fix(uno): close tcp connexions to the LibreOffice listener, better health check that take the restart into account

go.sum

@@ -212,8 +212,6 @@ github.com/matttproud/golang_protobuf_extensions v1.0.1 h1:4hp9jkHxhMHkqkrB3Ix0j
 github.com/matttproud/golang_protobuf_extensions v1.0.1/go.mod h1:D8He9yQNgCq6Z5Ld7szi9bcBfOoFv/3dc6xSMkL2PC0=
 github.com/mholt/archiver/v3 v3.5.1 h1:rDjOBX9JSF5BvoJGvjqK479aL70qh9DIpZCl+k7Clwo=
 github.com/mholt/archiver/v3 v3.5.1/go.mod h1:e3dqJ7H78uzsRSEACH1joayhuSyhnonssnDhppzS1L4=
-github.com/microcosm-cc/bluemonday v1.0.17 h1:Z1a//hgsQ4yjC+8zEkV8IWySkXnsxmdSY642CTFQb5Y=
-github.com/microcosm-cc/bluemonday v1.0.17/go.mod h1:Z0r70sCuXHig8YpBzCc5eGHAap2K7e/u082ZUpDRRqM=
 github.com/microcosm-cc/bluemonday v1.0.18 h1:6HcxvXDAi3ARt3slx6nTesbvorIc3QeTzBNRvWktHBo=
 github.com/microcosm-cc/bluemonday v1.0.18/go.mod h1:Z0r70sCuXHig8YpBzCc5eGHAap2K7e/u082ZUpDRRqM=
 github.com/modern-go/concurrent v0.0.0-20180228061459-e0a39a4cb421/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
@@ -231,8 +229,6 @@ github.com/orisano/pixelmatch v0.0.0-20210112091706-4fa4c7ba91d5/go.mod h1:nZgzb
 github.com/pdfcpu/pdfcpu v0.3.12 h1:B+MdKisilWNSk5OCO58Z9U6H93usH73xqk6hMOaZCls=
 github.com/pdfcpu/pdfcpu v0.3.12/go.mod h1:8XVBtVxuuIuSZL4Ez15Q4QoC+H8zeAaGnuiOEwAk8jA=
 github.com/pierrec/lz4/v4 v4.1.2/go.mod h1:gZWDp/Ze/IJXGXf23ltt2EXimqmTUXEy0GFuRQyBid4=
-github.com/pierrec/lz4/v4 v4.1.13 h1:/OvL3gfLjTf7nEATCYFLe4VeorMGI3nhLU5eb8FnEjU=
-github.com/pierrec/lz4/v4 v4.1.13/go.mod h1:gZWDp/Ze/IJXGXf23ltt2EXimqmTUXEy0GFuRQyBid4=
 github.com/pierrec/lz4/v4 v4.1.14 h1:+fL8AQEZtz/ijeNnpduH0bROTu0O3NZAlPjQxGn8LwE=
 github.com/pierrec/lz4/v4 v4.1.14/go.mod h1:gZWDp/Ze/IJXGXf23ltt2EXimqmTUXEy0GFuRQyBid4=
 github.com/pkg/errors v0.8.0/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0=
@@ -306,8 +302,6 @@ go.uber.org/goleak v1.1.11/go.mod h1:cwTWslyiVhfpKIDGSZEM2HlOvcqm+tG4zioyIeLoqMQ
 go.uber.org/multierr v1.6.0/go.mod h1:cdWPpRnG4AhwMwsgIHip0KRBQjJy5kYEpYjJxpXp9iU=
 go.uber.org/multierr v1.7.0 h1:zaiO/rmgFjbmCXdSYJWQcdvOCsthmdaHfr3Gm2Kx4Ec=
 go.uber.org/multierr v1.7.0/go.mod h1:7EAYxJLBy9rStEaz58O2t4Uvip6FSURkq8/ppBp95ak=
-go.uber.org/zap v1.20.0 h1:N4oPlghZwYG55MlU6LXk/Zp00FVNE9X9wrYO8CEs4lc=
-go.uber.org/zap v1.20.0/go.mod h1:wjWOCqI0f2ZZrJF/UufIOkiC8ii6tm1iqIsLo76RfJw=
 go.uber.org/zap v1.21.0 h1:WefMeulhovoZ2sYXz7st6K0sLj7bBhpiFaud4r4zST8=
 go.uber.org/zap v1.21.0/go.mod h1:wjWOCqI0f2ZZrJF/UufIOkiC8ii6tm1iqIsLo76RfJw=
 golang.org/x/crypto v0.0.0-20180904163835-0709b304e793/go.mod h1:6SG95UA2DQfeDnfUPMdvaQW0Q7yPrPDi9nlGo2tz2b4=

