openbilibili/app/job/main/member/model/queue/queue.go

/*
Copyright 2014 Workiva, LLC

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 queue includes a regular queue and a priority queue.
These queues rely on waitgroups to pause listening threads
on empty queues until a message is received.  If any thread
calls Dispose on the queue, any listeners are immediately returned
with an error.  Any subsequent put to the queue will return an error
as opposed to panicking as with channels.  Queues will grow with unbounded
behavior as opposed to channels which can be buffered but will pause
while a thread attempts to put to a full channel.

Recently added is a lockless ring buffer using the same basic C design as
found here:

http://www.1024cores.net/home/lock-free-algorithms/queues/bounded-mpmc-queue

Modified for use with Go with the addition of some dispose semantics providing
the capability to release blocked threads.  This works for both puts
and gets, either will return an error if they are blocked and the buffer
is disposed.  This could serve as a signal to kill a goroutine.  All threadsafety
is acheived using CAS operations, making this buffer pretty quick.

Benchmarks:
BenchmarkPriorityQueue-8	 		2000000	       782 ns/op
BenchmarkQueue-8	 		 		2000000	       671 ns/op
BenchmarkChannel-8	 		 		1000000	      2083 ns/op
BenchmarkQueuePut-8	   		   		20000	     84299 ns/op
BenchmarkQueueGet-8	   		   		20000	     80753 ns/op
BenchmarkExecuteInParallel-8	    20000	     68891 ns/op
BenchmarkRBLifeCycle-8				10000000	       177 ns/op
BenchmarkRBPut-8					30000000	        58.1 ns/op
BenchmarkRBGet-8					50000000	        26.8 ns/op

TODO: We really need a Fibonacci heap for the priority queue.
TODO: Unify the types of queue to the same interface.
*/
package queue

import (
	"runtime"
	"sync"
	"sync/atomic"
	"time"
)

type waiters []*sema

func (w *waiters) get() *sema {
	if len(*w) == 0 {
		return nil
	}

	sema := (*w)[0]
	copy((*w)[0:], (*w)[1:])
	(*w)[len(*w)-1] = nil // or the zero value of T
	*w = (*w)[:len(*w)-1]
	return sema
}

func (w *waiters) put(sema *sema) {
	*w = append(*w, sema)
}

func (w *waiters) remove(sema *sema) {
	if len(*w) == 0 {
		return
	}
	// build new slice, copy all except sema
	ws := *w
	newWs := make(waiters, 0, len(*w))
	for i := range ws {
		if ws[i] != sema {
			newWs = append(newWs, ws[i])
		}
	}
	*w = newWs
}

type items []interface{}

func (items *items) get(number int64) []interface{} {
	returnItems := make([]interface{}, 0, number)
	index := int64(0)
	for i := int64(0); i < number; i++ {
		if i >= int64(len(*items)) {
			break
		}

		returnItems = append(returnItems, (*items)[i])
		(*items)[i] = nil
		index++
	}

	*items = (*items)[index:]
	return returnItems
}

func (items *items) peek() (interface{}, bool) {
	length := len(*items)

	if length == 0 {
		return nil, false
	}

	return (*items)[0], true
}

func (items *items) getUntil(checker func(item interface{}) bool) []interface{} {
	length := len(*items)

	if len(*items) == 0 {
		// returning nil here actually wraps that nil in a list
		// of interfaces... thanks go
		return []interface{}{}
	}

	returnItems := make([]interface{}, 0, length)
	index := -1
	for i, item := range *items {
		if !checker(item) {
			break
		}

		returnItems = append(returnItems, item)
		index = i
		(*items)[i] = nil // prevent memory leak
	}

	*items = (*items)[index+1:]
	return returnItems
}

type sema struct {
	ready    chan bool
	response *sync.WaitGroup
}

func newSema() *sema {
	return &sema{
		ready:    make(chan bool, 1),
		response: &sync.WaitGroup{},
	}
}

// Queue is the struct responsible for tracking the state
// of the queue.
type Queue struct {
	waiters  waiters
	items    items
	lock     sync.Mutex
	disposed bool
}

// Put will add the specified items to the queue.
func (q *Queue) Put(items ...interface{}) error {
	if len(items) == 0 {
		return nil
	}

	q.lock.Lock()

	if q.disposed {
		q.lock.Unlock()
		return ErrDisposed
	}

	q.items = append(q.items, items...)
	for {
		sema := q.waiters.get()
		if sema == nil {
			break
		}
		sema.response.Add(1)
		select {
		case sema.ready <- true:
			sema.response.Wait()
		default:
			// This semaphore timed out.
		}
		if len(q.items) == 0 {
			break
		}
	}

	q.lock.Unlock()
	return nil
}

// Get retrieves items from the queue.  If there are some items in the
// queue, get will return a number UP TO the number passed in as a
// parameter.  If no items are in the queue, this method will pause
// until items are added to the queue.
func (q *Queue) Get(number int64) ([]interface{}, error) {
	return q.Poll(number, 0)
}

