tasks/service/history/timerQueueProcessor.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 history

import (
	"fmt"
	"sync"
	"sync/atomic"
	"time"

	"go.temporal.io/api/serviceerror"

	"go.temporal.io/server/api/matchingservice/v1"
	"go.temporal.io/server/client"
	"go.temporal.io/server/common"
	"go.temporal.io/server/common/cluster"
	"go.temporal.io/server/common/dynamicconfig"
	"go.temporal.io/server/common/log"
	"go.temporal.io/server/common/log/tag"
	"go.temporal.io/server/common/metrics"
	"go.temporal.io/server/common/persistence"
	"go.temporal.io/server/common/quotas"
	"go.temporal.io/server/service/history/configs"
	"go.temporal.io/server/service/history/queues"
	"go.temporal.io/server/service/history/shard"
	"go.temporal.io/server/service/history/tasks"
	"go.temporal.io/server/service/history/workflow"
	"go.temporal.io/server/service/worker/archiver"
)

var errUnknownTimerTask = serviceerror.NewInternal("unknown timer task")

type (
	timeNow                 func() time.Time
	updateTimerAckLevel     func(tasks.Key) error
	timerQueueShutdown      func() error
	timerQueueProcessorImpl struct {
		singleProcessor            bool
		currentClusterName         string
		shard                      shard.Context
		taskAllocator              taskAllocator
		config                     *configs.Config
		metricProvider             metrics.MetricsHandler
		metricsClient              metrics.Client
		workflowCache              workflow.Cache
		scheduler                  queues.Scheduler
		workflowDeleteManager      workflow.DeleteManager
		ackLevel                   tasks.Key
		hostRateLimiter            quotas.RateLimiter
		logger                     log.Logger
		clientBean                 client.Bean
		matchingClient             matchingservice.MatchingServiceClient
		status                     int32
		shutdownChan               chan struct{}
		shutdownWG                 sync.WaitGroup
		activeTimerProcessor       *timerQueueActiveProcessorImpl
		standbyTimerProcessorsLock sync.RWMutex
		standbyTimerProcessors     map[string]*timerQueueStandbyProcessorImpl
	}
)

func newTimerQueueProcessor(
	shard shard.Context,
	workflowCache workflow.Cache,
	scheduler queues.Scheduler,
	clientBean client.Bean,
	archivalClient archiver.Client,
	matchingClient matchingservice.MatchingServiceClient,
	metricProvider metrics.MetricsHandler,
	hostRateLimiter quotas.RateLimiter,
) queues.Processor {
	singleProcessor := !shard.GetClusterMetadata().IsGlobalNamespaceEnabled() ||
		shard.GetConfig().TimerProcessorEnableSingleCursor()

	currentClusterName := shard.GetClusterMetadata().GetCurrentClusterName()
	config := shard.GetConfig()
	logger := log.With(shard.GetLogger(), tag.ComponentTimerQueue)
	taskAllocator := newTaskAllocator(shard)
	workflowDeleteManager := workflow.NewDeleteManager(
		shard,
		workflowCache,
		config,
		archivalClient,
		shard.GetTimeSource(),
	)

	return &timerQueueProcessorImpl{
		singleProcessor:       singleProcessor,
		currentClusterName:    currentClusterName,
		shard:                 shard,
		taskAllocator:         taskAllocator,
		config:                config,
		metricProvider:        metricProvider,
		metricsClient:         shard.GetMetricsClient(),
		workflowCache:         workflowCache,
		scheduler:             scheduler,
		workflowDeleteManager: workflowDeleteManager,
		ackLevel:              shard.GetQueueAckLevel(tasks.CategoryTimer),
		hostRateLimiter:       hostRateLimiter,
		logger:                logger,
		clientBean:            clientBean,
		matchingClient:        matchingClient,
		status:                common.DaemonStatusInitialized,
		shutdownChan:          make(chan struct{}),
		activeTimerProcessor: newTimerQueueActiveProcessor(
			shard,
			workflowCache,
			scheduler,
			workflowDeleteManager,
			matchingClient,
			taskAllocator,
			clientBean,
			newQueueProcessorRateLimiter(
				hostRateLimiter,
				config.TimerProcessorMaxPollRPS,
			),
			logger,
			metricProvider,
			singleProcessor,
		),
		standbyTimerProcessors: make(map[string]*timerQueueStandbyProcessorImpl),
	}
}

func (t *timerQueueProcessorImpl) Start() {
	if !atomic.CompareAndSwapInt32(&t.status, common.DaemonStatusInitialized, common.DaemonStatusStarted) {
		return
	}
	t.activeTimerProcessor.Start()
	if !t.singleProcessor {
		t.listenToClusterMetadataChange()
	}

	t.shutdownWG.Add(1)
	go t.completeTimersLoop()
}

func (t *timerQueueProcessorImpl) Stop() {
	if !atomic.CompareAndSwapInt32(&t.status, common.DaemonStatusStarted, common.DaemonStatusStopped) {
		return
	}
	t.activeTimerProcessor.Stop()
	if !t.singleProcessor {
		t.shard.GetClusterMetadata().UnRegisterMetadataChangeCallback(t)
		t.standbyTimerProcessorsLock.RLock()
		for _, standbyTimerProcessor := range t.standbyTimerProcessors {
			standbyTimerProcessor.Stop()
		}
		t.standbyTimerProcessorsLock.RUnlock()
	}
	close(t.shutdownChan)
	common.AwaitWaitGroup(&t.shutdownWG, time.Minute)
}