pkg/modules/libreoffice/uno/listener.go

@@ -15,6 +15,7 @@ import (
 type listener interface {
 	start(logger *zap.Logger) error
 	stop(logger *zap.Logger) error
+	restart(logger *zap.Logger) error
 	lock(ctx context.Context, logger *zap.Logger) error
 	unlock(logger *zap.Logger) error
 	port() int
@@ -33,6 +34,8 @@ type libreOfficeListener struct {
 	cfgMu              sync.RWMutex
 
 	usage         int
+	restarting    bool
+	restartingMu  sync.RWMutex
 	queueLength   int
 	queueLengthMu sync.RWMutex
 	lockChan      chan struct{}
@@ -106,8 +109,12 @@ func (listener *libreOfficeListener) start(logger *zap.Logger) error {
 			return fmt.Errorf("waiting for the LibreOffice listener socket to be available: %w", ctx.Err())
 		}
 
-		_, err = net.DialTimeout("tcp", fmt.Sprintf("127.0.0.1:%d", port), time.Duration(1)*time.Second)
+		conn, err := net.DialTimeout("tcp", fmt.Sprintf("127.0.0.1:%d", port), time.Duration(1)*time.Second)
 		if err == nil {
+			err = conn.Close()
+			if err != nil {
+				logger.Debug(fmt.Sprintf("close connection after health checking the LibreOffice listener: %v", err))
+			}
 			break
 		}
 	}
@@ -149,6 +156,32 @@ func (listener *libreOfficeListener) stop(logger *zap.Logger) error {
 	return nil
 }
 
+func (listener *libreOfficeListener) restart(logger *zap.Logger) error {
+	listener.restartingMu.Lock()
+	listener.restarting = true
+	listener.restartingMu.Unlock()
+
+	defer func() {
+		listener.restartingMu.Lock()
+		listener.restarting = false
+		listener.restartingMu.Unlock()
+	}()
+
+	err := listener.stop(logger)
+	if err != nil {
+		return fmt.Errorf("stop LibreOffice listener: %w", err)
+	}
+
+	err = listener.start(logger)
+	if err != nil {
+		return fmt.Errorf("start LibreOffice listener: %w", err)
+	}
+
+	listener.usage = 0
+
+	return nil
+}
+
 func (listener *libreOfficeListener) lock(ctx context.Context, logger *zap.Logger) error {
 	listener.queueLengthMu.Lock()
 	listener.queueLength += 1
@@ -162,6 +195,17 @@ func (listener *libreOfficeListener) lock(ctx context.Context, logger *zap.Logge
 		listener.queueLength -= 1
 		listener.queueLengthMu.Unlock()
 
+		if !listener.healthy() {
+			logger.Debug("LibreOffice listener is unhealthy, restarting it...")
+
+			err := listener.restart(logger)
+			if err == nil {
+				return nil
+			}
+
+			return fmt.Errorf("restart LibreOffice listener: %w", err)
+		}
+
 		return nil
 	case <-ctx.Done():
 		logger.Debug("failed to acquire LibreOffice listener lock before deadline")
@@ -175,22 +219,6 @@ func (listener *libreOfficeListener) lock(ctx context.Context, logger *zap.Logge
 }
 
 func (listener *libreOfficeListener) unlock(logger *zap.Logger) error {
-	restart := func() error {
-		err := listener.stop(logger)
-		if err != nil {
-			return fmt.Errorf("stop LibreOffice listener: %w", err)
-		}
-
-		err = listener.start(logger)
-		if err != nil {
-			return fmt.Errorf("start LibreOffice listener: %w", err)
-		}
-
-		listener.usage = 0
-
-		return nil
-	}
-
 	defer func() {
 		<-listener.lockChan
 		logger.Debug("LibreOffice listener lock released")
@@ -199,7 +227,7 @@ func (listener *libreOfficeListener) unlock(logger *zap.Logger) error {
 	if !listener.healthy() {
 		logger.Debug("LibreOffice listener is unhealthy, restarting it...")
 
