freezer.go

// Copyright 2019 The go-ethereum Authors
// This file is part of the go-ethereum library.
//
// The go-ethereum library is free software: you can redistribute it and/or modify
// it under the terms of the GNU Lesser General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// The go-ethereum library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public License
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.

package rawdb

import (
	"errors"
	"fmt"
	"io/ioutil"
	"math"
	"os"
	"path/filepath"
	"sync"
	"sync/atomic"
	"time"

	"github.com/ethereum/go-ethereum/common"
	"github.com/ethereum/go-ethereum/ethdb"
	"github.com/ethereum/go-ethereum/log"
	"github.com/ethereum/go-ethereum/metrics"
	"github.com/ethereum/go-ethereum/params"
	"github.com/prometheus/tsdb/fileutil"
)

var (
	// errReadOnly is returned if the freezer is opened in read only mode. All the
	// mutations are disallowed.
	errReadOnly = errors.New("read only")

	// errUnknownTable is returned if the user attempts to read from a table that is
	// not tracked by the freezer.
	errUnknownTable = errors.New("unknown table")

	// errOutOrderInsertion is returned if the user attempts to inject out-of-order
	// binary blobs into the freezer.
	errOutOrderInsertion = errors.New("the append operation is out-order")

	// errSymlinkDatadir is returned if the ancient directory specified by user
	// is a symbolic link.
	errSymlinkDatadir = errors.New("symbolic link datadir is not supported")
)

const (
	// freezerRecheckInterval is the frequency to check the key-value database for
	// chain progression that might permit new blocks to be frozen into immutable
	// storage.
	freezerRecheckInterval = time.Minute

	// freezerBatchLimit is the maximum number of blocks to freeze in one batch
	// before doing an fsync and deleting it from the key-value store.
	freezerBatchLimit = 30000

	// freezerTableSize defines the maximum size of freezer data files.
	freezerTableSize = 2 * 1000 * 1000 * 1000
)

// freezer is a memory mapped append-only database to store immutable chain data
// into flat files:
//
// - The append only nature ensures that disk writes are minimized.
// - The memory mapping ensures we can max out system memory for caching without
//   reserving it for go-ethereum. This would also reduce the memory requirements
//   of Geth, and thus also GC overhead.
type freezer struct {
	// WARNING: The `frozen` field is accessed atomically. On 32 bit platforms, only
	// 64-bit aligned fields can be atomic. The struct is guaranteed to be so aligned,
	// so take advantage of that (https://golang.org/pkg/sync/atomic/#pkg-note-BUG).
	frozen    uint64 // Number of blocks already frozen
	tail      uint64 // Number of the first stored item in the freezer
	threshold uint64 // Number of recent blocks not to freeze (params.FullImmutabilityThreshold apart from tests)

	// This lock synchronizes writers and the truncate operation, as well as
	// the "atomic" (batched) read operations.
	writeLock  sync.RWMutex
	writeBatch *freezerBatch

	readonly     bool
	tables       map[string]*freezerTable // Data tables for storing everything
	instanceLock fileutil.Releaser        // File-system lock to prevent double opens

	trigger chan chan struct{} // Manual blocking freeze trigger, test determinism

	quit      chan struct{}
	wg        sync.WaitGroup
	closeOnce sync.Once
}

// newFreezer creates a chain freezer that moves ancient chain data into
// append-only flat file containers.
//
// The 'tables' argument defines the data tables. If the value of a map
// entry is true, snappy compression is disabled for the table.
func newFreezer(datadir string, namespace string, readonly bool, maxTableSize uint32, tables map[string]bool) (*freezer, error) {
	// Create the initial freezer object
	var (
		readMeter  = metrics.NewRegisteredMeter(namespace+"ancient/read", nil)
		writeMeter = metrics.NewRegisteredMeter(namespace+"ancient/write", nil)
		sizeGauge  = metrics.NewRegisteredGauge(namespace+"ancient/size", nil)
	)
	// Ensure the datadir is not a symbolic link if it exists.
	if info, err := os.Lstat(datadir); !os.IsNotExist(err) {
		if info.Mode()&os.ModeSymlink != 0 {
			log.Warn("Symbolic link ancient database is not supported", "path", datadir)
			return nil, errSymlinkDatadir
		}
	}
	// Leveldb uses LOCK as the filelock filename. To prevent the
	// name collision, we use FLOCK as the lock name.
	lock, _, err := fileutil.Flock(filepath.Join(datadir, "FLOCK"))
	if err != nil {
		return nil, err
	}
	// Open all the supported data tables
	freezer := &freezer{
		readonly:     readonly,
		threshold:    params.FullImmutabilityThreshold,
		tables:       make(map[string]*freezerTable),
		instanceLock: lock,
		trigger:      make(chan chan struct{}),
		quit:         make(chan struct{}),
	}

