XL/erasure-read: read disks in parallel. (#1975)

On read failure, fallback to reads from other
disks also happen in parallel.

erasure-readfile.go

@@ -26,42 +26,42 @@ import (
 	"github.com/klauspost/reedsolomon"
 )
 
-// isSuccessDecodeBlocks - do we have all the blocks to be successfully decoded?.
-// input disks here are expected to be ordered i.e parityBlocks
-// are preceded by dataBlocks. For for information look at getOrderedDisks().
-func isSuccessDecodeBlocks(disks []StorageAPI, dataBlocks int) bool {
+// isSuccessDecodeBlocks - do we have all the blocks to be
+// successfully decoded?. Input encoded blocks ordered matrix.
+func isSuccessDecodeBlocks(enBlocks [][]byte, dataBlocks int) bool {
 	// Count number of data and parity blocks that were read.
 	var successDataBlocksCount = 0
 	var successParityBlocksCount = 0
-	for index, disk := range disks {
-		if disk == nil {
+	for index := range enBlocks {
+		if enBlocks[index] == nil {
 			continue
 		}
+		// block index lesser than data blocks, update data block count.
 		if index < dataBlocks {
 			successDataBlocksCount++
 			continue
-		}
+		} // else { // update parity block count.
 		successParityBlocksCount++
 	}
 	// Returns true if we have atleast dataBlocks + 1 parity.
-	return successDataBlocksCount+successParityBlocksCount >= dataBlocks+1
+	return successDataBlocksCount == dataBlocks || successDataBlocksCount+successParityBlocksCount >= dataBlocks+1
 }
 
 // isSuccessDataBlocks - do we have all the data blocks?
-// input disks here are expected to be ordered i.e parityBlocks
-// are preceded by dataBlocks. For for information look at getOrderedDisks().
-func isSuccessDataBlocks(disks []StorageAPI, dataBlocks int) bool {
+// Input encoded blocks ordered matrix.
+func isSuccessDataBlocks(enBlocks [][]byte, dataBlocks int) bool {
 	// Count number of data blocks that were read.
 	var successDataBlocksCount = 0
-	for index, disk := range disks[:dataBlocks] {
-		if disk == nil {
+	for index := range enBlocks[:dataBlocks] {
+		if enBlocks[index] == nil {
 			continue
 		}
+		// block index lesser than data blocks, update data block count.
 		if index < dataBlocks {
 			successDataBlocksCount++
 		}
 	}
-	// Returns true if we have all the dataBlocks.
+	// Returns true if we have atleast the dataBlocks.
 	return successDataBlocksCount >= dataBlocks
 }
 
@@ -79,6 +79,52 @@ func getOrderedDisks(distribution []int, disks []StorageAPI, blockCheckSums []ch
 	return orderedDisks, orderedBlockCheckSums
 }
 
+// Return readable disks slice from which we can read parallely.
+func getReadDisks(orderedDisks []StorageAPI, index int, dataBlocks int) (readDisks []StorageAPI, nextIndex int, err error) {
+	readDisks = make([]StorageAPI, len(orderedDisks))
+	dataDisks := 0
+	parityDisks := 0
+	// Count already read data and parity chunks.
+	for i := 0; i < index; i++ {
+		if orderedDisks[i] == nil {
+			continue
+		}
+		if i < dataBlocks {
+			dataDisks++
+		} else {
+			parityDisks++
+		}
+	}
+
+	// Sanity checks - we should never have this situation.
+	if dataDisks == dataBlocks {
+		return nil, 0, errUnexpected
+	}
+	if dataDisks+parityDisks >= dataBlocks+1 {
+		return nil, 0, errUnexpected
+	}
+
+	// Find the disks from which next set of parallel reads should happen.
+	for i := index; i < len(orderedDisks); i++ {
+		if orderedDisks[i] == nil {
+			continue
+		}
+		if i < dataBlocks {
+			dataDisks++
+		} else {
+			parityDisks++
+		}
+		readDisks[i] = orderedDisks[i]
+		if dataDisks == dataBlocks {
+			return readDisks, i + 1, nil
+		}
+		if dataDisks+parityDisks == dataBlocks+1 {
+			return readDisks, i + 1, nil
+		}
+	}
+	return nil, 0, errXLReadQuorum
+}
+
 // erasureReadFile - read bytes from erasure coded files and writes to given writer.
 // Erasure coded files are read block by block as per given erasureInfo and data chunks
 // are decoded into a data block. Data block is trimmed for given offset and length,
@@ -95,9 +141,9 @@ func erasureReadFile(writer io.Writer, disks []StorageAPI, volume string, path s
 	// disks and rest will be parity.
 	orderedDisks, orderedBlockCheckSums := getOrderedDisks(eInfo.Distribution, disks, blockCheckSums)
 