// NotifyNewTasks - Notify the processor about the new active / standby timer arrival.
// This should be called each time new timer arrives, otherwise timers maybe fired unexpected.
func (t *timerQueueProcessorImpl) NotifyNewTasks(
	clusterName string,
	timerTasks []tasks.Task,
) {
	if clusterName == t.currentClusterName || t.singleProcessor {
		t.activeTimerProcessor.notifyNewTimers(timerTasks)
		return
	}

	t.standbyTimerProcessorsLock.RLock()
	standbyTimerProcessor, ok := t.standbyTimerProcessors[clusterName]
	t.standbyTimerProcessorsLock.RUnlock()
	if !ok {
		panic(fmt.Sprintf("Cannot find timer processor for %s.", clusterName))
	}
	standbyTimerProcessor.setCurrentTime(t.shard.GetCurrentTime(clusterName))
	standbyTimerProcessor.notifyNewTimers(timerTasks)
}

func (t *timerQueueProcessorImpl) FailoverNamespace(
	namespaceIDs map[string]struct{},
) {
	if t.singleProcessor {
		// TODO: we may want to reschedule all tasks for new active namespaces in buffer
		// so that they don't have to keeping waiting on the backoff timer
		return
	}

	// Failover queue is used to scan all inflight tasks, if queue processor is not
	// started, there's no inflight task and we don't need to create a failover processor.
	// Also the HandleAction will be blocked if queue processor processing loop is not running.
	if atomic.LoadInt32(&t.status) != common.DaemonStatusStarted {
		return
	}

	minLevel := t.shard.GetQueueClusterAckLevel(tasks.CategoryTimer, t.currentClusterName).FireTime
	standbyClusterName := t.currentClusterName
	for clusterName, info := range t.shard.GetClusterMetadata().GetAllClusterInfo() {
		if !info.Enabled {
			continue
		}

		ackLevel := t.shard.GetQueueClusterAckLevel(tasks.CategoryTimer, clusterName).FireTime
		if ackLevel.Before(minLevel) {
			minLevel = ackLevel
			standbyClusterName = clusterName
		}
	}
	// the ack manager is exclusive, so just add a cassandra min precision
	maxLevel := t.activeTimerProcessor.getReadLevel().FireTime.Add(1 * time.Millisecond)
	t.logger.Info("Timer Failover Triggered",
		tag.WorkflowNamespaceIDs(namespaceIDs),
		tag.MinLevel(minLevel.UnixNano()),
		tag.MaxLevel(maxLevel.UnixNano()))
	// we should consider make the failover idempotent
	updateShardAckLevel, failoverTimerProcessor := newTimerQueueFailoverProcessor(
		t.shard,
		t.workflowCache,
		t.scheduler,
		t.workflowDeleteManager,
		namespaceIDs,
		standbyClusterName,
		minLevel,
		maxLevel,
		t.matchingClient,
		t.taskAllocator,
		newQueueProcessorRateLimiter(
			t.hostRateLimiter,
			t.config.TimerProcessorFailoverMaxPollRPS,
		),
		t.logger,
		t.metricProvider,
	)

	// NOTE: READ REF BEFORE MODIFICATION
	// ref: historyEngine.go registerNamespaceFailoverCallback function
	err := updateShardAckLevel(tasks.NewKey(minLevel, 0))
	if err != nil {
		t.logger.Error("Error when update shard ack level", tag.Error(err))
	}
	failoverTimerProcessor.Start()
}

func (t *timerQueueProcessorImpl) LockTaskProcessing() {
	if t.singleProcessor {
		return
	}

	t.taskAllocator.lock()
}

func (t *timerQueueProcessorImpl) UnlockTaskProcessing() {
	if t.singleProcessor {
		return
	}

	t.taskAllocator.unlock()
}

func (t *timerQueueProcessorImpl) Category() tasks.Category {
	return tasks.CategoryTimer
}

func (t *timerQueueProcessorImpl) completeTimersLoop() {
	defer t.shutdownWG.Done()