	// Create the tables.
	for name, disableSnappy := range tables {
		table, err := newTable(datadir, name, readMeter, writeMeter, sizeGauge, maxTableSize, disableSnappy, readonly)
		if err != nil {
			for _, table := range freezer.tables {
				table.Close()
			}
			lock.Release()
			return nil, err
		}
		freezer.tables[name] = table
	}

	if freezer.readonly {
		// In readonly mode only validate, don't truncate.
		// validate also sets `freezer.frozen`.
		err = freezer.validate()
	} else {
		// Truncate all tables to common length.
		err = freezer.repair()
	}
	if err != nil {
		for _, table := range freezer.tables {
			table.Close()
		}
		lock.Release()
		return nil, err
	}

	// Create the write batch.
	freezer.writeBatch = newFreezerBatch(freezer)

	log.Info("Opened ancient database", "database", datadir, "readonly", readonly)
	return freezer, nil
}

// Close terminates the chain freezer, unmapping all the data files.
func (f *freezer) Close() error {
	f.writeLock.Lock()
	defer f.writeLock.Unlock()

	var errs []error
	f.closeOnce.Do(func() {
		close(f.quit)
		// Wait for any background freezing to stop
		f.wg.Wait()
		for _, table := range f.tables {
			if err := table.Close(); err != nil {
				errs = append(errs, err)
			}
		}
		if err := f.instanceLock.Release(); err != nil {
			errs = append(errs, err)
		}
	})
	if errs != nil {
		return fmt.Errorf("%v", errs)
	}
	return nil
}

// HasAncient returns an indicator whether the specified ancient data exists
// in the freezer.
func (f *freezer) HasAncient(kind string, number uint64) (bool, error) {
	if table := f.tables[kind]; table != nil {
		return table.has(number), nil
	}
	return false, nil
}

// Ancient retrieves an ancient binary blob from the append-only immutable files.
func (f *freezer) Ancient(kind string, number uint64) ([]byte, error) {
	if table := f.tables[kind]; table != nil {
		return table.Retrieve(number)
	}
	return nil, errUnknownTable
}

// AncientRange retrieves multiple items in sequence, starting from the index 'start'.
// It will return
//  - at most 'max' items,
//  - at least 1 item (even if exceeding the maxByteSize), but will otherwise
//   return as many items as fit into maxByteSize.
func (f *freezer) AncientRange(kind string, start, count, maxBytes uint64) ([][]byte, error) {
	if table := f.tables[kind]; table != nil {
		return table.RetrieveItems(start, count, maxBytes)
	}
	return nil, errUnknownTable
}

// Ancients returns the length of the frozen items.
func (f *freezer) Ancients() (uint64, error) {
	return atomic.LoadUint64(&f.frozen), nil
}

// Tail returns the number of first stored item in the freezer.
func (f *freezer) Tail() (uint64, error) {
	return atomic.LoadUint64(&f.tail), nil
}

// AncientSize returns the ancient size of the specified category.
func (f *freezer) AncientSize(kind string) (uint64, error) {
	// This needs the write lock to avoid data races on table fields.
	// Speed doesn't matter here, AncientSize is for debugging.
	f.writeLock.RLock()
	defer f.writeLock.RUnlock()

	if table := f.tables[kind]; table != nil {
		return table.size()
	}
	return 0, errUnknownTable
}

// ReadAncients runs the given read operation while ensuring that no writes take place
// on the underlying freezer.
func (f *freezer) ReadAncients(fn func(ethdb.AncientReader) error) (err error) {
	f.writeLock.RLock()
	defer f.writeLock.RUnlock()
	return fn(f)
}

// ModifyAncients runs the given write operation.
func (f *freezer) ModifyAncients(fn func(ethdb.AncientWriteOp) error) (writeSize int64, err error) {
	if f.readonly {
		return 0, errReadOnly
	}
	f.writeLock.Lock()
	defer f.writeLock.Unlock()

	// Roll back all tables to the starting position in case of error.
	prevItem := f.frozen
	defer func() {
		if err != nil {
			// The write operation has failed. Go back to the previous item position.
			for name, table := range f.tables {
				err := table.truncateHead(prevItem)
				if err != nil {
					log.Error("Freezer table roll-back failed", "table", name, "index", prevItem, "err", err)
				}
			}
		}
	}()

	f.writeBatch.reset()
	if err := fn(f.writeBatch); err != nil {
		return 0, err
	}
	item, writeSize, err := f.writeBatch.commit()
	if err != nil {
		return 0, err
	}
	atomic.StoreUint64(&f.frozen, item)
	return writeSize, nil
}

