[libbeat] Add configurable exponential backoff for disk queue write e…

…rrors (#21493)

libbeat/publisher/queue/diskqueue/config.go

@@ -21,6 +21,7 @@ import (
 	"errors"
 	"fmt"
 	"path/filepath"
+	"time"
 
 	"github.com/elastic/beats/v7/libbeat/common"
 	"github.com/elastic/beats/v7/libbeat/common/cfgtype"
@@ -61,16 +62,26 @@ type Settings struct {
 	// A listener that should be sent ACKs when an event is successfully
 	// written to disk.
 	WriteToDiskListener queue.ACKListener
+
+	// RetryInterval specifies how long to wait before retrying a fatal error
+	// writing to disk. If MaxRetryInterval is nonzero, subsequent retries will
+	// use exponential backoff up to the specified limit.
+	RetryInterval    time.Duration
+	MaxRetryInterval time.Duration
 }
 
 // userConfig holds the parameters for a disk queue that are configurable
 // by the end user in the beats yml file.
 type userConfig struct {
-	Path            string            `config:"path"`
-	MaxSize         cfgtype.ByteSize  `config:"max_size" validate:"required"`
-	SegmentSize     *cfgtype.ByteSize `config:"segment_size"`
-	ReadAheadLimit  *int              `config:"read_ahead"`
-	WriteAheadLimit *int              `config:"write_ahead"`
+	Path        string            `config:"path"`
+	MaxSize     cfgtype.ByteSize  `config:"max_size" validate:"required"`
+	SegmentSize *cfgtype.ByteSize `config:"segment_size"`
+
+	ReadAheadLimit  *int `config:"read_ahead"`
+	WriteAheadLimit *int `config:"write_ahead"`
+
+	RetryInterval    *time.Duration `config:"retry_interval" validate:"positive"`
+	MaxRetryInterval *time.Duration `config:"max_retry_interval" validate:"positive"`
 }
 
 func (c *userConfig) Validate() error {
@@ -96,6 +107,13 @@ func (c *userConfig) Validate() error {
 			"Disk queue segment_size (%d) cannot be less than 1MB", *c.SegmentSize)
 	}
 
+	if c.RetryInterval != nil && c.MaxRetryInterval != nil &&
+		*c.MaxRetryInterval < *c.RetryInterval {
+		return fmt.Errorf(
+			"Disk queue max_retry_interval (%v) can't be less than retry_interval (%v)",
+			*c.MaxRetryInterval, *c.RetryInterval)
+	}
+
 	return nil
 }
 
@@ -108,6 +126,9 @@ func DefaultSettings() Settings {
 
 		ReadAheadLimit:  512,
 		WriteAheadLimit: 2048,
+
+		RetryInterval:    1 * time.Second,
+		MaxRetryInterval: 30 * time.Second,
 	}
 }
 
@@ -137,6 +158,13 @@ func SettingsForUserConfig(config *common.Config) (Settings, error) {
 		settings.WriteAheadLimit = *userConfig.WriteAheadLimit
 	}
 
+	if userConfig.RetryInterval != nil {
+		settings.RetryInterval = *userConfig.RetryInterval
+	}
+	if userConfig.MaxRetryInterval != nil {
+		settings.MaxRetryInterval = *userConfig.RetryInterval
+	}
+
 	return settings, nil
 }
 
@@ -164,3 +192,17 @@ func (settings Settings) segmentPath(segmentID segmentID) string {
 func (settings Settings) maxSegmentOffset() segmentOffset {
 	return segmentOffset(settings.MaxSegmentSize - segmentHeaderSize)
 }
+
+// Given a retry interval, nextRetryInterval returns the next higher level
+// of backoff.
+func (settings Settings) nextRetryInterval(
+	currentInterval time.Duration,
+) time.Duration {
+	if settings.MaxRetryInterval > 0 {
+		currentInterval *= 2
+		if currentInterval > settings.MaxRetryInterval {
+			currentInterval = settings.MaxRetryInterval
+		}
+	}
+	return currentInterval
+}

libbeat/publisher/queue/diskqueue/deleter_loop.go

@@ -57,6 +57,7 @@ func newDeleterLoop(settings Settings) *deleterLoop {
 }
 
 func (dl *deleterLoop) run() {
+	currentRetryInterval := dl.settings.RetryInterval
 	for {
 		request, ok := <-dl.requestChan
 		if !ok {
@@ -87,10 +88,14 @@ func (dl *deleterLoop) run() {
 			// The delay can be interrupted if the request channel is closed,
 			// indicating queue shutdown.
 			select {
-			// TODO: make the retry interval configurable.
-			case <-time.After(time.Second):
+			case <-time.After(currentRetryInterval):
 			case <-dl.requestChan:
 			}
+			currentRetryInterval =
+				dl.settings.nextRetryInterval(currentRetryInterval)
+		} else {
+			// If we made progress, reset the retry interval.
+			currentRetryInterval = dl.settings.RetryInterval
 		}
 		dl.responseChan <- deleterLoopResponse{
 			results: results,