-		err := restart()
+		err := listener.restart(logger)
 		if err == nil {
 			return nil
 		}
@@ -214,7 +242,7 @@ func (listener *libreOfficeListener) unlock(logger *zap.Logger) error {
 
 	logger.Debug("LibreOffice listener threshold reached, restarting it...")
 
-	err := restart()
+	err := listener.restart(logger)
 	if err == nil {
 		return nil
 	}
@@ -237,9 +265,24 @@ func (listener *libreOfficeListener) queue() int {
 }
 
 func (listener *libreOfficeListener) healthy() bool {
-	_, err := net.DialTimeout("tcp", fmt.Sprintf("127.0.0.1:%d", listener.port()), time.Duration(1)*time.Second)
+	listener.restartingMu.RLock()
+	defer listener.restartingMu.RUnlock()
 
-	return err == nil
+	if listener.restarting {
+		return true
+	}
+
+	conn, err := net.DialTimeout("tcp", fmt.Sprintf("127.0.0.1:%d", listener.port()), time.Duration(1)*time.Second)
+	if err == nil {
+		err := conn.Close()
+		if err != nil {
+			listener.logger.Debug(fmt.Sprintf("close connection after health checking the LibreOffice listener: %v", err))
+		}
+
+		return true
+	}
+
+	return false
 }
 
 // Interface guards.

pkg/modules/libreoffice/uno/listener_test.go

@@ -2,7 +2,6 @@ package uno
 
 import (
 	"context"
-	"errors"
 	"os"
 	"testing"
 	"time"
@@ -74,7 +73,7 @@ func TestListener_stop(t *testing.T) {
 	}
 }
 
-func TestListener_lock(t *testing.T) {
+func TestListener_restart(t *testing.T) {
 	listener := newLibreOfficeListener(
 		zap.NewNop(),
 		os.Getenv("LIBREOFFICE_BIN_PATH"),
@@ -82,18 +81,112 @@ func TestListener_lock(t *testing.T) {
 		10,
 	)
 
-	ctx, cancel := context.WithTimeout(context.Background(), time.Duration(10)*time.Second)
+	err := listener.start(zap.NewNop())
-
-	err := listener.lock(ctx, zap.NewNop())
 	if err != nil {
-		t.Fatalf("expected no error from listener.lock(), but got: %v", err)
+		t.Fatalf("expected no error from listener.start(), but got: %v", err)
 	}
 
+	err = listener.restart(zap.NewNop())
+	if err != nil {
+		t.Errorf("expected no error from listener.stop(), but got: %v", err)
+	}
+
+	if !listener.healthy() {
+		t.Error("expected an healthy LibreOffice listener")
+	}
+}
+
+func TestListener_lock(t *testing.T) {
+	tests := []struct {
+		name          string
+		listener      listener
+		ctx           context.Context
+		teardown      func(listener listener) error
+		expectLockErr bool
+	}{
+		{
+			name: "nominal behavior",
+			listener: func() listener {
+				listener := newLibreOfficeListener(zap.NewNop(), os.Getenv("LIBREOFFICE_BIN_PATH"), time.Duration(10)*time.Second, 10)
+
+				err := listener.start(zap.NewNop())
+				if err != nil {
+					t.Fatalf("expected no error from listener.start(), but got: %v", err)
+				}
+
+				return listener
+			}(),
+			ctx: context.Background(),
+			teardown: func(listener listener) error {
+				return listener.stop(zap.NewNop())
+			},
+		},
+		{
+			name: "unhealthy listener",
+			listener: func() listener {
+				listener := newLibreOfficeListener(zap.NewNop(), os.Getenv("LIBREOFFICE_BIN_PATH"), time.Duration(10)*time.Second, 10)
+
+				err := listener.start(zap.NewNop())
+				if err != nil {
+					t.Fatalf("expected no error from listener.start(), but got: %v", err)
+				}
+
+				err = listener.stop(zap.NewNop())
+				if err != nil {
+					t.Fatalf("expected no error from listener.stop(), but got: %v", err)
+				}
+
+				return listener
+			}(),
+			ctx: context.Background(),
+			teardown: func(listener listener) error {
+				return listener.stop(zap.NewNop())
+			},
+		},
+		{
+			name: "context done",
+			listener: func() listener {
+				listener := newLibreOfficeListener(zap.NewNop(), os.Getenv("LIBREOFFICE_BIN_PATH"), time.Duration(10)*time.Second, 10)
+
+				err := listener.start(zap.NewNop())
+				if err != nil {
+					t.Fatalf("expected no error from listener.start(), but got: %v", err)
+				}
+
+				return listener
+			}(),
+			ctx: func() context.Context {
+				ctx, cancel := context.WithCancel(context.Background())
 				cancel()
 