// TruncateHead discards any recent data above the provided threshold number.
func (f *freezer) TruncateHead(items uint64) error {
	if f.readonly {
		return errReadOnly
	}
	f.writeLock.Lock()
	defer f.writeLock.Unlock()

	if atomic.LoadUint64(&f.frozen) <= items {
		return nil
	}
	for _, table := range f.tables {
		if err := table.truncateHead(items); err != nil {
			return err
		}
	}
	atomic.StoreUint64(&f.frozen, items)
	return nil
}

// TruncateTail discards any recent data below the provided threshold number.
func (f *freezer) TruncateTail(tail uint64) error {
	if f.readonly {
		return errReadOnly
	}
	f.writeLock.Lock()
	defer f.writeLock.Unlock()

	if atomic.LoadUint64(&f.tail) >= tail {
		return nil
	}
	for _, table := range f.tables {
		if err := table.truncateTail(tail); err != nil {
			return err
		}
	}
	atomic.StoreUint64(&f.tail, tail)
	return nil
}

// Sync flushes all data tables to disk.
func (f *freezer) Sync() error {
	var errs []error
	for _, table := range f.tables {
		if err := table.Sync(); err != nil {
			errs = append(errs, err)
		}
	}
	if errs != nil {
		return fmt.Errorf("%v", errs)
	}
	return nil
}

// validate checks that every table has the same length.
// Used instead of `repair` in readonly mode.
func (f *freezer) validate() error {
	if len(f.tables) == 0 {
		return nil
	}
	var (
		length uint64
		name   string
	)
	// Hack to get length of any table
	for kind, table := range f.tables {
		length = atomic.LoadUint64(&table.items)
		name = kind
		break
	}
	// Now check every table against that length
	for kind, table := range f.tables {
		items := atomic.LoadUint64(&table.items)
		if length != items {
			return fmt.Errorf("freezer tables %s and %s have differing lengths: %d != %d", kind, name, items, length)
		}
	}
	atomic.StoreUint64(&f.frozen, length)
	return nil
}

// repair truncates all data tables to the same length.
func (f *freezer) repair() error {
	var (
		head = uint64(math.MaxUint64)
		tail = uint64(0)
	)
	for _, table := range f.tables {
		items := atomic.LoadUint64(&table.items)
		if head > items {
			head = items
		}
		hidden := atomic.LoadUint64(&table.itemHidden)
		if hidden > tail {
			tail = hidden
		}
	}
	for _, table := range f.tables {
		if err := table.truncateHead(head); err != nil {
			return err
		}
		if err := table.truncateTail(tail); err != nil {
			return err
		}
	}
	atomic.StoreUint64(&f.frozen, head)
	atomic.StoreUint64(&f.tail, tail)
	return nil
}

// freeze is a background thread that periodically checks the blockchain for any
// import progress and moves ancient data from the fast database into the freezer.
//
// This functionality is deliberately broken off from block importing to avoid
// incurring additional data shuffling delays on block propagation.
func (f *freezer) freeze(db ethdb.KeyValueStore) {
	nfdb := &nofreezedb{KeyValueStore: db}

	var (
		backoff   bool
		triggered chan struct{} // Used in tests
	)
	for {
		select {
		case <-f.quit:
			log.Info("Freezer shutting down")
			return
		default:
		}
		if backoff {
			// If we were doing a manual trigger, notify it
			if triggered != nil {
				triggered <- struct{}{}
				triggered = nil
			}
			select {
			case <-time.NewTimer(freezerRecheckInterval).C:
				backoff = false
			case triggered = <-f.trigger:
				backoff = false
			case <-f.quit:
				return
			}
		}
		// Retrieve the freezing threshold.
		hash := ReadHeadBlockHash(nfdb)
		if hash == (common.Hash{}) {
			log.Debug("Current full block hash unavailable") // new chain, empty database
			backoff = true
			continue
		}
		number := ReadHeaderNumber(nfdb, hash)
		threshold := atomic.LoadUint64(&f.threshold)

		switch {
		case number == nil:
			log.Error("Current full block number unavailable", "hash", hash)
			backoff = true
			continue

		case *number < threshold:
			log.Debug("Current full block not old enough", "number", *number, "hash", hash, "delay", threshold)
			backoff = true
			continue

		case *number-threshold <= f.frozen:
			log.Debug("Ancient blocks frozen already", "number", *number, "hash", hash, "frozen", f.frozen)
			backoff = true
			continue
		}
		head := ReadHeader(nfdb, hash, *number)
		if head == nil {
			log.Error("Current full block unavailable", "number", *number, "hash", hash)
			backoff = true
			continue
		}

