作者:孙伟

1、进程/线程/协程基本概念

  • 一个进程可以有多个线程,一般情况下固定2MB内存块来做栈,用来保存当前被调用/挂起的函数内部的变量,CPU在执行调度的时候切换的是线程,如果下一个线程也是当前进程的,就只有线程切换,“很快”就能完成;如果下一个线程不是当前的进程,就需要切换进程,这就得费点时间了。
  • 线程分为内核态线程用户态线程,用户态线程需要绑定内核态线程,CPU并不能感知用户态线程的存在,它只知道它在运行1个线程,这个线程实际是内核态线程。
  • 用户态线程实际有个名字叫协程(co-routine),为了容易区分,我们使用协程指用户态线程,使用线程指内核态线程。
  • 协程跟线程是有区别的,线程由CPU调度是抢占式的,协程由用户态调度是协作式的,一个协程让出CPU后,才执行下一个协程。

协程和线程绑定关系有以下3种:

  • N:1,N个协程绑定1个线程,优点就是协程在用户态线程即完成切换,不会陷入到内核态,这种切换非常的轻量快速。但也有很大的缺点,1个进程的所有协程都绑定在1个线程上,一是某个程序用不了硬件的多核加速能力,二是一旦某协程阻塞,造成线程阻塞,本进程的其他协程都无法执行了,根本就没有并发的能力了。
  • 1:1,1个协程绑定1个线程,这种最容易实现。协程的调度都由CPU完成了,不存在N:1缺点,但有一个缺点是协程的创建、删除和切换的代价都由CPU完成,有点略显昂贵了。
  • M:N,M个协程绑定N个线程,是N:1和1:1类型的结合,克服了以上2种模型的缺点,但实现起来最为复杂。

2、Golang简介

2.1 Goroutine 概念

因为线程切换需要很大的上下文,这种切换消耗了大量CPU时间,所以Go的并行单元并不是传统意义上的线程,而是采用更轻量的协程(goroutine)来处理,大大提高了并行度,因此Go被称为“最并行的语言”。

2.2与其他并发模型的对比

  • Python等解释性语言采用的是多进程并发模型,进程的上下文是最大的,所以切换耗费巨大,同时由于多进程通信只能用socket通讯,或者专门设置共享内存,给编程带来了极大的困扰与不便;
  • C++等语言通常会采用多线程并发模型,相比进程,线程的上下文要小很多,而且多个线程之间本来就是共享内存的,所以编程相比要轻松很多。但是线程的启动和销毁,切换依然要耗费大量CPU时间;于是出现了线程池技术,将线程先储存起来,保持一定的数量,来避免频繁开启/关闭线程的时间消耗,但是这种初级的技术存在一些问题,比如有线程一直被IO阻塞,这样的话这个线程一直占据着坑位,导致后面的任务排不到队,拿不到线程来执行;
  • Go的并发较为复杂,Go采用了更轻量的数据结构来代替线程,这种数据结构相比线程更轻量,他有自己的栈,切换起来更快。然而真正执行并发的还是线程,Go通过调度器将goroutine调度到线程中执行,并适时地释放和创建新的线程,并且当一个正在运行的goroutine进入阻塞(常见场景就是等待IO)时,将其脱离占用的线程,将其他准备好运行的goroutine放在该线程上执行。通过较为复杂的调度手段,使得整个系统获得极高的并行度同时又不耗费大量的CPU资源。

2.3 Goroutine的特点

  • 非阻塞。Goroutine的引入是为了方便高并发程序的编写。一个Goroutine在进行阻塞操作(比如系统调用)时,会把当前线程中的其他Goroutine移交到其他线程中继续执行,从而避免了整个程序的阻塞。
  • 调度器。虽然Golang引入了垃圾回收(gc),在执行gc时就要求Goroutine是停止的,但Go通过自己实现调度器,也可以方便的实现该功能。 通过多个Goroutine来实现并发程序,既有异步IO的优势,又具有多线程、多进程编写程序的便利性。
  • 自己维护堆栈。当然引入Goroutine,也意味着引入了极大的复杂性。一个Goroutine既要包含要执行的代码,又要包含用于执行该代码的栈、PC(PC值=当前程序执行位置+8)和SP指针。堆栈指针需要保证各种模式下程序完成性。

既然每个Goroutine都有自己的栈,那么在创建Goroutine时,就要同时创建对应的栈。Goroutine在执行时,栈空间会不停增长。栈通常是连续增长的,由于每个进程中的各个线程共享虚拟内存空间,当有多个线程时,就需要为每个线程分配不同起始地址的栈。这就需要在分配栈之前先预估每个线程栈的大小。如果线程数量非常多,就很容易栈溢出。

为了解决这个问题,就有了Split Stacks 技术:创建栈时,只分配一块比较小的内存,如果进行某次函数调用导致栈空间不足时,就会在其他地方分配一块新的栈空间。新的空间不需要和老的栈空间连续。函数调用的参数会拷贝到新的栈空间中,接下来的函数执行都在新栈空间中进行。Golang的栈管理方式与此类似,但是为了更高的效率,使用了连续栈( Golang连续栈) 实现方式也是先分配一块固定大小的栈,在栈空间不足时,分配一块更大的栈,并把旧的栈全部拷贝到新栈中。这样避免了Split Stacks方法可能导致的频繁内存分配和释放。

Goroutine的执行是可以被抢占的。如果一个Goroutine一直占用CPU,长时间没有被调度过,就会被runtime抢占掉,把CPU时间交给其他Goroutine。 这个可以通过 debug/goroutine 阻塞实现。

