tasks/service/worker/migration/activities.go

// The MIT License
//
// Copyright (c) 2020 Temporal Technologies Inc.  All rights reserved.
//
// Copyright (c) 2020 Uber Technologies, Inc.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.

package migration

import (
	"context"
	"fmt"
	"math"
	"sort"
	"time"

	"github.com/pkg/errors"
	"google.golang.org/protobuf/types/known/timestamppb"

	commonpb "go.temporal.io/api/common/v1"
	replicationpb "go.temporal.io/api/replication/v1"
	"go.temporal.io/api/serviceerror"
	"go.temporal.io/api/workflowservice/v1"
	"go.temporal.io/sdk/activity"
	"go.temporal.io/sdk/temporal"

	"go.temporal.io/server/api/adminservice/v1"
	enumsspb "go.temporal.io/server/api/enums/v1"
	"go.temporal.io/server/api/historyservice/v1"
	replicationspb "go.temporal.io/server/api/replication/v1"
	serverClient "go.temporal.io/server/client"
	"go.temporal.io/server/client/admin"
	"go.temporal.io/server/common/definition"
	"go.temporal.io/server/common/headers"
	"go.temporal.io/server/common/log"
	"go.temporal.io/server/common/log/tag"
	"go.temporal.io/server/common/metrics"
	"go.temporal.io/server/common/namespace"
	"go.temporal.io/server/common/persistence"
	"go.temporal.io/server/common/quotas"
	"go.temporal.io/server/common/searchattribute"
)

type (
	activities struct {
		historyShardCount              int32
		executionManager               persistence.ExecutionManager
		taskManager                    persistence.TaskManager
		namespaceRegistry              namespace.Registry
		historyClient                  historyservice.HistoryServiceClient
		frontendClient                 workflowservice.WorkflowServiceClient
		clientFactory                  serverClient.Factory
		clientBean                     serverClient.Bean
		logger                         log.Logger
		metricsHandler                 metrics.Handler
		forceReplicationMetricsHandler metrics.Handler
		namespaceReplicationQueue      persistence.NamespaceReplicationQueue
	}

	SkippedWorkflowExecution struct {
		WorkflowExecution *commonpb.WorkflowExecution
		Reason            string
	}

	replicationTasksHeartbeatDetails struct {
		NextIndex                        int
		CheckPoint                       time.Time
		LastNotVerifiedWorkflowExecution *commonpb.WorkflowExecution
	}

	verifyStatus int
	verifyResult struct {
		status verifyStatus
		reason string
	}
)

const (
	reasonZombieWorkflow           = "Zombie workflow"
	reasonWorkflowNotFound         = "Workflow not found"
	reasonWorkflowCloseToRetention = "Workflow close to retention"

	notVerified verifyStatus = 0
	verified    verifyStatus = 1
	skipped     verifyStatus = 2
)

func (r verifyResult) isVerified() bool {
	return r.status == verified || r.status == skipped
}

// TODO: CallerTypePreemptablee should be set in activity background context for all migration activities.
// However, activity background context is per-worker, which means once set, all activities processed by the
// worker will use CallerTypePreemptable, including those not related to migration. This is not ideal.
// Using a different task queue and a dedicated worker for migration can solve the issue but requires
// changing all existing tooling around namespace migration to start workflows & activities on the new task queue.
// Another approach is to use separate workers for workflow tasks and activities and keep existing tooling unchanged.

