kvdb/leveldb/leveldb.go
// +build !js
// Package leveldb implements the key-value database layer based on LevelDB.
package leveldb
import (
"fmt"
"strconv"
"strings"
"sync"
"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/syndtr/goleveldb/leveldb"
"github.com/syndtr/goleveldb/leveldb/errors"
"github.com/syndtr/goleveldb/leveldb/filter"
"github.com/syndtr/goleveldb/leveldb/opt"
"github.com/syndtr/goleveldb/leveldb/util"
)
const (
// degradationWarnInterval specifies how often warning should be printed if the
// leveldb database cannot keep up with requested writes.
degradationWarnInterval = time.Minute
// minCache is the minimum amount of memory in megabytes to allocate to leveldb
// read and write caching, split half and half.
minCache = 16
// minHandles is the minimum number of files handles to allocate to the open
// database files.
minHandles = 16
// metricsGatheringInterval specifies the interval to retrieve leveldb database
// compaction, io and pause stats to report to the user.
metricsGatheringInterval = 3 * time.Second
)
// Database is a persistent key-value store. Apart from basic data storage
// functionality it also supports batch writes and iterating over the keyspace in
// binary-alphabetical order.
type Database struct {
fn string // filename for reporting
db *leveldb.DB // LevelDB instance
compTimeMeter metrics.Meter // Meter for measuring the total time spent in database compaction
compReadMeter metrics.Meter // Meter for measuring the data read during compaction
compWriteMeter metrics.Meter // Meter for measuring the data written during compaction
writeDelayNMeter metrics.Meter // Meter for measuring the write delay number due to database compaction
writeDelayMeter metrics.Meter // Meter for measuring the write delay duration due to database compaction
diskSizeGauge metrics.Gauge // Gauge for tracking the size of all the levels in the database
diskReadMeter metrics.Meter // Meter for measuring the effective amount of data read
diskWriteMeter metrics.Meter // Meter for measuring the effective amount of data written
quitLock sync.Mutex // Mutex protecting the quit channel access
quitChan chan chan error // Quit channel to stop the metrics collection before closing the database
log log.Logger // Contextual logger tracking the database path
onClose func() error
onDrop func()
}
// New returns a wrapped LevelDB object. The namespace is the prefix that the
// metrics reporting should use for surfacing internal stats.
func New(path string, cache int, handles int, namespace string, close func() error, drop func()) (*Database, error) {
// Ensure we have some minimal caching and file guarantees
if cache < minCache {
cache = minCache
}
if handles < minHandles {
handles = minHandles
}
logger := log.New("database", path)
logger.Info("Allocated cache and file handles", "cache", common.StorageSize(cache*1024*1024), "handles", handles)
// Open the db and recover any potential corruptions
db, err := leveldb.OpenFile(path, &opt.Options{
OpenFilesCacheCapacity: handles,
BlockCacheCapacity: cache / 2 * opt.MiB,
WriteBuffer: cache / 4 * opt.MiB, // Two of these are used internally
Filter: filter.NewBloomFilter(10),
})
if _, corrupted := err.(*errors.ErrCorrupted); corrupted {
db, err = leveldb.RecoverFile(path, nil)
}
if err != nil {
return nil, err
}
// Assemble the wrapper with all the registered metrics
ldb := &Database{
fn: path,
db: db,
log: logger,
quitChan: make(chan chan error),
}
ldb.compTimeMeter = metrics.NewRegisteredMeter(namespace+"compact/time", nil)
ldb.compReadMeter = metrics.NewRegisteredMeter(namespace+"compact/input", nil)
ldb.compWriteMeter = metrics.NewRegisteredMeter(namespace+"compact/output", nil)
ldb.diskSizeGauge = metrics.NewRegisteredGauge(namespace+"disk/size", nil)
ldb.diskReadMeter = metrics.NewRegisteredMeter(namespace+"disk/read", nil)
ldb.diskWriteMeter = metrics.NewRegisteredMeter(namespace+"disk/write", nil)
ldb.writeDelayMeter = metrics.NewRegisteredMeter(namespace+"compact/writedelay/duration", nil)
ldb.writeDelayNMeter = metrics.NewRegisteredMeter(namespace+"compact/writedelay/counter", nil)
ldb.onClose = close
ldb.onDrop = drop
// Start up the metrics gathering and return
go ldb.meter(metricsGatheringInterval)
return ldb, nil
}
// Close stops the metrics collection, flushes any pending data to disk and closes
// all io accesses to the underlying key-value store.
func (db *Database) Close() error {
db.quitLock.Lock()
defer db.quitLock.Unlock()
if db.db == nil {
panic("already closed")
}
if db.quitChan != nil {
errc := make(chan error)
db.quitChan <- errc
if err := <-errc; err != nil {
db.log.Error("Metrics collection failed", "err", err)
}
db.quitChan = nil
}
ldb := db.db
db.db = nil
if db.onClose != nil {
if err := db.onClose(); err != nil {
return err
}
db.onClose = nil
}
if err := ldb.Close(); err != nil {
return err
}
return nil
}
// Drop whole database.
func (db *Database) Drop() {
if db.db != nil {
panic("Close database first!")
