satellite/orders: bound RollupsWriteCache flushes

In the situation where the flushes take longer than the incoming
rate of writes, the RollupsWriteCache will take every connection
in the database pool and use them forever. Instead of doing that
and taking down satellite availability, bound the number of flush
operations that it will perform and drop incoming writes earlier
to keep memory usage constant.

Adds monitoring events for if any flushes or updates are lost.

Change-Id: I81b169b73501ee9b999f4b03d1e79645fc56f167

satellite/orders/rollups_write_cache.go

@@ -42,9 +42,9 @@ type RollupsWriteCache struct {
 
 	mu             sync.Mutex
 	pendingRollups RollupData
-	currentSize    int
 	latestTime     time.Time
 	stopped        bool
+	flushing       bool
 
 	nextFlushCompletion *sync2.Fence
 }
@@ -78,23 +78,42 @@ func (cache *RollupsWriteCache) UpdateBucketBandwidthSettle(ctx context.Context,
 // resetCache should only be called after you have acquired the cache lock. It
 // will reset the various cache values and return the pendingRollups,
 // latestTime, and currentSize.
-func (cache *RollupsWriteCache) resetCache() (RollupData, time.Time, int) {
+func (cache *RollupsWriteCache) resetCache() (RollupData, time.Time) {
 	pendingRollups := cache.pendingRollups
 	cache.pendingRollups = make(RollupData)
-	oldSize := cache.currentSize
+
-	cache.currentSize = 0
 	latestTime := cache.latestTime
 	cache.latestTime = time.Time{}
-	return pendingRollups, latestTime, oldSize
+
+	return pendingRollups, latestTime
 }
 
 // Flush resets cache then flushes the everything in the rollups write cache to the database.
 func (cache *RollupsWriteCache) Flush(ctx context.Context) {
 	defer mon.Task()(&ctx)(nil)
+
 	cache.mu.Lock()
-	pendingRollups, latestTime, oldSize := cache.resetCache()
+
+	// while we're already flushing, wait for it to complete.
+	for cache.flushing {
+		done := cache.nextFlushCompletion.Done()
 		cache.mu.Unlock()
-	cache.flush(ctx, pendingRollups, latestTime, oldSize)
+
+		select {
+		case <-done:
+		case <-ctx.Done():
+			return
+		}
+
+		cache.mu.Lock()
+	}
+
+	cache.flushing = true
+	pendingRollups, latestTime := cache.resetCache()
+
+	cache.mu.Unlock()
+
+	cache.flush(ctx, pendingRollups, latestTime)
 }
 
 // CloseAndFlush flushes anything in the cache and marks the cache as stopped.
@@ -102,6 +121,7 @@ func (cache *RollupsWriteCache) CloseAndFlush(ctx context.Context) error {
 	cache.mu.Lock()
 	cache.stopped = true
 	cache.mu.Unlock()
+
 	cache.wg.Wait()
 
 	cache.Flush(ctx)
@@ -109,10 +129,11 @@ func (cache *RollupsWriteCache) CloseAndFlush(ctx context.Context) error {
 }
 
 // flush flushes the everything in the rollups write cache to the database.
-func (cache *RollupsWriteCache) flush(ctx context.Context, pendingRollups RollupData, latestTime time.Time, oldSize int) {
+func (cache *RollupsWriteCache) flush(ctx context.Context, pendingRollups RollupData, latestTime time.Time) {
 	defer mon.Task()(&ctx)(nil)
 
-	rollups := make([]BucketBandwidthRollup, 0, oldSize)
+	if len(pendingRollups) > 0 {
+		rollups := make([]BucketBandwidthRollup, 0, len(pendingRollups))
 		for cacheKey, cacheData := range pendingRollups {
 			rollups = append(rollups, BucketBandwidthRollup{
 				ProjectID:  cacheKey.ProjectID,
@@ -126,14 +147,17 @@ func (cache *RollupsWriteCache) flush(ctx context.Context, pendingRollups Rollup
 
 		err := cache.DB.UpdateBucketBandwidthBatch(ctx, latestTime, rollups)
 		if err != nil {
+			mon.Event("rollups_write_cache_flush_lost")
 			cache.log.Error("MONEY LOST! Bucket bandwidth rollup batch flush failed.", zap.Error(err))
 		}
+	}
 