2.4 结构体

  • M:指go中的工作者线程,是真正执行代码的单元;
  • P:是一种调度goroutine的上下文,goroutine依赖于P进行调度,P是真正的并行单元;
  • G:即goroutine,是go语言中的一段代码(以一个函数的形式展现),最小的并行单元;

P必须绑定在M上才能运行,M必须绑定了P才能运行,而一般情况下,最多有MAXPROCS(通常等于CPU数量)个P,但是可能有很多个M,真正运行的只有绑定了M的P,所以P是真正的并行单元。

每个P有一个自己的runnableG队列,可以从里面拿出一个G来运行,同时也有一个全局的runnable G队列,G通过P依附在M上面执行。不单独使用全局的runnable G队列的原因是,分布式的队列有利于减小临界区大小,想一想多个线程同时请求可用的G的时候,如果只有全局的资源,那么这个全局的锁会导致多少线程一直在等待。

但是如果一个正在执行的G进入了阻塞,典型的例子就是等待IO,那么他和它所在的M会在那边等待,而上下文P会传递到其他可用的M上面,这样这个阻塞就不会影响程序的并行度。

G结构体

type g struct {
   // Stack parameters.
   // stack describes the actual stack memory: [stack.lo, stack.hi).
   // stackguard0 is the stack pointer compared in the Go stack growth prologue.
   // It is stack.lo+StackGuard normally, but can be StackPreempt to trigger a preemption.
   // stackguard1 is the stack pointer compared in the C stack growth prologue.
   // It is stack.lo+StackGuard on g0 and gsignal stacks.
   // It is ~0 on other goroutine stacks, to trigger a call to morestackc (and crash).
   stack       stack   // offset known to runtime/cgo //描述了真实的栈内存,包括上下界、
   stackguard0 uintptr // offset known to liblink
   stackguard1 uintptr // offset known to liblink

   _panic         *_panic // innermost panic - offset known to liblink
   _defer         *_defer // innermost defer
   m              *m      // current m; offset known to arm liblink  //当前的M
   sched          gobuf    //goroutine切换时,用于保存g的上下文
   syscallsp      uintptr        // if status==Gsyscall, syscallsp = sched.sp to use during gc
   syscallpc      uintptr        // if status==Gsyscall, syscallpc = sched.pc to use during gc
   stktopsp       uintptr        // expected sp at top of stack, to check in traceback
   param          unsafe.Pointer // passed parameter on wakeup 用于传递参数,睡眠时 其他goroutine可以设置param,唤醒时该goroutine可以获取
   atomicstatus   uint32
   stackLock      uint32 // sigprof/scang lock; TODO: fold in to atomicstatus
   goid           int64   //goroutine 的ID
   waitsince      int64  // approx time when the g become blocked  g被阻塞的 大概时间
   waitreason     string // if status==Gwaiting
   schedlink      guintptr
   preempt        bool     // preemption signal, duplicates stackguard0 = stackpreempt
   paniconfault   bool     // panic (instead of crash) on unexpected fault address
   preemptscan    bool     // preempted g does scan for gc
   gcscandone     bool     // g has scanned stack; protected by _Gscan bit in status
   gcscanvalid    bool     // false at start of gc cycle, true if G has not run since last scan; TODO: remove?
   throwsplit     bool     // must not split stack
   raceignore     int8     // ignore race detection events
   sysblocktraced bool     // StartTrace has emitted EvGoInSyscall about this goroutine
   sysexitticks   int64    // cputicks when syscall has returned (for tracing)
   traceseq       uint64   // trace event sequencer
   tracelastp     puintptr // last P emitted an event for this goroutine
   lockedm        muintptr    //G被锁定只能在这个M运行
   sig            uint32
   writebuf       []byte
   sigcode0       uintptr
   sigcode1       uintptr
   sigpc          uintptr
   gopc           uintptr // pc of go statement that created this goroutine
   startpc        uintptr // pc of goroutine function
   racectx        uintptr
   waiting        *sudog         // sudog structures this g is waiting on (that have a valid elem ptr); in lock order
   cgoCtxt        []uintptr      // cgo traceback context
   labels         unsafe.Pointer // profiler labels
   timer          *timer         // cached timer for time.Sleep
   selectDone     uint32         // are we participating in a select and did someone win the race?

   // Per-G GC state

   // gcAssistBytes is this G's GC assist credit in terms of
   // bytes allocated. If this is positive, then the G has credit
   // to allocate gcAssistBytes bytes without assisting. If this
   // is negative, then the G must correct this by performing
   // scan work. We track this in bytes to make it fast to update
   // and check for debt in the malloc hot path. The assist ratio
   // determines how this corresponds to scan work debt.
   gcAssistBytes int64
}

Gobuf结构体

type gobuf struct {
    sp   uintptr
    pc   uintptr
    g    guintptr
    ctxt unsafe.Pointer
    ret  sys.Uintreg
    lr   uintptr
    bp   uintptr // for GOEXPERIMENT=framepointer
}

其中最主要的当然是sched了,保存了goroutine的上下文。goroutine切换的时候不同于线程有OS来负责这部分数据,而是由一个gobuf对象来保存,这样能够更加轻量级,再来看看gobuf的结构

M结构体

type m struct {
    g0      *g     // 带有调度栈的goroutine
    gsignal       *g         // 处理信号的goroutine
    tls           [6]uintptr // thread-local storage
    mstartfn      func()
    curg          *g       // 当前运行的goroutine
    caughtsig     guintptr
    p             puintptr // 关联p和执行的go代码
    nextp         puintptr
    id            int32
    mallocing     int32 // 状态
    spinning      bool // m是否out of work
    blocked       bool // m是否被阻塞
    inwb          bool // m是否在执行写屏蔽
    printlock     int8
    incgo         bool // m在执行cgo吗
    fastrand      uint32
    ncgocall      uint64      // cgo调用的总数
    ncgo          int32       // 当前cgo调用的数目
    park          note
    alllink       *m // 用于链接allm
    schedlink     muintptr
    mcache        *mcache // 当前m的内存缓存
    lockedg       *g // 锁定g在当前m上执行,而不会切换到其他m
    createstack   [32]uintptr // thread创建的栈
}

