Files
tidb/pkg/disttask/framework/scheduler/interface.go

219 lines
10 KiB
Go

// Copyright 2023 PingCAP, Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package scheduler
import (
"context"
"github.com/pingcap/tidb/pkg/disttask/framework/proto"
"github.com/pingcap/tidb/pkg/disttask/framework/storage"
"github.com/pingcap/tidb/pkg/sessionctx"
"github.com/pingcap/tidb/pkg/util/syncutil"
)
// TaskManager defines the interface to access task table.
type TaskManager interface {
// GetTopUnfinishedTasks returns unfinished tasks, limited by MaxConcurrentTask*2,
// to make sure lower rank tasks can be scheduled if resource is enough.
// The returned tasks are sorted by task order, see proto.Task.
GetTopUnfinishedTasks(ctx context.Context) ([]*proto.TaskBase, error)
// GetAllSubtasks gets all subtasks with basic columns.
GetAllSubtasks(ctx context.Context) ([]*proto.SubtaskBase, error)
GetTasksInStates(ctx context.Context, states ...any) (task []*proto.Task, err error)
GetTaskByID(ctx context.Context, taskID int64) (task *proto.Task, err error)
GetTaskBaseByID(ctx context.Context, taskID int64) (task *proto.TaskBase, err error)
GCSubtasks(ctx context.Context) error
GetAllNodes(ctx context.Context) ([]proto.ManagedNode, error)
DeleteDeadNodes(ctx context.Context, nodes []string) error
// TransferTasks2History transfer tasks, and it's related subtasks to history tables.
TransferTasks2History(ctx context.Context, tasks []*proto.Task) error
// CancelTask updated task state to canceling.
CancelTask(ctx context.Context, taskID int64) error
// FailTask updates task state to Failed and updates task error.
FailTask(ctx context.Context, taskID int64, currentState proto.TaskState, taskErr error) error
// RevertTask updates task state to reverting, and task error.
RevertTask(ctx context.Context, taskID int64, taskState proto.TaskState, taskErr error) error
// RevertedTask updates task state to reverted.
RevertedTask(ctx context.Context, taskID int64) error
// PauseTask updated task state to pausing.
PauseTask(ctx context.Context, taskKey string) (bool, error)
// PausedTask updated task state to paused.
PausedTask(ctx context.Context, taskID int64) error
// ResumedTask updated task state from resuming to running.
ResumedTask(ctx context.Context, taskID int64) error
// SucceedTask updates a task to success state.
SucceedTask(ctx context.Context, taskID int64) error
// SwitchTaskStep switches the task to the next step and add subtasks in one
// transaction. It will change task state too if we're switch from InitStep to
// next step.
SwitchTaskStep(ctx context.Context, task *proto.Task, nextState proto.TaskState, nextStep proto.Step, subtasks []*proto.Subtask) error
// SwitchTaskStepInBatch similar to SwitchTaskStep, but it will insert subtasks
// in batch, and task step change will be in a separate transaction.
// Note: subtasks of this step must be stable, i.e. count, order and content
// should be the same on each try, else the subtasks inserted might be messed up.
// And each subtask of this step must be different, to handle the network
// partition or owner change.
SwitchTaskStepInBatch(ctx context.Context, task *proto.Task, nextState proto.TaskState, nextStep proto.Step, subtasks []*proto.Subtask) error
// GetUsedSlotsOnNodes returns the used slots on nodes that have subtask scheduled.
// subtasks of each task on one node is only accounted once as we don't support
// running them concurrently.
// we only consider pending/running subtasks, subtasks related to revert are
// not considered.
GetUsedSlotsOnNodes(ctx context.Context) (map[string]int, error)
// GetActiveSubtasks returns subtasks of the task that are in pending/running state.
GetActiveSubtasks(ctx context.Context, taskID int64) ([]*proto.SubtaskBase, error)
// GetSubtaskCntGroupByStates returns the count of subtasks of some step group by state.
GetSubtaskCntGroupByStates(ctx context.Context, taskID int64, step proto.Step) (map[proto.SubtaskState]int64, error)
ResumeSubtasks(ctx context.Context, taskID int64) error
GetSubtaskErrors(ctx context.Context, taskID int64) ([]error, error)
UpdateSubtasksExecIDs(ctx context.Context, subtasks []*proto.SubtaskBase) error
// GetManagedNodes returns the nodes managed by dist framework and can be used
// to execute tasks. If there are any nodes with background role, we use them,
// else we use nodes without role.
// returned nodes are sorted by node id(host:port).
GetManagedNodes(ctx context.Context) ([]proto.ManagedNode, error)
// GetAllSubtasksByStepAndState gets all subtasks by given states for one step.
GetAllSubtasksByStepAndState(ctx context.Context, taskID int64, step proto.Step, state proto.SubtaskState) ([]*proto.Subtask, error)
WithNewSession(fn func(se sessionctx.Context) error) error
WithNewTxn(ctx context.Context, fn func(se sessionctx.Context) error) error
}
// Extension is used to control the process operations for each task.
// it's used to extend functions of BaseScheduler.
