eth, les: rework chain tracer (ethereum#25143)

This PR simplifies the logic of chain tracer and also adds the unit tests.

The most important change has been made in this PR is the state management. Whenever a tracing state is acquired there is a corresponding release function be returned as well. It must be called once the state is used up, otherwise resource leaking can happen.

And also the logic of state management has been simplified a lot. Specifically, the state provider(eth backend, les backend) should ensure the state is available and referenced. State customers can use the state according to their own needs, or build other states based on the given state. But once the release function is called, there is no guarantee of the availability of the state.


Co-authored-by: Sina Mahmoodi <1591639+s1na@users.noreply.github.com>
Co-authored-by: Péter Szilágyi <peterke@gmail.com>

eth/api.go

@@ -411,10 +411,12 @@ func (api *DebugAPI) StorageRangeAt(blockHash common.Hash, txIndex int, contract
 	if block == nil {
 		return StorageRangeResult{}, fmt.Errorf("block %#x not found", blockHash)
 	}
-	_, _, statedb, err := api.eth.stateAtTransaction(block, txIndex, 0)
+	_, _, statedb, release, err := api.eth.stateAtTransaction(block, txIndex, 0)
 	if err != nil {
 		return StorageRangeResult{}, err
 	}
+	defer release()
+
 	st := statedb.StorageTrie(contractAddress)
 	if st == nil {
 		return StorageRangeResult{}, fmt.Errorf("account %x doesn't exist", contractAddress)

eth/api_backend.go

@@ -33,6 +33,7 @@ import (
 	"github.com/ethereum/go-ethereum/core/types"
 	"github.com/ethereum/go-ethereum/core/vm"
 	"github.com/ethereum/go-ethereum/eth/gasprice"
+	"github.com/ethereum/go-ethereum/eth/tracers"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/event"
 	"github.com/ethereum/go-ethereum/miner"
@@ -363,10 +364,10 @@ func (b *EthAPIBackend) StartMining(threads int) error {
 	return b.eth.StartMining(threads)
 }
 
-func (b *EthAPIBackend) StateAtBlock(ctx context.Context, block *types.Block, reexec uint64, base *state.StateDB, checkLive, preferDisk bool) (*state.StateDB, error) {
-	return b.eth.StateAtBlock(block, reexec, base, checkLive, preferDisk)
+func (b *EthAPIBackend) StateAtBlock(ctx context.Context, block *types.Block, reexec uint64, base *state.StateDB, readOnly bool, preferDisk bool) (*state.StateDB, tracers.StateReleaseFunc, error) {
+	return b.eth.StateAtBlock(block, reexec, base, readOnly, preferDisk)
 }
 
-func (b *EthAPIBackend) StateAtTransaction(ctx context.Context, block *types.Block, txIndex int, reexec uint64) (core.Message, vm.BlockContext, *state.StateDB, error) {
+func (b *EthAPIBackend) StateAtTransaction(ctx context.Context, block *types.Block, txIndex int, reexec uint64) (core.Message, vm.BlockContext, *state.StateDB, tracers.StateReleaseFunc, error) {
 	return b.eth.stateAtTransaction(block, txIndex, reexec)
 }

eth/state_accessor.go

@@ -26,77 +26,97 @@ import (
 	"github.com/ethereum/go-ethereum/core/state"
 	"github.com/ethereum/go-ethereum/core/types"
 	"github.com/ethereum/go-ethereum/core/vm"
+	"github.com/ethereum/go-ethereum/eth/tracers"
 	"github.com/ethereum/go-ethereum/log"
 	"github.com/ethereum/go-ethereum/trie"
 )
 