结构体M中有两个G是需要关注一下的:

  • 一个是curg,代表结构体M当前绑定的结构体G。
  • 另一个是g0,是带有调度栈的goroutine,这是一个比较特殊的goroutine。普通的goroutine的栈是在堆上分配的可增长的栈,而g0的栈是M对应的线程的栈。所有调度相关的代码,会先切换到该goroutine的栈中再执行。也就是说线程的栈也是用的g实现,而不是使用的OS的。

P结构体

type p struct {
    lock mutex
    id          int32
    status      uint32 // 状态,可以为pidle/prunning/...
    link        puintptr
    schedtick   uint32     // 每调度一次加1
    syscalltick uint32     // 每一次系统调用加1
    sysmontick  sysmontick
    m           muintptr   // 回链到关联的m
    mcache      *mcache
    racectx     uintptr
    goidcache    uint64 // goroutine的ID的缓存
    goidcacheend uint64
    // 可运行的goroutine的队列
    runqhead uint32
    runqtail uint32
    runq     [256]guintptr
    runnext guintptr // 下一个运行的g
    sudogcache []*sudog
    sudogbuf   [128]*sudog
    palloc persistentAlloc // per-P to avoid mutex
    pad [sys.CacheLineSize]byte
}

其中P的状态有Pidle, Prunning, Psyscall, Pgcstop, Pdead;在其内部队列runqhead里面有可运行的goroutine,P优先从内部获取执行的g,这样能够提高效率。

Schedt结构体

type schedt struct {
   goidgen  uint64
    lastpoll uint64
    lock mutex
    midle        muintptr // idle状态的m
    nmidle       int32    // idle状态的m个数
    nmidlelocked int32    // lockde状态的m个数
    mcount       int32    // 创建的m的总数
    maxmcount    int32    // m允许的最大个数
    ngsys uint32 // 系统中goroutine的数目,会自动更新
    pidle      puintptr // idle的p
    npidle     uint32
    nmspinning uint32
    // 全局的可运行的g队列
    runqhead guintptr
    runqtail guintptr
    runqsize int32
    // dead的G的全局缓存
    gflock       mutex
    gfreeStack   *g
    gfreeNoStack *g
    ngfree       int32
    // sudog的缓存中心
    sudoglock  mutex
    sudogcache *sudog
}

大多数需要的信息都已放在了结构体M、G和P中,schedt结构体只是一个壳。可以看到,其中有M的idle队列,P的idle队列,以及一个全局的就绪的G队列。schedt结构体中的Lock是非常必须的,如果M或P等做一些非局部的操作,它们一般需要先锁住调度器。

2.5具体函数

goroutine调度器的代码在/src/runtime/proc.go中,一些比较关键的函数分析如下。

2.5.1 schedule函数

schedule函数在runtime需要进行调度时执行,为当前的P寻找一个可以运行的G并执行它,寻找顺序如下:

  • 1) 调用runqget函数来从P自己的runnable G队列中得到一个可以执行的G;
  • 2) 如果1)失败,则调用findrunnable函数去寻找一个可以执行的G;
  • 3) 如果2)也没有得到可以执行的G,那么结束调度,从上次的现场继续执行。
  • 4) 注意)//偶尔会先检查一次全局可运行队列,以确保公平性。否则,两个goroutine可以完全占用本地runqueue。 通过 schedtick计数 %61来保证

代码如下:

// One round of scheduler: find a runnable goroutine and execute it.
// Never returns.
func schedule() {
   _g_ := getg()

   if _g_.m.locks != 0 {
      throw("schedule: holding locks")
   }

   if _g_.m.lockedg != 0 {
      stoplockedm()
      execute(_g_.m.lockedg.ptr(), false) // Never returns.
   }

   // We should not schedule away from a g that is executing a cgo call,
   // since the cgo call is using the m's g0 stack.
   if _g_.m.incgo {
      throw("schedule: in cgo")
   }

top:
   if sched.gcwaiting != 0 {
      gcstopm()
      goto top
   }
   if _g_.m.p.ptr().runSafePointFn != 0 {
      runSafePointFn()
   }

   var gp *g
   var inheritTime bool
   if trace.enabled || trace.shutdown {
      gp = traceReader()
      if gp != nil {
         casgstatus(gp, _Gwaiting, _Grunnable)
         traceGoUnpark(gp, 0)
      }
   }
   if gp == nil && gcBlackenEnabled != 0 {
      gp = gcController.findRunnableGCWorker(_g_.m.p.ptr())
   }
   if gp == nil {
      // Check the global runnable queue once in a while to ensure fairness.
      // Otherwise two goroutines can completely occupy the local runqueue
      // by constantly respawning each other.
      if _g_.m.p.ptr().schedtick%61 == 0 && sched.runqsize > 0 {
         lock(&sched.lock)
         gp = globrunqget(_g_.m.p.ptr(), 1)
         unlock(&sched.lock)
      }
   }
   if gp == nil {
      gp, inheritTime = runqget(_g_.m.p.ptr())
      if gp != nil && _g_.m.spinning {
         throw("schedule: spinning with local work")
      }
   }
   if gp == nil {
      gp, inheritTime = findrunnable() // blocks until work is available
   }

