Fix group migration when runconfig is corrupt (#1506)

Author: eleftherias

pkg/migration/default_group.go

@@ -7,6 +7,7 @@ import (
 	"github.com/stacklok/toolhive/pkg/config"
 	"github.com/stacklok/toolhive/pkg/groups"
 	"github.com/stacklok/toolhive/pkg/logger"
+	"github.com/stacklok/toolhive/pkg/runner"
 	"github.com/stacklok/toolhive/pkg/workloads"
 )
 
@@ -95,6 +96,21 @@ func (m *DefaultGroupMigrator) migrateWorkloadsToDefaultGroup(ctx context.Contex
 
 	migratedCount := 0
 	for _, workloadName := range workloadsWithoutGroup {
+		// Check the runconfig to validate if the workload is not in a group already
+		// ListWorkloadsInGroup doesn't check the runconfig, so we need an additional check here
+		runConfig, err := runner.LoadState(ctx, workloadName)
+		if err != nil {
+			logger.Warnf("Failed to migrate workload %s to default group due to missing or corrupt"+
+				" run configuration: %v. The workload may be unhealthy and need to be deleted.", workloadName, err)
+			continue
+		}
+
+		// Check if the workload is actually in the specified group
+		if runConfig.Group != "" {
+			logger.Debugf("Workload %s is already in group %s, skipping migration", workloadName, runConfig.Group)
+			continue
+		}
+
 		// Move workload to default group
 		if err := m.workloadsManager.MoveToGroup(ctx, []string{workloadName}, "", groups.DefaultGroup); err != nil {
 			logger.Warnf("Failed to migrate workload %s to default group: %v", workloadName, err)

pkg/workloads/statuses/file_status.go

@@ -127,6 +127,9 @@ func (f *fileStatusManager) ListWorkloads(ctx context.Context, listAll bool, lab
 		return nil, fmt.Errorf("failed to get workloads from files: %w", err)
 	}
 
+	// TODO: Fetch the runconfig if present to populate additional fields like package, tool type, group etc.
+	// There's currently an import cycle between this package and the runconfig package
+
 	// Create a map of runtime workloads by name for easy lookup
 	workloadMap := f.mergeRuntimeAndFileWorkloads(ctx, runtimeContainers, fileWorkloads)