Merge pull request #272 from hashicorp/f-apply-overlap

Plan queue apply overlaps plan verification with plan application to increase throughput

nomad/plan_apply.go

@@ -7,12 +7,41 @@ import (
 	"github.com/armon/go-metrics"
 	"github.com/hashicorp/nomad/nomad/state"
 	"github.com/hashicorp/nomad/nomad/structs"
+	"github.com/hashicorp/raft"
 )
 
 // planApply is a long lived goroutine that reads plan allocations from
 // the plan queue, determines if they can be applied safely and applies
 // them via Raft.
+//
+// Naively, we could simply dequeue a plan, verify, apply and then respond.
+// However, the plan application is bounded by the Raft apply time and
+// subject to some latency. This creates a stall condition, where we are
+// not evaluating, but simply waiting for a transaction to apply.
+//
+// To avoid this, we overlap verification with apply. This means once
+// we've verified plan N we attempt to apply it. However, while waiting
+// for apply, we begin to verify plan N+1 under the assumption that plan
+// N has succeeded.
+//
+// In this sense, we track two parallel versions of the world. One is
+// the pessimistic one driven by the Raft log which is replicated. The
+// other is optimistic and assumes our transactions will succeed. In the
+// happy path, this lets us do productive work during the latency of
+// apply.
+//
+// In the unhappy path (Raft transaction fails), effectively we only
+// wasted work during a time we would have been waiting anyways. However,
+// in anticipation of this case we cannot respond to the plan until
+// the Raft log is updated. This means our schedulers will stall,
+// but there are many of those and only a single plan verifier.
+//
 func (s *Server) planApply() {
+	// waitCh is used to track an outstanding application while snap
+	// holds an optimistic state which includes that plan application.
+	var waitCh chan struct{}
+	var snap *state.StateSnapshot
+
 	for {
 		// Pull the next pending plan, exit if we are no longer leader
 		pending, err := s.planQueue.Dequeue(0)
@@ -35,12 +64,23 @@ func (s *Server) planApply() {
 			continue
 		}
 
+		// Check if out last plan has completed
+		select {
+		case <-waitCh:
+			waitCh = nil
+			snap = nil
+		default:
+		}
+
 		// Snapshot the state so that we have a consistent view of the world
-		snap, err := s.fsm.State().Snapshot()
-		if err != nil {
-			s.logger.Printf("[ERR] nomad: failed to snapshot state: %v", err)
-			pending.respond(nil, err)
-			continue
+		// if no snapshot is available
+		if waitCh == nil || snap == nil {
+			snap, err = s.fsm.State().Snapshot()
+			if err != nil {
+				s.logger.Printf("[ERR] nomad: failed to snapshot state: %v", err)
+				pending.respond(nil, err)
+				continue
+			}
 		}
 
 		// Evaluate the plan
@@ -51,25 +91,40 @@ func (s *Server) planApply() {
 			continue
 		}
 
-		// Apply the plan if there is anything to do
-		if !result.IsNoOp() {
-			allocIndex, err := s.applyPlan(result)
+		// Fast-path the response if there is nothing to do
+		if result.IsNoOp() {
+			pending.respond(result, nil)
+			continue
+		}
+
+		// Ensure any parallel apply is complete before starting the next one.
+		// This also limits how out of date our snapshot can be.
+		if waitCh != nil {
+			<-waitCh
+			snap, err = s.fsm.State().Snapshot()
 			if err != nil {
-				s.logger.Printf("[ERR] nomad: failed to apply plan: %v", err)
+				s.logger.Printf("[ERR] nomad: failed to snapshot state: %v", err)
 				pending.respond(nil, err)
 				continue
 			}
-			result.AllocIndex = allocIndex
 		}
 
-		// Respond to the plan
-		pending.respond(result, nil)
+		// Dispatch the Raft transaction for the plan
+		future, err := s.applyPlan(result, snap)
+		if err != nil {
+			s.logger.Printf("[ERR] nomad: failed to submit plan: %v", err)
+			pending.respond(nil, err)
+			continue
+		}
+
+		// Respond to the plan in async
+		waitCh = make(chan struct{})
+		go s.asyncPlanWait(waitCh, future, result, pending)
 	}
 }
 
 // applyPlan is used to apply the plan result and to return the alloc index
-func (s *Server) applyPlan(result *structs.PlanResult) (uint64, error) {
-	defer metrics.MeasureSince([]string{"nomad", "plan", "apply"}, time.Now())
+func (s *Server) applyPlan(result *structs.PlanResult, snap *state.StateSnapshot) (raft.ApplyFuture, error) {
 	req := structs.AllocUpdateRequest{}
 	for _, updateList := range result.NodeUpdate {
 		req.Alloc = append(req.Alloc, updateList...)
@@ -79,8 +134,38 @@ func (s *Server) applyPlan(result *structs.PlanResult) (uint64, error) {
 	}
 	req.Alloc = append(req.Alloc, result.FailedAllocs...)
 