// GetMetadata returns history shard count and namespaceID for requested namespace.
func (a *activities) GetMetadata(_ context.Context, request metadataRequest) (*metadataResponse, error) {
	nsEntry, err := a.namespaceRegistry.GetNamespace(namespace.Name(request.Namespace))
	if err != nil {
		return nil, err
	}

	return &metadataResponse{
		ShardCount:  a.historyShardCount,
		NamespaceID: string(nsEntry.ID()),
	}, nil
}

// GetMaxReplicationTaskIDs returns max replication task id per shard
func (a *activities) GetMaxReplicationTaskIDs(ctx context.Context) (*replicationStatus, error) {
	ctx = headers.SetCallerInfo(ctx, headers.SystemPreemptableCallerInfo)

	resp, err := a.historyClient.GetReplicationStatus(ctx, &historyservice.GetReplicationStatusRequest{})
	if err != nil {
		return nil, err
	}
	result := &replicationStatus{MaxReplicationTaskIds: make(map[int32]int64)}
	for _, shard := range resp.Shards {
		result.MaxReplicationTaskIds[shard.ShardId] = shard.MaxReplicationTaskId
	}
	return result, nil
}

func (a *activities) WaitReplication(ctx context.Context, waitRequest waitReplicationRequest) error {
	ctx = headers.SetCallerInfo(ctx, headers.SystemPreemptableCallerInfo)

	for {
		done, err := a.checkReplicationOnce(ctx, waitRequest)
		if err != nil {
			return err
		}
		if done {
			return nil
		}
		// keep waiting and check again
		time.Sleep(time.Second)
		activity.RecordHeartbeat(ctx, nil)
	}
}

// Check if remote cluster has caught up on all shards on replication tasks
func (a *activities) checkReplicationOnce(ctx context.Context, waitRequest waitReplicationRequest) (bool, error) {

	resp, err := a.historyClient.GetReplicationStatus(ctx, &historyservice.GetReplicationStatusRequest{
		RemoteClusters: []string{waitRequest.RemoteCluster},
	})
	if err != nil {
		return false, err
	}
	if int(waitRequest.ShardCount) != len(resp.Shards) {
		return false, fmt.Errorf("GetReplicationStatus returns %d shards, expecting %d", len(resp.Shards), waitRequest.ShardCount)
	}

	// check that every shard has caught up
	readyShardCount := 0
	logged := false

	sort.SliceStable(resp.Shards, func(i, j int) bool {
		return resp.Shards[i].ShardId < resp.Shards[j].ShardId
	})

	for _, shard := range resp.Shards {
		clusterInfo, hasClusterInfo := shard.RemoteClusters[waitRequest.RemoteCluster]
		actualLag := shard.MaxReplicationTaskVisibilityTime.AsTime().Sub(clusterInfo.AckedTaskVisibilityTime.AsTime())
		if hasClusterInfo {
			// WE are all caught up
			if shard.MaxReplicationTaskId == clusterInfo.AckedTaskId {
				readyShardCount++
				continue
			}

			// Caught up to the last checked IDs, and within allowed lagging range
			if clusterInfo.AckedTaskId >= waitRequest.WaitForTaskIds[shard.ShardId] &&
				(shard.MaxReplicationTaskId-clusterInfo.AckedTaskId <= waitRequest.AllowedLaggingTasks ||
					actualLag <= waitRequest.AllowedLagging) {
				readyShardCount++
				continue
			}
		}

		// shard is not ready, log first non-ready shard
		if !logged {
			logged = true
			if !hasClusterInfo {
				a.logger.Info("Wait catchup missing remote cluster info", tag.ShardID(shard.ShardId), tag.ClusterName(waitRequest.RemoteCluster))
				// this is not expected, so fail activity to surface the error, but retryPolicy will keep retrying.
				return false, fmt.Errorf("GetReplicationStatus response for shard %d does not contains remote cluster %s", shard.ShardId, waitRequest.RemoteCluster)
			}
			a.logger.Info("Wait catchup not ready",
				tag.NewInt32("ShardId", shard.ShardId),
				tag.NewStringTag("RemoteCluster", waitRequest.RemoteCluster),
				tag.NewInt64("AckedTaskId", clusterInfo.AckedTaskId),
				tag.NewInt64("WaitForTaskId", waitRequest.WaitForTaskIds[shard.ShardId]),
				tag.NewDurationTag("AllowedLagging", waitRequest.AllowedLagging),
				tag.NewDurationTag("ActualLagging", actualLag),
				tag.NewInt64("MaxReplicationTaskId", shard.MaxReplicationTaskId),
				tag.NewTimePtrTag("MaxReplicationTaskVisibilityTime", shard.MaxReplicationTaskVisibilityTime),
				tag.NewTimePtrTag("AckedTaskVisibilityTime", clusterInfo.AckedTaskVisibilityTime),
				tag.NewInt64("AllowedLaggingTasks", waitRequest.AllowedLaggingTasks),
				tag.NewInt64("ActualLaggingTasks", shard.MaxReplicationTaskId-clusterInfo.AckedTaskId),
			)
		}
	}

	// emit metrics about how many shards are ready
	metrics.CatchUpReadyShardCountGauge.With(a.metricsHandler).Record(
		float64(readyShardCount),
		metrics.OperationTag(metrics.MigrationWorkflowScope),
		metrics.TargetClusterTag(waitRequest.RemoteCluster))

	return readyShardCount == len(resp.Shards), nil
}

func (a *activities) WaitHandover(ctx context.Context, waitRequest waitHandoverRequest) error {
	// Use the highest priority caller type for checking handover state
	// since during handover state namespace has no availability
	ctx = headers.SetCallerInfo(ctx, headers.NewCallerInfo(waitRequest.Namespace, headers.CallerTypeAPI, ""))

	for {
		done, err := a.checkHandoverOnce(ctx, waitRequest)
		if err != nil {
			return err
		}
		if done {
			return nil
		}
		// keep waiting and check again
		time.Sleep(time.Second)
		activity.RecordHeartbeat(ctx, nil)
	}
}

// Check if remote cluster has caught up on all shards on replication tasks
func (a *activities) checkHandoverOnce(ctx context.Context, waitRequest waitHandoverRequest) (bool, error) {

	resp, err := a.historyClient.GetReplicationStatus(ctx, &historyservice.GetReplicationStatusRequest{
		RemoteClusters: []string{waitRequest.RemoteCluster},
	})
	if err != nil {
		return false, err
	}
	if int(waitRequest.ShardCount) != len(resp.Shards) {
		return false, fmt.Errorf("GetReplicationStatus returns %d shards, expecting %d", len(resp.Shards), waitRequest.ShardCount)
	}

	readyShardCount := 0
	logged := false
	// check that every shard is ready to handover
	for _, shard := range resp.Shards {
		clusterInfo, hasClusterInfo := shard.RemoteClusters[waitRequest.RemoteCluster]
		handoverInfo, hasHandoverInfo := shard.HandoverNamespaces[waitRequest.Namespace]
		if hasClusterInfo && hasHandoverInfo {
			if clusterInfo.AckedTaskId >= handoverInfo.HandoverReplicationTaskId {
				readyShardCount++
				continue
			}
		}
		// shard is not ready, log first non-ready shard
		if !logged {
			logged = true
			if !hasClusterInfo {
				a.logger.Info("Wait handover missing remote cluster info", tag.ShardID(shard.ShardId), tag.ClusterName(waitRequest.RemoteCluster))
				// this is not expected, so fail activity to surface the error, but retryPolicy will keep retrying.
				return false, fmt.Errorf("GetReplicationStatus response for shard %d does not contains remote cluster %s", shard.ShardId, waitRequest.RemoteCluster)
			}

			if !hasHandoverInfo {
				// this could happen before namespace cache refresh
				a.logger.Info("Wait handover missing handover namespace info", tag.ShardID(shard.ShardId), tag.ClusterName(waitRequest.RemoteCluster), tag.WorkflowNamespace(waitRequest.Namespace))
			} else {
				a.logger.Info("Wait handover not ready",
					tag.NewInt32("ShardId", shard.ShardId),
					tag.NewInt64("AckedTaskId", clusterInfo.AckedTaskId),
					tag.NewInt64("HandoverTaskId", handoverInfo.HandoverReplicationTaskId),
					tag.NewStringTag("Namespace", waitRequest.Namespace),
					tag.NewStringTag("RemoteCluster", waitRequest.RemoteCluster),
					tag.NewInt64("MaxReplicationTaskId", shard.MaxReplicationTaskId),
				)
			}
		}
	}

