large workflow hot shard detection

common/dynamicconfig/constants.go

@@ -1346,7 +1346,21 @@ const (
 	// Value type: Int
 	// Default value: 100
 	SampleLoggingRate
-
+	// LargeShardHistorySizeMetricThreshold defines the threshold for what consititutes a large history storage size to alert on
+	// KeyName: system.largeShardHistorySizeMetricThreshold
+	// Value type: Int
+	// Default value: 10485760 (10mb)
+	LargeShardHistorySizeMetricThreshold
+	// LargeShardHistoryEventMetricThreshold defines the threshold for what consititutes a large history event size to alert on
+	// KeyName: system.largeShardHistoryEventMetricThreshold
+	// Value type: Int
+	// Default value: 50 * 1024
+	LargeShardHistoryEventMetricThreshold
+	// LargeShardHistoryBlobMetricThreshold defines the threshold for what consititutes a large history blob size to alert on
+	// KeyName: system.largeShardHistoryBlobMetricThreshold
+	// Value type: Int
+	// Default value: 262144
+	LargeShardHistoryBlobMetricThreshold
 	// LastIntKey must be the last one in this const group
 	LastIntKey
 )
@@ -3454,6 +3468,21 @@ var IntKeys = map[IntKey]DynamicInt{
 		Description:  "The rate for which sampled logs are logged at. 100 means 1/100 is logged",
 		DefaultValue: 100,
 	},
+	LargeShardHistorySizeMetricThreshold: DynamicInt{
+		KeyName:      "system.largeShardHistorySizeMetricThreshold",
+		Description:  "defines the threshold for what consititutes a large history size to alert on, default is 10mb",
+		DefaultValue: 10485760,
+	},
+	LargeShardHistoryEventMetricThreshold: DynamicInt{
+		KeyName:      "system.largeShardHistoryEventMetricThreshold",
+		Description:  "defines the threshold for what consititutes a large history event length to alert on, default is 10000",
+		DefaultValue: 50 * 1024,
+	},
+	LargeShardHistoryBlobMetricThreshold: DynamicInt{
+		KeyName:      "system.largeShardHistoryBlobMetricThreshold",
+		Description:  "defines the threshold for what consititutes a large history blob write to alert on, default is 1/4mb",
+		DefaultValue: 262144,
+	},
 }
 
 var BoolKeys = map[BoolKey]DynamicBool{

common/metrics/defs.go

@@ -1157,6 +1157,12 @@ const (
 	HistoryReplicationV2TaskScope
 	// SyncActivityTaskScope is the scope used by sync activity information processing
 	SyncActivityTaskScope
+	// LargeExecutionSizeShardScope is the scope to track large history size for hotshard detection
+	LargeExecutionSizeShardScope
+	// LargeExecutionCountShardScope is the scope to track large history count for hotshard detection
+	LargeExecutionCountShardScope
+	// LargeExecutionBlobShardScope is the scope to track large blobs for hotshard detection
+	LargeExecutionBlobShardScope
 
 	NumHistoryScopes
 )
@@ -1750,6 +1756,9 @@ var ScopeDefs = map[ServiceIdx]map[int]scopeDefinition{
 		FailoverMarkerScope:                                             {operation: "FailoverMarker"},
 		HistoryReplicationV2TaskScope:                                   {operation: "HistoryReplicationV2Task"},
 		SyncActivityTaskScope:                                           {operation: "SyncActivityTask"},
+		LargeExecutionSizeShardScope:                                    {operation: "LargeExecutionSizeShard"},
+		LargeExecutionCountShardScope:                                   {operation: "LargeExecutionCountShard"},
+		LargeExecutionBlobShardScope:                                    {operation: "LargeExecutionBlobShard"},
 	},
 	// Matching Scope Names
 	Matching: {
@@ -2247,6 +2256,9 @@ const (
 	HistoryFailoverCallbackCount
 	WorkflowVersionCount
 	WorkflowTypeCount
+	LargeHistoryBlobCount
+	LargeHistoryEventCount
+	LargeHistorySizeCount
 
 	NumHistoryMetrics
 )
@@ -2844,6 +2856,9 @@ var MetricDefs = map[ServiceIdx]map[int]metricDefinition{
 		ReplicationTasksCount:                                        {metricName: "replication_tasks_count", metricType: Timer},
 		WorkflowVersionCount:                                         {metricName: "workflow_version_count", metricType: Gauge},
 		WorkflowTypeCount:                                            {metricName: "workflow_type_count", metricType: Gauge},
+		LargeHistoryBlobCount:                                        {metricName: "large_history_blob_count", metricType: Counter},
+		LargeHistoryEventCount:                                       {metricName: "large_history_event_count", metricType: Counter},
+		LargeHistorySizeCount:                                        {metricName: "large_history_size_count", metricType: Counter},
 	},
 	Matching: {
 		PollSuccessPerTaskListCounter:            {metricName: "poll_success_per_tl", metricRollupName: "poll_success"},

common/persistence/dataManagerInterfaces.go

@@ -361,7 +361,10 @@ type (
 
 	// ExecutionStats is the statistics about workflow execution
 	ExecutionStats struct {
-		HistorySize int64
+		HistorySize     int64
+		OldHistorySize  int64
+		OldHistoryCount int64
+		BlobSize        int64
 	}
 