-	// bitrotVerify verifies if the file on a particular disk doesn't have bitrot
+	// bitRotVerify verifies if the file on a particular disk doesn't have bitrot
 	// by verifying the hash of the contents of the file.
-	bitrotVerify := func() func(diskIndex int) bool {
+	bitRotVerify := func() func(diskIndex int) bool {
 		verified := make([]bool, len(orderedDisks))
 		// Return closure so that we have reference to []verified and
 		// not recalculate the hash on it everytime the function is
@@ -117,9 +163,6 @@ func erasureReadFile(writer io.Writer, disks []StorageAPI, volume string, path s
 	// Total bytes written to writer
 	bytesWritten := int64(0)
 
-	// Each element of enBlocks holds curChunkSize'd amount of data read from its corresponding disk.
-	enBlocks := make([][]byte, len(orderedDisks))
-
 	// chunkSize is roughly BlockSize/DataBlocks.
 	// chunkSize is calculated such that chunkSize*DataBlocks accommodates BlockSize bytes.
 	// So chunkSize*DataBlocks can be slightly larger than BlockSize if BlockSize is not divisible by
@@ -133,11 +176,23 @@ func erasureReadFile(writer io.Writer, disks []StorageAPI, volume string, path s
 	// need to read parity disks. If one of the data disk is missing we need to read DataBlocks+1 number
 	// of disks. Once read, we Reconstruct() missing data if needed and write it to the given writer.
 	for block := startBlock; bytesWritten < length; block++ {
+		// Each element of enBlocks holds curChunkSize'd amount of data read from its corresponding disk.
+		enBlocks := make([][]byte, len(orderedDisks))
+
+		// enBlocks data can have 0-padding hence we need to figure the exact number
+		// of bytes we want to read from enBlocks.
+		blockSize := eInfo.BlockSize
+
 		// curChunkSize is chunkSize until end block.
 		curChunkSize := chunkSize
+
+		// We have endBlock, verify if we need to have padding.
 		if block == endBlock && (totalLength%eInfo.BlockSize != 0) {
 			// If this is the last block and size of the block is < BlockSize.
 			curChunkSize = getEncodedBlockLen(totalLength%eInfo.BlockSize, eInfo.DataBlocks)
+
+			// For the last block, the block size can be less than BlockSize.
+			blockSize = totalLength % eInfo.BlockSize
 		}
 
 		// Block offset.
@@ -146,148 +201,95 @@ func erasureReadFile(writer io.Writer, disks []StorageAPI, volume string, path s
 		// then it can result in wrong offset for the last block.
 		blockOffset := block * chunkSize
 
-		// Figure out the number of disks that are needed for the read.
-		// We will need DataBlocks number of disks if all the data disks are up.
-		// We will need DataBlocks+1 number of disks even if one of the data disks is down.
-		readableDiskCount := 0
-
-		// Count the number of data disks that are up.
-		for _, disk := range orderedDisks[:eInfo.DataBlocks] {
-			if disk == nil {
-				continue
+		// nextIndex - index from which next set of parallel reads
+		// should happen.
+		nextIndex := 0
+
+		// parallelRead() reads DataBlocks number of disks if all data
+		// disks are available or DataBlocks+1 number of disks if one
+		// of the data disks is missing. All the reads happen in parallel.
+		var parallelRead func() error
+		parallelRead = func() error {
+			// If enough blocks are available to do rs.Reconstruct()
+			if isSuccessDecodeBlocks(enBlocks, eInfo.DataBlocks) {
+				return nil
 			}
-			readableDiskCount++
-		}
-
-		// Readable disks..
-		if readableDiskCount < eInfo.DataBlocks {
-			// Not enough data disks up, so we need DataBlocks+1 number
-			// of disks for reed-solomon Reconstruct()
-			readableDiskCount = eInfo.DataBlocks + 1
-		}
-
-		// Initialize wait group.
-		var wg = &sync.WaitGroup{}
-
-		// Current disk index from which to read, this will be used later
-		// in case one of the parallel reads fails.
-		index := 0
-
-		// Read from the disks in parallel.
-		for _, disk := range orderedDisks {
-			if disk == nil {
-				index++
-				continue
+			if nextIndex == len(orderedDisks) {
+				// No more disks to read from.
+				return errXLReadQuorum
 			}
 
-			// Increment wait group.
-			wg.Add(1)
-
-			// Start reading from disk in a go-routine.
-			go func(index int, disk StorageAPI) {
-				defer wg.Done()
-
-				// Verify bit rot for this disk slice.
-				if !bitrotVerify(index) {
-					// So that we don't read from this disk for the next block.
-					orderedDisks[index] = nil
-					return
-				}
-
-				// Chunk writer.
-				chunkWriter := bytes.NewBuffer(make([]byte, 0, curChunkSize))
-
-				// CopyN copies until current chunk size.
-				err := copyN(chunkWriter, disk, volume, path, blockOffset, curChunkSize)
-				if err != nil {
-					// So that we don't read from this disk for the next block.
-					orderedDisks[index] = nil
-					return
-				}
-
-				// Copy the read blocks.
-				enBlocks[index] = chunkWriter.Bytes()
-
-				// Reset the buffer.
-				chunkWriter.Reset()
-
-				// Successfully read.
-			}(index, disk)
-
-			index++
-			readableDiskCount--
-			// We have read all the readable disks.
-			if readableDiskCount == 0 {
-				break
+			// readDisks - disks from which we need to read in parallel.
+			var readDisks []StorageAPI
+			var err error
+			readDisks, nextIndex, err = getReadDisks(orderedDisks, nextIndex, eInfo.DataBlocks)
+			if err != nil {
+				return err
 			}
-		}
 
