tasks/common/persistence/client/quotas.go

package client

import (
	"github.com/hanzoai/tasks/common/headers"
	"github.com/hanzoai/tasks/common/log"
	"github.com/hanzoai/tasks/common/metrics"
	p "github.com/hanzoai/tasks/common/persistence"
	"github.com/hanzoai/tasks/common/quotas"
	"github.com/hanzoai/tasks/service/history/tasks"
)

type (
	perShardPerNamespaceKey struct {
		namespaceID string
		shardID     int32
	}
)

var (
	CallerTypeDefaultPriority = map[string]int{
		headers.CallerTypeOperator:       0,
		headers.CallerTypeAPI:            2,
		headers.CallerTypeBackgroundHigh: 4,
		headers.CallerTypeBackgroundLow:  5,
		headers.CallerTypePreemptable:    6,
	}

	APITypeCallOriginPriorityOverride = map[string]int{
		"StartWorkflowExecution":           1,
		"SignalWithStartWorkflowExecution": 1,
		"SignalWorkflowExecution":          1,
		"RequestCancelWorkflowExecution":   1,
		"TerminateWorkflowExecution":       1,
		"GetWorkflowExecutionHistory":      1,
		"UpdateWorkflowExecution":          1,
	}

	BackgroundTypeAPIPriorityOverride = map[string]int{
		"GetOrCreateShard": 1,
		"UpdateShard":      1,

		// This is a preprequisite for checkpointing queue process progress
		p.ConstructHistoryTaskAPI("RangeCompleteHistoryTasks", tasks.CategoryTransfer):   1,
		p.ConstructHistoryTaskAPI("RangeCompleteHistoryTasks", tasks.CategoryTimer):      1,
		p.ConstructHistoryTaskAPI("RangeCompleteHistoryTasks", tasks.CategoryVisibility): 1,

		// Task resource isolation assumes task can always be loaded.
		// When one namespace has high load, all task processing goroutines
		// may be busy and consumes all persistence request tokens, preventing
		// tasks for other namespaces to be loaded. So give task loading a higher
		// priority than other background requests.
		// NOTE: we also don't want task loading to consume all persistence request tokens,
		// and blocks all other operations. This is done by setting the queue host rps limit
		// dynamic config.
		p.ConstructHistoryTaskAPI("GetHistoryTasks", tasks.CategoryTransfer):   3,
		p.ConstructHistoryTaskAPI("GetHistoryTasks", tasks.CategoryTimer):      3,
		p.ConstructHistoryTaskAPI("GetHistoryTasks", tasks.CategoryVisibility): 3,
	}

	RequestPrioritiesOrdered = []int{0, 1, 2, 3, 4, 5, 6}
)

func NewPriorityRateLimiter(
	hostMaxQPS PersistenceMaxQps,
	requestPriorityFn quotas.RequestPriorityFn,
	operatorRPSRatio OperatorRPSRatio,
	burstRatio PersistenceBurstRatio,
	healthSignals p.HealthSignalAggregator,
	dynamicParams DynamicRateLimitingParams,
	metricsHandler metrics.Handler,
	logger log.Logger,
) quotas.RequestRateLimiter {
	hostRateFn := func() float64 { return float64(hostMaxQPS()) }

	return quotas.NewMultiRequestRateLimiter(
		// host-level dynamic rate limiter
		newPriorityDynamicRateLimiter(
			hostRateFn,
			requestPriorityFn,
			operatorRPSRatio,
			burstRatio,
			healthSignals,
			dynamicParams,
			metricsHandler,
			logger,
		),
		// basic host-level rate limiter
		newPriorityRateLimiter(
			hostRateFn,
			requestPriorityFn,
			operatorRPSRatio,
			burstRatio,
		),
	)
}

func NewPriorityNamespaceRateLimiter(
	hostMaxQPS PersistenceMaxQps,
	namespaceMaxQPS PersistenceNamespaceMaxQps,
	requestPriorityFn quotas.RequestPriorityFn,
	operatorRPSRatio OperatorRPSRatio,
	burstRatio PersistenceBurstRatio,
) quotas.RequestRateLimiter {

	return newPriorityNamespaceRateLimiter(
		namespaceMaxQPS,
		hostMaxQPS,
		requestPriorityFn,
		operatorRPSRatio,
		burstRatio,
	)
}

func NewPriorityNamespaceShardRateLimiter(
	hostMaxQPS PersistenceMaxQps,
	perShardNamespaceMaxQPS PersistencePerShardNamespaceMaxQPS,
	requestPriorityFn quotas.RequestPriorityFn,
	operatorRPSRatio OperatorRPSRatio,
	burstRatio PersistenceBurstRatio,
) quotas.RequestRateLimiter {

	return newPerShardPerNamespacePriorityRateLimiter(
		perShardNamespaceMaxQPS,
		hostMaxQPS,
		requestPriorityFn,
		operatorRPSRatio,
		burstRatio,
	)
}

func newPerShardPerNamespacePriorityRateLimiter(
	perShardNamespaceMaxQPS PersistencePerShardNamespaceMaxQPS,
	hostMaxQPS PersistenceMaxQps,
	requestPriorityFn quotas.RequestPriorityFn,
	operatorRPSRatio OperatorRPSRatio,
	burstRatio PersistenceBurstRatio,
) quotas.RequestRateLimiter {
	return quotas.NewMapRequestRateLimiter(func(req quotas.Request) quotas.RequestRateLimiter {
		if hasCaller(req) && hasCallerSegment(req) {
			return newPriorityRateLimiter(func() float64 {
				if perShardNamespaceMaxQPS == nil || perShardNamespaceMaxQPS(req.Caller) <= 0 {
					return float64(hostMaxQPS())
				}
				return float64(perShardNamespaceMaxQPS(req.Caller))
			},
				requestPriorityFn,
				operatorRPSRatio,
				burstRatio,
			)
		}
		return quotas.NoopRequestRateLimiter
	},
		perShardPerNamespaceKeyFn,
	)
}