	// emit metrics about how many shards are ready
	metrics.HandoverReadyShardCountGauge.With(a.metricsHandler).Record(
		float64(readyShardCount),
		metrics.OperationTag(metrics.MigrationWorkflowScope),
		metrics.TargetClusterTag(waitRequest.RemoteCluster),
		metrics.NamespaceTag(waitRequest.Namespace))
	a.logger.Info("Wait handover ready shard count.",
		tag.NewInt("ReadyShards", readyShardCount),
		tag.NewStringTag("Namespace", waitRequest.Namespace),
		tag.NewStringTag("RemoteCluster", waitRequest.RemoteCluster))

	return readyShardCount == len(resp.Shards), nil
}

func (a *activities) generateWorkflowReplicationTask(ctx context.Context, rateLimiter quotas.RateLimiter, wKey definition.WorkflowKey) error {
	if err := rateLimiter.WaitN(ctx, 1); err != nil {
		return err
	}

	// will generate replication task
	ctx, cancel := context.WithTimeout(ctx, time.Second*10)
	defer cancel()

	resp, err := a.historyClient.GenerateLastHistoryReplicationTasks(ctx, &historyservice.GenerateLastHistoryReplicationTasksRequest{
		NamespaceId: wKey.NamespaceID,
		Execution: &commonpb.WorkflowExecution{
			WorkflowId: wKey.WorkflowID,
			RunId:      wKey.RunID,
		},
	})

	if err != nil {
		return err
	}

	// If workflow has many activity retries (bug in activity code e.g.,), the state transition count can be
	// large but the number of actual state transition that is applied on target cluster can be very small.
	// Take the minimum between StateTransitionCount and HistoryLength as heuristic to avoid unnecessary throttling
	// in such situation.
	count := min(resp.StateTransitionCount, resp.HistoryLength)
	for count > 0 {
		token := min(int(count), rateLimiter.Burst())
		count -= int64(token)
		_ = rateLimiter.ReserveN(time.Now(), token)
	}

	return nil
}

func (a *activities) UpdateNamespaceState(ctx context.Context, req updateStateRequest) error {
	// Use the highest priority caller type for updating namespace config
	// since during handover state, namespace has no availability
	ctx = headers.SetCallerInfo(ctx, headers.NewCallerInfo(req.Namespace, headers.CallerTypeAPI, ""))

	descResp, err := a.frontendClient.DescribeNamespace(ctx, &workflowservice.DescribeNamespaceRequest{
		Namespace: req.Namespace,
	})
	if err != nil {
		return err
	}
	if descResp.ReplicationConfig.State == req.NewState {
		return nil
	}

	_, err = a.frontendClient.UpdateNamespace(ctx, &workflowservice.UpdateNamespaceRequest{
		Namespace: req.Namespace,
		ReplicationConfig: &replicationpb.NamespaceReplicationConfig{
			State: req.NewState,
		},
	})

	return err
}

func (a *activities) UpdateActiveCluster(ctx context.Context, req updateActiveClusterRequest) error {
	// Use the highest priority caller type for updating namespace config
	// since when both clusters think namespace are standby, namespace has no availability
	ctx = headers.SetCallerInfo(ctx, headers.NewCallerInfo(req.Namespace, headers.CallerTypeAPI, ""))

	descResp, err := a.frontendClient.DescribeNamespace(ctx, &workflowservice.DescribeNamespaceRequest{
		Namespace: req.Namespace,
	})
	if err != nil {
		return err
	}
	if descResp.ReplicationConfig.GetActiveClusterName() == req.ActiveCluster {
		return nil
	}

