seaweedfs/weed/worker/tasks/erasure_coding/detection.go

package erasure_coding

import (
	"context"
	"fmt"
	"sort"
	"strings"
	"time"

	"github.com/seaweedfs/seaweedfs/weed/admin/topology"
	"github.com/seaweedfs/seaweedfs/weed/glog"
	"github.com/seaweedfs/seaweedfs/weed/operation"
	"github.com/seaweedfs/seaweedfs/weed/pb"
	"github.com/seaweedfs/seaweedfs/weed/pb/volume_server_pb"
	"github.com/seaweedfs/seaweedfs/weed/pb/worker_pb"
	"github.com/seaweedfs/seaweedfs/weed/storage/erasure_coding"
	"github.com/seaweedfs/seaweedfs/weed/storage/erasure_coding/placement"
	"github.com/seaweedfs/seaweedfs/weed/util"
	"github.com/seaweedfs/seaweedfs/weed/util/wildcard"
	"github.com/seaweedfs/seaweedfs/weed/worker/tasks/base"
	workerutil "github.com/seaweedfs/seaweedfs/weed/worker/tasks/util"
	"github.com/seaweedfs/seaweedfs/weed/worker/types"
)

const (
	minProposalsBeforeEarlyStop    = 10
	maxConsecutivePlanningFailures = 10
)

// Detection implements the detection logic for erasure coding tasks.
// It respects ctx cancellation and can stop early once maxResults is reached.
func Detection(ctx context.Context, metrics []*types.VolumeHealthMetrics, clusterInfo *types.ClusterInfo, config base.TaskConfig, maxResults int) ([]*types.TaskDetectionResult, bool, error) {
	if !config.IsEnabled() {
		return nil, false, nil
	}

	if maxResults < 0 {
		maxResults = 0
	}

	ecConfig := config.(*Config)
	var results []*types.TaskDetectionResult
	hasMore := false
	stoppedEarly := false
	now := time.Now()
	quietThreshold := time.Duration(ecConfig.QuietForSeconds) * time.Second
	minSizeBytes := uint64(ecConfig.MinSizeMB) * 1024 * 1024 // Configurable minimum

	debugCount := 0
	skippedAlreadyEC := 0
	skippedTooSmall := 0
	skippedCollectionFilter := 0
	skippedQuietTime := 0
	skippedFullness := 0
	skippedRemote := 0
	skippedTooFewNodes := 0
	consecutivePlanningFailures := 0

	var planner *ecPlacementPlanner

	allowedCollections := wildcard.CompileWildcardMatchers(ecConfig.CollectionFilter)

	// Cluster node count for the min-node safety gate (mirrors the shell ec.encode
	// guard that refuses to encode when nodes < parity shards, so shards cannot be
	// spread for fault tolerance).
	clusterNodeCount := countTopologyNodes(clusterInfo.ActiveTopology)

	// Group metrics by VolumeID to handle replicas and select canonical server
	volumeGroups := make(map[uint32][]*types.VolumeHealthMetrics)
	for _, metric := range metrics {
		if ctx != nil {
			if err := ctx.Err(); err != nil {
				return results, hasMore, err
			}
		}
		volumeGroups[metric.VolumeID] = append(volumeGroups[metric.VolumeID], metric)
	}

	groupKeys := make([]uint32, 0, len(volumeGroups))
	for volumeID := range volumeGroups {
		groupKeys = append(groupKeys, volumeID)
	}
	sort.Slice(groupKeys, func(i, j int) bool { return groupKeys[i] < groupKeys[j] })

	// Iterate over groups to check criteria and creation tasks
	for idx, volumeID := range groupKeys {
		if ctx != nil {
			if err := ctx.Err(); err != nil {
				return results, hasMore, err
			}
		}
		if maxResults > 0 && len(results) >= maxResults {
			if idx+1 < len(groupKeys) {
				hasMore = true
			}
			stoppedEarly = true
			break
		}

		groupMetrics := volumeGroups[volumeID]

		// Find canonical metric (lowest Server ID) to ensure consistent task deduplication
		metric := groupMetrics[0]
		for _, m := range groupMetrics {
			if m.Server < metric.Server {
				metric = m
			}
		}

		// Skip if already EC volume
		if metric.IsECVolume {
			skippedAlreadyEC++
			continue
		}

		// Handle the "stuck source" state from #9448: a previous encode
		// succeeded but the post-encode source-delete left a regular replica
		// behind, so the master heartbeats BOTH the replica AND its EC shards.
		// metric.IsECVolume above is set only for the EC-side metric path, so
		// the canonical metric we picked is the regular replica with
		// IsECVolume=false. Re-proposing an encode in that state collides with
		// the mounted shards on the targets ("ec volume %d is mounted; refusing
		// overwrite") and the detector re-queues forever.
		//
		// We only act when the EC shard set is COMPLETE — fewer than
		// totalShards present means the existing recovery branch below
		// (around the `existingECShards` block) should keep its chance to
		// fold the partial shards into the new task. Counting walks
		// EcIndexBits to handle a single info entry carrying multiple shards.
		if clusterInfo.ActiveTopology != nil {
			shardCount := countExistingEcShardsForVolume(clusterInfo.ActiveTopology, metric.VolumeID, metric.Collection)
			totalShards := erasure_coding.DataShardsCount + erasure_coding.ParityShardsCount
			if shardCount >= totalShards {
				glog.Warningf("EC Detection: Volume %d has all %d EC shards in topology; "+
					"source replica on %s is orphaned (#9448).",
					metric.VolumeID, totalShards, metric.Server)
				if clusterInfo.GrpcDialOption != nil {
					deleted, cleanupErr := cleanupOrphanSourceReplicas(ctx, clusterInfo, metric, totalShards)
					switch {
					case cleanupErr != nil:
						// Don't fall through to a re-encode — that would just
						// collide with the mounted shards again. Surface the
						// failure and wait for the next cycle; the source is
						// still safe.
						glog.Warningf("EC Detection: failed to auto-clean orphaned source for volume %d: %v", metric.VolumeID, cleanupErr)
					case deleted > 0:
						glog.Infof("EC Detection: auto-cleaned %d orphaned source replica(s) for volume %d after verifying all %d EC shards present", deleted, metric.VolumeID, totalShards)
					default:
						glog.V(1).Infof("EC Detection: no orphaned regular replicas found in topology for volume %d (collection %q)", metric.VolumeID, metric.Collection)
					}
				} else {
					glog.Warningf("EC Detection: no gRPC dial option available to auto-clean orphaned source for volume %d; "+
						"to clean up by hand, send a targeted VolumeDelete RPC to %s only — DO NOT use the cluster-wide `volume.delete` shell command, which would also delete the EC shards.",
						metric.VolumeID, metric.Server)
				}
				skippedAlreadyEC++
				continue
			}
		}