libbeat/publisher/queue/diskqueue/segments.go

@@ -207,10 +207,15 @@ func (segment *queueSegment) getWriter(
 // retry callback returns true. This is used for timed retries when
 // creating a queue segment from the writer loop.
 func (segment *queueSegment) getWriterWithRetry(
-	queueSettings Settings, retry func(error) bool,
+	queueSettings Settings, retry func(err error, firstTime bool) bool,
 ) (*os.File, error) {
+	firstTime := true
 	file, err := segment.getWriter(queueSettings)
-	for err != nil && retry(err) {
+	for err != nil && retry(err, firstTime) {
+		// Set firstTime to false so the retry callback can perform backoff
+		// etc if needed.
+		firstTime = false
+
 		// Try again
 		file, err = segment.getWriter(queueSettings)
 	}

libbeat/publisher/queue/diskqueue/util.go

@@ -69,16 +69,32 @@ func writeErrorIsRetriable(err error) bool {
 // "wrapped" field in-place as long as it isn't captured by the callback.
 type callbackRetryWriter struct {
 	wrapped io.Writer
-	retry   func(error) bool
+
+	// The retry callback is called with the error that was produced and whether
+	// this is the first (subsequent) error arising from this particular
+	// write call.
+	retry func(err error, firstTime bool) bool
 }
 
 func (w callbackRetryWriter) Write(p []byte) (int, error) {
+	// firstTime tracks whether the current error is the first subsequent error
+	// being passed to the retry callback. This is so that the callback can
+	// reset its internal counters in case it is using exponential backoff or
+	// a retry limit.
+	firstTime := true
 	bytesWritten := 0
 	writer := w.wrapped
 	n, err := writer.Write(p)
 	for n < len(p) {
-		if err != nil && !w.retry(err) {
-			return bytesWritten + n, err
+		if err != nil {
+			shouldRetry := w.retry(err, firstTime)
+			firstTime = false
+			if !shouldRetry {
+				return bytesWritten + n, err
+			}
+		} else {
+			// If we made progress without an error, reset firstTime.
+			firstTime = true
 		}
 		// Advance p and try again.
 		bytesWritten += n

libbeat/publisher/queue/diskqueue/writer_loop.go

@@ -82,6 +82,8 @@ type writerLoop struct {
 	// The file handle corresponding to currentSegment. When currentSegment
 	// changes, this handle is closed and a new one is created.
 	outputFile *os.File
+
+	currentRetryInterval time.Duration
 }
 
 func newWriterLoop(logger *logp.Logger, settings Settings) *writerLoop {
@@ -91,6 +93,8 @@ func newWriterLoop(logger *logp.Logger, settings Settings) *writerLoop {
 
 		requestChan:  make(chan writerLoopRequest, 1),
 		responseChan: make(chan writerLoopResponse),
+
+		currentRetryInterval: settings.RetryInterval,
 	}
 }
 
@@ -215,23 +219,39 @@ outerLoop:
 	return append(bytesWritten, curBytesWritten)
 }
 
-// retryCallback is called (by way of retryCallbackWriter) when there is
+func (wl *writerLoop) applyRetryBackoff() {
+	wl.currentRetryInterval =
+		wl.settings.nextRetryInterval(wl.currentRetryInterval)
+}
+
+func (wl *writerLoop) resetRetryBackoff() {
+	wl.currentRetryInterval = wl.settings.RetryInterval
+}
+
+// retryCallback is called (by way of callbackRetryWriter) when there is
 // an error writing to a segment file. It pauses for a configurable
 // interval and returns true if the operation should be retried (which
 // it always should, unless the queue is being closed).
-func (wl *writerLoop) retryCallback(err error) bool {
+func (wl *writerLoop) retryCallback(err error, firstTime bool) bool {
+	if firstTime {
+		// Reset any exponential backoff in the retry interval.
+		wl.resetRetryBackoff()
+	}
 	if writeErrorIsRetriable(err) {
 		return true
 	}
+	// If this error isn't immediately retriable, increase the exponential
+	// backoff afterwards.
+	defer wl.applyRetryBackoff()
+
 	// If the error is not immediately retriable, log the error
 	// and wait for the retry interval before trying again, but
 	// abort if the queue is closed (indicated by the request channel
 	// becoming unblocked).
 	wl.logger.Errorf("Writing to segment %v: %v",
 		wl.currentSegment.id, err)
 	select {
-	case <-time.After(time.Second):
-		// TODO: use a configurable interval here
+	case <-time.After(wl.currentRetryInterval):
 		return true
 	case <-wl.requestChan:
 		return false