}
if db.onDrop != nil {
db.onDrop()
}
}
// Has retrieves if a key is present in the key-value store.
func (db *Database) Has(key []byte) (bool, error) {
return db.db.Has(key, nil)
}
// Get retrieves the given key if it's present in the key-value store.
func (db *Database) Get(key []byte) ([]byte, error) {
dat, err := db.db.Get(key, nil)
if err != nil && err == leveldb.ErrNotFound {
return nil, nil
}
return dat, err
}
// Put inserts the given value into the key-value store.
func (db *Database) Put(key []byte, value []byte) error {
return db.db.Put(key, value, nil)
}
// Delete removes the key from the key-value store.
func (db *Database) Delete(key []byte) error {
return db.db.Delete(key, nil)
}
// NewBatch creates a write-only key-value store that buffers changes to its host
// database until a final write is called.
func (db *Database) NewBatch() ethdb.Batch {
return &batch{
db: db.db,
b: new(leveldb.Batch),
}
}
// NewIterator creates a binary-alphabetical iterator over a subset
// of database content with a particular key prefix, starting at a particular
// initial key (or after, if it does not exist).
func (db *Database) NewIterator(prefix []byte, start []byte) ethdb.Iterator {
return db.db.NewIterator(bytesPrefixRange(prefix, start), nil)
}
// Stat returns a particular internal stat of the database.
func (db *Database) Stat(property string) (string, error) {
return db.db.GetProperty(property)
}
// Compact flattens the underlying data store for the given key range. In essence,
// deleted and overwritten versions are discarded, and the data is rearranged to
// reduce the cost of operations needed to access them.
//
// A nil start is treated as a key before all keys in the data store; a nil limit
// is treated as a key after all keys in the data store. If both is nil then it
// will compact entire data store.
func (db *Database) Compact(start []byte, limit []byte) error {
return db.db.CompactRange(util.Range{Start: start, Limit: limit})
}
// Path returns the path to the database directory.
func (db *Database) Path() string {
return db.fn
}
// meter periodically retrieves internal leveldb counters and reports them to
// the metrics subsystem.
//
// This is how a LevelDB stats table looks like (currently):
// Compactions
// Level | Tables | Size(MB) | Time(sec) | Read(MB) | Write(MB)
// -------+------------+---------------+---------------+---------------+---------------
// 0 | 0 | 0.00000 | 1.27969 | 0.00000 | 12.31098
// 1 | 85 | 109.27913 | 28.09293 | 213.92493 | 214.26294
// 2 | 523 | 1000.37159 | 7.26059 | 66.86342 | 66.77884
// 3 | 570 | 1113.18458 | 0.00000 | 0.00000 | 0.00000
//
// This is how the write delay look like (currently):
// DelayN:5 Delay:406.604657ms Paused: false
//
// This is how the iostats look like (currently):
// Read(MB):3895.04860 Write(MB):3654.64712
func (db *Database) meter(refresh time.Duration) {
// Create the counters to store current and previous compaction values
compactions := make([][]float64, 2)
for i := 0; i < 2; i++ {
compactions[i] = make([]float64, 4)
}
// Create storage for iostats.
var iostats [2]float64
// Create storage and warning log tracer for write delay.
var (
delaystats [2]int64
lastWritePaused time.Time
)
var (
errc chan error
merr error
)
timer := time.NewTimer(refresh)
defer timer.Stop()
// Iterate ad infinitum and collect the stats
for i := 1; errc == nil && merr == nil; i++ {
// Retrieve the database stats
stats, err := db.db.GetProperty("leveldb.stats")
if err != nil {
db.log.Error("Failed to read database stats", "err", err)
merr = err
continue
}
// Find the compaction table, skip the header
lines := strings.Split(stats, "\n")
for len(lines) > 0 && strings.TrimSpace(lines[0]) != "Compactions" {
lines = lines[1:]
}
if len(lines) <= 3 {
db.log.Error("Compaction leveldbTable not found")
merr = errors.New("compaction leveldbTable not found")
continue
}
lines = lines[3:]
// Iterate over all the leveldbTable rows, and accumulate the entries
for j := 0; j < len(compactions[i%2]); j++ {
compactions[i%2][j] = 0
}
for _, line := range lines {
parts := strings.Split(line, "|")
if len(parts) != 6 {
break
}
for idx, counter := range parts[2:] {
value, err := strconv.ParseFloat(strings.TrimSpace(counter), 64)
if err != nil {
db.log.Error("Compaction entry parsing failed", "err", err)
merr = err
continue
}
compactions[i%2][idx] += value
}
}
// Update all the requested meters
if db.diskSizeGauge != nil {
db.diskSizeGauge.Update(int64(compactions[i%2][0] * 1024 * 1024))
}
if db.compTimeMeter != nil {
db.compTimeMeter.Mark(int64((compactions[i%2][1] - compactions[(i-1)%2][1]) * 1000 * 1000 * 1000))
}
if db.compReadMeter != nil {
db.compReadMeter.Mark(int64((compactions[i%2][2] - compactions[(i-1)%2][2]) * 1024 * 1024))
}
if db.compWriteMeter != nil {
db.compWriteMeter.Mark(int64((compactions[i%2][3] - compactions[(i-1)%2][3]) * 1024 * 1024))
}
// Retrieve the write delay statistic
writedelay, err := db.db.GetProperty("leveldb.writedelay")
if err != nil {
db.log.Error("Failed to read database write delay statistic", "err", err)
merr = err
continue
}
var (
delayN int64
delayDuration string
duration time.Duration
paused bool
)
if n, err := fmt.Sscanf(writedelay, "DelayN:%d Delay:%s Paused:%t", &delayN, &delayDuration, &paused); n != 3 || err != nil {
db.log.Error("Write delay statistic not found")
merr = err
continue
}
duration, err = time.ParseDuration(delayDuration)
if err != nil {
db.log.Error("Failed to parse delay duration", "err", err)
merr = err
continue
}
if db.writeDelayNMeter != nil {
db.writeDelayNMeter.Mark(delayN - delaystats[0])
}
if db.writeDelayMeter != nil {
db.writeDelayMeter.Mark(duration.Nanoseconds() - delaystats[1])
}
// If a warning that db is performing compaction has been displayed, any subsequent
// warnings will be withheld for one minute not to overwhelm the user.