-	var completion *sync2.Fence
 	cache.mu.Lock()
-	cache.nextFlushCompletion, completion = new(sync2.Fence), cache.nextFlushCompletion
+	defer cache.mu.Unlock()
-	cache.mu.Unlock()
+
-	completion.Release()
+	cache.nextFlushCompletion.Release()
+	cache.nextFlushCompletion = new(sync2.Fence)
+	cache.flushing = false
 }
 
 func (cache *RollupsWriteCache) updateCacheValue(ctx context.Context, projectID uuid.UUID, bucketName []byte, action pb.PieceAction, allocated, inline, settled int64, intervalStart time.Time) error {
@@ -146,35 +170,43 @@ func (cache *RollupsWriteCache) updateCacheValue(ctx context.Context, projectID
 		return Error.New("RollupsWriteCache is stopped")
 	}
 
-	if intervalStart.After(cache.latestTime) {
-		cache.latestTime = intervalStart
-	}
-
 	key := CacheKey{
 		ProjectID:  projectID,
 		BucketName: string(bucketName),
 		Action:     action,
 	}
 
+	// pevent unbounded memory memory growth if we're not flushing fast enough
+	// to keep up with incoming writes.
 	data, ok := cache.pendingRollups[key]
-	if !ok {
+	if !ok && len(cache.pendingRollups) >= cache.batchSize {
-		cache.currentSize++
+		mon.Event("rollups_write_cache_update_lost")
+		cache.log.Error("MONEY LOST! Flushing too slow to keep up with demand.")
+	} else {
+		if cache.latestTime.IsZero() || intervalStart.After(cache.latestTime) {
+			cache.latestTime = intervalStart
 		}
+
 		data.Allocated += allocated
 		data.Inline += inline
 		data.Settled += settled
 		cache.pendingRollups[key] = data
+	}
 
-	if cache.currentSize < cache.batchSize {
+	if len(cache.pendingRollups) < cache.batchSize {
 		return nil
 	}
-	pendingRollups, latestTime, oldSize := cache.resetCache()
+
+	if !cache.flushing {
+		cache.flushing = true
+		pendingRollups, latestTime := cache.resetCache()
 
 		cache.wg.Add(1)
 		go func() {
-		cache.flush(ctx, pendingRollups, latestTime, oldSize)
+			defer cache.wg.Done()
-		cache.wg.Done()
+			cache.flush(ctx, pendingRollups, latestTime)
 		}()
+	}
 
 	return nil
 }
@@ -183,22 +215,24 @@ func (cache *RollupsWriteCache) updateCacheValue(ctx context.Context, projectID
 // the returned channel.
 func (cache *RollupsWriteCache) OnNextFlush() <-chan struct{} {
 	cache.mu.Lock()
-	fence := cache.nextFlushCompletion
+	defer cache.mu.Unlock()
-	cache.mu.Unlock()
+
-	return fence.Done()
+	return cache.nextFlushCompletion.Done()
 }
 
 // CurrentSize returns the current size of the cache.
 func (cache *RollupsWriteCache) CurrentSize() int {
 	cache.mu.Lock()
 	defer cache.mu.Unlock()
-	return cache.currentSize
+
+	return len(cache.pendingRollups)
 }
 
 // CurrentData returns the contents of the cache.
 func (cache *RollupsWriteCache) CurrentData() RollupData {
 	cache.mu.Lock()
 	defer cache.mu.Unlock()
+
 	copyCache := RollupData{}
 	for k, v := range cache.pendingRollups {
 		copyCache[k] = v