Merge pull request #7335 from filecoin-project/feat/sched-ready-work

sealing: Improve scheduling of ready work

extern/sector-storage/sched.go

@@ -117,7 +117,8 @@ type activeResources struct {
 	gpuUsed    bool
 	cpuUse     uint64
 
-	cond *sync.Cond
+	cond    *sync.Cond
+	waiting int
 }
 
 type workerRequest struct {

extern/sector-storage/sched_resources.go

@@ -11,7 +11,9 @@ func (a *activeResources) withResources(id WorkerID, wr storiface.WorkerInfo, r
 		if a.cond == nil {
 			a.cond = sync.NewCond(locker)
 		}
+		a.waiting++
 		a.cond.Wait()
+		a.waiting--
 	}
 
 	a.add(wr.Resources, r)
@@ -19,13 +21,15 @@ func (a *activeResources) withResources(id WorkerID, wr storiface.WorkerInfo, r
 	err := cb()
 
 	a.free(wr.Resources, r)
-	if a.cond != nil {
-		a.cond.Broadcast()
-	}
 
 	return err
 }
 
+// must be called with the same lock as the one passed to withResources
+func (a *activeResources) hasWorkWaiting() bool {
+	return a.waiting > 0
+}
+
 func (a *activeResources) add(wr storiface.WorkerResources, r Resources) {
 	if r.CanGPU {
 		a.gpuUsed = true
@@ -42,6 +46,10 @@ func (a *activeResources) free(wr storiface.WorkerResources, r Resources) {
 	a.cpuUse -= r.Threads(wr.CPUs)
 	a.memUsedMin -= r.MinMemory
 	a.memUsedMax -= r.MaxMemory
+
+	if a.cond != nil {
+		a.cond.Broadcast()
+	}
 }
 
 // canHandleRequest evaluates if the worker has enough available resources to

extern/sector-storage/sched_worker.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"time"
 
+	"github.com/filecoin-project/lotus/extern/sector-storage/sealtasks"
 	"golang.org/x/xerrors"
 
 	"github.com/filecoin-project/lotus/extern/sector-storage/stores"
@@ -338,6 +339,11 @@ func (sw *schedWorker) workerCompactWindows() {
 }
 
 func (sw *schedWorker) processAssignedWindows() {
+	sw.assignReadyWork()
+	sw.assignPreparingWork()
+}
+
+func (sw *schedWorker) assignPreparingWork() {
 	worker := sw.worker
 
 assignLoop:
@@ -366,7 +372,7 @@ assignLoop:
 			todo := firstWindow.todo[tidx]
 
 			log.Debugf("assign worker sector %d", todo.sector.ID.Number)
-			err := sw.startProcessingTask(sw.taskDone, todo)
+			err := sw.startProcessingTask(todo)
 
 			if err != nil {
 				log.Errorf("startProcessingTask error: %+v", err)
@@ -387,7 +393,67 @@ assignLoop:
 	}
 }
 
