github.com/calmw/ethereum@v0.1.1/ethdb/leveldb/leveldb.go (about) 1 // Copyright 2018 The go-ethereum Authors 2 // This file is part of the go-ethereum library. 3 // 4 // The go-ethereum library is free software: you can redistribute it and/or modify 5 // it under the terms of the GNU Lesser General Public License as published by 6 // the Free Software Foundation, either version 3 of the License, or 7 // (at your option) any later version. 8 // 9 // The go-ethereum library is distributed in the hope that it will be useful, 10 // but WITHOUT ANY WARRANTY; without even the implied warranty of 11 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 // GNU Lesser General Public License for more details. 13 // 14 // You should have received a copy of the GNU Lesser General Public License 15 // along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>. 16 17 //go:build !js 18 // +build !js 19 20 // Package leveldb implements the key-value database layer based on LevelDB. 21 package leveldb 22 23 import ( 24 "fmt" 25 "strconv" 26 "strings" 27 "sync" 28 "time" 29 30 "github.com/calmw/ethereum/common" 31 "github.com/calmw/ethereum/ethdb" 32 "github.com/calmw/ethereum/log" 33 "github.com/calmw/ethereum/metrics" 34 "github.com/syndtr/goleveldb/leveldb" 35 "github.com/syndtr/goleveldb/leveldb/errors" 36 "github.com/syndtr/goleveldb/leveldb/filter" 37 "github.com/syndtr/goleveldb/leveldb/opt" 38 "github.com/syndtr/goleveldb/leveldb/util" 39 ) 40 41 const ( 42 // degradationWarnInterval specifies how often warning should be printed if the 43 // leveldb database cannot keep up with requested writes. 44 degradationWarnInterval = time.Minute 45 46 // minCache is the minimum amount of memory in megabytes to allocate to leveldb 47 // read and write caching, split half and half. 48 minCache = 16 49 50 // minHandles is the minimum number of files handles to allocate to the open 51 // database files. 52 minHandles = 16 53 54 // metricsGatheringInterval specifies the interval to retrieve leveldb database 55 // compaction, io and pause stats to report to the user. 56 metricsGatheringInterval = 3 * time.Second 57 ) 58 59 // Database is a persistent key-value store. Apart from basic data storage 60 // functionality it also supports batch writes and iterating over the keyspace in 61 // binary-alphabetical order. 62 type Database struct { 63 fn string // filename for reporting 64 db *leveldb.DB // LevelDB instance 65 66 compTimeMeter metrics.Meter // Meter for measuring the total time spent in database compaction 67 compReadMeter metrics.Meter // Meter for measuring the data read during compaction 68 compWriteMeter metrics.Meter // Meter for measuring the data written during compaction 69 writeDelayNMeter metrics.Meter // Meter for measuring the write delay number due to database compaction 70 writeDelayMeter metrics.Meter // Meter for measuring the write delay duration due to database compaction 71 diskSizeGauge metrics.Gauge // Gauge for tracking the size of all the levels in the database 72 diskReadMeter metrics.Meter // Meter for measuring the effective amount of data read 73 diskWriteMeter metrics.Meter // Meter for measuring the effective amount of data written 74 memCompGauge metrics.Gauge // Gauge for tracking the number of memory compaction 75 level0CompGauge metrics.Gauge // Gauge for tracking the number of table compaction in level0 76 nonlevel0CompGauge metrics.Gauge // Gauge for tracking the number of table compaction in non0 level 77 seekCompGauge metrics.Gauge // Gauge for tracking the number of table compaction caused by read opt 78 manualMemAllocGauge metrics.Gauge // Gauge to track the amount of memory that has been manually allocated (not a part of runtime/GC) 79 80 quitLock sync.Mutex // Mutex protecting the quit channel access 81 quitChan chan chan error // Quit channel to stop the metrics collection before closing the database 82 83 log log.Logger // Contextual logger tracking the database path 84 } 85 86 // New returns a wrapped LevelDB object. The namespace is the prefix that the 87 // metrics reporting should use for surfacing internal stats. 