if paused && delayN-delaystats[0] == 0 && duration.Nanoseconds()-delaystats[1] == 0 &&
time.Now().After(lastWritePaused.Add(degradationWarnInterval)) {
db.log.Warn("Database compacting, degraded performance")
lastWritePaused = time.Now()
}
delaystats[0], delaystats[1] = delayN, duration.Nanoseconds()
// Retrieve the database iostats.
ioStats, err := db.db.GetProperty("leveldb.iostats")
if err != nil {
db.log.Error("Failed to read database iostats", "err", err)
merr = err
continue
}
var nRead, nWrite float64
parts := strings.Split(ioStats, " ")
if len(parts) < 2 {
db.log.Error("Bad syntax of ioStats", "ioStats", ioStats)
merr = fmt.Errorf("bad syntax of ioStats %s", ioStats)
continue
}
if n, err := fmt.Sscanf(parts[0], "Read(MB):%f", &nRead); n != 1 || err != nil {
db.log.Error("Bad syntax of read entry", "entry", parts[0])
merr = err
continue
}
if n, err := fmt.Sscanf(parts[1], "Write(MB):%f", &nWrite); n != 1 || err != nil {
db.log.Error("Bad syntax of write entry", "entry", parts[1])
merr = err
continue
}
if db.diskReadMeter != nil {
db.diskReadMeter.Mark(int64((nRead - iostats[0]) * 1024 * 1024))
}
if db.diskWriteMeter != nil {
db.diskWriteMeter.Mark(int64((nWrite - iostats[1]) * 1024 * 1024))
}
iostats[0], iostats[1] = nRead, nWrite
// Sleep a bit, then repeat the stats collection
select {
case errc = <-db.quitChan:
// Quit requesting, stop hammering the database
case <-timer.C:
timer.Reset(refresh)
// Timeout, gather a new set of stats
}
}
if errc == nil {
errc = <-db.quitChan
}
errc <- merr
}
// batch is a write-only leveldb batch that commits changes to its host database
// when Write is called. A batch cannot be used concurrently.
type batch struct {
db *leveldb.DB
b *leveldb.Batch
size int
}
// Put inserts the given value into the batch for later committing.
func (b *batch) Put(key, value []byte) error {
b.b.Put(key, value)
b.size += len(value)
return nil
}
// Delete inserts the a key removal into the batch for later committing.
func (b *batch) Delete(key []byte) error {
b.b.Delete(key)
b.size++
return nil
}
// ValueSize retrieves the amount of data queued up for writing.
func (b *batch) ValueSize() int {
return b.size
}
// Write flushes any accumulated data to disk.
func (b *batch) Write() error {
return b.db.Write(b.b, nil)
}
// Reset resets the batch for reuse.
func (b *batch) Reset() {
b.b.Reset()
b.size = 0
}
// Replay replays the batch contents.
func (b *batch) Replay(w ethdb.KeyValueWriter) error {
return b.b.Replay(&replayer{writer: w})
}
// replayer is a small wrapper to implement the correct replay methods.
type replayer struct {
writer ethdb.KeyValueWriter
failure error
}
// Put inserts the given value into the key-value data store.
func (r *replayer) Put(key, value []byte) {
// If the replay already failed, stop executing ops
if r.failure != nil {
return
}
r.failure = r.writer.Put(key, value)
}
// Delete removes the key from the key-value data store.
func (r *replayer) Delete(key []byte) {
// If the replay already failed, stop executing ops
if r.failure != nil {
return
}
r.failure = r.writer.Delete(key)
}
// bytesPrefixRange returns key range that satisfy
// - the given prefix, and
// - the given seek position
func bytesPrefixRange(prefix, start []byte) *util.Range {
r := util.BytesPrefix(prefix)
r.Start = append(r.Start, start...)
return r
}