这是本系列文章的第三篇,第一篇在此golang并发三板斧系列之一:channel用于通信和同步,第二篇在此golang并发三板斧系列之二:goroutine池用于并发。
前文描述了手工作坊的时代和工业时代,现在我们进入信息时代。
万恶的资本主义
前文描述的工业时代其实是资本主义,来到世间的每个毛孔都在滴血。不信你看前文的代码,gen
之类的函数创建了一堆任务之后就扔给下游的works处理了,也不管他们要处理多久,是不是加班到深夜,是不是996ICU。
func BenchmarkCPUPool(b *testing.B) {
channum := 100
gonum := runtime.NumCPU()
for i := 0; i < b.N; i++ {
f := func(w *Work) {
if v, ok := w.input.(float64); ok {
cpubound(v)
}
}
list := benchmarkList()
c := genPoolChanBuffer(list)
p := InitPool(channum, gonum, f)
p.RunWorker()
p.FeedWorker(c)
p.Wait()
}
}
人民当家作主
感谢毛主席,解放之后我们进入了社会主义,我们有了工会这个关爱人民的组织。人民的工会爱人民,人民的工会力量大,工会可以在合适的时候给大家放假,让大家休息,对应到程序中就是按下了神圣的ctrl+c
。
如下是最粗暴的模型,工会一喊放假,大家都放下手上的工作开心的玩耍了。但是有的工作做了一半就放弃了,这确实是很没有职业操守的:
var WorkPool chan work
type work struct {
sth string
}
func (w work) Working() {
log.Printf("start %s", w.sth)
time.Sleep(2 * time.Second)
log.Printf("end %s", w.sth)
}
func RunWorkerSimple() {
workers := 2
for i := 0; i < workers; i++ {
go HandleWorkerSimple()
}
}
func HandleWorkerSimple() {
for {
select {
case work := <-WorkPool:
work.Working()
}
}
return
}
func TestWorkerSimple(t *testing.T) {
WorkPool = make(chan work)
go RunWorkerSimple()
go func() {
list := benchmarkList()
for _, l := range list {
WorkPool <- work{fmt.Sprint(l)}
}
}()
select {}
}
很明显从结果可以看出,有工作编号为2和3的没有完成就被扔掉了,其余没有启动的工作都放弃了。
C02S259EFVH3:go_concurrency baixiao$ go test -run TestWorkerSimple
2019/04/08 22:56:11 start 1
2019/04/08 22:56:11 start 0
2019/04/08 22:56:13 end 0
2019/04/08 22:56:13 start 2
2019/04/08 22:56:13 end 1
2019/04/08 22:56:13 start 3
^Csignal: interrupt
FAIL _/Users/baixiao/Go/src/github.com/baixiaoustc/go_concurrency 2.790s
要对得起这份工作
人民的工人为人民,所以即便工会保障了工人的权益,该做好的工作还是要认真做完啊。对应到程序中,收到ctrl+c
中断后,每个worker应该完成手上正在做的工作,并且由BOSS(主goroutine)把剩余队列中的工作保存起来(比如写数据库或者文件),留待明天上班继续做。
代码写起来就复杂多了,要监控系统的中断信号,要等待所有worker处理完手上的事情,最后再把剩余的事情保存起来。需要用两个chan来通信,stopChan 用于通知workers下班啦,stoppedChan 用于所有worker处理完之后告知BOSS,再由BOSS保存剩余的工作。
func (w work) Saving() {
log.Printf("save %s", w.sth)
}
func RunWorkerHold(stop, stopped chan struct{}) {
var wg sync.WaitGroup
workers := 2
for i := 0; i < workers; i++ {
wg.Add(1)
go HandleWorkerHold(stop, &wg)
}
wg.Wait()
stopped <- struct{}{}
}
func HandleWorkerHold(stop chan struct{}, wg *sync.WaitGroup) {
defer wg.Done()
for {
select {
case work := <-WorkPool:
work.Working()
case <-stop:
log.Println("worker: caller has told us to stop")
return
}
}
return
}
func TestWorkerHold(t *testing.T) {
WorkPool = make(chan work)
stopChan := make(chan struct{})
stoppedChan := make(chan struct{})
go RunWorkerHold(stopChan, stoppedChan)
go func() {
list := benchmarkList()
for _, l := range list {
WorkPool <- work{fmt.Sprint(l)}
}
}()
// listen for C-c
c := make(chan os.Signal, 1)
signal.Notify(c, os.Interrupt)
<-c
log.Println("main: received C-c - shutting down")
// tell the goroutine to stop
log.Println("main: telling workers to stop")
close(stopChan)
// and wait for them to reply back
<-stoppedChan
log.Println("main: workers has told us they've finished")
for work := range WorkPool {
work.Saving()
}
return
}
这里没有把打印贴完,最终的结果是所有工作都save好了的:
C02S259EFVH3:go_concurrency baixiao$ go test -run TestWorkerHold
2019/04/08 23:32:02 start 1
2019/04/08 23:32:02 start 0
2019/04/08 23:32:04 end 1
2019/04/08 23:32:04 start 2
2019/04/08 23:32:04 end 0
2019/04/08 23:32:04 start 3
^C2019/04/08 23:32:05 main: received C-c - shutting down
2019/04/08 23:32:05 main: telling workers to stop
