fix(container) correct readiness error message and reduce waitForReady complexity

commit: 198c8789ab5abbccd88ce7f7dd4308c5fc7fc479
author: bnema · 8mo ago · 2026-02-06 20:43 UTC
parent: 84879b9
signature: unsigned

1 file · +33 -23

M internal/usecase/container/service.go +33 -23


@@ -1087,7 +1087,7 @@ func (s *Service) pullImageWithRetry(ctx context.Context, pullFn func(context.Co
1087	1087		return nil
1088	1088		}
1089	1089		if !isInternal \|\| !isConnectionRefusedError(err) \|\| attempt == internalPullMaxAttempts {
1090		-	return err
	1090	+	return fmt.Errorf("internal image pull failed after %d attempts: %w", attempt, err)
1091	1091		}
1092	1092
1093	1093		backoff := time.Duration(attempt) * time.Second

@@ -1764,30 +1764,14 @@ func (s Service) findContainerByName(ctx context.Context, name string) domain.
1764	1764		func (s *Service) waitForReady(ctx context.Context, containerID string) error {
1765	1765		log := zerowrap.FromCtx(ctx)
1766	1766
1767		-	// Poll for container to be running (max 30 seconds)
1768		-	for i := 0; i < 30; i++ {
1769		-	running, err := s.runtime.IsContainerRunning(ctx, containerID)
1770		-	if err != nil {
1771		-	return err
1772		-	}
1773		-	if running {
1774		-	break
1775		-	}
1776		-	if i == 29 {
1777		-	return fmt.Errorf("container did not start within 30 seconds")
1778		-	}
1779		-	select {
1780		-	case <-time.After(time.Second):
1781		-	case <-ctx.Done():
1782		-	return ctx.Err()
1783		-	}
	1767	+	if err := s.pollContainerRunning(ctx, containerID); err != nil {
	1768	+	return err
1784	1769		}
1785	1770
1786	1771		s.mu.RLock()
1787	1772		cfg := s.config
1788	1773		s.mu.RUnlock()
1789	1774
1790		-	// Additional readiness delay (configurable, default 5 seconds)
1791	1775		delay := cfg.ReadinessDelay
1792	1776		if delay == 0 {
1793	1777		delay = 5 * time.Second

@@ -1809,8 +1793,34 @@ func (s *Service) waitForReady(ctx context.Context, containerID string) error {
1809	1793		return nil
1810	1794		}
1811	1795
1812		-	// Be tolerant to short startup flaps: retry within a bounded recovery window
1813		-	// before treating readiness as failed.
	1796	+	return s.waitForRecovery(ctx, containerID)
	1797	+	}
	1798	+
	1799	+	// pollContainerRunning polls until the container is running (max 30 seconds).
	1800	+	func (s *Service) pollContainerRunning(ctx context.Context, containerID string) error {
	1801	+	for i := 0; i < 30; i++ {
	1802	+	running, err := s.runtime.IsContainerRunning(ctx, containerID)
	1803	+	if err != nil {
	1804	+	return err
	1805	+	}
	1806	+	if running {
	1807	+	return nil
	1808	+	}
	1809	+	if i == 29 {
	1810	+	return fmt.Errorf("container did not start within 30 seconds")
	1811	+	}
	1812	+	select {
	1813	+	case <-time.After(time.Second):
	1814	+	case <-ctx.Done():
	1815	+	return ctx.Err()
	1816	+	}
	1817	+	}
	1818	+	return nil
	1819	+	}
	1820	+
	1821	+	// waitForRecovery waits for a container to recover within the recovery window.
	1822	+	func (s *Service) waitForRecovery(ctx context.Context, containerID string) error {
	1823	+	log := zerowrap.FromCtx(ctx)
1814	1824		log.Warn().
1815	1825		Dur("recovery_window", readinessRecoveryWindow).
1816	1826		Msg("container not running after readiness delay, waiting for recovery")

@@ -1823,7 +1833,7 @@ func (s *Service) waitForReady(ctx context.Context, containerID string) error {
1823	1833		return ctx.Err()
1824	1834		}
1825	1835
1826		-	running, err = s.runtime.IsContainerRunning(ctx, containerID)
	1836	+	running, err := s.runtime.IsContainerRunning(ctx, containerID)
1827	1837		if err != nil {
1828	1838		return err
1829	1839		}

@@ -1833,7 +1843,7 @@ func (s *Service) waitForReady(ctx context.Context, containerID string) error {
1833	1843		}
1834	1844		}
1835	1845
1836		-	return fmt.Errorf("container stopped during readiness delay")
	1846	+	return fmt.Errorf("container not running after readiness delay and recovery window")
1837	1847		}
1838	1848
1839	1849		// publishContainerDeployed publishes a container.deployed event.

 			return nil
 		}
 		if !isInternal || !isConnectionRefusedError(err) || attempt == internalPullMaxAttempts {
-			return err
+			return fmt.Errorf("internal image pull failed after %d attempts: %w", attempt, err)
 		}
 		backoff := time.Duration(attempt) * time.Second
 func (s *Service) waitForReady(ctx context.Context, containerID string) error {
 	log := zerowrap.FromCtx(ctx)
-	// Poll for container to be running (max 30 seconds)
+	if err := s.pollContainerRunning(ctx, containerID); err != nil {
-	for i := 0; i < 30; i++ {
+		return err
-		running, err := s.runtime.IsContainerRunning(ctx, containerID)
-		if err != nil {
-			return err
-		}
-		if running {
-			break
-		}
-		if i == 29 {
-			return fmt.Errorf("container did not start within 30 seconds")
-		}
-		select {
-		case <-time.After(time.Second):
-		case <-ctx.Done():
-			return ctx.Err()
-		}
 	}
 	s.mu.RLock()
 	cfg := s.config
 	s.mu.RUnlock()
-	// Additional readiness delay (configurable, default 5 seconds)
 	delay := cfg.ReadinessDelay
 	if delay == 0 {
 		delay = 5 * time.Second
 		return nil
 	}
-	// Be tolerant to short startup flaps: retry within a bounded recovery window
+	return s.waitForRecovery(ctx, containerID)
-	// before treating readiness as failed.
+}
+// pollContainerRunning polls until the container is running (max 30 seconds).
+func (s *Service) pollContainerRunning(ctx context.Context, containerID string) error {
+	for i := 0; i < 30; i++ {
+		running, err := s.runtime.IsContainerRunning(ctx, containerID)
+		if err != nil {
+			return err
+		}
+		if running {
+			return nil
+		}
+		if i == 29 {
+			return fmt.Errorf("container did not start within 30 seconds")
+		}
+		select {
+		case <-time.After(time.Second):
+		case <-ctx.Done():
+			return ctx.Err()
+		}
+	}
+	return nil
+}
+// waitForRecovery waits for a container to recover within the recovery window.
+func (s *Service) waitForRecovery(ctx context.Context, containerID string) error {
+	log := zerowrap.FromCtx(ctx)
 	log.Warn().
 		Dur("recovery_window", readinessRecoveryWindow).
 		Msg("container not running after readiness delay, waiting for recovery")
 			return ctx.Err()
 		}
-		running, err = s.runtime.IsContainerRunning(ctx, containerID)
+		running, err := s.runtime.IsContainerRunning(ctx, containerID)
 		if err != nil {
 			return err
 		}
 		}
 	}
-	return fmt.Errorf("container stopped during readiness delay")
+	return fmt.Errorf("container not running after readiness delay and recovery window")
 }
 // publishContainerDeployed publishes a container.deployed event.