-	err = listener.lock(ctx, zap.NewNop())
+				return ctx
-	if !errors.Is(err, context.Canceled) {
+			}(),
-		t.Errorf("expected %v error, but got: %v", context.Canceled, err)
+			expectLockErr: true,
+			teardown: func(listener listener) error {
+				return listener.stop(zap.NewNop())
+			},
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			defer func() {
+				err := tc.teardown(tc.listener)
+				if err != nil {
+					t.Errorf("expected no error from tc.teardown(), but got: %v", err)
+				}
+			}()
+
+			err := tc.listener.lock(tc.ctx, zap.NewNop())
+
+			if tc.expectLockErr && err == nil {
+				t.Fatalf("expected listener.lock() error, but got none")
+			}
+
+			if !tc.expectLockErr && err != nil {
+				t.Fatalf("expected no error from listener.lock(), but got: %v", err)
+			}
+		})
 	}
 }
 
@@ -112,6 +205,12 @@ func TestListener_unlock(t *testing.T) {
 				if err != nil {
 					t.Fatalf("expected no error from listener.start(), but got: %v", err)
 				}
+
+				err = listener.lock(context.Background(), zap.NewNop())
+				if err != nil {
+					t.Fatalf("expected no error from listener.lock(), but got: %v", err)
+				}
+
 				return listener
 			}(),
 			teardown: func(listener listener) error {
@@ -128,6 +227,11 @@ func TestListener_unlock(t *testing.T) {
 					t.Fatalf("expected no error from listener.start(), but got: %v", err)
 				}
 
+				err = listener.lock(context.Background(), zap.NewNop())
+				if err != nil {
+					t.Fatalf("expected no error from listener.lock(), but got: %v", err)
+				}
+
 				err = listener.stop(zap.NewNop())
 				if err != nil {
 					t.Fatalf("expected no error from listener.stop(), but got: %v", err)
@@ -148,6 +252,12 @@ func TestListener_unlock(t *testing.T) {
 				if err != nil {
 					t.Fatalf("expected no error from listener.start(), but got: %v", err)
 				}
+
+				err = listener.lock(context.Background(), zap.NewNop())
+				if err != nil {
+					t.Fatalf("expected no error from listener.lock(), but got: %v", err)
+				}
+
 				return listener
 			}(),
 			teardown: func(listener listener) error {
@@ -158,23 +268,17 @@ func TestListener_unlock(t *testing.T) {
 
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
-			ctx, cancel := context.WithTimeout(context.Background(), time.Duration(10)*time.Second)
+			defer func() {
-			defer cancel()
+				err := tc.teardown(tc.listener)
-
-			err := tc.listener.lock(ctx, zap.NewNop())
-			if err != nil {
-				t.Fatalf("expected no error from listener.lock(), but got: %v", err)
-			}
-
-			err = tc.listener.unlock(zap.NewNop())
-			if err != nil {
-				t.Errorf("expected no error from listener.unlock(), but got: %v", err)
-			}
-
-			err = tc.teardown(tc.listener)
 				if err != nil {
 					t.Errorf("expected no error from tc.teardown(), but got: %v", err)
 				}
+			}()
+
+			err := tc.listener.unlock(zap.NewNop())
+			if err != nil {
+				t.Errorf("expected no error from listener.unlock(), but got: %v", err)
+			}
 		})
 	}
 }
@@ -211,6 +315,18 @@ func TestListener_queue(t *testing.T) {
 		10,
 	)
 
+	err := listener.start(zap.NewNop())
+	if err != nil {
+		t.Fatalf("expected no error from listener.start(), but got: %v", err)
+	}
+
+	defer func() {
+		err := listener.stop(zap.NewNop())
+		if err != nil {
+			t.Errorf("expected no error from listener.stop(), but got: %v", err)
+		}
+	}()
+
 	queueLength := listener.queue()
 	if queueLength != 0 {
 		t.Fatalf("expected a zero value from listener.queue(), but got %d", queueLength)
@@ -218,7 +334,7 @@ func TestListener_queue(t *testing.T) {
 
 	ctx, cancel := context.WithTimeout(context.Background(), time.Duration(10)*time.Second)
 
