diff --git a/pkg/gotenberg/supervisor.go b/pkg/gotenberg/supervisor.go
index 63998fe7..c5b9bc79 100644
--- a/pkg/gotenberg/supervisor.go
+++ b/pkg/gotenberg/supervisor.go
@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"sync"
 	"sync/atomic"
+	"time"
 
 	"go.uber.org/zap"
 )
@@ -54,9 +55,9 @@ type ProcessSupervisor interface {
 
 	// Healthy checks and returns the health status of the managed [Process].
 	//
-	// If the process has not been started or is restarting, it is considered
-	// healthy and true is returned. Otherwise, it returns the health status of
-	// the actual process.
+	// A non-started process is considered healthy (startup is deferred until
+	// the first request). Returns false if the process is currently restarting
+	// or is reported unhealthy by the underlying [Process].
 	Healthy() bool
 
 	// Run executes a provided task while managing the state of the [Process].
@@ -76,14 +77,18 @@ type ProcessSupervisor interface {
 }
 
 type processSupervisor struct {
-	logger          *zap.Logger
-	process         Process
-	maxReqLimit     int64
-	maxQueueSize    int64
-	maxConcurrency  int64
-	semaphore       chan struct{}
-	firstStart      atomic.Bool
-	firstStartOnce  sync.Once
+	logger         *zap.Logger
+	process        Process
+	maxReqLimit    int64
+	maxQueueSize   int64
+	maxConcurrency int64
+	semaphore      chan struct{}
+	firstStart     atomic.Bool
+	firstStartOnce sync.Once
+	// firstStartErr stores the error from the first Launch attempt executed
+	// via firstStartOnce. Subsequent callers that enter the !firstStart block
+	// need to observe this value after the Once has completed, without
+	// re-executing the closure.
 	firstStartErr   error
 	reqCounter      atomic.Int64
 	reqQueueSize    atomic.Int64
@@ -146,8 +151,9 @@ func (s *processSupervisor) restart() error {
 
 	err := s.Shutdown()
 	if err != nil {
-		// No big deal? Chances are it's already stopped.
-		s.logger.Debug(fmt.Sprintf("stop process before restart: %s", err))
+		// Not necessarily critical — chances are the process is already stopped,
+		// but worth flagging in case it indicates a real issue.
+		s.logger.Warn(fmt.Sprintf("stop process before restart: %s", err))
 	}
 
 	err = s.Launch()
@@ -164,117 +170,75 @@ func (s *processSupervisor) restart() error {
 
 func (s *processSupervisor) Healthy() bool {
 	if !s.firstStart.Load() {
-		// A non-started process is always healthy.
+		// A non-started process is considered healthy: Gotenberg defers
+		// process startup until the first request to keep resource usage low.
+		// Reporting unhealthy here would cause container orchestrators to
+		// restart the pod before any request arrives.
 		return true
 	}
 
 	if s.isRestarting.Load() {
-		// A restarting process is always healthy.
-		return true
+		// A restarting process is not yet healthy — this gives load balancers
+		// honest information so they can avoid routing traffic to this node.
+		return false
 	}
 
 	return s.process.Healthy(s.logger)
 }
 
 func (s *processSupervisor) Run(ctx context.Context, logger *zap.Logger, task func() error) error {
-	// A user reported a potential issue:
-	//
-	// "Although the counting operation is atomic, nothing prevent 2 concurrent
-	// goroutines to retrieve the same 'currentQueueSize' and to compare its
-	// value against the max limit. Then, resulting queue size would be 1 above
-	// the allowed limit."
-	//
-	// However, he was unable to actually trigger this issue, even when sending
-	// a lot of requests.
-	//
-	// For now, the best option is to consider this issue to be unlikely to
-	// happen, and keep the code as it is because it is more readable this way.
-	//
-	// See https://github.com/gotenberg/gotenberg/issues/951.
-	currentQueueSize := s.reqQueueSize.Load()
-	if s.maxQueueSize > 0 && currentQueueSize >= s.maxQueueSize {
-		return ErrMaximumQueueSizeExceeded
+	// Atomically check and increment the queue size to avoid the TOCTOU race
+	// originally reported in https://github.com/gotenberg/gotenberg/issues/951.
+	for {
+		current := s.reqQueueSize.Load()
+		if s.maxQueueSize > 0 && current >= s.maxQueueSize {
+			return ErrMaximumQueueSizeExceeded
+		}
+		if s.reqQueueSize.CompareAndSwap(current, current+1) {
+			break
+		}
 	}
 