 	// ReplicationState represents mutable state information for global domains.

common/persistence/persistenceMetricClients.go

@@ -316,6 +316,7 @@ func (p *persistenceMetricsClientBase) callWithDomainAndShardScope(scope int, op
 
 	domainMetricsScope.RecordTimer(metrics.PersistenceLatencyPerDomain, duration)
 	shardOperationsMetricsScope.RecordTimer(metrics.PersistenceLatencyPerShard, duration)
+	shardOverallMetricsScope.RecordTimer(metrics.PersistenceLatencyPerShard, duration)
 
 	if p.enableLatencyHistogramMetrics {
 		domainMetricsScope.RecordHistogramDuration(metrics.PersistenceLatencyHistogram, duration)

service/history/config/config.go

@@ -317,8 +317,12 @@ type Config struct {
 	EnableDebugMode             bool // note that this value is initialized once on service start
 	EnableTaskInfoLogByDomainID dynamicconfig.BoolPropertyFnWithDomainIDFilter
 
-	SampleLoggingRate    dynamicconfig.IntPropertyFn
-	EnableShardIDMetrics dynamicconfig.BoolPropertyFn
+	// Hotshard stuff
+	SampleLoggingRate                     dynamicconfig.IntPropertyFn
+	EnableShardIDMetrics                  dynamicconfig.BoolPropertyFn
+	LargeShardHistorySizeMetricThreshold  dynamicconfig.IntPropertyFn
+	LargeShardHistoryEventMetricThreshold dynamicconfig.IntPropertyFn
+	LargeShardHistoryBlobMetricThreshold  dynamicconfig.IntPropertyFn
 }
 
 // New returns new service config with default values
@@ -556,8 +560,11 @@ func New(dc *dynamicconfig.Collection, numberOfShards int, maxMessageSize int, s
 		EnableDebugMode:             dc.GetBoolProperty(dynamicconfig.EnableDebugMode)(),
 		EnableTaskInfoLogByDomainID: dc.GetBoolPropertyFilteredByDomainID(dynamicconfig.HistoryEnableTaskInfoLogByDomainID),
 
-		SampleLoggingRate:    dc.GetIntProperty(dynamicconfig.SampleLoggingRate),
-		EnableShardIDMetrics: dc.GetBoolProperty(dynamicconfig.EnableShardIDMetrics),
+		SampleLoggingRate:                     dc.GetIntProperty(dynamicconfig.SampleLoggingRate),
+		EnableShardIDMetrics:                  dc.GetBoolProperty(dynamicconfig.EnableShardIDMetrics),
+		LargeShardHistorySizeMetricThreshold:  dc.GetIntProperty(dynamicconfig.LargeShardHistorySizeMetricThreshold),
+		LargeShardHistoryEventMetricThreshold: dc.GetIntProperty(dynamicconfig.LargeShardHistoryEventMetricThreshold),
+		LargeShardHistoryBlobMetricThreshold:  dc.GetIntProperty(dynamicconfig.LargeShardHistoryBlobMetricThreshold),
 	}
 
 	return cfg

service/history/execution/context.go

@@ -194,7 +194,10 @@ func NewContext(
 		metricsClient:     shard.GetMetricsClient(),
 		mutex:             locks.NewMutex(),
 		stats: &persistence.ExecutionStats{
-			HistorySize: 0,
+			HistorySize:     0,
+			OldHistorySize:  0,
+			OldHistoryCount: 0,
+			BlobSize:        0,
 		},
 	}
 }
@@ -211,7 +214,10 @@ func (c *contextImpl) Clear() {
 	c.metricsClient.IncCounter(metrics.WorkflowContextScope, metrics.WorkflowContextCleared)
 	c.mutableState = nil
 	c.stats = &persistence.ExecutionStats{
-		HistorySize: 0,
+		HistorySize:     0,
+		OldHistorySize:  0,
+		OldHistoryCount: 0,
+		BlobSize:        0,
 	}
 }
 
