oc-datacenter/infrastructure/kubernetes/watchdog.go

package kubernetes

import (
	"context"
	"fmt"
	"regexp"
	"strings"
	"time"

	"oc-datacenter/conf"
	"oc-datacenter/infrastructure"
	"oc-datacenter/infrastructure/admiralty"
	"oc-datacenter/infrastructure/storage"

	oclib "cloud.o-forge.io/core/oc-lib"
	"cloud.o-forge.io/core/oc-lib/dbs"
	bookingmodel "cloud.o-forge.io/core/oc-lib/models/booking"
	"cloud.o-forge.io/core/oc-lib/models/workflow_execution"
	"cloud.o-forge.io/core/oc-lib/tools"
	v1 "k8s.io/api/core/v1"
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
)

// uuidNsPattern matches Kubernetes namespace names that are execution UUIDs.
var uuidNsPattern = regexp.MustCompile(`^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$`)

// WatchInfra is a safety-net watchdog that periodically scans Kubernetes for
// execution namespaces whose WorkflowExecution has reached a terminal state
// but whose infra was never torn down (e.g. because WORKFLOW_DONE_EVENT was
// missed due to oc-monitord or oc-datacenter crash/restart).
//
// Must be launched in a goroutine from main.
func (s *KubernetesService) Watch() {
	logger := oclib.GetLogger()
	logger.Info().Msg("InfraWatchdog: started")
	ticker := time.NewTicker(5 * time.Minute)
	defer ticker.Stop()
	for range ticker.C {
		if err := s.scanOrphaned(); err != nil {
			logger.Error().Msg("InfraWatchdog: " + err.Error())
		}
		if err := s.scanOrphanedMinio(); err != nil {
			logger.Error().Msg("InfraWatchdog(minio): " + err.Error())
		}
		if err := s.scanOrphanedAdmiraltyNodes(); err != nil {
			logger.Error().Msg("InfraWatchdog(admiralty-nodes): " + err.Error())
		}
		if err := s.scanOrphanedPVC(); err != nil {
			logger.Error().Msg("InfraWatchdog(pvc): " + err.Error())
		}
	}
}

// scanOrphanedInfra lists all UUID-named Kubernetes namespaces, looks up their
// WorkflowExecution in the DB, and triggers teardown for any that are in a
// terminal state. Namespaces already in Terminating phase are skipped.
func (s *KubernetesService) scanOrphaned() error {
	logger := oclib.GetLogger()

	serv, err := tools.NewKubernetesService(
		conf.GetConfig().KubeHost+":"+conf.GetConfig().KubePort,
		conf.GetConfig().KubeCA,
		conf.GetConfig().KubeCert,
		conf.GetConfig().KubeData,
	)
	if err != nil {
		return fmt.Errorf("failed to init k8s service: %w", err)
	}

	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
	defer cancel()

	nsList, err := serv.Set.CoreV1().Namespaces().List(ctx, metav1.ListOptions{})
	if err != nil {
		return fmt.Errorf("failed to list namespaces: %w", err)
	}

	myself, err := oclib.GetMySelf()
	if err != nil {
		return fmt.Errorf("could not resolve local peer: %w", err)
	}
	peerID := myself.GetID()

	for _, ns := range nsList.Items {
		executionsID := ns.Name
		if !uuidNsPattern.MatchString(executionsID) {
			continue
		}
		// Skip namespaces already being deleted by a previous teardown.
		if ns.Status.Phase == v1.NamespaceTerminating {
			continue
		}

		exec := findTerminalExecution(executionsID, peerID)
		if exec == nil {
			continue
		}

		logger.Info().Msgf("InfraWatchdog: orphaned infra detected for execution %s (state=%v) → teardown",
			executionsID, exec.State)
		go s.TeardownForExecution(exec.GetID())
	}
	return nil
}

// scanOrphanedMinio scans LIVE_STORAGE bookings for executions that are in a
// terminal state and triggers Minio teardown for each unique executionsID found.
// This covers the case where the Kubernetes namespace is already gone (manual
// deletion, prior partial teardown) but Minio SA and bucket were never revoked.
func (s *KubernetesService) scanOrphanedMinio() error {
	logger := oclib.GetLogger()

	myself, err := oclib.GetMySelf()
	if err != nil {
		return fmt.Errorf("could not resolve local peer: %w", err)
	}
	peerID := myself.GetID()

	res := oclib.NewRequest(oclib.LibDataEnum(oclib.BOOKING), "", peerID, []string{}, nil).
		Search(&dbs.Filters{
			And: map[string][]dbs.Filter{
				"resource_type": {{Operator: dbs.EQUAL.String(), Value: tools.LIVE_STORAGE.EnumIndex()}},
			},
		}, "", false)

	if res.Err != "" {
		return fmt.Errorf("failed to search LIVE_STORAGE bookings: %s", res.Err)
	}