		// Check minimum size requirement
		if metric.Size < minSizeBytes {
			skippedTooSmall++
			continue
		}

		// Check collection filter if specified
		if len(allowedCollections) > 0 && !wildcard.MatchesAnyWildcard(allowedCollections, metric.Collection) {
			skippedCollectionFilter++
			continue
		}

		// Skip remote/tiered volumes: encoding them would lose the tiering. The
		// shell ec.encode excludes remote volumes for the same reason.
		if metric.HasRemoteCopy {
			skippedRemote++
			continue
		}

		// Min-node safety gate: don't encode when the cluster has fewer nodes than
		// this collection's parity shards — shards could not be spread to tolerate
		// failures. Mirrors the shell ec.encode guard (node count < parity shards).
		if clusterNodeCount > 0 && clusterNodeCount < erasure_coding.ParityShardsCount {
			skippedTooFewNodes++
			continue
		}

		// Check quiet duration and fullness criteria
		if metric.Age >= quietThreshold && metric.FullnessRatio >= ecConfig.FullnessRatio {
			if ctx != nil {
				if err := ctx.Err(); err != nil {
					return results, hasMore, err
				}
			}
			glog.Infof("EC Detection: Volume %d meets all criteria, attempting to create task", metric.VolumeID)

			// Generate task ID for ActiveTopology integration
			taskID := fmt.Sprintf("ec_vol_%d_%d", metric.VolumeID, now.Unix())

			result := &types.TaskDetectionResult{
				TaskID:     taskID, // Link to ActiveTopology pending task
				TaskType:   types.TaskTypeErasureCoding,
				VolumeID:   metric.VolumeID,
				Server:     metric.Server,
				Collection: metric.Collection,
				Priority:   types.TaskPriorityLow, // EC is not urgent
				Reason: fmt.Sprintf("Volume meets EC criteria: quiet for %.1fs (>%ds), fullness=%.1f%% (>%.1f%%), size=%.1fMB (>%dMB)",
					metric.Age.Seconds(), ecConfig.QuietForSeconds, metric.FullnessRatio*100, ecConfig.FullnessRatio*100,
					float64(metric.Size)/(1024*1024), ecConfig.MinSizeMB),
				ScheduleAt: now,
			}

			// Plan EC destinations if ActiveTopology is available
			if clusterInfo.ActiveTopology != nil {
				// Check if ANY task already exists in ActiveTopology for this volume
				if clusterInfo.ActiveTopology.HasAnyTask(metric.VolumeID) {
					glog.V(2).Infof("EC Detection: Skipping volume %d, task already exists in ActiveTopology", metric.VolumeID)
					continue
				}

				glog.Infof("EC Detection: ActiveTopology available, planning destinations for volume %d", metric.VolumeID)
				if planner == nil {
					planner = newECPlacementPlanner(clusterInfo.ActiveTopology, ecConfig.PreferredTags)
				}
				dataShards := erasure_coding.DataShardsCount
				parityShards := erasure_coding.ParityShardsCount
				multiPlan, err := planECDestinations(planner, metric, ecConfig, dataShards, parityShards)
				if err != nil {
					glog.V(2).Infof("Failed to plan EC destinations for volume %d: %v", metric.VolumeID, err)
					consecutivePlanningFailures++
					if len(results) >= minProposalsBeforeEarlyStop && consecutivePlanningFailures >= maxConsecutivePlanningFailures {
						glog.Warningf("EC Detection: stopping early after %d consecutive placement failures with %d proposals already planned", consecutivePlanningFailures, len(results))
						hasMore = true
						stoppedEarly = true
						break
					}
					continue // Skip this volume if destination planning fails
				}
				consecutivePlanningFailures = 0
				glog.Infof("EC Detection: Successfully planned %d destinations for volume %d", len(multiPlan.Plans), metric.VolumeID)

				// Calculate expected shard size for EC operation
				// Each data shard will be approximately volumeSize / dataShards
				expectedShardSize := uint64(metric.Size) / uint64(dataShards)

				// Add pending EC shard task to ActiveTopology for capacity management

				// Extract shard destinations from multiPlan
				var shardDestinations []string
				var shardDiskIDs []uint32
				for _, plan := range multiPlan.Plans {
					shardDestinations = append(shardDestinations, plan.TargetNode)
					shardDiskIDs = append(shardDiskIDs, plan.TargetDisk)
				}

