commit for backup

Author: Samuel Archambault

libpod/oci_conmon_common.go

@@ -49,6 +49,12 @@ const (
 
 	// Healthcheck message type from conmon (using negative to avoid PID conflicts)
 	HealthCheckMsgStatusUpdate = -100
+
+	// Healthcheck status values sent by conmon (added to base message type -100)
+	HealthCheckStatusNone      = 0
+	HealthCheckStatusStarting  = 1
+	HealthCheckStatusHealthy   = 2
+	HealthCheckStatusUnhealthy = 3
 )
 
 // ConmonOCIRuntime is an OCI runtime managed by Conmon.
@@ -1508,75 +1514,61 @@ func readConmonHealthCheckPipeData(ctr *Container, pipe *os.File) {
 			return
 		}
 
-		// Log the raw JSON string received from conmon
-		logrus.Debugf("HEALTHCHECK: Raw JSON received from conmon for container %s: %q", ctr.ID(), string(b))
-		logrus.Debugf("HEALTHCHECK: JSON length: %d bytes", len(b))
+		// Log the raw message received from conmon
+		logrus.Debugf("HEALTHCHECK: Raw message received from conmon for container %s: %q", ctr.ID(), string(b))
+		logrus.Debugf("HEALTHCHECK: Message length: %d bytes", len(b))
 
-		// Parse the JSON
-		var si syncInfo
-		if err := json.Unmarshal(b, &si); err != nil {
-			logrus.Errorf("HEALTHCHECK: Failed to parse JSON from conmon for container %s: %v", ctr.ID(), err)
+		// Parse the message as a simple integer (no JSON)
+		messageStr := strings.TrimSpace(string(b))
+		messageType, err := strconv.Atoi(messageStr)
+		if err != nil {
+			logrus.Errorf("HEALTHCHECK: Failed to parse message as integer for container %s: %v", ctr.ID(), err)
 			continue
 		}
 
-		logrus.Debugf("HEALTHCHECK: Parsed sync info for container %s: Data=%d, Message=%q", ctr.ID(), si.Data, si.Message)
-
-		// Handle healthcheck status updates (negative message types)
-		if si.Data == HealthCheckMsgStatusUpdate && si.Message != "" {
-			logrus.Infof("HEALTHCHECK: Received healthcheck status update for container %s: %s", ctr.ID(), si.Message)
-			// Process the healthcheck status update
-			if err := handleHealthCheckStatusUpdate(ctr, si.Message); err != nil {
-				logrus.Errorf("HEALTHCHECK: Failed to process healthcheck status update for container %s: %v", ctr.ID(), err)
+		logrus.Debugf("HEALTHCHECK: Parsed message type for container %s: %d", ctr.ID(), messageType)
+
+		// Handle healthcheck status updates based on your new encoding scheme
+		// Base message type is -100, status values are added to it:
+		// -100 + 0 (none) = -100
+		// -100 + 1 (starting) = -99
+		// -100 + 2 (healthy) = -98
+		// -100 + 3 (unhealthy) = -97
+		if messageType >= HealthCheckMsgStatusUpdate && messageType <= HealthCheckMsgStatusUpdate+HealthCheckStatusUnhealthy {
+			statusValue := messageType - HealthCheckMsgStatusUpdate // Convert back to status value
+			var status string
+
+			switch statusValue {
+			case HealthCheckStatusNone:
+				status = define.HealthCheckReset // "reset" or "none"
+			case HealthCheckStatusStarting:
+				status = define.HealthCheckStarting // "starting"
+			case HealthCheckStatusHealthy:
+				status = define.HealthCheckHealthy // "healthy"
+			case HealthCheckStatusUnhealthy:
+				status = define.HealthCheckUnhealthy // "unhealthy"
+			default:
+				logrus.Errorf("HEALTHCHECK: Unknown status value %d for container %s", statusValue, ctr.ID())
+				continue
 			}
-		} else if si.Data == HealthCheckMsgStatusUpdate {
-			logrus.Debugf("HEALTHCHECK: Received healthcheck message type %d for container %s but no message content", si.Data, ctr.ID())
-		} else if si.Data < 0 {
-			// This might be a healthcheck message with a different negative type number
-			logrus.Debugf("HEALTHCHECK: Received negative message type %d for container %s - might be healthcheck related", si.Data, ctr.ID())
-		} else if si.Data > 0 {
-			// This might be a PID or other positive message - log but don't process as healthcheck
-			logrus.Debugf("HEALTHCHECK: Received positive message type %d for container %s - not healthcheck related", si.Data, ctr.ID())
-		}
-	}
-}
 