	_, err = a.frontendClient.UpdateNamespace(ctx, &workflowservice.UpdateNamespaceRequest{
		Namespace: req.Namespace,
		ReplicationConfig: &replicationpb.NamespaceReplicationConfig{
			ActiveClusterName: req.ActiveCluster,
		},
	})

	return err
}

func (a *activities) ListWorkflows(ctx context.Context, request *workflowservice.ListWorkflowExecutionsRequest) (*listWorkflowsResponse, error) {
	ctx = headers.SetCallerInfo(ctx, headers.NewCallerInfo(request.Namespace, headers.CallerTypePreemptable, ""))

	// modify query to include all namespace divisions
	request.Query = searchattribute.QueryWithAnyNamespaceDivision(request.Query)

	resp, err := a.frontendClient.ListWorkflowExecutions(ctx, request)
	if err != nil {
		return nil, err
	}
	var lastCloseTime, lastStartTime *timestamppb.Timestamp

	executions := make([]*commonpb.WorkflowExecution, len(resp.Executions))
	for i, e := range resp.Executions {
		executions[i] = e.Execution

		if e.CloseTime != nil {
			lastCloseTime = e.CloseTime
		}

		if e.StartTime != nil {
			lastStartTime = e.StartTime
		}
	}
	return &listWorkflowsResponse{Executions: executions, NextPageToken: resp.NextPageToken, LastCloseTime: lastCloseTime.AsTime(), LastStartTime: lastStartTime.AsTime()}, nil
}

func (a *activities) GenerateReplicationTasks(ctx context.Context, request *generateReplicationTasksRequest) error {
	ctx = a.setCallerInfoForGenReplicationTask(ctx, namespace.ID(request.NamespaceID))
	rateLimiter := quotas.NewRateLimiter(request.RPS, int(math.Ceil(request.RPS)))

	start := time.Now()
	defer func() {
		metrics.GenerateReplicationTasksLatency.With(a.forceReplicationMetricsHandler).Record(time.Since(start))
	}()

	startIndex := 0
	if activity.HasHeartbeatDetails(ctx) {
		var finishedIndex int
		if err := activity.GetHeartbeatDetails(ctx, &finishedIndex); err == nil {
			startIndex = finishedIndex + 1 // start from next one
		}
	}

	for i := startIndex; i < len(request.Executions); i++ {
		we := request.Executions[i]
		if err := a.generateWorkflowReplicationTask(ctx, rateLimiter, definition.NewWorkflowKey(request.NamespaceID, we.WorkflowId, we.RunId)); err != nil {
			if !isNotFoundServiceError(err) {
				a.logger.Error("force-replication failed to generate replication task", tag.WorkflowNamespaceID(request.NamespaceID), tag.WorkflowID(we.WorkflowId), tag.WorkflowRunID(we.RunId), tag.Error(err))
				return err
			}
		}

		activity.RecordHeartbeat(ctx, i)
	}

	return nil
}

func (a *activities) setCallerInfoForGenReplicationTask(
	ctx context.Context,
	namespaceID namespace.ID,
) context.Context {
	nsName, err := a.namespaceRegistry.GetNamespaceName(namespaceID)
	if err != nil {
		a.logger.Error("Failed to get namespace name when generating replication task",
			tag.WorkflowNamespaceID(namespaceID.String()),
			tag.Error(err),
		)
		nsName = namespace.EmptyName
	}
	return headers.SetCallerInfo(ctx, headers.NewPreemptableCallerInfo(nsName.String()))
}

type seedReplicationQueueWithUserDataEntriesHeartbeatDetails struct {
	NextPageToken []byte
	IndexInPage   int
}

func (a *activities) SeedReplicationQueueWithUserDataEntries(ctx context.Context, params TaskQueueUserDataReplicationParamsWithNamespace) error {
	if len(params.Namespace) == 0 {
		return temporal.NewNonRetryableApplicationError("namespace is required", "InvalidArgument", nil)
	}
	if params.PageSize == 0 {
		params.PageSize = defaultPageSizeForTaskQueueUserDataReplication
	}
	if params.RPS == 0 {
		params.RPS = defaultRPSForTaskQueueUserDataReplication
	}