	// Collect unique executionsIDs to avoid redundant teardowns.
	seen := map[string]bool{}
	ctx := context.Background()

	for _, dbo := range res.Data {
		b, ok := dbo.(*bookingmodel.Booking)
		if !ok || seen[b.ExecutionsID] {
			continue
		}

		exec := findTerminalExecution(b.ExecutionsID, peerID)
		if exec == nil {
			continue
		}

		seen[b.ExecutionsID] = true
		minio := storage.NewMinioSetter(b.ExecutionsID, b.ResourceID)
		// Determine this peer's role and call the appropriate teardown.
		if b.DestPeerID == peerID {
			logger.Info().Msgf("InfraWatchdog(minio): orphaned target resources for exec %s → TeardownAsTarget", b.ExecutionsID)
			event := storage.MinioDeleteEvent{
				ExecutionsID: b.ExecutionsID,
				MinioID:      b.ResourceID,
				SourcePeerID: b.DestPeerID,
				DestPeerID:   peerID,
			}
			go minio.TeardownAsTarget(ctx, event)
		} else {
			logger.Info().Msgf("InfraWatchdog(minio): orphaned source resources for exec %s → TeardownAsSource", b.ExecutionsID)
			event := storage.MinioDeleteEvent{
				ExecutionsID: b.ExecutionsID,
				MinioID:      b.ResourceID,
				SourcePeerID: peerID,
				DestPeerID:   b.DestPeerID,
			}
			go minio.TeardownAsSource(ctx, event)
		}
	}
	return nil
}

// scanOrphanedAdmiraltyNodes lists all Kubernetes nodes, identifies Admiralty
// virtual nodes (name prefix "admiralty-{UUID}-") that are NotReady, and
// explicitly deletes them when their WorkflowExecution is in a terminal state.
//
// This covers the gap where the namespace is already gone (or Terminating) but
// the virtual node was never cleaned up by the Admiralty controller — which can
// happen when the node goes NotReady before the AdmiraltyTarget CRD is deleted.
func (s *KubernetesService) scanOrphanedAdmiraltyNodes() error {
	logger := oclib.GetLogger()

	serv, err := tools.NewKubernetesService(
		conf.GetConfig().KubeHost+":"+conf.GetConfig().KubePort,
		conf.GetConfig().KubeCA,
		conf.GetConfig().KubeCert,
		conf.GetConfig().KubeData,
	)
	if err != nil {
		return fmt.Errorf("failed to init k8s service: %w", err)
	}

	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
	defer cancel()

	nodeList, err := serv.Set.CoreV1().Nodes().List(ctx, metav1.ListOptions{})
	if err != nil {
		return fmt.Errorf("failed to list nodes: %w", err)
	}

	myself, err := oclib.GetMySelf()
	if err != nil {
		return fmt.Errorf("could not resolve local peer: %w", err)
	}
	peerID := myself.GetID()

	for _, node := range nodeList.Items {
		// Admiralty virtual nodes are named: admiralty-{executionID}-target-{...}
		rest := strings.TrimPrefix(node.Name, "admiralty-")
		if rest == node.Name {
			continue // not an admiralty node
		}
		// UUID is exactly 36 chars: 8-4-4-4-12
		if len(rest) < 36 {
			continue
		}
		executionsID := rest[:36]
		if !uuidNsPattern.MatchString(executionsID) {
			continue
		}

		// Only act on NotReady nodes.
		ready := false
		for _, cond := range node.Status.Conditions {
			if cond.Type == v1.NodeReady {
				ready = cond.Status == v1.ConditionTrue
				break
			}
		}
		if ready {
			continue
		}

		exec := findTerminalExecution(executionsID, peerID)
		if exec == nil {
			continue
		}

		logger.Info().Msgf("InfraWatchdog(admiralty-nodes): NotReady orphaned node %s for terminal execution %s → deleting",
			node.Name, executionsID)
		if delErr := serv.Set.CoreV1().Nodes().Delete(ctx, node.Name, metav1.DeleteOptions{}); delErr != nil {
			logger.Error().Msgf("InfraWatchdog(admiralty-nodes): failed to delete node %s: %v", node.Name, delErr)
		}
	}
	return nil
}