+// noopReleaser is returned in case there is no operation expected
+// for releasing state.
+var noopReleaser = tracers.StateReleaseFunc(func() {})
+
 // StateAtBlock retrieves the state database associated with a certain block.
 // If no state is locally available for the given block, a number of blocks
 // are attempted to be reexecuted to generate the desired state. The optional
-// base layer statedb can be passed then it's regarded as the statedb of the
+// base layer statedb can be provided which is regarded as the statedb of the
 // parent block.
+//
+// An additional release function will be returned if the requested state is
+// available. Release is expected to be invoked when the returned state is no longer needed.
+// Its purpose is to prevent resource leaking. Though it can be noop in some cases.
+//
 // Parameters:
-// - block: The block for which we want the state (== state at the stateRoot of the parent)
-// - reexec: The maximum number of blocks to reprocess trying to obtain the desired state
-// - base: If the caller is tracing multiple blocks, the caller can provide the parent state
-//         continuously from the callsite.
-// - checklive: if true, then the live 'blockchain' state database is used. If the caller want to
-//        perform Commit or other 'save-to-disk' changes, this should be set to false to avoid
-//        storing trash persistently
-// - preferDisk: this arg can be used by the caller to signal that even though the 'base' is provided,
-//        it would be preferable to start from a fresh state, if we have it on disk.
-func (eth *Ethereum) StateAtBlock(block *types.Block, reexec uint64, base *state.StateDB, checkLive bool, preferDisk bool) (statedb *state.StateDB, err error) {
+// - block:      The block for which we want the state(state = block.Root)
+// - reexec:     The maximum number of blocks to reprocess trying to obtain the desired state
+// - base:       If the caller is tracing multiple blocks, the caller can provide the parent
+//               state continuously from the callsite.
+// - readOnly:   If true, then the live 'blockchain' state database is used. No mutation should
+//               be made from caller, e.g. perform Commit or other 'save-to-disk' changes.
+//               Otherwise, the trash generated by caller may be persisted permanently.
+// - preferDisk: this arg can be used by the caller to signal that even though the 'base' is
+//               provided, it would be preferable to start from a fresh state, if we have it
+//               on disk.
+func (eth *Ethereum) StateAtBlock(block *types.Block, reexec uint64, base *state.StateDB, readOnly bool, preferDisk bool) (statedb *state.StateDB, release tracers.StateReleaseFunc, err error) {
 	var (
 		current  *types.Block
 		database state.Database
 		report   = true
 		origin   = block.NumberU64()
 	)
-	// Check the live database first if we have the state fully available, use that.
-	if checkLive {
-		statedb, err = eth.blockchain.StateAt(block.Root())
-		if err == nil {
-			return statedb, nil
+	// The state is only for reading purposes, check the state presence in
+	// live database.
+	if readOnly {
+		// The state is available in live database, create a reference
+		// on top to prevent garbage collection and return a release
+		// function to deref it.
+		if statedb, err = eth.blockchain.StateAt(block.Root()); err == nil {
+			statedb.Database().TrieDB().Reference(block.Root(), common.Hash{})
+			return statedb, func() {
+				statedb.Database().TrieDB().Dereference(block.Root())
+			}, nil
 		}
 	}
+	// The state is both for reading and writing, or it's unavailable in disk,
+	// try to construct/recover the state over an ephemeral trie.Database for
+	// isolating the live one.
 	if base != nil {
 		if preferDisk {
 			// Create an ephemeral trie.Database for isolating the live one. Otherwise
 			// the internal junks created by tracing will be persisted into the disk.
 			database = state.NewDatabaseWithConfig(eth.chainDb, &trie.Config{Cache: 16})
 			if statedb, err = state.New(block.Root(), database, nil); err == nil {
 				log.Info("Found disk backend for state trie", "root", block.Root(), "number", block.Number())
-				return statedb, nil
+				return statedb, noopReleaser, nil
 			}
 		}
 		// The optional base statedb is given, mark the start point as parent block
 		statedb, database, report = base, base.Database(), false
 		current = eth.blockchain.GetBlock(block.ParentHash(), block.NumberU64()-1)
 	} else {
-		// Otherwise try to reexec blocks until we find a state or reach our limit
+		// Otherwise, try to reexec blocks until we find a state or reach our limit
 		current = block
 