-	s.reqQueueSize.Add(1)
-
 	for {
 		err := func() error {
-			select {
-			case s.semaphore <- struct{}{}:
-				logger.Debug("process lock acquired")
-
-				// If a restart drain is in progress, release the slot
-				// immediately so the drain can acquire it instead.
-				if s.isRestarting.Load() {
-					<-s.semaphore
-					return ErrProcessAlreadyRestarting
-				}
-
-				s.reqQueueSize.Add(-1)
-				s.reqCounter.Add(1)
-				s.activeTasks.Add(1)
-				releaseSemaphore := true
-
-				defer func() {
-					s.activeTasks.Add(-1)
-					if releaseSemaphore {
-						logger.Debug("process lock released")
-						<-s.semaphore
-					}
-				}()
-
-				if !s.firstStart.Load() {
-					s.firstStartOnce.Do(func() {
-						s.firstStartErr = s.runWithDeadline(ctx, func() error {
-							return s.Launch()
-						})
-					})
-					if s.firstStartErr != nil {
-						return fmt.Errorf("process first start: %w", s.firstStartErr)
-					}
-				}
-
-				if !s.Healthy() {
-					s.logger.Debug("process is unhealthy, cannot handle task, restarting...")
-					err := s.doRestart(ctx)
-					if err != nil {
-						return fmt.Errorf("process restart before task: %w", err)
-					}
-				}
-
-				err := s.runWithDeadline(ctx, task)
-
-				if s.maxReqLimit > 0 && s.reqCounter.Load() >= s.maxReqLimit {
-					// Only one goroutine should trigger the restart.
-					if s.restartMutex.TryLock() {
-						s.logger.Debug("max request limit reached, restarting eagerly...")
-						releaseSemaphore = false
-
-						go func() {
-							restartErr := s.doRestartLocked(context.Background())
-							s.restartMutex.Unlock()
-							if restartErr != nil {
-								s.logger.Error(fmt.Sprintf("process restart after task: %v", restartErr))
-							}
-							logger.Debug("process lock released")
-							<-s.semaphore
-						}()
-					}
-				}
-
-				// Note: no error wrapping because it leaks on Chromium console exceptions output.
+			if err := s.acquireSlot(ctx, logger); err != nil {
 				return err
-			case <-ctx.Done():
-				logger.Debug("failed to acquire process lock before deadline")
-				s.reqQueueSize.Add(-1)
-
-				return fmt.Errorf("acquire process lock: %w", ctx.Err())
 			}
+
+			s.reqQueueSize.Add(-1)
+			s.reqCounter.Add(1)
+			s.activeTasks.Add(1)
+			semaphoreOwned := true
+
+			defer func() {
+				s.activeTasks.Add(-1)
+				if semaphoreOwned {
+					logger.Debug("process lock released")
+					<-s.semaphore
+				}
+			}()
+
+			if err := s.ensureStarted(ctx); err != nil {
+				return err
+			}
+
+			if err := s.ensureHealthy(ctx); err != nil {
+				return err
+			}
+
+			err := s.runWithDeadline(ctx, task)
+
+			if s.maybeRestartAfterTask(logger) {
+				semaphoreOwned = false
+			}
+
+			// Note: no error wrapping because it leaks on Chromium console exceptions output.
+			return err
 		}()
 
 		if errors.Is(err, ErrProcessAlreadyRestarting) {
 			logger.Debug("process is already restarting, trying to acquire process lock again...")
+			time.Sleep(10 * time.Millisecond)
 			continue
 		}
 
@@ -283,6 +247,94 @@ func (s *processSupervisor) Run(ctx context.Context, logger *zap.Logger, task fu
 	}
 }
 