@@ -706,7 +712,6 @@ func (c *contextImpl) UpdateWorkflowExecutionWithNew(
 	currentWorkflowTransactionPolicy TransactionPolicy,
 	newWorkflowTransactionPolicy *TransactionPolicy,
 ) (retError error) {
-
 	defer func() {
 		if retError != nil {
 			c.Clear()
@@ -720,11 +725,12 @@ func (c *contextImpl) UpdateWorkflowExecutionWithNew(
 	if err != nil {
 		return err
 	}
-
 	var persistedBlobs events.PersistedBlobs
-	currentWorkflowSize := c.GetHistorySize()
+	currentWorkflowSize, oldWorkflowSize := c.GetHistorySize(), c.GetHistorySize()
+	currentWorkflowHistoryCount, oldWorkflowHistoryCount := c.mutableState.GetNextEventID()-1, c.mutableState.GetNextEventID()-1
 	for _, workflowEvents := range currentWorkflowEventsSeq {
 		blob, err := c.PersistNonStartWorkflowBatchEvents(ctx, workflowEvents)
+		currentWorkflowHistoryCount += int64(len(workflowEvents.Events))
 		if err != nil {
 			return err
 		}
@@ -852,6 +858,10 @@ func (c *contextImpl) UpdateWorkflowExecutionWithNew(
 		domainName,
 		resp.MutableStateUpdateSessionStats,
 	)
+	c.stats.BlobSize = currentWorkflowSize - oldWorkflowSize
+	c.stats.OldHistoryCount = oldWorkflowHistoryCount
+	c.stats.OldHistorySize = oldWorkflowSize
+	c.emitLargeWorkflowShardIDStats()
 	// emit workflow completion stats if any
 	if currentWorkflow.ExecutionInfo.State == persistence.WorkflowStateCompleted {
 		if event, err := c.mutableState.GetCompletionEvent(ctx); err == nil {

service/history/execution/context_util.go

@@ -21,6 +21,7 @@
 package execution
 
 import (
+	"strconv"
 	"time"
 
 	"github.com/uber/cadence/common/log"
@@ -30,6 +31,52 @@ import (
 	"github.com/uber/cadence/common/types"
 )
 
+func (c *contextImpl) emitLargeWorkflowShardIDStats() {
+	if c.shard.GetConfig().EnableShardIDMetrics() {
+		shardIDStr := strconv.Itoa(c.shard.GetShardID())
+
+		var blobSizeWarn int64
+		if c.shard.GetConfig().LargeShardHistoryBlobMetricThreshold() < c.shard.GetConfig().BlobSizeLimitWarn(c.GetDomainName()) {
+			blobSizeWarn = int64(c.shard.GetConfig().LargeShardHistoryBlobMetricThreshold())
+		} else {
+			blobSizeWarn = int64(c.shard.GetConfig().BlobSizeLimitWarn(c.GetDomainName()))
+		}
+		// check if blob size is larger than threshold in Dynamic config if so alert on it every time
+		if c.stats.BlobSize > blobSizeWarn {
+			c.logger.SampleInfo("Workflow writing a large blob", c.shard.GetConfig().SampleLoggingRate(), tag.WorkflowDomainName(c.GetDomainName()),
+				tag.WorkflowID(c.workflowExecution.GetWorkflowID()), tag.ShardID(c.shard.GetShardID()))
+			c.metricsClient.Scope(metrics.LargeExecutionBlobShardScope, metrics.ShardIDTag(shardIDStr)).IncCounter(metrics.LargeHistoryBlobCount)
+		}
+
+		var historyCountWarn int64
+		if c.shard.GetConfig().LargeShardHistoryEventMetricThreshold() < c.shard.GetConfig().HistoryCountLimitWarn(c.GetDomainName()) {
+			historyCountWarn = int64(c.shard.GetConfig().LargeShardHistoryEventMetricThreshold())
+		} else {
+			historyCountWarn = int64(c.shard.GetConfig().HistoryCountLimitWarn(c.GetDomainName()))
+		}
+		// check if the new history count is greater than our threshold and only count/log it once when it passes it
+		// this might sometimes double count if the workflow is extremely fast but should be ok to get a rough idea and identify bad actors
+		if c.stats.OldHistoryCount < historyCountWarn && (c.mutableState.GetNextEventID()-1) >= historyCountWarn {
+			c.logger.Warn("Workflow history event count is reaching dangerous levels", tag.WorkflowDomainName(c.GetDomainName()),
+				tag.WorkflowID(c.workflowExecution.GetWorkflowID()), tag.ShardID(c.shard.GetShardID()))
+			c.metricsClient.Scope(metrics.LargeExecutionCountShardScope, metrics.ShardIDTag(shardIDStr)).IncCounter(metrics.LargeHistoryEventCount)
+		}
+
+		var historySizeWarn int64
+		if c.shard.GetConfig().LargeShardHistorySizeMetricThreshold() < c.shard.GetConfig().HistorySizeLimitWarn(c.GetDomainName()) {
+			historySizeWarn = int64(c.shard.GetConfig().LargeShardHistorySizeMetricThreshold())
+		} else {
+			historySizeWarn = int64(c.shard.GetConfig().HistorySizeLimitWarn(c.GetDomainName()))
+		}
+		// check if the new history size is greater than our threshold and only count/log it once when it passes it
+		if c.stats.OldHistorySize < historySizeWarn && c.stats.HistorySize >= historySizeWarn {
+			c.logger.Warn("Workflow history event size is reaching dangerous levels", tag.WorkflowDomainName(c.GetDomainName()),
+				tag.WorkflowID(c.workflowExecution.GetWorkflowID()), tag.ShardID(c.shard.GetShardID()))
+			c.metricsClient.Scope(metrics.LargeExecutionSizeShardScope, metrics.ShardIDTag(shardIDStr)).IncCounter(metrics.LargeHistorySizeCount)
+		}
+	}
+}
+
 func emitWorkflowHistoryStats(
 	metricsClient metrics.Client,
 	domainName string,