-	_, index, err := s.raftApply(structs.AllocUpdateRequestType, &req)
-	return index, err
+	// Dispatch the Raft transaction
+	future, err := s.raftApplyFuture(structs.AllocUpdateRequestType, &req)
+	if err != nil {
+		return nil, err
+	}
+
+	// Optimistically apply to our state view
+	if snap != nil {
+		nextIdx := s.raft.AppliedIndex() + 1
+		if err := snap.UpsertAllocs(nextIdx, req.Alloc); err != nil {
+			return future, err
+		}
+	}
+	return future, nil
+}
+
+// asyncPlanWait is used to apply and respond to a plan async
+func (s *Server) asyncPlanWait(waitCh chan struct{}, future raft.ApplyFuture,
+	result *structs.PlanResult, pending *pendingPlan) {
+	defer metrics.MeasureSince([]string{"nomad", "plan", "apply"}, time.Now())
+	defer close(waitCh)
+
+	// Wait for the plan to apply
+	if err := future.Error(); err != nil {
+		s.logger.Printf("[ERR] nomad: failed to apply plan: %v", err)
+		pending.respond(nil, err)
+		return
+	}
+
+	// Respond to the plan
+	result.AllocIndex = future.Index()
+	pending.respond(result, nil)
 }
 
 // evaluatePlan is used to determine what portions of a plan

nomad/plan_apply_test.go

@@ -7,8 +7,17 @@ import (
 	"github.com/hashicorp/nomad/nomad/mock"
 	"github.com/hashicorp/nomad/nomad/structs"
 	"github.com/hashicorp/nomad/testutil"
+	"github.com/hashicorp/raft"
 )
 
+// planWaitFuture is used to wait for the Raft future to complete
+func planWaitFuture(future raft.ApplyFuture) (uint64, error) {
+	if err := future.Error(); err != nil {
+		return 0, err
+	}
+	return future.Index(), nil
+}
+
 func testRegisterNode(t *testing.T, s *Server, n *structs.Node) {
 	// Create the register request
 	req := &structs.NodeRegisterRequest{
@@ -45,8 +54,25 @@ func TestPlanApply_applyPlan(t *testing.T) {
 		FailedAllocs: []*structs.Allocation{allocFail},
 	}
 
+	// Snapshot the state
+	snap, err := s1.State().Snapshot()
+	if err != nil {
+		t.Fatalf("err: %v", err)
+	}
+
 	// Apply the plan
-	index, err := s1.applyPlan(plan)
+	future, err := s1.applyPlan(plan, snap)
+	if err != nil {
+		t.Fatalf("err: %v", err)
+	}
+
+	// Verify our optimistic snapshot is updated
+	if out, err := snap.AllocByID(alloc.ID); err != nil || out == nil {
+		t.Fatalf("bad: %v %v", out, err)
+	}
+
+	// Check plan does apply cleanly
+	index, err := planWaitFuture(future)
 	if err != nil {
 		t.Fatalf("err: %v", err)
 	}
@@ -86,8 +112,25 @@ func TestPlanApply_applyPlan(t *testing.T) {
 		},
 	}
 
+	// Snapshot the state
+	snap, err = s1.State().Snapshot()
+	if err != nil {
+		t.Fatalf("err: %v", err)
+	}
+
 	// Apply the plan
-	index, err = s1.applyPlan(plan)
+	future, err = s1.applyPlan(plan, snap)
+	if err != nil {
+		t.Fatalf("err: %v", err)
+	}
+
+	// Check that our optimistic view is updated
+	if out, _ := snap.AllocByID(allocEvict.ID); out.DesiredStatus != structs.AllocDesiredStatusEvict {
+		t.Fatalf("bad: %#v", out)
+	}
+
+	// Verify plan applies cleanly
+	index, err = planWaitFuture(future)
 	if err != nil {
 		t.Fatalf("err: %v", err)
 	}

nomad/rpc.go

@@ -13,6 +13,7 @@ import (
 	"github.com/hashicorp/net-rpc-msgpackrpc"
 	"github.com/hashicorp/nomad/nomad/state"
 	"github.com/hashicorp/nomad/nomad/structs"
+	"github.com/hashicorp/raft"
 	"github.com/hashicorp/yamux"
 )
 
@@ -225,12 +226,11 @@ func (s *Server) forwardRegion(region, method string, args interface{}, reply in
 	return s.connPool.RPC(region, server.Addr, server.Version, method, args, reply)
 }
 
-// raftApply is used to encode a message, run it through raft, and return
-// the FSM response along with any errors
-func (s *Server) raftApply(t structs.MessageType, msg interface{}) (interface{}, uint64, error) {
+// raftApplyFuture is used to encode a message, run it through raft, and return the Raft future.
+func (s *Server) raftApplyFuture(t structs.MessageType, msg interface{}) (raft.ApplyFuture, error) {
 	buf, err := structs.Encode(t, msg)
 	if err != nil {
-		return nil, 0, fmt.Errorf("Failed to encode request: %v", err)
+		return nil, fmt.Errorf("Failed to encode request: %v", err)
 	}
 
 	// Warn if the command is very large
@@ -239,10 +239,19 @@ func (s *Server) raftApply(t structs.MessageType, msg interface{}) (interface{},
 	}
 
 	future := s.raft.Apply(buf, enqueueLimit)
+	return future, nil
+}
+
+// raftApply is used to encode a message, run it through raft, and return
+// the FSM response along with any errors
+func (s *Server) raftApply(t structs.MessageType, msg interface{}) (interface{}, uint64, error) {
+	future, err := s.raftApplyFuture(t, msg)
+	if err != nil {
+		return nil, 0, err
+	}
 	if err := future.Error(); err != nil {
 		return nil, 0, err
 	}
-
 	return future.Response(), future.Index(), nil
 }