				// Find all volume replica locations (server + disk) from topology
				glog.Infof("EC Detection: Looking for replica locations for volume %d", metric.VolumeID)
				replicaLocations := findVolumeReplicaLocations(clusterInfo.ActiveTopology, metric.VolumeID, metric.Collection)
				if len(replicaLocations) == 0 {
					glog.Warningf("No replica locations found for volume %d, skipping EC", metric.VolumeID)
					continue
				}
				glog.Infof("EC Detection: Found %d replica locations for volume %d", len(replicaLocations), metric.VolumeID)

				// Find existing EC shards from previous failed attempts
				existingECShards := findExistingECShards(clusterInfo.ActiveTopology, metric.VolumeID, metric.Collection)

				// Combine volume replicas and existing EC shards for cleanup
				var sources []topology.TaskSourceSpec

				// Add volume replicas (will free volume slots)
				for _, replica := range replicaLocations {
					sources = append(sources, topology.TaskSourceSpec{
						ServerID:    replica.ServerID,
						DiskID:      replica.DiskID,
						DataCenter:  replica.DataCenter,
						Rack:        replica.Rack,
						CleanupType: topology.CleanupVolumeReplica,
					})
				}

				// Add existing EC shards (will free shard slots)
				duplicateCheck := make(map[string]bool)
				for _, replica := range replicaLocations {
					key := fmt.Sprintf("%s:%d", replica.ServerID, replica.DiskID)
					duplicateCheck[key] = true
				}

				for _, shard := range existingECShards {
					key := fmt.Sprintf("%s:%d", shard.ServerID, shard.DiskID)
					if !duplicateCheck[key] { // Avoid duplicates if EC shards are on same disk as volume replicas
						sources = append(sources, topology.TaskSourceSpec{
							ServerID:    shard.ServerID,
							DiskID:      shard.DiskID,
							DataCenter:  shard.DataCenter,
							Rack:        shard.Rack,
							CleanupType: topology.CleanupECShards,
							ShardIds:    append([]uint32(nil), shard.ShardIds...),
						})
						duplicateCheck[key] = true
					}
				}

				glog.V(2).Infof("Found %d volume replicas and %d existing EC shards for volume %d (total %d cleanup sources)",
					len(replicaLocations), len(existingECShards), metric.VolumeID, len(sources))

				// Convert shard destinations to TaskDestinationSpec. With fewer
				// disks than shards a destination holds several shards, so reserve
				// capacity for the actual per-disk shard count (round-robin matches
				// createECTargets) rather than assuming one shard each.
				destinations := make([]topology.TaskDestinationSpec, len(shardDestinations))
				shardsPerDest := distributeECShards(dataShards+parityShards, len(shardDestinations))
				for i, dest := range shardDestinations {
					shardCount := len(shardsPerDest[i])
					shardImpact := topology.CalculateECShardStorageImpact(int32(shardCount), int64(expectedShardSize))
					destSize := int64(expectedShardSize) * int64(shardCount)
					destinations[i] = topology.TaskDestinationSpec{
						ServerID:      dest,
						DiskID:        shardDiskIDs[i],
						StorageImpact: &shardImpact,
						EstimatedSize: &destSize,
					}
				}

				// Convert sources before mutating topology
				sourcesProto, err := convertTaskSourcesToProtobuf(sources, metric.VolumeID, clusterInfo.ActiveTopology)
				if err != nil {
					glog.Warningf("Failed to convert sources for EC task on volume %d: %v, skipping", metric.VolumeID, err)
					continue
				}

				err = clusterInfo.ActiveTopology.AddPendingTask(topology.TaskSpec{
					TaskID:       taskID,
					TaskType:     topology.TaskTypeErasureCoding,
					VolumeID:     metric.VolumeID,
					VolumeSize:   int64(metric.Size),
					Sources:      sources,
					Destinations: destinations,
				})
				if err != nil {
					glog.Warningf("Failed to add pending EC shard task to ActiveTopology for volume %d: %v", metric.VolumeID, err)
					continue // Skip this volume if topology task addition fails
				}

				if planner != nil {
					planner.applyTaskReservations(int64(metric.Size), sources, destinations)
				}

				glog.V(2).Infof("Added pending EC shard task %s to ActiveTopology for volume %d with %d cleanup sources and %d shard destinations",
					taskID, metric.VolumeID, len(sources), len(multiPlan.Plans))

				// Create unified sources and targets for EC task
				result.TypedParams = &worker_pb.TaskParams{
					TaskId:     taskID, // Link to ActiveTopology pending task
					VolumeId:   metric.VolumeID,
					Collection: metric.Collection,
					VolumeSize: metric.Size, // Store original volume size for tracking changes

					// Unified sources - all sources that will be processed/cleaned up
					Sources: sourcesProto,

					// Unified targets - all EC shard destinations
					Targets: createECTargets(multiPlan, dataShards, parityShards),

					TaskParams: &worker_pb.TaskParams_ErasureCodingParams{
						ErasureCodingParams: createECTaskParams(dataShards, parityShards, metric.DiskType),
					},
				}

				glog.V(1).Infof("Planned EC destinations for volume %d: %d shards across %d racks, %d DCs",
					metric.VolumeID, len(multiPlan.Plans), multiPlan.SuccessfulRack, multiPlan.SuccessfulDCs)
			} else {
				glog.Warningf("No ActiveTopology available for destination planning in EC detection")
				continue // Skip this volume if no topology available
			}

			glog.Infof("EC Detection: Successfully created EC task for volume %d, adding to results", metric.VolumeID)
			results = append(results, result)
		} else {
			// Count debug reasons
			if metric.Age < quietThreshold {
				skippedQuietTime++
			}
			if metric.FullnessRatio < ecConfig.FullnessRatio {
				skippedFullness++
			}

			if debugCount < 5 { // Limit to avoid spam
				// Logic moved outside
			}
			debugCount++
		}
	}

	// Log debug summary if no tasks were created
	if len(results) == 0 && len(metrics) > 0 && !stoppedEarly {
		totalVolumes := len(metrics)
		glog.V(1).Infof("EC detection: No tasks created for %d volumes (skipped: %d already EC, %d too small, %d filtered, %d not quiet, %d not full, %d remote, %d too few nodes)",
			totalVolumes, skippedAlreadyEC, skippedTooSmall, skippedCollectionFilter, skippedQuietTime, skippedFullness, skippedRemote, skippedTooFewNodes)

		// Show details for first few volumes
		for i, metric := range metrics {
			if i >= 3 || metric.IsECVolume { // Limit to first 3 non-EC volumes
				continue
			}
			sizeMB := float64(metric.Size) / (1024 * 1024)
			glog.V(1).Infof("ERASURE CODING: Volume %d: size=%.1fMB (need ≥%dMB), age=%s (need ≥%s), fullness=%.1f%% (need ≥%.1f%%)",
				metric.VolumeID, sizeMB, ecConfig.MinSizeMB, metric.Age.Truncate(time.Minute), quietThreshold.Truncate(time.Minute),
				metric.FullnessRatio*100, ecConfig.FullnessRatio*100)
		}
	}

	return results, hasMore, nil
}

type ecDiskState struct {
	baseAvailable      int64
	reservedVolumes    int32
	reservedShardSlots int32
}

type ecPlacementPlanner struct {
	activeTopology *topology.ActiveTopology
	candidates     []*placement.DiskCandidate
	candidateByKey map[string]*placement.DiskCandidate
	diskStates     map[string]*ecDiskState
	diskTags       map[string][]string
	preferredTags  []string
}

func newECPlacementPlanner(activeTopology *topology.ActiveTopology, preferredTags []string) *ecPlacementPlanner {
	if activeTopology == nil {
		return nil
	}

	disks := activeTopology.GetDisksWithEffectiveCapacity(topology.TaskTypeErasureCoding, "", 0)
	candidates := diskInfosToCandidates(disks)
	tagsByKey := collectDiskTags(disks)
	normalizedPreferredTags := util.NormalizeTagList(preferredTags)
	if len(candidates) == 0 {
		return &ecPlacementPlanner{
			activeTopology: activeTopology,
			candidates:     candidates,
			candidateByKey: map[string]*placement.DiskCandidate{},
			diskStates:     map[string]*ecDiskState{},
			diskTags:       tagsByKey,
			preferredTags:  normalizedPreferredTags,
		}
	}

	candidateByKey := make(map[string]*placement.DiskCandidate, len(candidates))
	diskStates := make(map[string]*ecDiskState, len(candidates))
	for _, candidate := range candidates {
		key := ecDiskKey(candidate.NodeID, candidate.DiskID)
		candidateByKey[key] = candidate
		diskStates[key] = &ecDiskState{
			baseAvailable: int64(candidate.FreeSlots),
		}
	}

	return &ecPlacementPlanner{
		activeTopology: activeTopology,
		candidates:     candidates,
		candidateByKey: candidateByKey,
		diskStates:     diskStates,
		diskTags:       tagsByKey,
		preferredTags:  normalizedPreferredTags,
	}
}

func (p *ecPlacementPlanner) selectDestinations(sourceRack, sourceDC, sourceDiskType string, shardsNeeded int) ([]*placement.DiskCandidate, error) {
	if p == nil || p.activeTopology == nil {
		return nil, fmt.Errorf("ec placement planner is not initialized")
	}
	if shardsNeeded <= 0 {
		return nil, fmt.Errorf("invalid shardsNeeded %d", shardsNeeded)
	}

	config := placement.PlacementRequest{
		ShardsNeeded:           shardsNeeded,
		MaxShardsPerServer:     0,
		MaxShardsPerRack:       0,
		MaxTaskLoad:            topology.MaxTaskLoadForECPlacement,
		PreferDifferentServers: true,
		PreferDifferentRacks:   true,
		// Bias placement toward disks matching the source volume's disk
		// type; placement spills to other types only if the preferred
		// pool can't satisfy ShardsNeeded (#9423).
		PreferredDiskType: sourceDiskType,
	}

	var lastErr error
	for _, candidates := range p.buildCandidateSets(shardsNeeded) {
		if len(candidates) == 0 {
			continue
		}
		result, err := placement.SelectDestinations(candidates, config)
		if err == nil {
			if result.SpilledToOtherDiskType {
				glog.Warningf("EC placement spilled to disks outside preferred disk type %q to reach %d shards (source rack=%s dc=%s)",
					sourceDiskType, shardsNeeded, sourceRack, sourceDC)
			}
			return result.SelectedDisks, nil
		}
		lastErr = err
	}
	if lastErr == nil {
		lastErr = fmt.Errorf("no EC placement candidates available")
	}
	return nil, lastErr
}

func (p *ecPlacementPlanner) applyTaskReservations(volumeSize int64, sources []topology.TaskSourceSpec, destinations []topology.TaskDestinationSpec) {
	if p == nil {
		return
	}

	touched := make(map[string]bool)

	for _, source := range sources {
		impact := p.sourceImpact(source, volumeSize)
		p.applyImpact(source.ServerID, source.DiskID, impact)
		p.bumpShardCount(source.ServerID, source.DiskID, impact.ShardSlots)
		key := ecDiskKey(source.ServerID, source.DiskID)
		if !touched[key] {
			p.bumpLoad(source.ServerID, source.DiskID)
			touched[key] = true
		}
	}

	for _, dest := range destinations {
		impact := p.destinationImpact(dest, volumeSize)
		p.applyImpact(dest.ServerID, dest.DiskID, impact)
		p.bumpShardCount(dest.ServerID, dest.DiskID, impact.ShardSlots)
		key := ecDiskKey(dest.ServerID, dest.DiskID)
		if !touched[key] {
			p.bumpLoad(dest.ServerID, dest.DiskID)
			touched[key] = true
		}
	}
}

func (p *ecPlacementPlanner) sourceImpact(source topology.TaskSourceSpec, volumeSize int64) topology.StorageSlotChange {
	if source.StorageImpact != nil {
		return *source.StorageImpact
	}
	if source.CleanupType == topology.CleanupECShards {
		return topology.CalculateECShardCleanupImpact(volumeSize)
	}
	impact, _ := topology.CalculateTaskStorageImpact(topology.TaskTypeErasureCoding, volumeSize)
	return impact
}

func (p *ecPlacementPlanner) destinationImpact(dest topology.TaskDestinationSpec, volumeSize int64) topology.StorageSlotChange {
	if dest.StorageImpact != nil {
		return *dest.StorageImpact
	}
	_, impact := topology.CalculateTaskStorageImpact(topology.TaskTypeErasureCoding, volumeSize)
	return impact
}

func (p *ecPlacementPlanner) applyImpact(nodeID string, diskID uint32, impact topology.StorageSlotChange) {
	if impact.IsZero() {
		return
	}
	key := ecDiskKey(nodeID, diskID)
	state, ok := p.diskStates[key]
	if !ok {
		return
	}

	state.reservedVolumes += impact.VolumeSlots
	state.reservedShardSlots += impact.ShardSlots

	available := state.baseAvailable - int64(state.reservedVolumes) - int64(state.reservedShardSlots)/int64(topology.ShardsPerVolumeSlot)
	if available < 0 {
		available = 0
	}

	if candidate, ok := p.candidateByKey[key]; ok {
		candidate.FreeSlots = int(available)
		candidate.VolumeCount = candidate.MaxVolumeCount - available
	}
}

func (p *ecPlacementPlanner) bumpLoad(nodeID string, diskID uint32) {
	key := ecDiskKey(nodeID, diskID)
	if candidate, ok := p.candidateByKey[key]; ok {
		candidate.LoadCount++
	}
}

func (p *ecPlacementPlanner) bumpShardCount(nodeID string, diskID uint32, delta int32) {
	if delta == 0 {
		return
	}
	key := ecDiskKey(nodeID, diskID)
	if candidate, ok := p.candidateByKey[key]; ok {
		candidate.ShardCount += int(delta)
		if candidate.ShardCount < 0 {
			candidate.ShardCount = 0
		}
	}
}

func ecDiskKey(nodeID string, diskID uint32) string {
	return fmt.Sprintf("%s:%d", nodeID, diskID)
}

func collectDiskTags(disks []*topology.DiskInfo) map[string][]string {
	tagMap := make(map[string][]string, len(disks))
	for _, disk := range disks {
		if disk == nil || disk.DiskInfo == nil {
			continue
		}
		key := ecDiskKey(disk.NodeID, disk.DiskID)
		tags := util.NormalizeTagList(disk.DiskInfo.Tags)
		if len(tags) > 0 {
			tagMap[key] = tags
		}
	}
	return tagMap
}

func diskHasTag(tags []string, tag string) bool {
	if tag == "" || len(tags) == 0 {
		return false
	}
	for _, candidate := range tags {
		if candidate == tag {
			return true
		}
	}
	return false
}

// buildCandidateSets builds tiered candidate sets for preferred-tag prioritized placement.
// For a planner with preferredTags, it accumulates disks matching each tag in order into
// progressively larger tiers. It emits a candidate set once a tier reaches shardsNeeded,
// then continues accumulating for subsequent tags. Finally, it falls back to the full
// p.candidates set if preferred-tag tiers are insufficient. This ensures tagged disks
// are selected first before falling back to all available candidates.
func (p *ecPlacementPlanner) buildCandidateSets(shardsNeeded int) [][]*placement.DiskCandidate {
	if p == nil {
		return nil
	}
	if len(p.preferredTags) == 0 {
		return [][]*placement.DiskCandidate{p.candidates}
	}
	selected := make(map[string]bool, len(p.candidates))
	var tier []*placement.DiskCandidate
	var candidateSets [][]*placement.DiskCandidate
	for _, tag := range p.preferredTags {
		for _, candidate := range p.candidates {
			key := ecDiskKey(candidate.NodeID, candidate.DiskID)
			if selected[key] {
				continue
			}
			if diskHasTag(p.diskTags[key], tag) {
				selected[key] = true
				tier = append(tier, candidate)
			}
		}
		if shardsNeeded > 0 && len(tier) >= shardsNeeded {
			candidateSets = append(candidateSets, append([]*placement.DiskCandidate(nil), tier...))
		}
	}
	// Defensive check: selectDestinations always ensures shardsNeeded > 0 before calling
	// buildCandidateSets, but this branch handles direct callers and edge cases.
	if shardsNeeded <= 0 && len(tier) > 0 {
		candidateSets = append(candidateSets, append([]*placement.DiskCandidate(nil), tier...))
	}
	if len(tier) < len(p.candidates) {
		candidateSets = append(candidateSets, p.candidates)
	} else if len(candidateSets) == 0 {
		candidateSets = append(candidateSets, p.candidates)
	}
	return candidateSets
}

// planECDestinations plans the destinations for erasure coding operation.
// dataShards/parityShards are parameters so callers can drive non-10+4 ratios.
// countTopologyNodes counts volume-server nodes in the active topology, used by
// the min-node safety gate.
func countTopologyNodes(at *topology.ActiveTopology) int {
	if at == nil {
		return 0
	}
	topo := at.GetTopologyInfo()
	if topo == nil {
		return 0
	}
	n := 0
	for _, dc := range topo.DataCenterInfos {
		for _, rack := range dc.RackInfos {
			n += len(rack.DataNodeInfos)
		}
	}
	return n
}

func planECDestinations(planner *ecPlacementPlanner, metric *types.VolumeHealthMetrics, ecConfig *Config, dataShards, parityShards int) (*topology.MultiDestinationPlan, error) {
	if planner == nil || planner.activeTopology == nil {
		return nil, fmt.Errorf("active topology not available for EC placement")
	}
	if dataShards <= 0 || parityShards <= 0 {
		return nil, fmt.Errorf("invalid EC ratio: dataShards=%d parityShards=%d", dataShards, parityShards)
	}
	totalShards := dataShards + parityShards
	// Survive losing one disk: each disk holds at most parityShards shards,
	// so we need at least ceil(totalShards / parityShards) disks.
	minTotalDisks := (totalShards + parityShards - 1) / parityShards
	expectedShardSize := uint64(metric.Size) / uint64(dataShards)

	// Get source node information from topology
	var sourceRack, sourceDC string

	// Extract rack and DC from topology info
	topologyInfo := planner.activeTopology.GetTopologyInfo()
	if topologyInfo != nil {
		for _, dc := range topologyInfo.DataCenterInfos {
			for _, rack := range dc.RackInfos {
				for _, dataNodeInfo := range rack.DataNodeInfos {
					if dataNodeInfo.Id == metric.Server {
						sourceDC = dc.Id
						sourceRack = rack.Id
						break
					}
				}
				if sourceRack != "" {
					break
				}
			}
			if sourceDC != "" {
				break
			}
		}
	}

	// Select best disks for EC placement with rack/DC diversity using the cached planner.
	// Pass source disk type so placement prefers matching-type disks (#9423).
	selectedDisks, err := planner.selectDestinations(sourceRack, sourceDC, metric.DiskType, totalShards)
	if err != nil {
		return nil, err
	}
	if len(selectedDisks) < minTotalDisks {
		return nil, fmt.Errorf("found %d disks, but EC %d+%d needs at least %d disks so no disk holds more than %d shards",
			len(selectedDisks), dataShards, parityShards, minTotalDisks, parityShards)
	}
	// Fewer than totalShards disks is fine: createECTargets round-robins the
	// shards across the available disks, packing several distinct shards onto a
	// disk when needed (matching ec.encode's "spread as 4,4,3,3" fallback for
	// small clusters). A disk holding several shards of one volume is safe —
	// each is a separate .ecNN file and ReceiveFile keys by that extension. The
	// minTotalDisks floor above keeps any single disk under parityShards shards,
	// so the volume still survives losing any one disk.
	if len(selectedDisks) < totalShards {
		glog.V(1).Infof("EC volume %d: only %d disks for %d shards, packing up to %d shards per disk",
			metric.VolumeID, len(selectedDisks), totalShards, (totalShards+len(selectedDisks)-1)/len(selectedDisks))
	}

	var plans []*topology.DestinationPlan
	rackCount := make(map[string]int)
	dcCount := make(map[string]int)

	for _, disk := range selectedDisks {
		// Get the target server address
		targetAddress, err := workerutil.ResolveServerAddress(disk.NodeID, planner.activeTopology)
		if err != nil {
			return nil, fmt.Errorf("failed to resolve address for target server %s: %v", disk.NodeID, err)
		}

		plan := &topology.DestinationPlan{
			TargetNode:     disk.NodeID,
			TargetAddress:  targetAddress,
			TargetDisk:     disk.DiskID,
			TargetRack:     disk.Rack,
			TargetDC:       disk.DataCenter,
			ExpectedSize:   expectedShardSize, // Set calculated EC shard size
			PlacementScore: calculateECScoreCandidate(disk, sourceRack, sourceDC),
		}
		plans = append(plans, plan)

		// Count rack and DC diversity
		rackKey := fmt.Sprintf("%s:%s", disk.DataCenter, disk.Rack)
		rackCount[rackKey]++
		dcCount[disk.DataCenter]++
	}

	// Log capacity utilization information using ActiveTopology's encapsulated logic
	totalEffectiveCapacity := int64(0)
	for _, plan := range plans {
		key := ecDiskKey(plan.TargetNode, plan.TargetDisk)
		if candidate, ok := planner.candidateByKey[key]; ok {
			totalEffectiveCapacity += int64(candidate.FreeSlots)
		}
	}

	glog.V(1).Infof("Planned EC destinations for volume %d (size=%d bytes): expected shard size=%d bytes, %d shards across %d racks, %d DCs, total effective capacity=%d slots",
		metric.VolumeID, metric.Size, expectedShardSize, len(plans), len(rackCount), len(dcCount), totalEffectiveCapacity)

	// Log storage impact for EC task (source only - EC has multiple targets handled individually)
	sourceChange, _ := topology.CalculateTaskStorageImpact(topology.TaskTypeErasureCoding, int64(metric.Size))
	glog.V(2).Infof("EC task capacity management: source_reserves_with_zero_impact={VolumeSlots:%d, ShardSlots:%d}, %d_targets_will_receive_shards, estimated_size=%d",
		sourceChange.VolumeSlots, sourceChange.ShardSlots, len(plans), metric.Size)
	glog.V(2).Infof("EC source reserves capacity but with zero StorageSlotChange impact")

	return &topology.MultiDestinationPlan{
		Plans:          plans,
		TotalShards:    len(plans),
		SuccessfulRack: len(rackCount),
		SuccessfulDCs:  len(dcCount),
	}, nil
}