-// handleHealthCheckStatusUpdate processes healthcheck status updates from conmon
-func handleHealthCheckStatusUpdate(ctr *Container, message string) error {
-	logrus.Debugf("HEALTHCHECK: Processing healthcheck status update for container %s: %s", ctr.ID(), message)
+			logrus.Infof("HEALTHCHECK: Received healthcheck status update for container %s: %s (message type: %d, status value: %d)",
+				ctr.ID(), status, messageType, statusValue)
 
-	// Parse the healthcheck status JSON from conmon
-	type healthCheckStatus struct {
-		Type        string `json:"type"`
-		ContainerID string `json:"container_id"`
-		Status      string `json:"status"`
-		ExitCode    int    `json:"exit_code"`
-		Timestamp   int64  `json:"timestamp"`
-	}
-
-	var hcStatus healthCheckStatus
-	if err := json.Unmarshal([]byte(message), &hcStatus); err != nil {
-		logrus.Errorf("HEALTHCHECK: Failed to parse healthcheck status JSON for container %s: %v", ctr.ID(), err)
-		return fmt.Errorf("failed to parse healthcheck status from conmon: %w", err)
-	}
-
-	logrus.Debugf("HEALTHCHECK: Parsed healthcheck status for container %s: Type=%s, ContainerID=%s, Status=%s, ExitCode=%d, Timestamp=%d",
-		ctr.ID(), hcStatus.Type, hcStatus.ContainerID, hcStatus.Status, hcStatus.ExitCode, hcStatus.Timestamp)
-
-	// Verify this is for the correct container
-	if hcStatus.ContainerID != ctr.ID() {
-		logrus.Errorf("HEALTHCHECK: Healthcheck status for wrong container: expected %s, got %s", ctr.ID(), hcStatus.ContainerID)
-		return fmt.Errorf("healthcheck status for wrong container: expected %s, got %s", ctr.ID(), hcStatus.ContainerID)
-	}
-
-	// Update the container's healthcheck status
-	logrus.Debugf("HEALTHCHECK: Updating healthcheck status for container %s to %s", ctr.ID(), hcStatus.Status)
-	if err := ctr.updateHealthStatus(hcStatus.Status); err != nil {
-		logrus.Errorf("HEALTHCHECK: Failed to update healthcheck status for container %s: %v", ctr.ID(), err)
-		return fmt.Errorf("failed to update healthcheck status for container %s: %w", ctr.ID(), err)
+			// Update the container's healthcheck status
+			if err := ctr.updateHealthStatus(status); err != nil {
+				logrus.Errorf("HEALTHCHECK: Failed to update healthcheck status for container %s: %v", ctr.ID(), err)
+			} else {
+				logrus.Infof("HEALTHCHECK: Successfully updated healthcheck status for container %s to %s", ctr.ID(), status)
+			}
+		} else if messageType < 0 {
+			// Other negative message types - might be healthcheck related but not recognized
+			logrus.Debugf("HEALTHCHECK: Received unrecognized negative message type %d for container %s - might be healthcheck related", messageType, ctr.ID())
+		} else if messageType > 0 {
+			// Positive message types - not healthcheck related
+			logrus.Debugf("HEALTHCHECK: Received positive message type %d for container %s - not healthcheck related", messageType, ctr.ID())
+		}
 	}
-
-	logrus.Infof("HEALTHCHECK: Successfully updated healthcheck status for container %s to %s (exit code: %d)", ctr.ID(), hcStatus.Status, hcStatus.ExitCode)
-	return nil
 }
 
 // writeConmonPipeData writes nonce data to a pipe