fix(gitter): Better handling for context cancel (#4918)

Add proper handling for context cancellation in semaphore and our longer
running git operations.

Author: Ly-Joey

go/cmd/gitter/gitter.go

@@ -117,8 +117,11 @@ func prepareCmd(ctx context.Context, dir string, env []string, name string, args
 	}
 	// Use SIGINT instead of SIGKILL for graceful shutdown of subprocesses
 	cmd.Cancel = func() error {
+		logger.DebugContext(ctx, "SIGINT sent to command", slog.String("cmd", name), slog.Any("args", args))
 		return cmd.Process.Signal(syscall.SIGINT)
 	}
+	// Ensure it eventually dies if it ignores SIGINT
+	cmd.WaitDelay = shutdownTimeout / 2
 
 	return cmd
 }
@@ -369,8 +372,15 @@ func gitHandler(w http.ResponseWriter, r *http.Request) {
 		}
 	}
 
-	semaphore <- struct{}{}
-	defer func() { <-semaphore }()
+	select {
+	case semaphore <- struct{}{}:
+		defer func() { <-semaphore }()
+	case <-ctx.Done():
+		logger.WarnContext(ctx, "Request cancelled while waiting for semaphore")
+		http.Error(w, "Server context cancelled", http.StatusServiceUnavailable)
+
+		return
+	}
 	logger.DebugContext(ctx, "Concurrent requests", slog.Int("count", len(semaphore)))
 
 	// Fetch repo first
@@ -441,8 +451,15 @@ func cacheHandler(w http.ResponseWriter, r *http.Request) {
 	ctx := context.WithValue(r.Context(), urlKey, url)
 	logger.InfoContext(ctx, "Received request: /cache")
 
-	semaphore <- struct{}{}
-	defer func() { <-semaphore }()
+	select {
+	case semaphore <- struct{}{}:
+		defer func() { <-semaphore }()
+	case <-ctx.Done():
+		logger.WarnContext(ctx, "Request cancelled while waiting for semaphore")
+		http.Error(w, "Server context cancelled", http.StatusServiceUnavailable)
+
+		return
+	}
 	logger.DebugContext(ctx, "Concurrent requests", slog.Int("count", len(semaphore)))
 
 	// Fetch repo if it's not fresh

go/cmd/gitter/repository.go

@@ -146,6 +146,10 @@ func (r *Repository) buildCommitGraph(ctx context.Context, cache *pb.RepositoryC
 
 	scanner := bufio.NewScanner(file)
 	for scanner.Scan() {
+		// Handle context cancel within the loop to exit faster if we're processing a very large repo
+		if ctx.Err() != nil {
+			return nil, ctx.Err()
+		}
 		// Example of a line of git log output
 		// 4c9bdbf0e2d45a5297cc080c3ebe809c0cca3581		bc0e4b4c4dbf7932fab7d264929d4d820e82c817 65be82edcdbc5aa6eeea23655cc96b5c84547d3b		upstream/master, upstream/HEAD\n
 		// Corresponds to: commit hash \t parent hashes (space delimited) \t refs (comma delimited)
@@ -316,6 +320,10 @@ func (r *Repository) calculatePatchIDsWorker(ctx context.Context, chunk []SHA1)
 	go func() {
 		defer in.Close()
 		for _, hash := range chunk {
+			// Handle context cancel
+			if ctx.Err() != nil {
+				return
+			}
 			fmt.Fprintf(in, "%s\n", hex.EncodeToString(hash[:]))
 		}
 	}()
@@ -330,6 +338,10 @@ func (r *Repository) calculatePatchIDsWorker(ctx context.Context, chunk []SHA1)
 	// Read results from stdout of git patch-id
 	scanner := bufio.NewScanner(out)
 	for scanner.Scan() {
+		// Handle context cancel to exit faster as this can be a long process
+		if ctx.Err() != nil {
+			return ctx.Err()
+		}
 		// Format of git patch-id result: <patch ID> <commit hash>
 		line := scanner.Text()