 		// Create an ephemeral trie.Database for isolating the live one. Otherwise
 		// the internal junks created by tracing will be persisted into the disk.
 		database = state.NewDatabaseWithConfig(eth.chainDb, &trie.Config{Cache: 16})
 
-		// If we didn't check the dirty database, do check the clean one, otherwise
-		// we would rewind past a persisted block (specific corner case is chain
-		// tracing from the genesis).
-		if !checkLive {
+		// If we didn't check the live database, do check state over ephemeral database,
+		// otherwise we would rewind past a persisted block (specific corner case is
+		// chain tracing from the genesis).
+		if !readOnly {
 			statedb, err = state.New(current.Root(), database, nil)
 			if err == nil {
-				return statedb, nil
+				return statedb, noopReleaser, nil
 			}
 		}
 		// Database does not have the state for the given block, try to regenerate
 		for i := uint64(0); i < reexec; i++ {
 			if current.NumberU64() == 0 {
-				return nil, errors.New("genesis state is missing")
+				return nil, nil, errors.New("genesis state is missing")
 			}
 			parent := eth.blockchain.GetBlock(current.ParentHash(), current.NumberU64()-1)
 			if parent == nil {
-				return nil, fmt.Errorf("missing block %v %d", current.ParentHash(), current.NumberU64()-1)
+				return nil, nil, fmt.Errorf("missing block %v %d", current.ParentHash(), current.NumberU64()-1)
 			}
 			current = parent
 
@@ -108,13 +128,14 @@ func (eth *Ethereum) StateAtBlock(block *types.Block, reexec uint64, base *state
 		if err != nil {
 			switch err.(type) {
 			case *trie.MissingNodeError:
-				return nil, fmt.Errorf("required historical state unavailable (reexec=%d)", reexec)
+				return nil, nil, fmt.Errorf("required historical state unavailable (reexec=%d)", reexec)
 			default:
-				return nil, err
+				return nil, nil, err
 			}
 		}
 	}
-	// State was available at historical point, regenerate
+	// State is available at historical point, re-execute the blocks on top for
+	// the desired state.
 	var (
 		start  = time.Now()
 		logged time.Time
@@ -129,22 +150,24 @@ func (eth *Ethereum) StateAtBlock(block *types.Block, reexec uint64, base *state
 		// Retrieve the next block to regenerate and process it
 		next := current.NumberU64() + 1
 		if current = eth.blockchain.GetBlockByNumber(next); current == nil {
-			return nil, fmt.Errorf("block #%d not found", next)
+			return nil, nil, fmt.Errorf("block #%d not found", next)
 		}
 		_, _, _, err := eth.blockchain.Processor().Process(current, statedb, vm.Config{})
 		if err != nil {
-			return nil, fmt.Errorf("processing block %d failed: %v", current.NumberU64(), err)
+			return nil, nil, fmt.Errorf("processing block %d failed: %v", current.NumberU64(), err)
 		}
 		// Finalize the state so any modifications are written to the trie
 		root, err := statedb.Commit(eth.blockchain.Config().IsEIP158(current.Number()))
 		if err != nil {
-			return nil, fmt.Errorf("stateAtBlock commit failed, number %d root %v: %w",
+			return nil, nil, fmt.Errorf("stateAtBlock commit failed, number %d root %v: %w",
 				current.NumberU64(), current.Root().Hex(), err)
 		}
 		statedb, err = state.New(root, database, nil)
 		if err != nil {
-			return nil, fmt.Errorf("state reset after block %d failed: %v", current.NumberU64(), err)
+			return nil, nil, fmt.Errorf("state reset after block %d failed: %v", current.NumberU64(), err)
 		}
+		// Hold the state reference and also drop the parent state
+		// to prevent accumulating too many nodes in memory.
 		database.TrieDB().Reference(root, common.Hash{})
 		if parent != (common.Hash{}) {
 			database.TrieDB().Dereference(parent)
@@ -155,28 +178,28 @@ func (eth *Ethereum) StateAtBlock(block *types.Block, reexec uint64, base *state
 		nodes, imgs := database.TrieDB().Size()
 		log.Info("Historical state regenerated", "block", current.NumberU64(), "elapsed", time.Since(start), "nodes", nodes, "preimages", imgs)
 	}
-	return statedb, nil
+	return statedb, func() { database.TrieDB().Dereference(block.Root()) }, nil
 }
 