// as golang doesn't support inheritance, we embed this interface in Scheduler
// to simulate abstract method as in other OO languages.
type Extension interface {
// OnTick is used to handle the ticker event, if business impl need to do some periodical work, you can
// do it here, but don't do too much work here, because the ticker interval is small, and it will block
// the event is generated every CheckTaskRunningInterval, and only when the task NOT FINISHED and NO ERROR.
OnTick(ctx context.Context, task *proto.Task)
// OnNextSubtasksBatch is used to generate batch of subtasks for next stage
// NOTE: don't change task.State inside, framework will manage it.
// it's called when:
// 1. task is pending and entering it's first step.
// 2. subtasks scheduled has all finished with no error.
// when next step is StepDone, it should return nil, nil.
OnNextSubtasksBatch(ctx context.Context, h storage.TaskHandle, task *proto.Task, execIDs []string, step proto.Step) (subtaskMetas [][]byte, err error)
// OnDone is called when task is done, either finished successfully or failed
// with error.
// if the task is failed when initializing scheduler, or it's an unknown task,
// we don't call this function.
OnDone(ctx context.Context, h storage.TaskHandle, task *proto.Task) error
// GetEligibleInstances is used to get the eligible instances for the task.
// on certain condition we may want to use some instances to do the task, such as instances with more disk.
// if returned instances is empty, it means all instances are eligible.
// TODO: run import from server disk using framework makes this logic complicated,
// the instance might not be managed by framework.
GetEligibleInstances(ctx context.Context, task *proto.Task) ([]string, error)
// IsRetryableErr is used to check whether the error occurred in scheduler is retryable.
IsRetryableErr(err error) bool
// GetNextStep is used to get the next step for the task.
// if task runs successfully, it should go from StepInit to business steps,
// then to StepDone, then scheduler will mark it as finished.
// NOTE: don't depend on task meta to decide the next step, if it's really needed,
// initialize required fields on scheduler.Init
GetNextStep(task *proto.TaskBase) proto.Step
}
// Param is used to pass parameters when creating scheduler.
type Param struct {
taskMgr TaskManager
nodeMgr *NodeManager
slotMgr *SlotManager
serverID string
allocatedSlots bool
}
// schedulerFactoryFn is used to create a scheduler.
type schedulerFactoryFn func(ctx context.Context, task *proto.Task, param Param) Scheduler
var schedulerFactoryMap = struct {
syncutil.RWMutex
m map[proto.TaskType]schedulerFactoryFn
}{
m: make(map[proto.TaskType]schedulerFactoryFn),
}
// RegisterSchedulerFactory is used to register the scheduler factory.
// normally scheduler ctor should be registered before the server start.
// and should be called in a single routine, such as in init().
// after the server start, there's should be no write to the map.
// but for index backfill, the register call stack is so deep, not sure
// if it's safe to do so, so we use a lock here.
func RegisterSchedulerFactory(taskType proto.TaskType, ctor schedulerFactoryFn) {
schedulerFactoryMap.Lock()
defer schedulerFactoryMap.Unlock()
schedulerFactoryMap.m[taskType] = ctor
}
// getSchedulerFactory is used to get the scheduler factory.
func getSchedulerFactory(taskType proto.TaskType) schedulerFactoryFn {
schedulerFactoryMap.RLock()
defer schedulerFactoryMap.RUnlock()
return schedulerFactoryMap.m[taskType]
}
// ClearSchedulerFactory is only used in test.
func ClearSchedulerFactory() {
schedulerFactoryMap.Lock()
defer schedulerFactoryMap.Unlock()
schedulerFactoryMap.m = make(map[proto.TaskType]schedulerFactoryFn)
}
// CleanUpRoutine is used for the framework to do some clean up work if the task is finished.
type CleanUpRoutine interface {
// CleanUp do the cleanup work.
// task.Meta can be updated here, such as redacting some sensitive info.
CleanUp(ctx context.Context, task *proto.Task) error
}
type cleanUpFactoryFn func() CleanUpRoutine
var cleanUpFactoryMap = struct {
syncutil.RWMutex
m map[proto.TaskType]cleanUpFactoryFn
}{
m: make(map[proto.TaskType]cleanUpFactoryFn),
}
// RegisterSchedulerCleanUpFactory is used to register the scheduler clean up factory.
// normally scheduler cleanup is used in the scheduler_manager gcTaskLoop to do clean up
// works when tasks are finished.
func RegisterSchedulerCleanUpFactory(taskType proto.TaskType, ctor cleanUpFactoryFn) {
cleanUpFactoryMap.Lock()
defer cleanUpFactoryMap.Unlock()
cleanUpFactoryMap.m[taskType] = ctor
}
// getSchedulerCleanUpFactory is used to get the scheduler factory.
func getSchedulerCleanUpFactory(taskType proto.TaskType) cleanUpFactoryFn {
cleanUpFactoryMap.RLock()
defer cleanUpFactoryMap.RUnlock()
return cleanUpFactoryMap.m[taskType]
}
// ClearSchedulerCleanUpFactory is only used in test.
func ClearSchedulerCleanUpFactory() {
cleanUpFactoryMap.Lock()
defer cleanUpFactoryMap.Unlock()
cleanUpFactoryMap.m = make(map[proto.TaskType]cleanUpFactoryFn)
}