		// Seems we have data ready to be frozen, process in usable batches
		var (
			start    = time.Now()
			first, _ = f.Ancients()
			limit    = *number - threshold
		)
		if limit-first > freezerBatchLimit {
			limit = first + freezerBatchLimit
		}
		ancients, err := f.freezeRange(nfdb, first, limit)
		if err != nil {
			log.Error("Error in block freeze operation", "err", err)
			backoff = true
			continue
		}

		// Batch of blocks have been frozen, flush them before wiping from leveldb
		if err := f.Sync(); err != nil {
			log.Crit("Failed to flush frozen tables", "err", err)
		}

		// Wipe out all data from the active database
		batch := db.NewBatch()
		for i := 0; i < len(ancients); i++ {
			// Always keep the genesis block in active database
			if first+uint64(i) != 0 {
				DeleteBlockWithoutNumber(batch, ancients[i], first+uint64(i))
				DeleteCanonicalHash(batch, first+uint64(i))
			}
		}
		if err := batch.Write(); err != nil {
			log.Crit("Failed to delete frozen canonical blocks", "err", err)
		}
		batch.Reset()

		// Wipe out side chains also and track dangling side chains
		var dangling []common.Hash
		for number := first; number < f.frozen; number++ {
			// Always keep the genesis block in active database
			if number != 0 {
				dangling = ReadAllHashes(db, number)
				for _, hash := range dangling {
					log.Trace("Deleting side chain", "number", number, "hash", hash)
					DeleteBlock(batch, hash, number)
				}
			}
		}
		if err := batch.Write(); err != nil {
			log.Crit("Failed to delete frozen side blocks", "err", err)
		}
		batch.Reset()

		// Step into the future and delete and dangling side chains
		if f.frozen > 0 {
			tip := f.frozen
			for len(dangling) > 0 {
				drop := make(map[common.Hash]struct{})
				for _, hash := range dangling {
					log.Debug("Dangling parent from freezer", "number", tip-1, "hash", hash)
					drop[hash] = struct{}{}
				}
				children := ReadAllHashes(db, tip)
				for i := 0; i < len(children); i++ {
					// Dig up the child and ensure it's dangling
					child := ReadHeader(nfdb, children[i], tip)
					if child == nil {
						log.Error("Missing dangling header", "number", tip, "hash", children[i])
						continue
					}
					if _, ok := drop[child.ParentHash]; !ok {
						children = append(children[:i], children[i+1:]...)
						i--
						continue
					}
					// Delete all block data associated with the child
					log.Debug("Deleting dangling block", "number", tip, "hash", children[i], "parent", child.ParentHash)
					DeleteBlock(batch, children[i], tip)
				}
				dangling = children
				tip++
			}
			if err := batch.Write(); err != nil {
				log.Crit("Failed to delete dangling side blocks", "err", err)
			}
		}

		// Log something friendly for the user
		context := []interface{}{
			"blocks", f.frozen - first, "elapsed", common.PrettyDuration(time.Since(start)), "number", f.frozen - 1,
		}
		if n := len(ancients); n > 0 {
			context = append(context, []interface{}{"hash", ancients[n-1]}...)
		}
		log.Info("Deep froze chain segment", context...)

		// Avoid database thrashing with tiny writes
		if f.frozen-first < freezerBatchLimit {
			backoff = true
		}
	}
}

func (f *freezer) freezeRange(nfdb *nofreezedb, number, limit uint64) (hashes []common.Hash, err error) {
	hashes = make([]common.Hash, 0, limit-number)

	_, err = f.ModifyAncients(func(op ethdb.AncientWriteOp) error {
		for ; number <= limit; number++ {
			// Retrieve all the components of the canonical block.
			hash := ReadCanonicalHash(nfdb, number)
			if hash == (common.Hash{}) {
				return fmt.Errorf("canonical hash missing, can't freeze block %d", number)
			}
			header := ReadHeaderRLP(nfdb, hash, number)
			if len(header) == 0 {
				return fmt.Errorf("block header missing, can't freeze block %d", number)
			}
			body := ReadBodyRLP(nfdb, hash, number)
			if len(body) == 0 {
				return fmt.Errorf("block body missing, can't freeze block %d", number)
			}
			receipts := ReadReceiptsRLP(nfdb, hash, number)
			if len(receipts) == 0 {
				return fmt.Errorf("block receipts missing, can't freeze block %d", number)
			}
			td := ReadTdRLP(nfdb, hash, number)
			if len(td) == 0 {
				return fmt.Errorf("total difficulty missing, can't freeze block %d", number)
			}