// scanOrphanedPVC scans LIVE_STORAGE bookings for executions that are in a
// terminal state and triggers PVC teardown for each one where this peer holds
// the local storage. This covers the case where the Kubernetes namespace was
// already deleted (or its teardown was partial) but the PersistentVolume
// (cluster-scoped) was never reclaimed.
//
// A LIVE_STORAGE booking is treated as a local PVC only when ResolveStorageName
// returns a non-empty name — the same guard used by teardownPVCForExecution.
func (s *KubernetesService) scanOrphanedPVC() error {
	logger := oclib.GetLogger()

	myself, err := oclib.GetMySelf()
	if err != nil {
		return fmt.Errorf("could not resolve local peer: %w", err)
	}
	peerID := myself.GetID()

	res := oclib.NewRequest(oclib.LibDataEnum(oclib.BOOKING), "", peerID, []string{}, nil).
		Search(&dbs.Filters{
			And: map[string][]dbs.Filter{
				"resource_type": {{Operator: dbs.EQUAL.String(), Value: tools.LIVE_STORAGE.EnumIndex()}},
			},
		}, "", false)

	if res.Err != "" {
		return fmt.Errorf("failed to search LIVE_STORAGE bookings: %s", res.Err)
	}

	seen := map[string]bool{}
	ctx := context.Background()

	for _, dbo := range res.Data {
		b, ok := dbo.(*bookingmodel.Booking)
		if !ok || seen[b.ExecutionsID+b.ResourceID] {
			continue
		}

		storageName := storage.ResolveStorageName(b.ResourceID, peerID)
		if storageName == "" {
			continue // not a local PVC booking
		}

		exec := findTerminalExecution(b.ExecutionsID, peerID)
		if exec == nil {
			continue
		}

		seen[b.ExecutionsID+b.ResourceID] = true

		logger.Info().Msgf("InfraWatchdog(pvc): orphaned PVC for exec %s storage %s → TeardownAsSource",
			b.ExecutionsID, b.ResourceID)
		event := storage.PVCDeleteEvent{
			ExecutionsID: b.ExecutionsID,
			StorageID:    b.ResourceID,
			StorageName:  storageName,
			SourcePeerID: peerID,
			DestPeerID:   b.DestPeerID,
		}
		go storage.NewPVCSetter(b.ExecutionsID, b.ResourceID).TeardownAsSource(ctx, event)
	}
	return nil
}

// findTerminalExecution returns the WorkflowExecution for the given executionsID
// if it exists in the DB and is in a terminal state, otherwise nil.
func findTerminalExecution(executionsID string, peerID string) *workflow_execution.WorkflowExecution {
	res := oclib.NewRequest(oclib.LibDataEnum(oclib.WORKFLOW_EXECUTION), "", peerID, []string{}, nil).
		Search(&dbs.Filters{
			And: map[string][]dbs.Filter{
				"executions_id": {{Operator: dbs.EQUAL.String(), Value: executionsID}},
			},
		}, "", false)

	if res.Err != "" || len(res.Data) == 0 {
		return nil
	}

	exec, ok := res.Data[0].(*workflow_execution.WorkflowExecution)
	if !ok {
		return nil
	}

	if !infrastructure.ClosingStates[exec.State] {
		return nil
	}
	return exec
}

// teardownInfraForExecution handles infrastructure cleanup when a workflow terminates.
// oc-datacenter is responsible only for infra here — booking/execution state
// is managed by oc-scheduler.
func (s *KubernetesService) TeardownForExecution(executionID string) {
	logger := oclib.GetLogger()

	myself, err := oclib.GetMySelf()
	if err != nil || myself == nil {
		return
	}
	selfPeerID := myself.GetID()

	adminReq := &tools.APIRequest{Admin: true}
	res, _, loadErr := workflow_execution.NewAccessor(adminReq).LoadOne(executionID)
	if loadErr != nil || res == nil {
		logger.Warn().Msgf("teardownInfraForExecution: execution %s not found", executionID)
		return
	}
	exec := res.(*workflow_execution.WorkflowExecution)

	ctx := context.Background()
	admiralty.NewAdmiraltySetter(s.ExecutionsID).TeardownIfRemote(exec, selfPeerID)
	storage.NewMinioSetter(s.ExecutionsID, "").TeardownForExecution(ctx, selfPeerID)
	storage.NewPVCSetter(s.ExecutionsID, "").TeardownForExecution(ctx, selfPeerID)
	s.CleanupImages(ctx)
}