	timer := time.NewTimer(t.config.TimerProcessorCompleteTimerInterval())
	defer timer.Stop()
	for {
		select {
		case <-t.shutdownChan:
			// before shutdown, make sure the ack level is up to date
			t.completeTimers() //nolint:errcheck
			return
		case <-timer.C:
		CompleteLoop:
			for attempt := 1; attempt <= t.config.TimerProcessorCompleteTimerFailureRetryCount(); attempt++ {
				err := t.completeTimers()
				if err != nil {
					t.logger.Info("Failed to complete timers.", tag.Error(err))
					if shard.IsShardOwnershipLostError(err) {
						// shard is unloaded, timer processor should quit as well
						go t.Stop()
						return
					}
					backoff := time.Duration((attempt - 1) * 100)
					time.Sleep(backoff * time.Millisecond)
				} else {
					break CompleteLoop
				}
			}
			timer.Reset(t.config.TimerProcessorCompleteTimerInterval())
		}
	}
}

func (t *timerQueueProcessorImpl) completeTimers() error {
	lowerAckLevel := t.ackLevel
	upperAckLevel := t.activeTimerProcessor.getAckLevel()

	if !t.singleProcessor {
		t.standbyTimerProcessorsLock.RLock()
		for _, standbyTimerProcessor := range t.standbyTimerProcessors {
			ackLevel := standbyTimerProcessor.getAckLevel()
			if upperAckLevel.CompareTo(ackLevel) > 0 {
				upperAckLevel = ackLevel
			}
		}
		t.standbyTimerProcessorsLock.RUnlock()

		for _, failoverInfo := range t.shard.GetAllFailoverLevels(tasks.CategoryTimer) {
			if !upperAckLevel.FireTime.Before(failoverInfo.MinLevel.FireTime) {
				upperAckLevel = failoverInfo.MinLevel
			}
		}
	}

	t.logger.Debug("Start completing timer task", tag.AckLevel(lowerAckLevel), tag.AckLevel(upperAckLevel))
	if lowerAckLevel.CompareTo(upperAckLevel) > 0 {
		return nil
	}

	t.metricsClient.IncCounter(metrics.TimerQueueProcessorScope, metrics.TaskBatchCompleteCounter)

	if lowerAckLevel.FireTime.Before(upperAckLevel.FireTime) {
		ctx, cancel := newQueueIOContext()
		defer cancel()

		err := t.shard.GetExecutionManager().RangeCompleteHistoryTasks(ctx, &persistence.RangeCompleteHistoryTasksRequest{
			ShardID:             t.shard.GetShardID(),
			TaskCategory:        tasks.CategoryTimer,
			InclusiveMinTaskKey: tasks.NewKey(lowerAckLevel.FireTime, 0),
			ExclusiveMaxTaskKey: tasks.NewKey(upperAckLevel.FireTime, 0),
		})
		if err != nil {
			return err
		}
	}

	t.ackLevel = upperAckLevel

	return t.shard.UpdateQueueAckLevel(tasks.CategoryTimer, t.ackLevel)
}

func (t *timerQueueProcessorImpl) listenToClusterMetadataChange() {
	t.shard.GetClusterMetadata().RegisterMetadataChangeCallback(
		t,
		t.handleClusterMetadataUpdate,
	)
}

func (t *timerQueueProcessorImpl) handleClusterMetadataUpdate(
	oldClusterMetadata map[string]*cluster.ClusterInformation,
	newClusterMetadata map[string]*cluster.ClusterInformation,
) {
	t.standbyTimerProcessorsLock.Lock()
	defer t.standbyTimerProcessorsLock.Unlock()
	for clusterName := range oldClusterMetadata {
		if clusterName == t.currentClusterName {
			continue
		}
		// The metadata triggers a update when the following fields update: 1. Enabled 2. Initial Failover Version 3. Cluster address
		// The callback covers three cases:
		// Case 1: Remove a cluster Case 2: Add a new cluster Case 3: Refresh cluster metadata.
		if processor, ok := t.standbyTimerProcessors[clusterName]; ok {
			// Case 1 and Case 3
			processor.Stop()
			delete(t.standbyTimerProcessors, clusterName)
		}
		if clusterInfo := newClusterMetadata[clusterName]; clusterInfo != nil && clusterInfo.Enabled {
			// Case 2 and Case 3
			processor := newTimerQueueStandbyProcessor(
				t.shard,
				t.workflowCache,
				t.scheduler,
				t.workflowDeleteManager,
				t.matchingClient,
				clusterName,
				t.taskAllocator,
				t.clientBean,
				newQueueProcessorRateLimiter(
					t.hostRateLimiter,
					t.config.TimerProcessorMaxPollRPS,
				),
				t.logger,
				t.metricProvider,
			)
			processor.Start()
			t.standbyTimerProcessors[clusterName] = processor
		}
	}
}

func newQueueProcessorRateLimiter(
	hostRateLimiter quotas.RateLimiter,
	shardMaxPollRPS dynamicconfig.IntPropertyFn,
) quotas.RateLimiter {
	return quotas.NewMultiRateLimiter(
		[]quotas.RateLimiter{
			quotas.NewDefaultOutgoingRateLimiter(
				func() float64 {
					return float64(shardMaxPollRPS())
				},
			),
			hostRateLimiter,
		},
	)
}