2019/04/08 23:32:06 end 2
2019/04/08 23:32:06 worker: caller has told us to stop
2019/04/08 23:32:06 end 3
2019/04/08 23:32:06 start 4
2019/04/08 23:32:08 end 4
2019/04/08 23:32:08 start 5
2019/04/08 23:32:10 end 5
2019/04/08 23:32:10 worker: caller has told us to stop
2019/04/08 23:32:10 main: workers has told us they've finished
2019/04/08 23:32:10 save 6
2019/04/08 23:32:10 save 7
2019/04/08 23:32:10 save 8
2019/04/08 23:32:10 save 9
2019/04/08 23:32:10 save 10
2019/04/08 23:32:10 save 11
2019/04/08 23:32:10 save 12
2019/04/08 23:32:10 save 13
2019/04/08 23:32:10 save 14
2019/04/08 23:32:10 save 15
。。。
但是值得注意的是,不是close(stopChan)
一执行,马上所有的worker都能结束工作了。如上其中一个worker在接着完成work4和work5之后才走了退出流程,是因为对于select
来讲,如果多个chan
都准备好了的话,是随机选择其中一个,所以会有概率一直接着work的。
更进一步
上面的模式还是有缺陷,如果worker下面还有徒弟怎么办(又新开了goroutine)?最后BOSS在做剩余工作的保存时也不想耽误太久怎么办?保存工作写数据库也想受控制(比如database/sql
包提供了相应支持)怎么办?
终于我们的主角登场了,golang提供了context
模式用于解决goroutine的高效且安全退出问题,教程在网上很多了,不用细讲,只贴一下主要函数:
// 带cancel返回值的Context,一旦cancel被调用,即取消该创建的context
func WithCancel(parent Context) (ctx Context, cancel CancelFunc)
// 带有效期cancel返回值的Context,即必须到达指定时间点调用的cancel方法才会被执行
func WithDeadline(parent Context, deadline time.Time) (Context, CancelFunc)
// 带超时时间cancel返回值的Context,类似Deadline,前者是时间点,后者为时间间隔
// 相当于WithDeadline(parent, time.Now().Add(timeout)).
func WithTimeout(parent Context, timeout time.Duration) (Context, CancelFunc)
最后进化到我们的代码,注意SavingDB()
方法只是一个伪代码示意:
import "database/sql"
//fake, just a example
func (w work) SavingDB(ctx context.Context) {
log.Printf("save %s", w.sth)
stmt := "select name from db.table"
db := sql.DB{}
conn, _ := db.Conn(ctx)
rows, err := conn.QueryContext(ctx, stmt)
if err != nil {
if err == context.DeadlineExceeded {
// context canceled
}
return
}
var name string
for rows.Next() {
if err := rows.Scan(&name); err != nil {
if err == context.DeadlineExceeded {
log.Println("scan canceled")
}
}
}
}
func RunWorkerContext(ctx context.Context, stopped chan struct{}) {
var wg sync.WaitGroup
workers := 2
for i := 0; i < workers; i++ {
wg.Add(1)
go HandleWorkerContext(ctx, &wg)
}
wg.Wait()
stopped <- struct{}{}
}
func HandleWorkerContext(ctx context.Context, wg *sync.WaitGroup) {
defer wg.Done()
for {
select {
case work := <-WorkPool:
work.Working()
case <-ctx.Done():
log.Println("worker: caller has told us to stop")
return
}
}
return
}
func TestWorkerContext(t *testing.T) {
WorkPool = make(chan work)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
stoppedChan := make(chan struct{})
go RunWorkerContext(ctx, stoppedChan)
go func() {
list := benchmarkList()
for _, l := range list {
WorkPool <- work{fmt.Sprint(l)}
}
}()
// listen for C-c
c := make(chan os.Signal, 1)
signal.Notify(c, os.Interrupt)
<-c
log.Println("main: received C-c - shutting down")
// tell the goroutine to stop
log.Println("main: telling workers to stop")
cancel()
// and wait for them to reply back
<-stoppedChan
log.Println("main: workers has told us they've finished")
ctxTimeout, cancelTimeout := context.WithTimeout(ctx, 100*time.Microsecond)
defer cancelTimeout()
for {
select {
case work := <-WorkPool:
work.SavingDB(ctxTimeout)
case <-ctxTimeout.Done():
log.Println("main: cann't wait any more")
return
}
}
return
}
通过cancel()
方法通知该context.Context
其下的所有goroutine进入退出流程,并可以启动带timeout的ctx开始保存工作流程,所有流程都是受控的。
这像不像是现代企业的扁平化管理,BOSS直接控制所有员工,提升所有的工作效率?
所有代码都在https://github.com/baixiaoustc/go_concurrency/blob/master/third_post_test.go中能找到。