Files
temporal/service/matching/physical_task_queue_manager_interface.go
Kannan 44ae6d566d Fix poller scaling decisions for backlog tasks and rate-limited queues (#11618)
## What
- Don't send scale down signal when task is matched from backlog even if
the poll wait time is high.
- Do not send scale up signal when task queue is rate limited.

## Why
- When a task comes from DB backlog, the poll wait time reflects DB read
path latency, not excess pollers — the -1 is not appropriate. Instead we
want to apply the normal scale up check.

- Similarly, when dispatch is bottlenecked by a task queue rate limit,
scaling up pollers won't help.

## How did you test it?
Unit tests

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-08-19 18:57:58 +00:00

80 lines
4.4 KiB
Go

//go:generate mockgen -package $GOPACKAGE -source $GOFILE -destination physical_task_queue_manager_mock.go
package matching
import (
"context"
"time"
enumspb "go.temporal.io/api/enums/v1"
taskqueuepb "go.temporal.io/api/taskqueue/v1"
enumsspb "go.temporal.io/server/api/enums/v1"
"go.temporal.io/server/api/matchingservice/v1"
persistencespb "go.temporal.io/server/api/persistence/v1"
taskqueuespb "go.temporal.io/server/api/taskqueue/v1"
)
type (
physicalTaskQueueManager interface {
Start()
Stop(unloadCause)
WaitUntilInitialized(context.Context) error
// StartScaleManager is called by backlog manager after it's loaded metadata from the
// default queue. (New matcher only.)
StartScaleManager(*persistencespb.PartitionScaleState)
// UpdateScaleState is called from the scale manager to update the scale state in the db.
UpdateScaleState(*persistencespb.PartitionScaleState, bool) error
SetupDraining()
// FinishedDraining is called by a draining backlog manager when it has fully drained.
FinishedDraining()
// ReprocessRedirectedTasksAfterStop sends tasks in the matcher that came from other
// physical queues back to be reprocessed. Note this is called after Stop.
ReprocessRedirectedTasksAfterStop()
// PollTask blocks waiting for a task Returns error when context deadline is exceeded
// maxDispatchPerSecond is the max rate at which tasks are allowed to be dispatched
// from this task queue to pollers
PollTask(ctx context.Context, pollMetadata *pollMetadata) (*internalTask, error)
// MarkAlive updates the liveness timer to keep this physicalTaskQueueManager alive.
MarkAlive()
// TrySyncMatch tries to match task to a local or remote poller. Returns a syncMatchOutcome
// indicating success or the reason the match failed.
TrySyncMatch(ctx context.Context, task *internalTask) (syncMatchOutcome, error)
// SpoolTask spools a task to persistence to be matched asynchronously when a poller is available.
SpoolTask(taskInfo *persistencespb.TaskInfo) error
// TODO(pri): old matcher cleanup
ProcessSpooledTask(ctx context.Context, task *internalTask) error
// DispatchSpooledTask dispatches a task to a poller. When there are no pollers to pick
// up the task, this method will return error. Task will not be persisted to db
// TODO(pri): old matcher cleanup
DispatchSpooledTask(ctx context.Context, task *internalTask, userDataChanged <-chan struct{}) error
AddSpooledTask(task *internalTask) error
AddSpooledTaskToMatcher(task *internalTask) error
UserDataChanged()
// DispatchQueryTask will dispatch query to local or remote poller. If forwarded then result or error is returned,
// if dispatched to local poller then nil and nil is returned.
DispatchQueryTask(ctx context.Context, task *internalTask) (*matchingservice.QueryWorkflowResponse, error)
// DispatchNexusTask dispatches a nexus task to a local or remote poller. If forwarded then result or
// error is returned, if dispatched to local poller then nil and nil is returned.
DispatchNexusTask(ctx context.Context, task *internalTask) (*matchingservice.DispatchNexusTaskResponse, error)
UpdatePollerInfo(pollerIdentity, *pollMetadata)
RemovePoller(pollerIdentity)
GetAllPollerInfo() []*taskqueuepb.PollerInfo
HasPollerAfter(accessTime time.Time) bool
// LegacyDescribeTaskQueue returns pollers info and legacy TaskQueueStatus for this physical queue
LegacyDescribeTaskQueue(includeTaskQueueStatus bool) *matchingservice.DescribeTaskQueueResponse
GetStatsByPriority(includeRates bool) map[int32]*taskqueuepb.TaskQueueStats
GetInternalTaskQueueStatus() []*taskqueuespb.InternalTaskQueueStatus
UnloadFromPartitionManager(unloadCause)
QueueKey() *PhysicalTaskQueueKey
// MakePollerScalingDecision makes a decision on whether to scale pollers up or down based on the current state
// of the task queue and the task about to be returned.
MakePollerScalingDecision(ctx context.Context, pollStartTime time.Time, taskSource enumsspb.TaskSource) *taskqueuepb.PollerScalingDecision
// GetFairnessWeightOverrides returns current fairness weight overrides for this queue.
GetFairnessWeightOverrides() fairnessWeightOverrides
UpdateRemotePriorityBacklogs(remotePriorityBacklogSet)
// RecordTaskAdd records the outcome of a task add to this physical queue using
// the queue's tagged metrics handler, so all per-physical-queue labels are included.
RecordTaskAdd(result string, forwarded bool, behavior enumspb.VersioningBehavior)
}
)