core, eth, trie: expose more detailed dirty ram tracking for diff layers (#27971)

core/blockchain.go

@@ -1021,7 +1021,7 @@ func (bc *BlockChain) Stop() {
 			for !bc.triegc.Empty() {
 				triedb.Dereference(bc.triegc.PopItem())
 			}
-			if size, _ := triedb.Size(); size != 0 {
+			if _, nodes, _ := triedb.Size(); nodes != 0 { // all memory is contained within the nodes return for hashdb
 				log.Error("Dangling trie nodes after full cleanup")
 			}
 		}
@@ -1429,8 +1429,8 @@ func (bc *BlockChain) writeBlockWithState(block *types.Block, receipts []*types.
 	}
 	// If we exceeded our memory allowance, flush matured singleton nodes to disk
 	var (
-		nodes, imgs = bc.triedb.Size()
-		limit       = common.StorageSize(bc.cacheConfig.TrieDirtyLimit) * 1024 * 1024
+		_, nodes, imgs = bc.triedb.Size() // all memory is contained within the nodes return for hashdb
+		limit          = common.StorageSize(bc.cacheConfig.TrieDirtyLimit) * 1024 * 1024
 	)
 	if nodes > limit || imgs > 4*1024*1024 {
 		bc.triedb.Cap(limit - ethdb.IdealBatchSize)
@@ -1866,8 +1866,12 @@ func (bc *BlockChain) insertChain(chain types.Blocks, setHead bool) (int, error)
 		stats.processed++
 		stats.usedGas += usedGas
 
-		dirty, _ := bc.triedb.Size()
-		stats.report(chain, it.index, dirty, setHead)
+		var snapDiffItems, snapBufItems common.StorageSize
+		if bc.snaps != nil {
+			snapDiffItems, snapBufItems = bc.snaps.Size()
+		}
+		trieDiffNodes, trieBufNodes, _ := bc.triedb.Size()
+		stats.report(chain, it.index, snapDiffItems, snapBufItems, trieDiffNodes, trieBufNodes, setHead)
 
 		if !setHead {
 			// After merge we expect few side chains. Simply count

core/blockchain_insert.go

@@ -39,7 +39,7 @@ const statsReportLimit = 8 * time.Second
 
 // report prints statistics if some number of blocks have been processed
 // or more than a few seconds have passed since the last message.
-func (st *insertStats) report(chain []*types.Block, index int, dirty common.StorageSize, setHead bool) {
+func (st *insertStats) report(chain []*types.Block, index int, snapDiffItems, snapBufItems, trieDiffNodes, triebufNodes common.StorageSize, setHead bool) {
 	// Fetch the timings for the batch
 	var (
 		now     = mclock.Now()
@@ -63,7 +63,16 @@ func (st *insertStats) report(chain []*types.Block, index int, dirty common.Stor
 		if timestamp := time.Unix(int64(end.Time()), 0); time.Since(timestamp) > time.Minute {
 			context = append(context, []interface{}{"age", common.PrettyAge(timestamp)}...)
 		}
-		context = append(context, []interface{}{"dirty", dirty}...)
+		if snapDiffItems != 0 || snapBufItems != 0 { // snapshots enabled
+			context = append(context, []interface{}{"snapdiffs", snapDiffItems}...)
+			if snapBufItems != 0 { // future snapshot refactor
+				context = append(context, []interface{}{"snapdirty", snapBufItems}...)
+			}
+		}
+		if trieDiffNodes != 0 { // pathdb
+			context = append(context, []interface{}{"triediffs", trieDiffNodes}...)
+		}
+		context = append(context, []interface{}{"triedirty", triebufNodes}...)
 
 		if st.queued > 0 {
 			context = append(context, []interface{}{"queued", st.queued}...)

core/blockchain_test.go

@@ -1844,7 +1844,7 @@ func TestTrieForkGC(t *testing.T) {
 		chain.TrieDB().Dereference(blocks[len(blocks)-1-i].Root())
 		chain.TrieDB().Dereference(forks[len(blocks)-1-i].Root())
 	}
-	if nodes, _ := chain.TrieDB().Size(); nodes > 0 {
+	if _, nodes, _ := chain.TrieDB().Size(); nodes > 0 { // all memory is returned in the nodes return for hashdb
 		t.Fatalf("stale tries still alive after garbase collection")
 	}
 }

core/state/snapshot/snapshot.go

@@ -852,3 +852,21 @@ func (t *Tree) DiskRoot() common.Hash {
 
 	return t.diskRoot()
 }
+
+// Size returns the memory usage of the diff layers above the disk layer and the
+// dirty nodes buffered in the disk layer. Currently, the implementation uses a
+// special diff layer (the first) as an aggregator simulating a dirty buffer, so
+// the second return will always be 0. However, this will be made consistent with
+// the pathdb, which will require a second return.
+func (t *Tree) Size() (diffs common.StorageSize, buf common.StorageSize) {
+	t.lock.RLock()
+	defer t.lock.RUnlock()
+
+	var size common.StorageSize
+	for _, layer := range t.layers {
+		if layer, ok := layer.(*diffLayer); ok {
+			size += common.StorageSize(layer.memory)
+		}
+	}
+	return size, 0
+}