   // This thread is going to run a goroutine and is not spinning anymore,
   // so if it was marked as spinning we need to reset it now and potentially
   // start a new spinning M.
   if _g_.m.spinning {
      resetspinning()
   }

   if gp.lockedm != 0 {
      // Hands off own p to the locked m,
      // then blocks waiting for a new p.
      startlockedm(gp)
      goto top
   }

   execute(gp, inheritTime)
}

2.5.2 findrunnable函数

findrunnable函数负责给一个P寻找可以执行的G,它的寻找顺序如下:

  • 1) 调用runqget函数来从P自己的runnable G队列中得到一个可以执行的G;
  • 2) 如果1)失败,调用globrunqget函数从全局runnableG队列中得到一个可以执行的G;
  • 3) 如果2)失败,调用netpoll(非阻塞)函数取一个异步回调的G
  • 4) 如果3)失败,尝试从其他P那里偷取一半数量的G过来;
  • 5) 如果4)失败,再次调用globrunqget函数从全局runnableG队列中得到一个可以执行的G;
  • 6) 如果5)失败,调用netpoll(阻塞)函数取一个异步回调的G;
  • 7) 如果6)仍然没有取到G,那么调用stopm函数停止这个M。

代码如下:

// Finds a runnable goroutine to execute.
// Tries to steal from other P's, get g from global queue, poll network.
func findrunnable() (gp *g, inheritTime bool) {
   _g_ := getg()

   // The conditions here and in handoffp must agree: if
   // findrunnable would return a G to run, handoffp must start
   // an M.

top:
   _p_ := _g_.m.p.ptr()
   if sched.gcwaiting != 0 {
      gcstopm()
      goto top
   }
   if _p_.runSafePointFn != 0 {
      runSafePointFn()
   }
   if fingwait && fingwake {
      if gp := wakefing(); gp != nil {
         ready(gp, 0, true)
      }
   }
   if *cgo_yield != nil {
      asmcgocall(*cgo_yield, nil)
   }

   // local runq
   if gp, inheritTime := runqget(_p_); gp != nil {
      return gp, inheritTime
   }

   // global runq
   if sched.runqsize != 0 {
      lock(&sched.lock)
      gp := globrunqget(_p_, 0)
      unlock(&sched.lock)
      if gp != nil {
         return gp, false
      }
   }

   // Poll network.
   // This netpoll is only an optimization before we resort to stealing.
   // We can safely skip it if there are no waiters or a thread is blocked
   // in netpoll already. If there is any kind of logical race with that
   // blocked thread (e.g. it has already returned from netpoll, but does
   // not set lastpoll yet), this thread will do blocking netpoll below
   // anyway.
   if netpollinited() && atomic.Load(&netpollWaiters) > 0 && atomic.Load64(&sched.lastpoll) != 0 {
      if gp := netpoll(false); gp != nil { // non-blocking
         // netpoll returns list of goroutines linked by schedlink.
         injectglist(gp.schedlink.ptr())
         casgstatus(gp, _Gwaiting, _Grunnable)
         if trace.enabled {
            traceGoUnpark(gp, 0)
         }
         return gp, false
      }
   }

   // Steal work from other P's.
   procs := uint32(gomaxprocs)
   if atomic.Load(&sched.npidle) == procs-1 {
      // Either GOMAXPROCS=1 or everybody, except for us, is idle already.
      // New work can appear from returning syscall/cgocall, network or timers.
      // Neither of that submits to local run queues, so no point in stealing.
      goto stop
   }
   // If number of spinning M's >= number of busy P's, block.
   // This is necessary to prevent excessive CPU consumption
   // when GOMAXPROCS>>1 but the program parallelism is low.
   if !_g_.m.spinning && 2*atomic.Load(&sched.nmspinning) >= procs-atomic.Load(&sched.npidle) {
      goto stop
   }
   if !_g_.m.spinning {
      _g_.m.spinning = true
      atomic.Xadd(&sched.nmspinning, 1)
   }
   for i := 0; i < 4; i++ {
      for enum := stealOrder.start(fastrand()); !enum.done(); enum.next() {
         if sched.gcwaiting != 0 {
            goto top
         }
         stealRunNextG := i > 2 // first look for ready queues with more than 1 g
         if gp := runqsteal(_p_, allp[enum.position()], stealRunNextG); gp != nil {
            return gp, false
         }
      }
   }

stop:

   // We have nothing to do. If we're in the GC mark phase, can
   // safely scan and blacken objects, and have work to do, run
   // idle-time marking rather than give up the P.
   if gcBlackenEnabled != 0 && _p_.gcBgMarkWorker != 0 && gcMarkWorkAvailable(_p_) {
      _p_.gcMarkWorkerMode = gcMarkWorkerIdleMode
      gp := _p_.gcBgMarkWorker.ptr()
      casgstatus(gp, _Gwaiting, _Grunnable)
      if trace.enabled {
         traceGoUnpark(gp, 0)
      }
      return gp, false
   }

   // Before we drop our P, make a snapshot of the allp slice,
   // which can change underfoot once we no longer block
   // safe-points. We don't need to snapshot the contents because
   // everything up to cap(allp) is immutable.
   allpSnapshot := allp

   // return P and block
   lock(&sched.lock)
   if sched.gcwaiting != 0 || _p_.runSafePointFn != 0 {
      unlock(&sched.lock)
      goto top
   }
   if sched.runqsize != 0 {
      gp := globrunqget(_p_, 0)
      unlock(&sched.lock)
      return gp, false
   }
   if releasep() != _p_ {
      throw("findrunnable: wrong p")
   }
   pidleput(_p_)
   unlock(&sched.lock)