// distributeECShards assigns shard ids 0..totalShards-1 across numTargets
// targets round-robin, so each target holds either floor or ceil of
// totalShards/numTargets shards. When numTargets < totalShards this packs
// several shards onto a target; planECDestinations guarantees numTargets is at
// least ceil(totalShards/parityShards), so no target exceeds parityShards shards.
func distributeECShards(totalShards, numTargets int) [][]uint32 {
	targetShards := make([][]uint32, numTargets)
	for i := range targetShards {
		targetShards[i] = make([]uint32, 0)
	}
	for shardId := 0; shardId < totalShards; shardId++ {
		targetIndex := shardId % numTargets
		targetShards[targetIndex] = append(targetShards[targetIndex], uint32(shardId))
	}
	return targetShards
}

// createECTargets builds TaskTargets, round-robining shards across the plan
// entries. With fewer disks than shards a target receives several shard ids.
func createECTargets(multiPlan *topology.MultiDestinationPlan, dataShards, parityShards int) []*worker_pb.TaskTarget {
	var targets []*worker_pb.TaskTarget
	numTargets := len(multiPlan.Plans)
	totalShards := dataShards + parityShards

	targetShards := distributeECShards(totalShards, numTargets)

	for i, plan := range multiPlan.Plans {
		target := &worker_pb.TaskTarget{
			Node:          plan.TargetAddress,
			DiskId:        plan.TargetDisk,
			Rack:          plan.TargetRack,
			DataCenter:    plan.TargetDC,
			ShardIds:      targetShards[i],
			EstimatedSize: plan.ExpectedSize,
		}
		targets = append(targets, target)

		assignedData := make([]uint32, 0)
		assignedParity := make([]uint32, 0)
		for _, shardId := range targetShards[i] {
			if int(shardId) < dataShards {
				assignedData = append(assignedData, shardId)
			} else {
				assignedParity = append(assignedParity, shardId)
			}
		}
		glog.V(2).Infof("EC planning: target %s assigned shards %v (data: %v, parity: %v)",
			plan.TargetNode, targetShards[i], assignedData, assignedParity)
	}

	glog.V(1).Infof("EC planning: distributed %d shards across %d targets using round-robin (data shards 0-%d, parity shards %d-%d)",
		totalShards, numTargets, dataShards-1, dataShards, totalShards-1)
	return targets
}

// convertTaskSourcesToProtobuf converts topology.TaskSourceSpec to worker_pb.TaskSource
func convertTaskSourcesToProtobuf(sources []topology.TaskSourceSpec, volumeID uint32, activeTopology *topology.ActiveTopology) ([]*worker_pb.TaskSource, error) {
	var protobufSources []*worker_pb.TaskSource

	for _, source := range sources {
		serverAddress, err := workerutil.ResolveServerAddress(source.ServerID, activeTopology)
		if err != nil {
			return nil, fmt.Errorf("failed to resolve address for source server %s: %v", source.ServerID, err)
		}

		pbSource := &worker_pb.TaskSource{
			Node:       serverAddress,
			DiskId:     source.DiskID,
			DataCenter: source.DataCenter,
			Rack:       source.Rack,
		}

		// Convert storage impact to estimated size
		if source.EstimatedSize != nil {
			pbSource.EstimatedSize = uint64(*source.EstimatedSize)
		}

		// Populated ShardIds is the wire-level marker that flags an
		// EC-shard cleanup source; the worker routes it through
		// cleanupStaleEcShards and skips it in getReplicas.
		switch source.CleanupType {
		case topology.CleanupVolumeReplica:
			pbSource.VolumeId = volumeID
		case topology.CleanupECShards:
			pbSource.VolumeId = volumeID
			pbSource.ShardIds = append([]uint32(nil), source.ShardIds...)
		}

		protobufSources = append(protobufSources, pbSource)
	}

	return protobufSources, nil
}

// createECTaskParams creates clean EC task parameters (destinations now in unified targets).
// sourceDiskType is forwarded to VolumeEcShardsMount so the resulting EC volume
// reports under the source's disk type rather than the target location's (#9423).
func createECTaskParams(dataShards, parityShards int, sourceDiskType string) *worker_pb.ErasureCodingTaskParams {
	return &worker_pb.ErasureCodingTaskParams{
		DataShards:     int32(dataShards),
		ParityShards:   int32(parityShards),
		SourceDiskType: sourceDiskType,
	}
}

// diskInfosToCandidates converts topology.DiskInfo slice to placement.DiskCandidate slice
func diskInfosToCandidates(disks []*topology.DiskInfo) []*placement.DiskCandidate {
	var candidates []*placement.DiskCandidate
	for _, disk := range disks {
		if disk.DiskInfo == nil {
			continue
		}

		// Calculate free slots (using default max if not set)
		freeSlots := int(disk.DiskInfo.MaxVolumeCount - disk.DiskInfo.VolumeCount)
		if freeSlots < 0 {
			freeSlots = 0
		}

		// Calculate EC shard count for this specific disk
		// EcShardInfos contains all shards, so we need to filter by DiskId and sum actual shard counts
		ecShardCount := 0
		if disk.DiskInfo.EcShardInfos != nil {
			for _, shardInfo := range disk.DiskInfo.EcShardInfos {
				if shardInfo.DiskId == disk.DiskID {
					ecShardCount += erasure_coding.GetShardCount(shardInfo)
				}
			}
		}

		candidates = append(candidates, &placement.DiskCandidate{
			NodeID:         disk.NodeID,
			DiskID:         disk.DiskID,
			DataCenter:     disk.DataCenter,
			Rack:           disk.Rack,
			DiskType:       disk.DiskType,
			VolumeCount:    disk.DiskInfo.VolumeCount,
			MaxVolumeCount: disk.DiskInfo.MaxVolumeCount,
			ShardCount:     ecShardCount,
			FreeSlots:      freeSlots,
			LoadCount:      disk.LoadCount,
		})
	}
	return candidates
}