eth/state_accessor.go

@@ -168,7 +168,7 @@ func (eth *Ethereum) hashState(ctx context.Context, block *types.Block, reexec u
 		parent = root
 	}
 	if report {
-		nodes, imgs := triedb.Size()
+		_, nodes, imgs := triedb.Size() // all memory is contained within the nodes return in hashdb
 		log.Info("Historical state regenerated", "block", current.NumberU64(), "elapsed", time.Since(start), "nodes", nodes, "preimages", imgs)
 	}
 	return statedb, func() { triedb.Dereference(block.Root()) }, nil

eth/tracers/api.go

@@ -369,8 +369,8 @@ func (api *API) traceChain(start, end *types.Block, config *TraceConfig, closed
 			// if the relevant state is available in disk.
 			var preferDisk bool
 			if statedb != nil {
-				s1, s2 := statedb.Database().TrieDB().Size()
-				preferDisk = s1+s2 > defaultTracechainMemLimit
+				s1, s2, s3 := statedb.Database().TrieDB().Size()
+				preferDisk = s1+s2+s3 > defaultTracechainMemLimit
 			}
 			statedb, release, err = api.backend.StateAtBlock(ctx, block, reexec, statedb, false, preferDisk)
 			if err != nil {

trie/database.go

@@ -55,9 +55,12 @@ type backend interface {
 	// according to the state scheme.
 	Initialized(genesisRoot common.Hash) bool
 
-	// Size returns the current storage size of the memory cache in front of the
-	// persistent database layer.
-	Size() common.StorageSize
+	// Size returns the current storage size of the diff layers on top of the
+	// disk layer and the storage size of the nodes cached in the disk layer.
+	//
+	// For hash scheme, there is no differentiation between diff layer nodes
+	// and dirty disk layer nodes, so both are merged into the second return.
+	Size() (common.StorageSize, common.StorageSize)
 
 	// Update performs a state transition by committing dirty nodes contained
 	// in the given set in order to update state from the specified parent to
@@ -165,18 +168,19 @@ func (db *Database) Commit(root common.Hash, report bool) error {
 	return db.backend.Commit(root, report)
 }
 
-// Size returns the storage size of dirty trie nodes in front of the persistent
-// database and the size of cached preimages.
-func (db *Database) Size() (common.StorageSize, common.StorageSize) {
+// Size returns the storage size of diff layer nodes above the persistent disk
+// layer, the dirty nodes buffered within the disk layer, and the size of cached
+// preimages.
+func (db *Database) Size() (common.StorageSize, common.StorageSize, common.StorageSize) {
 	var (
-		storages  common.StorageSize
-		preimages common.StorageSize
+		diffs, nodes common.StorageSize
+		preimages    common.StorageSize
 	)
-	storages = db.backend.Size()
+	diffs, nodes = db.backend.Size()
 	if db.preimages != nil {
 		preimages = db.preimages.size()
 	}
-	return storages, preimages
+	return diffs, nodes, preimages
 }
 
 // Initialized returns an indicator if the state data is already initialized

trie/triedb/hashdb/database.go

@@ -624,7 +624,10 @@ func (db *Database) Update(root common.Hash, parent common.Hash, block uint64, n
 
 // Size returns the current storage size of the memory cache in front of the
 // persistent database layer.
-func (db *Database) Size() common.StorageSize {
+//
+// The first return will always be 0, representing the memory stored in unbounded
+// diff layers above the dirty cache. This is only available in pathdb.
+func (db *Database) Size() (common.StorageSize, common.StorageSize) {
 	db.lock.RLock()
 	defer db.lock.RUnlock()
 
@@ -632,7 +635,7 @@ func (db *Database) Size() common.StorageSize {
 	// the total memory consumption, the maintenance metadata is also needed to be
 	// counted.
 	var metadataSize = common.StorageSize(len(db.dirties) * cachedNodeSize)
-	return db.dirtiesSize + db.childrenSize + metadataSize
+	return 0, db.dirtiesSize + db.childrenSize + metadataSize
 }
 
 // Close closes the trie database and releases all held resources.

trie/triedb/pathdb/database.go

@@ -383,16 +383,16 @@ func (db *Database) Close() error {
 
 // Size returns the current storage size of the memory cache in front of the
 // persistent database layer.
-func (db *Database) Size() (size common.StorageSize) {
+func (db *Database) Size() (diffs common.StorageSize, nodes common.StorageSize) {
 	db.tree.forEach(func(layer layer) {
 		if diff, ok := layer.(*diffLayer); ok {
-			size += common.StorageSize(diff.memory)
+			diffs += common.StorageSize(diff.memory)
 		}
 		if disk, ok := layer.(*diskLayer); ok {
-			size += disk.size()
+			nodes += disk.size()
 		}
 	})
-	return size
+	return diffs, nodes
 }
 
 // Initialized returns an indicator if the state data is already