Ensure Workhorse log writers are closed to avoid Goroutine leaks

We've seen in production the number of Goroutines in Workhorse exceed
4000. The pprof output suggests that many of these came from the
logrus Goroutines lingering.

When Workhorse forks `gitlab-zip-cat` and `gitlab-zip-metadata` it
attempts to log all stderr via a logrus Writer. logrus launches a
Goroutine to listen from a pipe. We should ensure this Writer is
closed so that the Goroutines shut down as soon as possible.

Changelog: fixed

workhorse/internal/artifacts/entry.go

@@ -71,12 +71,19 @@ func unpackFileFromZip(ctx context.Context, archivePath, encodedFilename string,
 		return err
 	}
 
+	logWriter := log.ContextLogger(ctx).Writer()
+	defer func() {
+		if closeErr := logWriter.Close(); closeErr != nil {
+			log.ContextLogger(ctx).WithError(closeErr).Error("failed to close gitlab-zip-cat log writer")
+		}
+	}()
+
 	catFile := exec.Command("gitlab-zip-cat")
 	catFile.Env = append(os.Environ(),
 		"ARCHIVE_PATH="+archivePath,
 		"ENCODED_FILE_NAME="+encodedFilename,
 	)
-	catFile.Stderr = log.ContextLogger(ctx).Writer()
+	catFile.Stderr = logWriter
 	catFile.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
 	stdout, err := catFile.StdoutPipe()
 	if err != nil {

workhorse/internal/upload/artifacts_uploader.go

@@ -72,8 +72,15 @@ func (a *artifactsUploadProcessor) generateMetadataFromZip(ctx context.Context,
 		fileName = file.RemoteURL
 	}
 
+	logWriter := log.ContextLogger(ctx).Writer()
+	defer func() {
+		if closeErr := logWriter.Close(); closeErr != nil {
+			log.ContextLogger(ctx).WithError(closeErr).Error("failed to close gitlab-zip-metadata log writer")
+		}
+	}()
+
 	zipMd := exec.CommandContext(ctx, "gitlab-zip-metadata", "-zip-reader-limit", strconv.FormatInt(readerLimit, 10), fileName)
-	zipMd.Stderr = log.ContextLogger(ctx).Writer()
+	zipMd.Stderr = logWriter
 	zipMd.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
 
 	zipMdOut, err := zipMd.StdoutPipe()