-func (sw *schedWorker) startProcessingTask(taskDone chan struct{}, req *workerRequest) error {
+func (sw *schedWorker) assignReadyWork() {
+	worker := sw.worker
+
+	worker.lk.Lock()
+	defer worker.lk.Unlock()
+
+	if worker.active.hasWorkWaiting() {
+		// prepared tasks have priority
+		return
+	}
+
+assignLoop:
+	// process windows in order
+	for len(worker.activeWindows) > 0 {
+		firstWindow := worker.activeWindows[0]
+
+		// process tasks within a window, preferring tasks at lower indexes
+		for len(firstWindow.todo) > 0 {
+			tidx := -1
+
+			for t, todo := range firstWindow.todo {
+				if todo.taskType != sealtasks.TTCommit1 && todo.taskType != sealtasks.TTCommit2 { // todo put in task
+					continue
+				}
+
+				needRes := ResourceTable[todo.taskType][todo.sector.ProofType]
+				if worker.active.canHandleRequest(needRes, sw.wid, "startPreparing", worker.info) {
+					tidx = t
+					break
+				}
+			}
+
+			if tidx == -1 {
+				break assignLoop
+			}
+
+			todo := firstWindow.todo[tidx]
+
+			log.Debugf("assign worker sector %d (ready)", todo.sector.ID.Number)
+			err := sw.startProcessingReadyTask(todo)
+
+			if err != nil {
+				log.Errorf("startProcessingTask error: %+v", err)
+				go todo.respond(xerrors.Errorf("startProcessingTask error: %w", err))
+			}
+
+			// Note: we're not freeing window.allocated resources here very much on purpose
+			copy(firstWindow.todo[tidx:], firstWindow.todo[tidx+1:])
+			firstWindow.todo[len(firstWindow.todo)-1] = nil
+			firstWindow.todo = firstWindow.todo[:len(firstWindow.todo)-1]
+		}
+
+		copy(worker.activeWindows, worker.activeWindows[1:])
+		worker.activeWindows[len(worker.activeWindows)-1] = nil
+		worker.activeWindows = worker.activeWindows[:len(worker.activeWindows)-1]
+
+		sw.windowsRequested--
+	}
+}
+
+func (sw *schedWorker) startProcessingTask(req *workerRequest) error {
 	w, sh := sw.worker, sw.sched
 
 	needRes := ResourceTable[req.taskType][req.sector.ProofType]
@@ -406,9 +472,10 @@ func (sw *schedWorker) startProcessingTask(taskDone chan struct{}, req *workerRe
 			w.lk.Unlock()
 
 			select {
-			case taskDone <- struct{}{}:
+			case sw.taskDone <- struct{}{}:
 			case <-sh.closing:
 				log.Warnf("scheduler closed while sending response (prepare error: %+v)", err)
+			default: // there is a notification pending already
 			}
 
 			select {
@@ -428,8 +495,9 @@ func (sw *schedWorker) startProcessingTask(taskDone chan struct{}, req *workerRe
 			defer w.lk.Lock() // we MUST return locked from this function
 
 			select {
-			case taskDone <- struct{}{}:
+			case sw.taskDone <- struct{}{}:
 			case <-sh.closing:
+			default: // there is a notification pending already
 			}
 
 			// Do the work!
@@ -457,6 +525,47 @@ func (sw *schedWorker) startProcessingTask(taskDone chan struct{}, req *workerRe
 	return nil
 }
 
+func (sw *schedWorker) startProcessingReadyTask(req *workerRequest) error {
+	w, sh := sw.worker, sw.sched
+
+	needRes := ResourceTable[req.taskType][req.sector.ProofType]
+
+	w.active.add(w.info.Resources, needRes)
+
+	go func() {
+		// Do the work!
+		err := req.work(req.ctx, sh.workTracker.worker(sw.wid, w.info, w.workerRpc))
+
+		select {
+		case req.ret <- workerResponse{err: err}:
+		case <-req.ctx.Done():
+			log.Warnf("request got cancelled before we could respond")
+		case <-sh.closing:
+			log.Warnf("scheduler closed while sending response")
+		}
+
+		w.lk.Lock()
+
+		w.active.free(w.info.Resources, needRes)
+
+		select {
+		case sw.taskDone <- struct{}{}:
+		case <-sh.closing:
+			log.Warnf("scheduler closed while sending response (prepare error: %+v)", err)
+		default: // there is a notification pending already
+		}
+
+		w.lk.Unlock()
+
+		// This error should always be nil, since nothing is setting it, but just to be safe:
+		if err != nil {
+			log.Errorf("error executing worker (ready): %+v", err)
+		}
+	}()
+
+	return nil
+}
+
 func (sh *scheduler) workerCleanup(wid WorkerID, w *workerHandle) {
 	select {
 	case <-w.closingMgr: