Add structured logging and better error handling

internal/fetch/downloader.go

@@ -3,6 +3,7 @@ package fetch
 import (
 	"context"
 	"errors"
+	"log/slog"
 	"net/http"
 	"os"
 	"path/filepath"
@@ -34,6 +35,14 @@ type Request struct {
 	State         *State
 	Progress      Progress
 	Checkpoint    func(State)
+	Logger        *slog.Logger
+}
+
+func (request Request) logger() *slog.Logger {
+	if request.Logger != nil {
+		return request.Logger
+	}
+	return slog.Default()
 }
 
 type Downloader struct {
@@ -90,11 +99,19 @@ func (downloader *Downloader) Download(ctx context.Context, request Request) (st
 }
 
 func (downloader *Downloader) prepare(ctx context.Context, source *remoteFile, request Request, partPath string) (State, *http.Response, error) {
-	if request.State != nil && request.State.valid() && fileExists(partPath) {
-		state := *request.State
-		source.validator = state.Validator
-		source.totalBytes = state.TotalBytes
-		return state, nil, nil
+	if request.State != nil {
+		if request.State.valid() && fileExists(partPath) {
+			state := *request.State
+			source.validator = state.Validator
+			source.totalBytes = state.TotalBytes
+			request.logger().Info("resuming from saved progress",
+				"bytes_done", state.DoneBytes(),
+				"bytes_total", state.TotalBytes,
+				"parts", len(state.Segments),
+			)
+			return state, nil, nil
+		}
+		request.logger().Warn("saved progress is unusable, starting over", "part_file_exists", fileExists(partPath))
 	}
 
 	probe, err := downloader.open(ctx, source, 0, 0)
@@ -113,6 +130,13 @@ func (downloader *Downloader) prepare(ctx context.Context, source *remoteFile, r
 		TotalBytes: source.totalBytes,
 		Segments:   planSegments(source.totalBytes, connections),
 	}
+	request.logger().Info("connected to server",
+		"file", state.FileName,
+		"bytes_total", state.TotalBytes,
+		"supports_ranges", probe.Header.Get("Accept-Ranges") == "bytes",
+		"has_validator", state.Validator != "",
+		"parts", len(state.Segments),
+	)
 	return state, probe, nil
 }
 

internal/fetch/session.go

@@ -123,6 +123,8 @@ func (session *session) saveCheckpoint() {
 		return
 	}
 	state := session.snapshot()
-	session.file.Sync()
+	if err := session.file.Sync(); err != nil {
+		session.request.logger().Warn("could not flush partial file to disk", "error", err)
+	}
 	session.request.Checkpoint(state)
 }

internal/fetch/worker.go

@@ -68,11 +68,26 @@ func (worker *segmentWorker) record(count int64) error {
 
 func (worker *segmentWorker) reconnect(ctx context.Context, cause error) (*http.Response, error) {
 	progress := worker.session.request.Progress
+	logger := worker.session.request.logger().With("part_start", worker.segment.start)
 	for {
 		worker.failedAttempts++
-		if ctx.Err() != nil || !isRetryable(cause) || worker.failedAttempts > maxRetriesWithoutProgress {
+		if ctx.Err() != nil {
 			return nil, cause
 		}
+		if !isRetryable(cause) {
+			logger.Debug("error cannot be retried", "error", cause)
+			return nil, cause
+		}
+		if worker.failedAttempts > maxRetriesWithoutProgress {
+			logger.Error("giving up after repeated connection failures", "attempts", maxRetriesWithoutProgress, "error", cause)
+			return nil, fmt.Errorf("gave up after %d reconnect attempts: %w", maxRetriesWithoutProgress, cause)
+		}
+
+		logger.Warn("connection lost, reconnecting",
+			"attempt", worker.failedAttempts,
+			"position", worker.segment.position(),
+			"error", cause,
+		)
 		progress.SetNotice(fmt.Sprintf("Connection lost, reconnecting (attempt %d of %d)", worker.failedAttempts, maxRetriesWithoutProgress))
 		if err := waitBeforeRetry(ctx, worker.failedAttempts); err != nil {
 			return nil, err
@@ -80,6 +95,7 @@ func (worker *segmentWorker) reconnect(ctx context.Context, cause error) (*http.
 
 		response, err := worker.open(ctx)
 		if err == nil {
+			logger.Info("reconnected", "position", worker.segment.position())
 			progress.SetNotice("")
 			return response, nil
 		}