   // Delicate dance: thread transitions from spinning to non-spinning state,
   // potentially concurrently with submission of new goroutines. We must
   // drop nmspinning first and then check all per-P queues again (with
   // #StoreLoad memory barrier in between). If we do it the other way around,
   // another thread can submit a goroutine after we've checked all run queues
   // but before we drop nmspinning; as the result nobody will unpark a thread
   // to run the goroutine.
   // If we discover new work below, we need to restore m.spinning as a signal
   // for resetspinning to unpark a new worker thread (because there can be more
   // than one starving goroutine). However, if after discovering new work
   // we also observe no idle Ps, it is OK to just park the current thread:
   // the system is fully loaded so no spinning threads are required.
   // Also see "Worker thread parking/unparking" comment at the top of the file.
   wasSpinning := _g_.m.spinning
   if _g_.m.spinning {
      _g_.m.spinning = false
      if int32(atomic.Xadd(&sched.nmspinning, -1)) < 0 {
         throw("findrunnable: negative nmspinning")
      }
   }

   // check all runqueues once again
   for _, _p_ := range allpSnapshot {
      if !runqempty(_p_) {
         lock(&sched.lock)
         _p_ = pidleget()
         unlock(&sched.lock)
         if _p_ != nil {
            acquirep(_p_)
            if wasSpinning {
               _g_.m.spinning = true
               atomic.Xadd(&sched.nmspinning, 1)
            }
            goto top
         }
         break
      }
   }

   // Check for idle-priority GC work again.
   if gcBlackenEnabled != 0 && gcMarkWorkAvailable(nil) {
      lock(&sched.lock)
      _p_ = pidleget()
      if _p_ != nil && _p_.gcBgMarkWorker == 0 {
         pidleput(_p_)
         _p_ = nil
      }
      unlock(&sched.lock)
      if _p_ != nil {
         acquirep(_p_)
         if wasSpinning {
            _g_.m.spinning = true
            atomic.Xadd(&sched.nmspinning, 1)
         }
         // Go back to idle GC check.
         goto stop
      }
   }

   // poll network
   if netpollinited() && atomic.Load(&netpollWaiters) > 0 && atomic.Xchg64(&sched.lastpoll, 0) != 0 {
      if _g_.m.p != 0 {
         throw("findrunnable: netpoll with p")
      }
      if _g_.m.spinning {
         throw("findrunnable: netpoll with spinning")
      }
      gp := netpoll(true) // block until new work is available
      atomic.Store64(&sched.lastpoll, uint64(nanotime()))
      if gp != nil {
         lock(&sched.lock)
         _p_ = pidleget()
         unlock(&sched.lock)
         if _p_ != nil {
            acquirep(_p_)
            injectglist(gp.schedlink.ptr())
            casgstatus(gp, _Gwaiting, _Grunnable)
            if trace.enabled {
               traceGoUnpark(gp, 0)
            }
            return gp, false
         }
         injectglist(gp)
      }
   }
   stopm()
   goto top
}

2.5.3 newproc函数

newproc函数负责创建一个可以运行的G并将其放在当前的P的runnable G队列中,它是类似”go func() { … }”语句真正被编译器翻译后的调用,核心代码在newproc1函数。这个函数执行顺序如下:

  • 1) 获得当前的G所在的 P,然后从free G队列中取出一个G;
  • 2) 如果1)取到则对这个G进行参数配置,否则新建一个G;
  • 3) 将G加入P的runnable G队列。

代码如下:

// Go1.10.8版本默认stack大小为2KB

_StackMin = 2048
// 创建一个g对象,然后放到g队列
// 等待被执行

// Create a new g running fn with narg bytes of arguments starting
// at argp. callerpc is the address of the go statement that created
// this. The new g is put on the queue of g's waiting to run.
func newproc1(fn *funcval, argp *uint8, narg int32, callerpc uintptr) {
   _g_ := getg()

   if fn == nil {
      _g_.m.throwing = -1 // do not dump full stacks
      throw("go of nil func value")
   }
   _g_.m.locks++ // disable preemption because it can be holding p in a local var
   siz := narg
   siz = (siz + 7) &^ 7

   // We could allocate a larger initial stack if necessary.
   // Not worth it: this is almost always an error.
   // 4*sizeof(uintreg): extra space added below
   // sizeof(uintreg): caller's LR (arm) or return address (x86, in gostartcall).
   if siz >= _StackMin-4*sys.RegSize-sys.RegSize {
      throw("newproc: function arguments too large for new goroutine")
   }

   _p_ := _g_.m.p.ptr()
   newg := gfget(_p_)
   if newg == nil {
      newg = malg(_StackMin)
      casgstatus(newg, _Gidle, _Gdead)
      allgadd(newg) // publishes with a g->status of Gdead so GC scanner doesn't look at uninitialized stack.
   }
   if newg.stack.hi == 0 {
      throw("newproc1: newg missing stack")
   }

   if readgstatus(newg) != _Gdead {
      throw("newproc1: new g is not Gdead")
   }

   totalSize := 4*sys.RegSize + uintptr(siz) + sys.MinFrameSize // extra space in case of reads slightly beyond frame
   totalSize += -totalSize & (sys.SpAlign - 1)                  // align to spAlign
   sp := newg.stack.hi - totalSize
   spArg := sp
   if usesLR {
      // caller's LR
      *(*uintptr)(unsafe.Pointer(sp)) = 0
      prepGoExitFrame(sp)
      spArg += sys.MinFrameSize
   }
   if narg > 0 {
      memmove(unsafe.Pointer(spArg), unsafe.Pointer(argp), uintptr(narg))
      // This is a stack-to-stack copy. If write barriers
      // are enabled and the source stack is grey (the
      // destination is always black), then perform a
      // barrier copy. We do this *after* the memmove
      // because the destination stack may have garbage on
      // it.
      if writeBarrier.needed && !_g_.m.curg.gcscandone {
         f := findfunc(fn.fn)
         stkmap := (*stackmap)(funcdata(f, _FUNCDATA_ArgsPointerMaps))
         // We're in the prologue, so it's always stack map index 0.
         bv := stackmapdata(stkmap, 0)
         bulkBarrierBitmap(spArg, spArg, uintptr(narg), 0, bv.bytedata)
      }
   }

   memclrNoHeapPointers(unsafe.Pointer(&newg.sched), unsafe.Sizeof(newg.sched))
   newg.sched.sp = sp
   newg.stktopsp = sp
   newg.sched.pc = funcPC(goexit) + sys.PCQuantum // +PCQuantum so that previous instruction is in same function
   newg.sched.g = guintptr(unsafe.Pointer(newg))
   gostartcallfn(&newg.sched, fn)
   newg.gopc = callerpc
   newg.startpc = fn.fn
   if _g_.m.curg != nil {
      newg.labels = _g_.m.curg.labels
   }
   if isSystemGoroutine(newg) {
      atomic.Xadd(&sched.ngsys, +1)
   }
   newg.gcscanvalid = false
   casgstatus(newg, _Gdead, _Grunnable)

   if _p_.goidcache == _p_.goidcacheend {
      // Sched.goidgen is the last allocated id,
      // this batch must be [sched.goidgen+1, sched.goidgen+GoidCacheBatch].
      // At startup sched.goidgen=0, so main goroutine receives goid=1.
      _p_.goidcache = atomic.Xadd64(&sched.goidgen, _GoidCacheBatch)
      _p_.goidcache -= _GoidCacheBatch - 1
      _p_.goidcacheend = _p_.goidcache + _GoidCacheBatch
   }
   newg.goid = int64(_p_.goidcache)
   _p_.goidcache++
   if raceenabled {
      newg.racectx = racegostart(callerpc)
   }
   if trace.enabled {
      traceGoCreate(newg, newg.startpc)
   }
   runqput(_p_, newg, true)

   if atomic.Load(&sched.npidle) != 0 && atomic.Load(&sched.nmspinning) == 0 && mainStarted {
      wakep()
   }
   _g_.m.locks--
   if _g_.m.locks == 0 && _g_.preempt { // restore the preemption request in case we've cleared it in newstack
      _g_.stackguard0 = stackPreempt
   }
}

2.5.4 goexit0函数

goexit函数是当G退出时调用的。这个函数对G进行一些设置后,将它放入free G列表中,供以后复用,之后调用schedule函数调度。

// goexit continuation on g0.
func goexit0(gp *g) {
   _g_ := getg()

   //设置g的 status从 _Grunning变为 _Gdead
   casgstatus(gp, _Grunning, _Gdead)
   if isSystemGoroutine(gp) {
      atomic.Xadd(&sched.ngsys, -1)
   }
   //对该g 进行释放设置 基本为nil /0
   gp.m = nil
   locked := gp.lockedm != 0
   gp.lockedm = 0
   _g_.m.lockedg = 0
   gp.paniconfault = false
   gp._defer = nil // should be true already but just in case.
   gp._panic = nil // non-nil for Goexit during panic. points at stack-allocated data.
   gp.writebuf = nil
   gp.waitreason = ""
   gp.param = nil
   gp.labels = nil
   gp.timer = nil

   if gcBlackenEnabled != 0 && gp.gcAssistBytes > 0 {
      // Flush assist credit to the global pool. This gives
      // better information to pacing if the application is
      // rapidly creating an exiting goroutines.
      scanCredit := int64(gcController.assistWorkPerByte * float64(gp.gcAssistBytes))
      atomic.Xaddint64(&gcController.bgScanCredit, scanCredit)
      gp.gcAssistBytes = 0
   }

   // Note that gp's stack scan is now "valid" because it has no
   // stack.
   gp.gcscanvalid = true
   dropg()

   if _g_.m.lockedInt != 0 {
      print("invalid m->lockedInt = ", _g_.m.lockedInt, "\n")
      throw("internal lockOSThread error")
   }
   _g_.m.lockedExt = 0
   //把这个g 推到free G 列表
   gfput(_g_.m.p.ptr(), gp)
   if locked {
      // The goroutine may have locked this thread because
      // it put it in an unusual kernel state. Kill it
      // rather than returning it to the thread pool.

      // Return to mstart, which will release the P and exit
      // the thread.
      if GOOS != "plan9" { // See golang.org/issue/22227.
         gogo(&_g_.m.g0.sched)
      }
   }
   schedule()
}

2.5.5 handoffp函数

handoffp函数将P从系统调用或阻塞的M中传递出去,如果P还有runnable G队列,那么新开一个M,调用startm函数,新开的M不空旋。

// Hands off P from syscall or locked M.
// Always runs without a P, so write barriers are not allowed.
//go:nowritebarrierrec
func handoffp(_p_ *p) {
   // handoffp must start an M in any situation where
   // findrunnable would return a G to run on _p_.