func perShardPerNamespaceKeyFn(req quotas.Request) perShardPerNamespaceKey {
	return perShardPerNamespaceKey{
		namespaceID: req.Caller,
		shardID:     req.CallerSegment,
	}
}

func newPriorityNamespaceRateLimiter(
	namespaceMaxQPS PersistenceNamespaceMaxQps,
	hostMaxQPS PersistenceMaxQps,
	requestPriorityFn quotas.RequestPriorityFn,
	operatorRPSRatio OperatorRPSRatio,
	burstRatio PersistenceBurstRatio,
) quotas.RequestRateLimiter {
	return quotas.NewNamespaceRequestRateLimiter(func(req quotas.Request) quotas.RequestRateLimiter {
		if hasCaller(req) {
			return newPriorityRateLimiter(
				func() float64 {
					if namespaceMaxQPS == nil {
						return float64(hostMaxQPS())
					}

					namespaceQPS := float64(namespaceMaxQPS(req.Caller))
					if namespaceQPS <= 0 {
						return float64(hostMaxQPS())
					}

					return namespaceQPS
				},
				requestPriorityFn,
				operatorRPSRatio,
				burstRatio,
			)
		}
		return quotas.NoopRequestRateLimiter
	})
}

func newPriorityRateLimiter(
	rateFn quotas.RateFn,
	requestPriorityFn quotas.RequestPriorityFn,
	operatorRPSRatio OperatorRPSRatio,
	burstRatio PersistenceBurstRatio,
) quotas.RequestRateLimiter {
	rateLimiters := make(map[int]quotas.RequestRateLimiter)
	for priority := range RequestPrioritiesOrdered {
		if priority == CallerTypeDefaultPriority[headers.CallerTypeOperator] {
			rateLimiters[priority] = quotas.NewRequestRateLimiterAdapter(
				quotas.NewDefaultRateLimiter(
					operatorRateFn(rateFn, operatorRPSRatio),
					quotas.BurstRatioFn(burstRatio),
				),
			)
		} else {
			rateLimiters[priority] = quotas.NewRequestRateLimiterAdapter(
				quotas.NewDefaultRateLimiter(
					rateFn,
					quotas.BurstRatioFn(burstRatio),
				),
			)
		}
	}

	return quotas.NewPriorityRateLimiter(
		requestPriorityFn,
		rateLimiters,
	)
}

func newPriorityDynamicRateLimiter(
	rateFn quotas.RateFn,
	requestPriorityFn quotas.RequestPriorityFn,
	operatorRPSRatio OperatorRPSRatio,
	burstRatio PersistenceBurstRatio,
	healthSignals p.HealthSignalAggregator,
	dynamicParams DynamicRateLimitingParams,
	metricsHandler metrics.Handler,
	logger log.Logger,
) quotas.RequestRateLimiter {
	rateLimiters := make(map[int]quotas.RequestRateLimiter)
	for priority := range RequestPrioritiesOrdered {
		// TODO: refactor this so dynamic rate adjustment is global for all priorities
		if priority == CallerTypeDefaultPriority[headers.CallerTypeOperator] {
			rateLimiters[priority] = NewHealthRequestRateLimiterImpl(
				healthSignals,
				operatorRateFn(rateFn, operatorRPSRatio),
				dynamicParams,
				burstRatio,
				metricsHandler,
				logger,
			)
		} else {
			rateLimiters[priority] = NewHealthRequestRateLimiterImpl(
				healthSignals,
				rateFn,
				dynamicParams,
				burstRatio,
				metricsHandler,
				logger,
			)
		}
	}

	return quotas.NewPriorityRateLimiter(
		requestPriorityFn,
		rateLimiters,
	)
}

func RequestPriorityFn(req quotas.Request) int {
	switch req.CallerType {
	case headers.CallerTypeOperator:
		return CallerTypeDefaultPriority[req.CallerType]
	case headers.CallerTypeAPI:
		if priority, ok := APITypeCallOriginPriorityOverride[req.Initiation]; ok {
			return priority
		}
		return CallerTypeDefaultPriority[req.CallerType]
	case headers.CallerTypeBackgroundHigh, headers.CallerTypeBackgroundLow:
		if priority, ok := BackgroundTypeAPIPriorityOverride[req.API]; ok {
			return priority
		}
		return CallerTypeDefaultPriority[req.CallerType]
	case headers.CallerTypePreemptable:
		return CallerTypeDefaultPriority[req.CallerType]
	default:
		// default requests to API priority to be consistent with existing behavior
		return CallerTypeDefaultPriority[headers.CallerTypeAPI]
	}
}

func operatorRateFn(rateFn quotas.RateFn, operatorRPSRatio OperatorRPSRatio) quotas.RateFn {
	return func() float64 {
		return operatorRPSRatio() * rateFn()
	}
}

func hasCaller(req quotas.Request) bool {
	return req.Caller != "" && req.Caller != headers.CallerNameSystem
}

func hasCallerSegment(req quotas.Request) bool {
	return req.CallerSegment > 0 && req.CallerSegment != p.CallerSegmentMissing
}