feat: implement exponential backoff strategy for error handling in extractor and loader processes; enhance retry configuration options

internal/app/custom_errors/backoff.go

@@ -0,0 +1,61 @@
+package custom_errors
+
+import (
+	"context"
+	"math/rand"
+	"time"
+)
+
+func computeBackoffDelay(retryCounter int, baseDelayMs int, maxDelayMs int, maxJitterMs int) time.Duration {
+	if retryCounter < 0 {
+		retryCounter = 0
+	}
+
+	delay := max(time.Duration(baseDelayMs)*time.Millisecond, 0)
+
+	maxDelay := time.Duration(maxDelayMs) * time.Millisecond
+	for i := 0; i < retryCounter; i++ {
+		if maxDelayMs > 0 && delay >= maxDelay {
+			delay = maxDelay
+			break
+		}
+		if delay == 0 {
+			break
+		}
+		delay *= 2
+	}
+
+	if maxDelayMs > 0 && delay > maxDelay {
+		delay = maxDelay
+	}
+
+	if maxJitterMs > 0 {
+		jitter := time.Duration(rand.Intn(maxJitterMs+1)) * time.Millisecond
+		delay += jitter
+	}
+
+	if delay < 0 {
+		delay = 0
+	}
+
+	return delay
+}
+
+func requeueWithBackoff(ctx context.Context, delay time.Duration, enqueue func()) {
+	if delay <= 0 {
+		enqueue()
+		return
+	}
+
+	go func() {
+		timer := time.NewTimer(delay)
+		defer timer.Stop()
+
+		select {
+		case <-ctx.Done():
+			return
+		case <-timer.C:
+			enqueue()
+		}
+	}()
+}

internal/app/custom_errors/extractor.error.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"sync"
 
+	"git.ksdemosapps.com/kylesoda/go-migrate/internal/app/config"
 	"git.ksdemosapps.com/kylesoda/go-migrate/internal/app/models"
 	"github.com/google/uuid"
 )
@@ -22,7 +23,7 @@ func (e *ExtractorError) Error() string {
 
 func ExtractorErrorHandler(
 	ctx context.Context,
-	maxRetryAttempts int,
+	retryConfig config.RetryConfig,
 	chErrorsIn <-chan ExtractorError,
 	chPartitionsOut chan<- models.Partition,
 	chJobErrorsOut chan<- JobError,
@@ -42,11 +43,11 @@ func ExtractorErrorHandler(
 				return
 			}
 
-			if err.Partition.RetryCounter >= maxRetryAttempts {
+			if err.Partition.RetryCounter >= retryConfig.Attempts {
 				wgActivePartitions.Done()
 				jobError := JobError{
 					ShouldCancelJob: false,
-					Msg:             fmt.Sprintf("Partition %v reached max retries (%d)", err.Partition.Id, maxRetryAttempts),
+					Msg:             fmt.Sprintf("Partition %v reached max retries (%d)", err.Partition.Id, retryConfig.Attempts),
 					Prev:            &err,
 				}
 
@@ -74,6 +75,13 @@ func ExtractorErrorHandler(
 			newPartition := err.Partition
 			newPartition.RetryCounter++
 
+			delay := computeBackoffDelay(
+				newPartition.RetryCounter,
+				retryConfig.BaseDelayMs,
+				retryConfig.MaxDelayMs,
+				retryConfig.MaxJitterMs,
+			)
+
 			if err.HasLastId {
 				newPartition.ParentId = err.Partition.Id
 				newPartition.Id = uuid.New()
@@ -81,11 +89,13 @@ func ExtractorErrorHandler(
 				newPartition.IsLowerLimitInclusive = false
 			}
 
-			select {
-			case chPartitionsOut <- newPartition:
-			case <-ctx.Done():
-				return
-			}
+			requeueWithBackoff(ctx, delay, func() {
+				select {
+				case chPartitionsOut <- newPartition:
+				case <-ctx.Done():
+					return
+				}
+			})
 		}
 	}
 }

internal/app/custom_errors/loader.error.go

@@ -5,12 +5,13 @@ import (
 	"fmt"
 	"sync"
 
+	"git.ksdemosapps.com/kylesoda/go-migrate/internal/app/config"
 	"git.ksdemosapps.com/kylesoda/go-migrate/internal/app/models"
 )
 
 type LoaderError struct {
-	models.Batch
-	Msg string
+	Batch models.Batch
+	Msg   string
 }
 
 func (e *LoaderError) Error() string {
@@ -19,7 +20,7 @@ func (e *LoaderError) Error() string {
 
 func LoaderErrorHandler(
 	ctx context.Context,
-	maxRetryAttempts int,
+	retryConfig config.RetryConfig,
 	chErrorsIn <-chan LoaderError,
 	chBatchesOut chan<- models.Batch,
 	chJobErrorsOut chan<- JobError,
@@ -39,11 +40,11 @@ func LoaderErrorHandler(
 				return
 			}
 
-			if err.RetryCounter >= maxRetryAttempts {
+			if err.Batch.RetryCounter >= retryConfig.Attempts {
 				wgActiveBatches.Done()
 				jobError := JobError{
 					ShouldCancelJob: false,
-					Msg:             fmt.Sprintf("Batch %v reached max retries (%d)", err.Id, maxRetryAttempts),
+					Msg:             fmt.Sprintf("Batch %v reached max retries (%d)", err.Batch.Id, retryConfig.Attempts),
 					Prev:            &err,
 				}
 
@@ -68,13 +69,21 @@ func LoaderErrorHandler(
 				}
 			}
 
-			err.RetryCounter++
+			err.Batch.RetryCounter++
+			delay := computeBackoffDelay(
+				err.Batch.RetryCounter,
+				retryConfig.BaseDelayMs,
+				retryConfig.MaxDelayMs,
+				retryConfig.MaxJitterMs,
+			)
 
-			select {
-			case chBatchesOut <- err.Batch:
-			case <-ctx.Done():
-				return
-			}
+			requeueWithBackoff(ctx, delay, func() {
+				select {
+				case chBatchesOut <- err.Batch:
+				case <-ctx.Done():
+					return
+				}
+			})
 		}
 	}
 }