   //如果这个P的队列不为空或调度内的size不为空 那么 进行startm 且不空旋
   if !runqempty(_p_) || sched.runqsize != 0 {
      startm(_p_, false)
      return
   }
   //如果正在进行GC处理  同上
   if gcBlackenEnabled != 0 && gcMarkWorkAvailable(_p_) {
      startm(_p_, false)
      return
   }
   //如果没活可做了,检查下有没有 空闲/自旋的 M
   //否则 不需要我们做自旋
   if atomic.Load(&sched.nmspinning)+atomic.Load(&sched.npidle) == 0 && atomic.Cas(&sched.nmspinning, 0, 1) { // TODO: fast atomic
      startm(_p_, true)
      return
   }
   //调度上锁  将这个P 摘除走
   lock(&sched.lock)
   if sched.gcwaiting != 0 {
      _p_.status = _Pgcstop
      sched.stopwait--
      if sched.stopwait == 0 {
         notewakeup(&sched.stopnote)
      }
      unlock(&sched.lock)
      return
   }
   if _p_.runSafePointFn != 0 && atomic.Cas(&_p_.runSafePointFn, 1, 0) {
      sched.safePointFn(_p_)
      sched.safePointWait--
      if sched.safePointWait == 0 {
         notewakeup(&sched.safePointNote)
      }
   }
   if sched.runqsize != 0 {
      unlock(&sched.lock)
      startm(_p_, false)
      return
   }
   // If this is the last running P and nobody is polling network,
   // need to wakeup another M to poll network.
   if sched.npidle == uint32(gomaxprocs-1) && atomic.Load64(&sched.lastpoll) != 0 {
      unlock(&sched.lock)
      startm(_p_, false)
      return
   }
   pidleput(_p_)
   unlock(&sched.lock)
}

2.5.6 startm函数

startm函数调度一个M或者必要时创建一个M来运行指定的P。

// Schedules some M to run the p (creates an M if necessary).
// If p==nil, tries to get an idle P, if no idle P's does nothing.
// May run with m.p==nil, so write barriers are not allowed.
// If spinning is set, the caller has incremented nmspinning and startm will
// either decrement nmspinning or set m.spinning in the newly started M.
//go:nowritebarrierrec
func startm(_p_ *p, spinning bool) {
   //加锁
   lock(&sched.lock)
   if _p_ == nil {

      _p_ = pidleget()
      if _p_ == nil {
         unlock(&sched.lock)
         if spinning {
            // The caller incremented nmspinning, but there are no idle Ps,
            // so it's okay to just undo the increment and give up.
            if int32(atomic.Xadd(&sched.nmspinning, -1)) < 0 {
               throw("startm: negative nmspinning")
            }
         }
         return
      }
   }

   mp := mget()
   unlock(&sched.lock)
   if mp == nil {
      var fn func()
      if spinning {
         // The caller incremented nmspinning, so set m.spinning in the new M.
         fn = mspinning
      }
      newm(fn, _p_)
      return
   }

   if mp.spinning {
      throw("startm: m is spinning")
   }
   if mp.nextp != 0 {
      throw("startm: m has p")
   }
   if spinning && !runqempty(_p_) {
      throw("startm: p has runnable gs")
   }
   // The caller incremented nmspinning, so set m.spinning in the new M.
   mp.spinning = spinning
   mp.nextp.set(_p_)
   notewakeup(&mp.park)
}

2.5.7 sysmon函数

sysmon函数是Go runtime启动时创建的,负责监控所有goroutine的状态,判断是否需要GC,进行netpoll等操作。sysmon函数中会调用retake函数进行抢占式调度。

// Always runs without a P, so write barriers are not allowed.
//
//go:nowritebarrierrec
func sysmon() {
   lock(&sched.lock)
   sched.nmsys++
   checkdead()
   unlock(&sched.lock)

   // If a heap span goes unused for 5 minutes after a garbage collection,
   // we hand it back to the operating system.
   scavengelimit := int64(5 * 60 * 1e9)

   if debug.scavenge > 0 {
      // Scavenge-a-lot for testing.
      forcegcperiod = 10 * 1e6
      scavengelimit = 20 * 1e6
   }

   lastscavenge := nanotime()
   nscavenge := 0

   lasttrace := int64(0)
   idle := 0 // how many cycles in succession we had not wokeup somebody
   delay := uint32(0)
   for {
      if idle == 0 { // start with 20us sleep...
         delay = 20
      } else if idle > 50 { // start doubling the sleep after 1ms...
         delay *= 2
      }
      if delay > 10*1000 { // up to 10ms
         delay = 10 * 1000
      }
      usleep(delay)
      if debug.schedtrace <= 0 && (sched.gcwaiting != 0 || atomic.Load(&sched.npidle) == uint32(gomaxprocs)) {
         lock(&sched.lock)
         if atomic.Load(&sched.gcwaiting) != 0 || atomic.Load(&sched.npidle) == uint32(gomaxprocs) {
            atomic.Store(&sched.sysmonwait, 1)
            unlock(&sched.lock)
            // Make wake-up period small enough
            // for the sampling to be correct.
            maxsleep := forcegcperiod / 2
            if scavengelimit < forcegcperiod {
               maxsleep = scavengelimit / 2
            }
            shouldRelax := true
            if osRelaxMinNS > 0 {
               next := timeSleepUntil()
               now := nanotime()
               if next-now < osRelaxMinNS {
                  shouldRelax = false
               }
            }
            if shouldRelax {
               osRelax(true)
            }
            notetsleep(&sched.sysmonnote, maxsleep)
            if shouldRelax {
               osRelax(false)
            }
            lock(&sched.lock)
            atomic.Store(&sched.sysmonwait, 0)
            noteclear(&sched.sysmonnote)
            idle = 0
            delay = 20
         }
         unlock(&sched.lock)
      }
      // trigger libc interceptors if needed
      if *cgo_yield != nil {
         asmcgocall(*cgo_yield, nil)
      }
      // poll network if not polled for more than 10ms
      lastpoll := int64(atomic.Load64(&sched.lastpoll))
      now := nanotime()
      if netpollinited() && lastpoll != 0 && lastpoll+10*1000*1000 < now {
         atomic.Cas64(&sched.lastpoll, uint64(lastpoll), uint64(now))
         gp := netpoll(false) // non-blocking - returns list of goroutines
         if gp != nil {
            // Need to decrement number of idle locked M's
            // (pretending that one more is running) before injectglist.
            // Otherwise it can lead to the following situation:
            // injectglist grabs all P's but before it starts M's to run the P's,
            // another M returns from syscall, finishes running its G,
            // observes that there is no work to do and no other running M's
            // and reports deadlock.
            incidlelocked(-1)
            injectglist(gp)
            incidlelocked(1)
         }
      }
      // retake P's blocked in syscalls
      // and preempt long running G's
      if retake(now) != 0 {
         idle = 0
      } else {
         idle++
      }
      // check if we need to force a GC
      if t := (gcTrigger{kind: gcTriggerTime, now: now}); t.test() && atomic.Load(&forcegc.idle) != 0 {
         lock(&forcegc.lock)
         forcegc.idle = 0
         forcegc.g.schedlink = 0
         injectglist(forcegc.g)
         unlock(&forcegc.lock)
      }
      // scavenge heap once in a while
      if lastscavenge+scavengelimit/2 < now {
         mheap_.scavenge(int32(nscavenge), uint64(now), uint64(scavengelimit))
         lastscavenge = now
         nscavenge++
      }
      if debug.schedtrace > 0 && lasttrace+int64(debug.schedtrace)*1000000 <= now {
         lasttrace = now
         schedtrace(debug.scheddetail > 0)
      }
   }
}

2.5.8 retake函数

枚举所有的P 如果P在系统调用中(_Psyscall), 且经过了一次sysmon循环(20us~10ms), 则抢占这个P, 调用handoffp解除M和P之间的关联, 如果P在运行中(_Prunning), 且经过了一次sysmon循环并且G运行时间超过forcePreemptNS(10ms), 则抢占这个P

并设置g.preempt = true,g.stackguard0 = stackPreempt。

为什么设置了stackguard就可以实现抢占?
因为这个值用于检查当前栈空间是否足够, go函数的开头会比对这个值判断是否需要扩张栈。
newstack函数判断g.stackguard0等于stackPreempt, 就知道这是抢占触发的, 这时会再检查一遍是否要抢占。
抢占机制保证了不会有一个G长时间的运行导致其他G无法运行的情况发生。

func retake(now int64) uint32 {
   n := 0
   // Prevent allp slice changes. This lock will be completely
   // uncontended unless we're already stopping the world.
   lock(&allpLock)
   // We can't use a range loop over allp because we may
   // temporarily drop the allpLock. Hence, we need to re-fetch
   // allp each time around the loop.
   for i := 0; i < len(allp); i++ {
      _p_ := allp[i]
      if _p_ == nil {
         // This can happen if procresize has grown
         // allp but not yet created new Ps.
         continue
      }
      pd := &_p_.sysmontick
      s := _p_.status
      if s == _Psyscall {
         // Retake P from syscall if it's there for more than 1 sysmon tick (at least 20us).
         t := int64(_p_.syscalltick)
         if int64(pd.syscalltick) != t {
            pd.syscalltick = uint32(t)
            pd.syscallwhen = now
            continue
         }
         // On the one hand we don't want to retake Ps if there is no other work to do,
         // but on the other hand we want to retake them eventually
         // because they can prevent the sysmon thread from deep sleep.
         if runqempty(_p_) && atomic.Load(&sched.nmspinning)+atomic.Load(&sched.npidle) > 0 && pd.syscallwhen+10*1000*1000 > now {
            continue
         }
         // Drop allpLock so we can take sched.lock.
         unlock(&allpLock)
         // Need to decrement number of idle locked M's
         // (pretending that one more is running) before the CAS.
         // Otherwise the M from which we retake can exit the syscall,
         // increment nmidle and report deadlock.
         incidlelocked(-1)
         if atomic.Cas(&_p_.status, s, _Pidle) {
            if trace.enabled {
               traceGoSysBlock(_p_)
               traceProcStop(_p_)
            }
            n++
            _p_.syscalltick++
            handoffp(_p_)
         }
         incidlelocked(1)
         lock(&allpLock)
      } else if s == _Prunning {
         // Preempt G if it's running for too long.
         t := int64(_p_.schedtick)
         if int64(pd.schedtick) != t {
            pd.schedtick = uint32(t)
            pd.schedwhen = now
            continue
         }
         if pd.schedwhen+forcePreemptNS > now {
            continue
         }
         preemptone(_p_)
      }
   }
   unlock(&allpLock)
   return uint32(n)
}

3、调度器总结

3.1 调度器的两大思想

  • 复用线程:协程本身就是运行在一组线程之上,不需要频繁的创建、销毁线程,而是对线程的复用。在调度器中复用线程还有2个体现:1)work stealing,当本线程无可运行的G时,尝试从其他线程绑定的P偷取G,而不是销毁线程。2)handoff,当本线程因为G进行系统调用阻塞时,线程释放绑定的P,把P转移给其他空闲的线程执行。
  • 利用并行:GOMAXPROCS设置P的数量,当GOMAXPROCS大于1时,就最多有GOMAXPROCS个线程处于运行状态,这些线程可能分布在多个CPU核上同时运行,使得并发利用并行。另外,GOMAXPROCS也限制了并发的程度,比如GOMAXPROCS = 核数/2,则最多利用了一半的CPU核进行并行。

3.2调度器的两小策略:

  • 抢占:在coroutine中要等待一个协程主动让出CPU才执行下一个协程,在Go中,一个goroutine最多占用CPU 10ms,防止其他goroutine被饿死,这就是goroutine不同于coroutine的一个地方。
  • 全局G队列:在新的调度器中依然有全局G队列,但功能已经被弱化了,当M执行work stealing从其他P偷不到G时,它可以从全局G队列获取G。

4、参考资料

04-07 00:15