+		for pb.Next() {
+			n := counter.Add(1)
+			entry := &filer.Entry{
+				FullPath: util.NewFullPath("/benchmark_concurrent_batch", fmt.Sprintf("%d.txt", n)),
+				Attr: filer.Attr{
+					Mode:  0644,
+					Uid:   1000,
+					Gid:   1000,
+					Mtime: time.Now(),
+				},
+			}
+			err := store.InsertEntry(ctx, entry)
+			if err != nil {
+				b.Fatalf("InsertEntry failed: %v", err)
+			}
+		}
+	})
+}
+
 // Helper functions
 func getTestClusterFile() string {
 	clusterFile := os.Getenv("FDB_CLUSTER_FILE")
@@ -349,6 +464,32 @@ func createBenchmarkStore(b *testing.B) *FoundationDBStore {
 	return store
 }

+// createBenchmarkStoreWithBatching creates a store with specific batching configuration
+// for comparing performance between batched and non-batched modes
+func createBenchmarkStoreWithBatching(b *testing.B, batchEnabled bool, batchSize int, batchInterval time.Duration) *FoundationDBStore {
+	clusterFile := getTestClusterFile()
+	if _, err := os.Stat(clusterFile); os.IsNotExist(err) {
+		b.Skip("FoundationDB cluster file not found, skipping benchmark")
+	}
+
+	store := &FoundationDBStore{
+		batchEnabled:    batchEnabled,
+		batchSize:       batchSize,
+		batchInterval:   batchInterval,
+		directoryPrefix: "benchmark",
+		timeout:         5 * time.Second,
+		maxRetryDelay:   1 * time.Second,
+	}
+	err := store.initialize(clusterFile, 740)
+	if err != nil {
+		b.Skipf("Failed to initialize FoundationDB store: %v", err)
+	}
+
+	// Note: initialize() already creates the batcher if batchEnabled is true
+
+	return store
+}
+
 func getTestStore(t *testing.T) *FoundationDBStore {
 	t.Helper()