Use rampMeter for Executor (hypermodeinc#5503)

Currently we have rampMeter while applying proposals. This rampMeter puts a throttling on
memory used by underlying mutations present in proposals to be applied. However this works
correctly only in normal mode.
In ludicrous mode, we apply mutations for these proposals asynchronously. Hence above
rampMeter doesn't work as expected and this results in lot of memory usage in cases when there
are many proposals to apply.
This PR introduces a rampMeter for executor as well. Executor is responsible for applying mutations
in ludicrous mode.

Testing:
I tested this PR by running live loader on 21M dataset in ludicrous mode. On this PR live loader completes in ~5-6 minutes. However when I changed maxPendingEdgesSize to 64KB(instead of 64MB currently) in this PR, live loader completes in ~10-11 minutes.

Benchmarking:
I benchmarked using data generated by below script:

package main

import (
  "bytes"
  "fmt"
  "os"
)

var (
  total int = 100000000
  pred      = 1024
)

func main() {
  f, err := os.OpenFile("test.rdf", os.O_CREATE|os.O_RDWR, 0755)
  if err != nil {
    panic(err)
  }
  defer f.Close()

  totalPerPred := total / pred

  buf := bytes.NewBuffer(nil)
  count := 1
  for i := 0; i < totalPerPred; i++ {
    for j := 0; j < pred; j++ {
      rec := fmt.Sprintf(`_:record_%d <pred_%d> "value_%d" .`, count, j, count)
      buf.WriteString(rec)
      buf.WriteString("\n")
      count++
      if count%100000 == 0 { 
        buf.WriteTo(f)
        buf.Reset()
      }   
    }   
  }

  buf.WriteTo(f)
  if err := f.Sync(); err != nil {
    panic(err)
  }
  fmt.Println("Done writing to file: ", count)
}
Above scripts generates ~100M records with 97K records/predicate and total of 1024 predicates.

Master:
Time to finish live loader: 8m44s
Alpha RAM(RES) usage: 10.9 GB

This PR with maxPendingEdgesSize = 64KB:
Time to finish live loader: 8m48s
Alpha RAM(RES) usage: 9.6 GB

This PR with maxPendingEdgesSize = 64MB:
Time to finish live loader: 8m32s
Alpha RAM(RES) usage: 10.5 GB

Author: ashish-goswami

worker/draft.go

@@ -849,17 +849,20 @@ func (n *node) proposeSnapshot(discardN int) error {
 	return n.Raft().Propose(n.ctx, data)
 }
 
-const maxPendingSize int64 = 64 << 20 // in bytes.
+const (
+	maxPendingSize int64 = 64 << 20 // in bytes.
+	nodeApplyChan        = "raft node applyCh"
+)
 
-func (n *node) rampMeter() {
+func rampMeter(address *int64, maxSize int64, component string) {
 	start := time.Now()
 	defer func() {
 		if dur := time.Since(start); dur > time.Second {
-			glog.Infof("Blocked pushing to applyCh for %v", dur.Round(time.Millisecond))
+			glog.Infof("Blocked pushing to %s for %v", component, dur.Round(time.Millisecond))
 		}
 	}()
 	for {
-		if atomic.LoadInt64(&n.pendingSize) <= maxPendingSize {
+		if atomic.LoadInt64(address) <= maxSize {
 			return
 		}
 		time.Sleep(3 * time.Millisecond)
@@ -1168,7 +1171,7 @@ func (n *node) Run() {
 				// Apply the meter this before adding size to pending size so some crazy big
 				// proposal can be pushed to applyCh. If this do this after adding its size to
 				// pending size, we could block forever in rampMeter.
-				n.rampMeter()
+				rampMeter(&n.pendingSize, maxPendingSize, nodeApplyChan)
 				var pendingSize int64
 				for _, p := range proposals {
 					pendingSize += int64(p.Size())

worker/executor.go

@@ -21,6 +21,7 @@ package worker
 import (
 	"context"
 	"sync"
+	"sync/atomic"
 
 	"github.com/dgraph-io/badger/v2/y"
 	"github.com/dgraph-io/dgraph/posting"
@@ -35,6 +36,8 @@ type subMutation struct {
 }
 
 type executor struct {
+	pendingSize int64
+
 	sync.RWMutex
 	predChan map[string]chan *subMutation
 	closer   *y.Closer
@@ -54,8 +57,10 @@ func (e *executor) processMutationCh(ch chan *subMutation) {
 
 	writer := posting.NewTxnWriter(pstore)
 	for payload := range ch {
+		var esize int64
 		ptxn := posting.NewTxn(payload.startTs)
 		for _, edge := range payload.edges {
+			esize += int64(edge.Size())
 			for {
 				err := runMutation(payload.ctx, edge, ptxn)
 				if err == nil {
@@ -74,6 +79,8 @@ func (e *executor) processMutationCh(ch chan *subMutation) {
 		if err := writer.Wait(); err != nil {
 			glog.Errorf("Error while waiting for writes: %v", err)
 		}
+
+		atomic.AddInt64(&e.pendingSize, -esize)
 	}
 }
 
@@ -99,9 +106,16 @@ func (e *executor) getChannelUnderLock(pred string) (ch chan *subMutation) {
 	return ch
 }
 
+const (
+	maxPendingEdgesSize int64 = 64 << 20
+	executorAddEdges          = "executor.addEdges"
+)
+
 func (e *executor) addEdges(ctx context.Context, startTs uint64, edges []*pb.DirectedEdge) {
-	payloadMap := make(map[string]*subMutation)
+	rampMeter(&e.pendingSize, maxPendingEdgesSize, executorAddEdges)
 
+	payloadMap := make(map[string]*subMutation)
+	var esize int64
 	for _, edge := range edges {
 		payload, ok := payloadMap[edge.Attr]
 		if !ok {
@@ -112,6 +126,7 @@ func (e *executor) addEdges(ctx context.Context, startTs uint64, edges []*pb.Dir
 			payload = payloadMap[edge.Attr]
 		}
 		payload.edges = append(payload.edges, edge)
+		esize += int64(edge.Size())
 	}
 
 	// Lock() in case the channel gets closed from underneath us.
@@ -127,4 +142,5 @@ func (e *executor) addEdges(ctx context.Context, startTs uint64, edges []*pb.Dir
 		}
 	}
 
+	atomic.AddInt64(&e.pendingSize, esize)
 }