 // stateAtTransaction returns the execution environment of a certain transaction.
-func (eth *Ethereum) stateAtTransaction(block *types.Block, txIndex int, reexec uint64) (core.Message, vm.BlockContext, *state.StateDB, error) {
+func (eth *Ethereum) stateAtTransaction(block *types.Block, txIndex int, reexec uint64) (core.Message, vm.BlockContext, *state.StateDB, tracers.StateReleaseFunc, error) {
 	// Short circuit if it's genesis block.
 	if block.NumberU64() == 0 {
-		return nil, vm.BlockContext{}, nil, errors.New("no transaction in genesis")
+		return nil, vm.BlockContext{}, nil, nil, errors.New("no transaction in genesis")
 	}
 	// Create the parent state database
 	parent := eth.blockchain.GetBlock(block.ParentHash(), block.NumberU64()-1)
 	if parent == nil {
-		return nil, vm.BlockContext{}, nil, fmt.Errorf("parent %#x not found", block.ParentHash())
+		return nil, vm.BlockContext{}, nil, nil, fmt.Errorf("parent %#x not found", block.ParentHash())
 	}
 	// Lookup the statedb of parent block from the live database,
 	// otherwise regenerate it on the flight.
-	statedb, err := eth.StateAtBlock(parent, reexec, nil, true, false)
+	statedb, release, err := eth.StateAtBlock(parent, reexec, nil, true, false)
 	if err != nil {
-		return nil, vm.BlockContext{}, nil, err
+		return nil, vm.BlockContext{}, nil, nil, err
 	}
 	if txIndex == 0 && len(block.Transactions()) == 0 {
-		return nil, vm.BlockContext{}, statedb, nil
+		return nil, vm.BlockContext{}, statedb, release, nil
 	}
 	// Recompute transactions up to the target index.
 	signer := types.MakeSigner(eth.blockchain.Config(), block.Number())
@@ -186,17 +209,17 @@ func (eth *Ethereum) stateAtTransaction(block *types.Block, txIndex int, reexec
 		txContext := core.NewEVMTxContext(msg)
 		context := core.NewEVMBlockContext(block.Header(), eth.blockchain, nil)
 		if idx == txIndex {
-			return msg, context, statedb, nil
+			return msg, context, statedb, release, nil
 		}
 		// Not yet the searched for transaction, execute on top of the current state
 		vmenv := vm.NewEVM(context, txContext, statedb, eth.blockchain.Config(), vm.Config{})
 		statedb.Prepare(tx.Hash(), idx)
 		if _, err := core.ApplyMessage(vmenv, msg, new(core.GasPool).AddGas(tx.Gas())); err != nil {
-			return nil, vm.BlockContext{}, nil, fmt.Errorf("transaction %#x failed: %v", tx.Hash(), err)
+			return nil, vm.BlockContext{}, nil, nil, fmt.Errorf("transaction %#x failed: %v", tx.Hash(), err)
 		}
 		// Ensure any modifications are committed to the state
 		// Only delete empty objects if EIP158/161 (a.k.a Spurious Dragon) is in effect
 		statedb.Finalise(vmenv.ChainConfig().IsEIP158(block.Number()))
 	}
-	return nil, vm.BlockContext{}, nil, fmt.Errorf("transaction index %d out of range for block %#x", txIndex, block.Hash())
+	return nil, vm.BlockContext{}, nil, nil, fmt.Errorf("transaction index %d out of range for block %#x", txIndex, block.Hash())
 }