// Poll retrieves items from the queue.  If there are some items in the queue,
// Poll will return a number UP TO the number passed in as a parameter.  If no
// items are in the queue, this method will pause until items are added to the
// queue or the provided timeout is reached.  A non-positive timeout will block
// until items are added.  If a timeout occurs, ErrTimeout is returned.
func (q *Queue) Poll(number int64, timeout time.Duration) ([]interface{}, error) {
	if number < 1 {
		// thanks again go
		return []interface{}{}, nil
	}

	q.lock.Lock()

	if q.disposed {
		q.lock.Unlock()
		return nil, ErrDisposed
	}

	var items []interface{}

	if len(q.items) == 0 {
		sema := newSema()
		q.waiters.put(sema)
		q.lock.Unlock()

		var timeoutC <-chan time.Time
		if timeout > 0 {
			timeoutC = time.After(timeout)
		}
		select {
		case <-sema.ready:
			// we are now inside the put's lock
			if q.disposed {
				return nil, ErrDisposed
			}
			items = q.items.get(number)
			sema.response.Done()
			return items, nil
		case <-timeoutC:
			// cleanup the sema that was added to waiters
			select {
			case sema.ready <- true:
				// we called this before Put() could
				// Remove sema from waiters.
				q.lock.Lock()
				q.waiters.remove(sema)
				q.lock.Unlock()
			default:
				// Put() got it already, we need to call Done() so Put() can move on
				sema.response.Done()
			}
			return nil, ErrTimeout
		}
	}

	items = q.items.get(number)
	q.lock.Unlock()
	return items, nil
}

// Peek returns a the first item in the queue by value
// without modifying the queue.
func (q *Queue) Peek() (interface{}, error) {
	q.lock.Lock()
	defer q.lock.Unlock()

	if q.disposed {
		return nil, ErrDisposed
	}

	peekItem, ok := q.items.peek()
	if !ok {
		return nil, ErrEmptyQueue
	}

	return peekItem, nil
}

// TakeUntil takes a function and returns a list of items that
// match the checker until the checker returns false.  This does not
// wait if there are no items in the queue.
func (q *Queue) TakeUntil(checker func(item interface{}) bool) ([]interface{}, error) {
	if checker == nil {
		return nil, nil
	}

	q.lock.Lock()

	if q.disposed {
		q.lock.Unlock()
		return nil, ErrDisposed
	}

	result := q.items.getUntil(checker)
	q.lock.Unlock()
	return result, nil
}

// Empty returns a bool indicating if this bool is empty.
func (q *Queue) Empty() bool {
	q.lock.Lock()
	defer q.lock.Unlock()

	return len(q.items) == 0
}

// Len returns the number of items in this queue.
func (q *Queue) Len() int64 {
	q.lock.Lock()
	defer q.lock.Unlock()

	return int64(len(q.items))
}

// Disposed returns a bool indicating if this queue
// has had disposed called on it.
func (q *Queue) Disposed() bool {
	q.lock.Lock()
	defer q.lock.Unlock()

	return q.disposed
}

// Dispose will dispose of this queue and returns
// the items disposed. Any subsequent calls to Get
// or Put will return an error.
func (q *Queue) Dispose() []interface{} {
	q.lock.Lock()
	defer q.lock.Unlock()

	q.disposed = true
	for _, waiter := range q.waiters {
		waiter.response.Add(1)
		select {
		case waiter.ready <- true:
			// release Poll immediately
		default:
			// ignore if it's a timeout or in the get
		}
	}

	disposedItems := q.items

	q.items = nil
	q.waiters = nil

	return disposedItems
}

// New is a constructor for a new threadsafe queue.
func New(hint int64) *Queue {
	return &Queue{
		items: make([]interface{}, 0, hint),
	}
}

// ExecuteInParallel will (in parallel) call the provided function
// with each item in the queue until the queue is exhausted.  When the queue
// is exhausted execution is complete and all goroutines will be killed.
// This means that the queue will be disposed so cannot be used again.
func ExecuteInParallel(q *Queue, fn func(interface{})) {
	if q == nil {
		return
	}

	q.lock.Lock() // so no one touches anything in the middle
	// of this process
	todo, done := uint64(len(q.items)), int64(-1)
	// this is important or we might face an infinite loop
	if todo == 0 {
		return
	}

	numCPU := 1
	if runtime.NumCPU() > 1 {
		numCPU = runtime.NumCPU() - 1
	}

	var wg sync.WaitGroup
	wg.Add(numCPU)
	items := q.items

	for i := 0; i < numCPU; i++ {
		go func() {
			for {
				index := atomic.AddInt64(&done, 1)
				if index >= int64(todo) {
					wg.Done()
					break
				}

				fn(items[index])
				items[index] = 0
			}
		}()
	}
	wg.Wait()
	q.lock.Unlock()
	q.Dispose()
}