// calculateECScoreCandidate calculates placement score for EC operations.
// Used for logging and plan metadata.
func calculateECScoreCandidate(disk *placement.DiskCandidate, sourceRack, sourceDC string) float64 {
	if disk == nil {
		return 0.0
	}

	score := 0.0

	// Prefer disks with available capacity (primary factor)
	if disk.MaxVolumeCount > 0 {
		utilization := float64(disk.VolumeCount) / float64(disk.MaxVolumeCount)
		score += (1.0 - utilization) * 60.0 // Up to 60 points for available capacity
	}

	// Consider current load (secondary factor)
	score += (10.0 - float64(disk.LoadCount)) // Up to 10 points for low load

	return score
}

// findVolumeReplicaLocations finds all replica locations (server + disk) for the specified volume
// Uses O(1) indexed lookup for optimal performance on large clusters.
func findVolumeReplicaLocations(activeTopology *topology.ActiveTopology, volumeID uint32, collection string) []topology.VolumeReplica {
	if activeTopology == nil {
		return nil
	}
	return activeTopology.GetVolumeLocations(volumeID, collection)
}

// findExistingECShards finds existing EC shards for a volume (from previous failed EC attempts)
// Uses O(1) indexed lookup for optimal performance on large clusters.
func findExistingECShards(activeTopology *topology.ActiveTopology, volumeID uint32, collection string) []topology.VolumeReplica {
	if activeTopology == nil {
		return nil
	}
	return activeTopology.GetECShardLocations(volumeID, collection)
}

// cleanupOrphanSourceReplicas deletes any regular volume replicas still
// present in the topology for (volumeID, collection) after re-verifying that
// the full EC shard set is intact. Caller must hold expectedShards equal to
// the configured totalShards count. Issues VolumeDelete RPC to each replica
// server's address — that endpoint only touches the regular volume on the
// targeted server, never EC shards (those live in a separate store path).
// The cluster-wide `volume.delete` shell command is what would have nuked
// the EC shards too; the targeted RPC used here is safe by construction.
// Returns the count of replicas successfully deleted plus any error.
func cleanupOrphanSourceReplicas(ctx context.Context, clusterInfo *types.ClusterInfo, metric *types.VolumeHealthMetrics, expectedShards int) (int, error) {
	if clusterInfo == nil || clusterInfo.ActiveTopology == nil {
		return 0, fmt.Errorf("active topology unavailable")
	}
	if clusterInfo.GrpcDialOption == nil {
		return 0, fmt.Errorf("grpc dial option unavailable")
	}

	// Re-verify shard completeness right before acting. Defensive: detection
	// processes many volumes sequentially and the topology snapshot we built
	// at start-of-detection could have lost shards in between (a volume
	// server going down between iterations). Refusing to delete the source
	// when we can no longer prove the shards are complete is the safer
	// failure mode — the source replica is the only complete copy.
	actualShards := countExistingEcShardsForVolume(clusterInfo.ActiveTopology, metric.VolumeID, metric.Collection)
	if actualShards < expectedShards {
		return 0, fmt.Errorf("EC shard set shrank between detection and cleanup (%d < %d); refusing to delete source replica", actualShards, expectedShards)
	}

	replicas := findVolumeReplicaLocations(clusterInfo.ActiveTopology, metric.VolumeID, metric.Collection)
	if len(replicas) == 0 {
		return 0, nil
	}

	deleted := 0
	var deleteErrors []string
	for _, replica := range replicas {
		serverAddress := replica.ServerID
		err := operation.WithVolumeServerClient(false, pb.ServerAddress(serverAddress), clusterInfo.GrpcDialOption,
			func(client volume_server_pb.VolumeServerClient) error {
				_, deleteErr := client.VolumeDelete(ctx, &volume_server_pb.VolumeDeleteRequest{
					VolumeId:  metric.VolumeID,
					OnlyEmpty: false,
				})
				return deleteErr
			})
		if err != nil {
			deleteErrors = append(deleteErrors, fmt.Sprintf("server %s: %v", serverAddress, err))
			continue
		}
		deleted++
		glog.V(1).Infof("EC Detection: deleted orphan regular replica for volume %d on %s", metric.VolumeID, serverAddress)
	}

	if len(deleteErrors) > 0 {
		return deleted, fmt.Errorf("%d of %d replica delete(s) failed: %s", len(deleteErrors), len(replicas), strings.Join(deleteErrors, "; "))
	}
	return deleted, nil
}

// countExistingEcShardsForVolume returns the number of distinct EC shard IDs
// for (volumeID, collection) present in the topology. Walks every disk's
// EcIndexBits bitmap rather than trusting len(EcShardInfos), because a single
// info entry can carry multiple shards. Used by the #9448 guard to decide
// whether the EC shard set is complete enough that the orphaned regular
// replica is safe to delete.
func countExistingEcShardsForVolume(activeTopology *topology.ActiveTopology, volumeID uint32, collection string) int {
	if activeTopology == nil {
		return 0
	}
	topologyInfo := activeTopology.GetTopologyInfo()
	if topologyInfo == nil {
		return 0
	}
	var seen erasure_coding.ShardBits
	for _, dc := range topologyInfo.DataCenterInfos {
		for _, rack := range dc.RackInfos {
			for _, node := range rack.DataNodeInfos {
				for _, diskInfo := range node.DiskInfos {
					for _, ecShardInfo := range diskInfo.EcShardInfos {
						if ecShardInfo.Id != volumeID || ecShardInfo.Collection != collection {
							continue
						}
						seen |= erasure_coding.ShardBits(ecShardInfo.EcIndexBits)
					}
				}
			}
		}
	}
	return seen.Count()
}