port async batch commit code from ipfs

(ipfs/kubo#4296)

1. Modern storage devices (i.e., SSDs) tend to be highly parallel.
2. Allows us to read and write at the same time (avoids pausing while flushing).

fixes ipfs/kubo#898 (comment)

batch.go

@@ -1,9 +1,16 @@
 package format
 
 import (
+	"runtime"
+
 	cid "github.com/ipfs/go-cid"
 )
 
+// ParallelBatchCommits is the number of batch commits that can be in-flight before blocking.
+// TODO(ipfs/go-ipfs#4299): Experiment with multiple datastores, storage
+// devices, and CPUs to find the right value/formula.
+var ParallelBatchCommits = runtime.NumCPU() * 2
+
 // NewBatch returns a node buffer (Batch) that buffers nodes internally and
 // commits them to the underlying DAGService in batches. Use this if you intend
 // to add a lot of nodes all at once.
@@ -15,37 +22,91 @@ func NewBatch(ds DAGService) *Batch {
 		// By default, only batch up to 128 nodes at a time.
 		// The current implementation of flatfs opens this many file
 		// descriptors at the same time for the optimized batch write.
-		MaxBlocks: 128,
+		MaxNodes: 128,
 	}
 }
 
+// Batch is a buffer for batching adds to a dag.
 type Batch struct {
 	ds DAGService
 
-	// TODO: try to re-use memory.
-	nodes     []Node
-	size      int
-	MaxSize   int
-	MaxBlocks int
+	activeCommits int
+	commitError   error
+	commitResults chan error
+
+	nodes []Node
+	size  int
+
+	MaxSize  int
+	MaxNodes int
+}
+
+func (t *Batch) processResults() {
+	for t.activeCommits > 0 && t.commitError == nil {
+		select {
+		case err := <-t.commitResults:
+			t.activeCommits--
+			if err != nil {
+				t.commitError = err
+			}
+		default:
+			return
+		}
+	}
+}
+
+func (t *Batch) asyncCommit() {
+	numBlocks := len(t.nodes)
+	if numBlocks == 0 || t.commitError != nil {
+		return
+	}
+	if t.activeCommits >= ParallelBatchCommits {
+		err := <-t.commitResults
+		t.activeCommits--
+
+		if err != nil {
+			t.commitError = err
+			return
+		}
+	}
+	go func(b []Node) {
+		_, err := t.ds.AddMany(b)
+		t.commitResults <- err
+	}(t.nodes)
+
+	t.activeCommits++
+	t.nodes = make([]Node, 0, numBlocks)
+	t.size = 0
+
+	return
 }
 
-// Add a node to this batch of nodes, potentially committing the set of batched
-// nodes to the underlying DAGService.
+// Add adds a node to the batch and commits the batch if necessary.
 func (t *Batch) Add(nd Node) (*cid.Cid, error) {
+	// Not strictly necessary but allows us to catch errors early.
+	t.processResults()
+	if t.commitError != nil {
+		return nil, t.commitError
+	}
+
 	t.nodes = append(t.nodes, nd)
 	t.size += len(nd.RawData())
-	if t.size > t.MaxSize || len(t.nodes) > t.MaxBlocks {
-		return nd.Cid(), t.Commit()
+	if t.size > t.MaxSize || len(t.nodes) > t.MaxNodes {
+		t.asyncCommit()
 	}
-	return nd.Cid(), nil
+	return nd.Cid(), t.commitError
 }
 
-// Commit commits the buffered of nodes to the underlying DAGService.
-// Make sure to call this after you're done adding nodes to the batch to ensure
-// that they're actually added to the DAGService.
+// Commit commits batched nodes.
 func (t *Batch) Commit() error {
-	_, err := t.ds.AddMany(t.nodes)
-	t.nodes = nil
-	t.size = 0
-	return err
+	t.asyncCommit()
+	for t.activeCommits > 0 && t.commitError == nil {
+		err := <-t.commitResults
+		t.activeCommits--
+		if err != nil {
+			t.commitError = err
+		}
+	}
+
+	return t.commitError
 }