			// Write to the batch.
			if err := op.AppendRaw(freezerHashTable, number, hash[:]); err != nil {
				return fmt.Errorf("can't write hash to freezer: %v", err)
			}
			if err := op.AppendRaw(freezerHeaderTable, number, header); err != nil {
				return fmt.Errorf("can't write header to freezer: %v", err)
			}
			if err := op.AppendRaw(freezerBodiesTable, number, body); err != nil {
				return fmt.Errorf("can't write body to freezer: %v", err)
			}
			if err := op.AppendRaw(freezerReceiptTable, number, receipts); err != nil {
				return fmt.Errorf("can't write receipts to freezer: %v", err)
			}
			if err := op.AppendRaw(freezerDifficultyTable, number, td); err != nil {
				return fmt.Errorf("can't write td to freezer: %v", err)
			}

			hashes = append(hashes, hash)
		}
		return nil
	})

	return hashes, err
}

// convertLegacyFn takes a raw freezer entry in an older format and
// returns it in the new format.
type convertLegacyFn = func([]byte) ([]byte, error)

// MigrateTable processes the entries in a given table in sequence
// converting them to a new format if they're of an old format.
func (f *freezer) MigrateTable(kind string, convert convertLegacyFn) error {
	if f.readonly {
		return errReadOnly
	}
	f.writeLock.Lock()
	defer f.writeLock.Unlock()

	table, ok := f.tables[kind]
	if !ok {
		return errUnknownTable
	}
	// forEach iterates every entry in the table serially and in order, calling `fn`
	// with the item as argument. If `fn` returns an error the iteration stops
	// and that error will be returned.
	forEach := func(t *freezerTable, fn func(uint64, []byte) error) error {
		var (
			items     = atomic.LoadUint64(&t.items)
			batchSize = uint64(1024)
			maxBytes  = uint64(1024 * 1024)
		)
		for i := uint64(0); i < items; {
			if i+batchSize > items {
				batchSize = items - i
			}
			data, err := t.RetrieveItems(i, batchSize, maxBytes)
			if err != nil {
				return err
			}
			for j, item := range data {
				if err := fn(i+uint64(j), item); err != nil {
					return err
				}
			}
			i += uint64(len(data))
		}
		return nil
	}
	if table.itemOffset > 0 || table.itemHidden > 0 {
		return fmt.Errorf("migration not supported for tail-deleted freezers")
	}
	ancientsPath := filepath.Dir(table.index.Name())
	// Set up new dir for the migrated table, the content of which
	// we'll at the end move over to the ancients dir.
	migrationPath := filepath.Join(ancientsPath, "migration")
	newTable, err := NewFreezerTable(migrationPath, kind, FreezerNoSnappy[kind], false)
	if err != nil {
		return err
	}
	var (
		batch  = newTable.newBatch()
		out    []byte
		start  = time.Now()
		logged = time.Now()
	)
	// Iterate through entries and transform them
	forEach(table, func(i uint64, blob []byte) error {
		if i%10000 == 0 && time.Since(logged) > 16*time.Second {
			log.Info("Processing legacy elements", "count", i, "elapsed", common.PrettyDuration(time.Since(start)))
			logged = time.Now()
		}
		out, err = convert(blob)
		if err != nil {
			return err
		}
		if err := batch.AppendRaw(i, out); err != nil {
			return err
		}
		return nil
	})
	if err := batch.commit(); err != nil {
		return err
	}
	log.Info("Replacing old table files with migrated ones", "elapsed", common.PrettyDuration(time.Since(start)))
	// Release and delete old table files. Note this won't
	// delete the index file.
	table.releaseFilesAfter(0, true)

	if err := newTable.Close(); err != nil {
		return err
	}
	files, err := ioutil.ReadDir(migrationPath)
	if err != nil {
		return err
	}
	// Move migrated files to ancients dir.
	for _, f := range files {
		// This will replace the old index file as a side-effect.
		if err := os.Rename(filepath.Join(migrationPath, f.Name()), filepath.Join(ancientsPath, f.Name())); err != nil {
			return err
		}
	}
	// Delete by now empty dir.
	if err := os.Remove(migrationPath); err != nil {
		return err
	}

	return nil
}