	describeResponse, err := a.frontendClient.DescribeNamespace(ctx, &workflowservice.DescribeNamespaceRequest{
		Namespace: params.Namespace,
	})
	if err != nil {
		return err
	}

	rateLimiter := quotas.NewRateLimiter(params.RPS, int(math.Ceil(params.RPS)))
	heartbeatDetails := seedReplicationQueueWithUserDataEntriesHeartbeatDetails{}

	if activity.HasHeartbeatDetails(ctx) {
		if err := activity.GetHeartbeatDetails(ctx, &heartbeatDetails); err != nil {
			return temporal.NewNonRetryableApplicationError("failed to load previous heartbeat details", "TypeError", err)
		}
	}

	for {
		if err := rateLimiter.Wait(ctx); err != nil {
			return err
		}

		request := &persistence.ListTaskQueueUserDataEntriesRequest{
			NamespaceID:   describeResponse.GetNamespaceInfo().Id,
			NextPageToken: heartbeatDetails.NextPageToken,
			PageSize:      params.PageSize,
		}
		response, err := a.taskManager.ListTaskQueueUserDataEntries(ctx, request)
		if err != nil {
			a.logger.Error("List task queue user data failed", tag.WorkflowNamespaceID(request.NamespaceID), tag.Error(err))
			return err
		}
		for idx, entry := range response.Entries {
			if heartbeatDetails.IndexInPage > idx {
				continue
			}
			heartbeatDetails.IndexInPage = idx
			activity.RecordHeartbeat(ctx, heartbeatDetails)
			err = a.namespaceReplicationQueue.Publish(ctx, &replicationspb.ReplicationTask{
				TaskType: enumsspb.REPLICATION_TASK_TYPE_TASK_QUEUE_USER_DATA,
				Attributes: &replicationspb.ReplicationTask_TaskQueueUserDataAttributes{
					TaskQueueUserDataAttributes: &replicationspb.TaskQueueUserDataAttributes{
						NamespaceId:   request.NamespaceID,
						TaskQueueName: entry.TaskQueue,
						UserData:      entry.UserData.GetData(),
					},
				},
			})
			if err != nil {
				a.logger.Error("Inserting into namespace replication queue failed", tag.WorkflowNamespaceID(request.NamespaceID), tag.Error(err))
				return err
			}
		}
		if len(response.NextPageToken) == 0 {
			return nil
		}
		heartbeatDetails.NextPageToken = response.NextPageToken
		heartbeatDetails.IndexInPage = 0
		activity.RecordHeartbeat(ctx, heartbeatDetails)
	}
}

func isNotFoundServiceError(err error) bool {
	_, ok := err.(*serviceerror.NotFound)
	return ok
}

func isCloseToCurrentTime(t time.Time, duration time.Duration) bool {
	currentTime := time.Now()
	diff := currentTime.Sub(t)

	// check both before and after current time in case:
	//   - workflow deletion time has passed (slow delete)
	//   - workflow is abort to be deleted (target may run with a faster clock)
	if diff < -duration || diff > duration {
		return false
	}

	return true
}

func (a *activities) checkSkipWorkflowExecution(
	ctx context.Context,
	request *verifyReplicationTasksRequest,
	we *commonpb.WorkflowExecution,
	ns *namespace.Namespace,
) (verifyResult, error) {
	namespaceID := request.NamespaceID
	tags := []tag.Tag{tag.WorkflowNamespaceID(namespaceID), tag.WorkflowID(we.WorkflowId), tag.WorkflowRunID(we.RunId)}
	resp, err := a.historyClient.DescribeMutableState(ctx, &historyservice.DescribeMutableStateRequest{
		NamespaceId: namespaceID,
		Execution:   we,
	})