+// acquireSlot attempts to acquire a semaphore slot, yielding it back if a
+// restart drain is in progress.
+func (s *processSupervisor) acquireSlot(ctx context.Context, logger *zap.Logger) error {
+	select {
+	case s.semaphore <- struct{}{}:
+		// If a restart drain is in progress, release the slot
+		// immediately so the drain can acquire it instead.
+		if s.isRestarting.Load() {
+			<-s.semaphore
+			return ErrProcessAlreadyRestarting
+		}
+
+		logger.Debug("process lock acquired")
+
+		return nil
+	case <-ctx.Done():
+		logger.Debug("failed to acquire process lock before deadline")
+		s.reqQueueSize.Add(-1)
+
+		return fmt.Errorf("acquire process lock: %w", ctx.Err())
+	}
+}
+
+// ensureStarted performs a one-time lazy launch of the process on its first
+// use. Subsequent calls are no-ops.
+func (s *processSupervisor) ensureStarted(ctx context.Context) error {
+	if s.firstStart.Load() {
+		return nil
+	}
+
+	s.firstStartOnce.Do(func() {
+		s.firstStartErr = s.runWithDeadline(ctx, func() error {
+			return s.Launch()
+		})
+	})
+
+	if s.firstStartErr != nil {
+		return fmt.Errorf("process first start: %w", s.firstStartErr)
+	}
+
+	return nil
+}
+
+// ensureHealthy checks the underlying process health and triggers a
+// synchronous restart if the process is unhealthy. Skips the check if a
+// restart is already in progress.
+func (s *processSupervisor) ensureHealthy(ctx context.Context) error {
+	if s.isRestarting.Load() || s.process.Healthy(s.logger) {
+		return nil
+	}
+
+	s.logger.Debug("process is unhealthy, cannot handle task, restarting...")
+
+	if err := s.doRestart(ctx); err != nil {
+		return fmt.Errorf("process restart before task: %w", err)
+	}
+
+	return nil
+}
+
+// maybeRestartAfterTask checks if the maximum request limit has been reached
+// and, if so, triggers an asynchronous restart. If a restart is initiated, it
+// takes ownership of the caller's semaphore slot (the caller must not release
+// it). Returns true if ownership was taken.
+func (s *processSupervisor) maybeRestartAfterTask(logger *zap.Logger) bool {
+	if s.maxReqLimit <= 0 || s.reqCounter.Load() < s.maxReqLimit {
+		return false
+	}
+
+	if !s.restartMutex.TryLock() {
+		return false
+	}
+
+	s.logger.Debug("max request limit reached, restarting eagerly...")
+
+	go func() {
+		restartErr := s.doRestartLocked(context.Background())
+		s.restartMutex.Unlock()
+		if restartErr != nil {
+			s.logger.Error(fmt.Sprintf("process restart after task: %v", restartErr))
+		}
+		logger.Debug("process lock released")
+		<-s.semaphore
+	}()
+
+	return true
+}
+
 // doRestart coordinates a process restart, draining all active concurrent
 // tasks before stopping and restarting the process.
 func (s *processSupervisor) doRestart(ctx context.Context) error {
diff --git a/pkg/gotenberg/supervisor_test.go b/pkg/gotenberg/supervisor_test.go
index 8b57150f..b5bf8754 100644
--- a/pkg/gotenberg/supervisor_test.go
+++ b/pkg/gotenberg/supervisor_test.go
@@ -170,15 +170,15 @@ func TestProcessSupervisor_Healthy(t *testing.T) {
 		expectHealthy       bool
 	}{
 		{
-			scenario:         "non-started process is always healthy",
+			scenario:         "non-started process is healthy",
 			initiallyStarted: false,
 			expectHealthy:    true,
 		},
 		{
-			scenario:            "restarting process is always healthy",
+			scenario:            "restarting process is not healthy",
 			initiallyStarted:    true,
 			initiallyRestarting: true,
-			expectHealthy:       true,
+			expectHealthy:       false,
 		},
 		{
 			scenario:         "process reports as healthy",
@@ -552,6 +552,61 @@ func TestProcessSupervisor_ReqQueueSize(t *testing.T) {
 	}
 }
 
+func TestProcessSupervisor_QueueSizeCAS(t *testing.T) {
+	logger := zap.NewNop()
+
+	process := &ProcessMock{
+		StartMock: func(logger *zap.Logger) error {
+			return nil
+		},
+		HealthyMock: func(logger *zap.Logger) bool {
+			return true
+		},
+	}
+
+	maxQueueSize := int64(50)
+	// maxConcurrency=1 so all goroutines block on the semaphore, exercising queue logic.
+	ps := NewProcessSupervisor(logger, process, 0, maxQueueSize, 1).(*processSupervisor)
+
+	// Simulating a lock so that all goroutines queue up.
+	ps.semaphore <- struct{}{}
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	goroutines := 100
+	var wg sync.WaitGroup
+	var exceeded atomic.Int64
+
+	for range goroutines {
+		wg.Go(func() {
+			err := ps.Run(ctx, logger, func() error {
+				return nil
+			})
+			if err != nil {
+				if errors.Is(err, ErrMaximumQueueSizeExceeded) {
+					exceeded.Add(1)
+				}
+			}
+		})
+	}
+
+	// Wait a bit for goroutines to queue up.
+	time.Sleep(50 * time.Millisecond)
+
+	currentQueue := ps.ReqQueueSize()
+	if currentQueue > maxQueueSize {
+		t.Fatalf("queue size %d exceeded max %d", currentQueue, maxQueueSize)
+	}
+
+	cancel()
+	wg.Wait()
+
+	if exceeded.Load() < int64(goroutines)-maxQueueSize {
+		t.Errorf("expected at least %d rejections, got %d", goroutines-int(maxQueueSize), exceeded.Load())
+	}
+}
+
 func TestProcessSupervisor_RestartsCount(t *testing.T) {
 	for _, tc := range []struct {
 		scenario              string