-		// Wait for all the reads to finish.
-		wg.Wait()
-
-		// FIXME: make this parallel.
-
-		// If we have all the data blocks no need to decode.
-		if !isSuccessDataBlocks(orderedDisks, eInfo.DataBlocks) {
-			// If we don't have DataBlocks number of data blocks we
-			// will have to read enough parity blocks such that we
-			// have DataBlocks+1 number for blocks for rs.Reconstruct().
-			// index is either dataBlocks or dataBlocks + 1.
-			for ; index < len(orderedDisks); index++ {
-				// We have enough blocks to decode, break out.
-				if isSuccessDecodeBlocks(orderedDisks, eInfo.DataBlocks) {
-					// We have DataBlocks+1 blocks, enough for rs.Reconstruct()
-					break
-				}
-
-				// This disk was previously set to nil and ignored, do not read again.
-				if orderedDisks[index] == nil {
-					continue
-				}
+			// WaitGroup to synchronise the read go-routines.
+			wg := &sync.WaitGroup{}
 
-				// Verify bit-rot for this index.
-				if !bitrotVerify(index) {
-					// Mark nil so that we don't read from this disk for the next block.
-					orderedDisks[index] = nil
+			// Read disks in parallel.
+			for index := range readDisks {
+				if readDisks[index] == nil {
 					continue
 				}
+				wg.Add(1)
+				// Reads chunk from readDisk[index] in routine.
+				go func(index int) {
+					defer wg.Done()
+
+					// Verify bit rot for the file on this disk.
+					if !bitRotVerify(index) {
+						// So that we don't read from this disk for the next block.
+						orderedDisks[index] = nil
+						return
+					}
+
+					// Chunk writer.
+					chunkWriter := bytes.NewBuffer(make([]byte, 0, curChunkSize))
+
+					// CopyN - copies until current chunk size.
+					err := copyN(chunkWriter, readDisks[index], volume, path, blockOffset, curChunkSize)
+					if err != nil {
+						// So that we don't read from this disk for the next block.
+						orderedDisks[index] = nil
+						return
+					}
+
+					// Copy the read blocks.
+					enBlocks[index] = chunkWriter.Bytes()
+
+					// Reset the buffer.
+					chunkWriter.Reset()
+
+					// Successfully read.
+				}(index)
+			}
 
-				// Chunk writer.
-				chunkWriter := bytes.NewBuffer(make([]byte, 0, curChunkSize))
+			// Waiting for first routines to finish.
+			wg.Wait()
 
-				// CopyN copies until current chunk size.
-				err := copyN(chunkWriter, orderedDisks[index], volume, path, blockOffset, curChunkSize)
-				if err != nil {
-					// ERROR: Mark nil so that we don't read from
-					// this disk for the next block.
-					orderedDisks[index] = nil
-					continue
-				}
-
-				// Copy the read blocks.
-				chunkWriter.Read(enBlocks[index])
+			// Continue to read the rest of the blocks in parallel.
+			return parallelRead()
+		}
 
-				// Reset the buffer.
-				chunkWriter.Reset()
-			}
+		// Start reading all blocks in parallel.
+		err := parallelRead()
+		if err != nil {
+			return bytesWritten, err
+		}
 
+		// If we have all the data blocks no need to decode, continue to write.
+		if !isSuccessDataBlocks(enBlocks, eInfo.DataBlocks) {
 			// Reconstruct the missing data blocks.
-			err := decodeData(enBlocks, eInfo.DataBlocks, eInfo.ParityBlocks)
-			if err != nil {
+			if err = decodeData(enBlocks, eInfo.DataBlocks, eInfo.ParityBlocks); err != nil {
 				return bytesWritten, err
 			}
-			// Success.
 		}
 
 		var outSize, outOffset int64
-		// enBlocks data can have 0-padding hence we need to figure the exact number
-		// of bytes we want to read from enBlocks.
-		blockSize := eInfo.BlockSize
-		if block == endBlock && totalLength%eInfo.BlockSize != 0 {
-			// For the last block, the block size can be less than BlockSize.
-			blockSize = totalLength % eInfo.BlockSize
-		}
-
 		// If this is start block, skip unwanted bytes.
 		if block == startBlock {
 			outOffset = bytesToSkip