	if err != nil {
		if isNotFoundServiceError(err) {
			// The outstanding workflow execution may be deleted (due to retention) on source cluster after replication tasks were generated.
			// Since retention runs on both source/target clusters, such execution may also be deleted (hence not found) from target cluster.
			metrics.EncounterNotFoundWorkflowCount.With(a.forceReplicationMetricsHandler).Record(1)
			return verifyResult{
				status: skipped,
				reason: reasonWorkflowNotFound,
			}, nil
		}

		return verifyResult{
			status: notVerified,
		}, err
	}

	// Zombie workflow should be a transient state. However, if there is Zombie workflow on the source cluster,
	// it is skipped to avoid such workflow being processed on the target cluster.
	if resp.GetDatabaseMutableState().GetExecutionState().GetState() == enumsspb.WORKFLOW_EXECUTION_STATE_ZOMBIE {
		metrics.EncounterZombieWorkflowCount.With(a.forceReplicationMetricsHandler).Record(1)
		a.logger.Info("createReplicationTasks skip Zombie workflow", tags...)
		return verifyResult{
			status: skipped,
			reason: reasonZombieWorkflow,
		}, nil
	}

	// Skip verifying workflow which has already passed retention time.
	if closeTime := resp.GetDatabaseMutableState().GetExecutionInfo().GetCloseTime(); closeTime != nil && ns != nil && ns.Retention() > 0 {
		deleteTime := closeTime.AsTime().Add(ns.Retention())
		if deleteTime.Before(time.Now()) {
			metrics.EncounterPassRetentionWorkflowCount.With(a.forceReplicationMetricsHandler).Record(1)
			return verifyResult{
				status: skipped,
				reason: reasonWorkflowCloseToRetention,
			}, nil
		}
	}

	return verifyResult{
		status: notVerified,
	}, nil
}

func (a *activities) verifySingleReplicationTask(
	ctx context.Context,
	request *verifyReplicationTasksRequest,
	remoteClient adminservice.AdminServiceClient,
	ns *namespace.Namespace,
	we *commonpb.WorkflowExecution,
) (result verifyResult, rerr error) {
	s := time.Now()
	// Check if execution exists on remote cluster
	_, err := remoteClient.DescribeMutableState(ctx, &adminservice.DescribeMutableStateRequest{
		Namespace: request.Namespace,
		Execution: we,
	})
	metrics.VerifyDescribeMutableStateLatency.With(a.forceReplicationMetricsHandler).Record(time.Since(s))

	switch err.(type) {
	case nil:
		metrics.VerifyReplicationTaskSuccess.With(a.forceReplicationMetricsHandler.WithTags(metrics.NamespaceTag(request.Namespace))).Record(1)
		return verifyResult{
			status: verified,
		}, nil

	case *serviceerror.NotFound:
		metrics.VerifyReplicationTaskNotFound.With(a.forceReplicationMetricsHandler.WithTags(metrics.NamespaceTag(request.Namespace))).Record(1)
		// Calling checkSkipWorkflowExecution for every NotFound is sub-optimal as most common case to skip is workfow being deleted due to retention.
		// A better solution is to only check the existence for workflow which is close to retention period.
		return a.checkSkipWorkflowExecution(ctx, request, we, ns)

	case *serviceerror.NamespaceNotFound:
		return verifyResult{
			status: notVerified,
		}, temporal.NewNonRetryableApplicationError("remoteClient.DescribeMutableState call failed", "NamespaceNotFound", err)

	default:
		metrics.VerifyReplicationTaskFailed.With(a.forceReplicationMetricsHandler.
			WithTags(metrics.NamespaceTag(request.Namespace), metrics.ServiceErrorTypeTag(err))).Record(1)

		return verifyResult{
			status: notVerified,
		}, errors.WithMessage(err, "remoteClient.DescribeMutableState call failed")
	}
}

func (a *activities) verifyReplicationTasks(
	ctx context.Context,
	request *verifyReplicationTasksRequest,
	details *replicationTasksHeartbeatDetails,
	remoteClient adminservice.AdminServiceClient,
	ns *namespace.Namespace,
	heartbeat func(details replicationTasksHeartbeatDetails),
) (bool, error) {
	start := time.Now()
	progress := false
	defer func() {
		if progress {
			// Update CheckPoint when there is a progress
			details.CheckPoint = time.Now()
		}