-	err := listener.lock(ctx, zap.NewNop())
+	err = listener.lock(ctx, zap.NewNop())
 	if err != nil {
 		t.Fatalf("expected no error from listener.lock(), but got: %v", err)
 	}
@@ -261,12 +377,13 @@ func TestListener_queue(t *testing.T) {
 }
 
 func TestListener_healthy(t *testing.T) {
-	listener := newLibreOfficeListener(
+	listener := &libreOfficeListener{
-		zap.NewNop(),
+		binPath:      os.Getenv("LIBREOFFICE_BIN_PATH"),
-		os.Getenv("LIBREOFFICE_BIN_PATH"),
+		startTimeout: time.Duration(10) * time.Second,
-		time.Duration(10)*time.Second,
+		threshold:    10,
-		10,
+		lockChan:     make(chan struct{}, 1),
-	)
+		logger:       zap.NewNop(),
+	}
 
 	err := listener.start(zap.NewNop())
 	if err != nil {
@@ -285,4 +402,10 @@ func TestListener_healthy(t *testing.T) {
 	if listener.healthy() {
 		t.Errorf("expected a non-healthy LibreOffice listener")
 	}
+
+	listener.restarting = true
+
+	if !listener.healthy() {
+		t.Error("expected an healthy LibreOffice listener")
+	}
 }

pkg/modules/libreoffice/uno/uno.go

@@ -179,10 +179,10 @@ func (mod UNO) Start() error {
 // StartupMessage returns a custom startup message.
 func (mod UNO) StartupMessage() string {
 	if mod.libreOfficeRestartThreshold == 0 {
-		return "Long-running LibreOffice listener disabled"
+		return "long-running LibreOffice listener disabled"
 	}
 
-	return "Long-running LibreOffice listener started"
+	return "long-running LibreOffice listener started"
 }
 
 // Stop stops the long-running LibreOffice Listener if it exists.
@@ -288,10 +288,6 @@ func (mod UNO) Checks() ([]health.CheckerOption, error) {
 
 				return errors.New("long-running LibreOffice listener unhealthy")
 			},
-			// The long-running LibreOffice listener may be restarting, so we
-			// wait a given amount of time until we consider the module
-			// unavailable.
-			MaxTimeInError: mod.libreOfficeStartTimeout,
 		}),
 	}, nil
 }
@@ -339,6 +335,8 @@ func (mod UNO) PDF(ctx context.Context, logger *zap.Logger, inputPath, outputPat
 				err := mod.listener.unlock(logger)
 				if err != nil {
 					mod.logger.Error(fmt.Sprintf("unlock long-running LibreOffice listener: %v", err))
+
+					return
 				}
 			}()
 		}()

pkg/modules/libreoffice/uno/uno_test.go

@@ -258,14 +258,14 @@ func TestUNO_StartupMessage(t *testing.T) {
 			mod: UNO{
 				libreOfficeRestartThreshold: 10,
 			},
-			expectMessage: "Long-running LibreOffice listener started",
+			expectMessage: "long-running LibreOffice listener started",
 		},
 		{
 			name: "long-running LibreOffice listener disabled",
 			mod: UNO{
 				libreOfficeRestartThreshold: 0,
 			},
-			expectMessage: "Long-running LibreOffice listener disabled",
+			expectMessage: "long-running LibreOffice listener disabled",
 		},
 	}
 
@@ -755,7 +755,7 @@ func TestUNO_PDF(t *testing.T) {
 				},
 				logger: zap.NewNop(),
 			},
-			ctx:          nil,
+			ctx:          context.Background(),
 			logger:       zap.NewNop(),
 			inputPath:    "/tests/test/testdata/libreoffice/sample1.docx",
 			expectPDFErr: true,
@@ -824,6 +824,7 @@ func TestUNO_UNO(t *testing.T) {
 type listenerMock struct {
 	startMock   func(logger *zap.Logger) error
 	stopMock    func(logger *zap.Logger) error
+	restartMock func(logger *zap.Logger) error
 	lockMock    func(ctx context.Context, logger *zap.Logger) error
 	unlockMock  func(logger *zap.Logger) error
 	portMock    func() int
@@ -839,6 +840,10 @@ func (listener listenerMock) stop(logger *zap.Logger) error {
 	return listener.stopMock(logger)
 }
 
+func (listener listenerMock) restart(logger *zap.Logger) error {
+	return listener.restartMock(logger)
+}
+
 func (listener listenerMock) lock(ctx context.Context, logger *zap.Logger) error {
 	return listener.lockMock(ctx, logger)
 }