		heartbeat(*details)
		metrics.VerifyReplicationTasksLatency.With(a.forceReplicationMetricsHandler).Record(time.Since(start))
	}()

	for ; details.NextIndex < len(request.Executions); details.NextIndex++ {
		we := request.Executions[details.NextIndex]
		r, err := a.verifySingleReplicationTask(ctx, request, remoteClient, ns, we)
		if err != nil {
			return false, err
		}

		if !r.isVerified() {
			details.LastNotVerifiedWorkflowExecution = we
			return false, nil
		}

		heartbeat(*details)
		progress = true
	}

	return true, nil
}

const (
	defaultNoProgressNotRetryableTimeout = 30 * time.Minute
)

func (a *activities) VerifyReplicationTasks(ctx context.Context, request *verifyReplicationTasksRequest) (verifyReplicationTasksResponse, error) {
	ctx = headers.SetCallerInfo(ctx, headers.NewPreemptableCallerInfo(request.Namespace))
	var response verifyReplicationTasksResponse
	var remoteClient adminservice.AdminServiceClient
	var err error

	if len(request.TargetClusterName) > 0 {
		remoteClient, err = a.clientBean.GetRemoteAdminClient(request.TargetClusterName)
		if err != nil {
			return response, err
		}
	} else {
		// TODO: remove once TargetClusterEndpoint is no longer used.
		remoteClient = a.clientFactory.NewRemoteAdminClientWithTimeout(
			request.TargetClusterEndpoint,
			admin.DefaultTimeout,
			admin.DefaultLargeTimeout,
		)
	}

	var details replicationTasksHeartbeatDetails
	if activity.HasHeartbeatDetails(ctx) {
		if err := activity.GetHeartbeatDetails(ctx, &details); err != nil {
			return response, err
		}
	} else {
		details.NextIndex = 0
		details.CheckPoint = time.Now()
		activity.RecordHeartbeat(ctx, details)
	}

	nsEntry, err := a.namespaceRegistry.GetNamespace(namespace.Name(request.Namespace))
	if err != nil {
		return response, err
	}

	// Verify if replication tasks exist on target cluster. There are several cases where execution was not found on target cluster.
	//  1. replication lag
	//  2. Zombie workflow execution
	//  3. workflow execution was deleted (due to retention) after replication task was created
	//  4. workflow execution was not applied succesfully on target cluster (i.e, bug)
	//
	// The verification step is retried for every VerifyInterval to handle #1. Verification progress
	// is recorded in activity heartbeat. The verification is considered of making progress if there was at least one new execution
	// being verified. If no progress is made for long enough, then
	//  - more than checkSkipThreshold, it checks if outstanding workflow execution can be skipped locally (#2 and #3)
	//  - more than NonRetryableTimeout, it means potentially #4. The activity returns
	//    non-retryable error and force-replication will fail.
	for {
		// Since replication has a lag, sleep first.
		time.Sleep(request.VerifyInterval)

		verified, err := a.verifyReplicationTasks(ctx, request, &details, remoteClient, nsEntry,
			func(d replicationTasksHeartbeatDetails) {
				activity.RecordHeartbeat(ctx, d)
			})
		if err != nil {
			return response, err
		}

		if verified == true {
			return response, nil
		}

		diff := time.Now().Sub(details.CheckPoint)
		if diff > defaultNoProgressNotRetryableTimeout {
			// Potentially encountered a missing execution, return non-retryable error
			return response, temporal.NewNonRetryableApplicationError(
				fmt.Sprintf("verifyReplicationTasks was not able to make progress for more than %v minutes (not retryable). Not found WorkflowExecution: %v, Checkpoint: %v",
					diff.Minutes(),
					details.LastNotVerifiedWorkflowExecution, details.CheckPoint),
				"", nil)
		}
	}
}