88 func New(file string, cache int, handles int, namespace string, readonly bool) (*Database, error) { 89 return NewCustom(file, namespace, func(options *opt.Options) { 90 // Ensure we have some minimal caching and file guarantees 91 if cache < minCache { 92 cache = minCache 93 } 94 if handles < minHandles { 95 handles = minHandles 96 } 97 // Set default options 98 options.OpenFilesCacheCapacity = handles 99 options.BlockCacheCapacity = cache / 2 * opt.MiB 100 options.WriteBuffer = cache / 4 * opt.MiB // Two of these are used internally 101 if readonly { 102 options.ReadOnly = true 103 } 104 }) 105 } 106 107 // NewCustom returns a wrapped LevelDB object. The namespace is the prefix that the 108 // metrics reporting should use for surfacing internal stats. 109 // The customize function allows the caller to modify the leveldb options. 110 func NewCustom(file string, namespace string, customize func(options *opt.Options)) (*Database, error) { 111 options := configureOptions(customize) 112 logger := log.New("database", file) 113 usedCache := options.GetBlockCacheCapacity() + options.GetWriteBuffer()*2 114 logCtx := []interface{}{"cache", common.StorageSize(usedCache), "handles", options.GetOpenFilesCacheCapacity()} 115 if options.ReadOnly { 116 logCtx = append(logCtx, "readonly", "true") 117 } 118 logger.Info("Allocated cache and file handles", logCtx...) 119 120 // Open the db and recover any potential corruptions 121 db, err := leveldb.OpenFile(file, options) 122 if _, corrupted := err.(*errors.ErrCorrupted); corrupted { 123 db, err = leveldb.RecoverFile(file, nil) 124 } 125 if err != nil { 126 return nil, err 127 } 128 // Assemble the wrapper with all the registered metrics 129 ldb := &Database{ 130 fn: file, 131 db: db, 132 log: logger, 133 quitChan: make(chan chan error), 134 } 135 ldb.compTimeMeter = metrics.NewRegisteredMeter(namespace+"compact/time", nil) 136 ldb.compReadMeter = metrics.NewRegisteredMeter(namespace+"compact/input", nil) 137 ldb.compWriteMeter = metrics.NewRegisteredMeter(namespace+"compact/output", nil) 138 ldb.diskSizeGauge = metrics.NewRegisteredGauge(namespace+"disk/size", nil) 139 ldb.diskReadMeter = metrics.NewRegisteredMeter(namespace+"disk/read", nil) 140 ldb.diskWriteMeter = metrics.NewRegisteredMeter(namespace+"disk/write", nil) 141 ldb.writeDelayMeter = metrics.NewRegisteredMeter(namespace+"compact/writedelay/duration", nil) 142 ldb.writeDelayNMeter = metrics.NewRegisteredMeter(namespace+"compact/writedelay/counter", nil) 143 ldb.memCompGauge = metrics.NewRegisteredGauge(namespace+"compact/memory", nil) 144 ldb.level0CompGauge = metrics.NewRegisteredGauge(namespace+"compact/level0", nil) 145 ldb.nonlevel0CompGauge = metrics.NewRegisteredGauge(namespace+"compact/nonlevel0", nil) 146 ldb.seekCompGauge = metrics.NewRegisteredGauge(namespace+"compact/seek", nil) 147 ldb.manualMemAllocGauge = metrics.NewRegisteredGauge(namespace+"memory/manualalloc", nil) 148 149 // Start up the metrics gathering and return 150 go ldb.meter(metricsGatheringInterval) 151 return ldb, nil 152 } 153 154 // configureOptions sets some default options, then runs the provided setter. 155 func configureOptions(customizeFn func(*opt.Options)) *opt.Options { 156 // Set default options 157 options := &opt.Options{ 158 Filter: filter.NewBloomFilter(10), 159 DisableSeeksCompaction: true, 160 } 161 // Allow caller to make custom modifications to the options 162 if customizeFn != nil { 163 customizeFn(options) 164 } 165 return options 166 } 167 168 // Close stops the metrics collection, flushes any pending data to disk and closes 169 // all io accesses to the underlying key-value store. 170 func (db *Database) Close() error { 171 db.quitLock.Lock() 172 defer db.quitLock.Unlock() 173 174 if db.quitChan != nil { 175 errc := make(chan error) 176 db.quitChan <- errc 177 if err := <-errc; err != nil { 178 db.log.Error("Metrics collection failed", "err", err) 179 } 180 db.quitChan = nil 181 } 182 return db.db.Close() 183 } 184 185 // Has retrieves if a key is present in the key-value store. 186 func (db *Database) Has(key []byte) (bool, error) { 187 return db.db.Has(key, nil) 188 } 189 190 // Get retrieves the given key if it's present in the key-value store. 191 func (db *Database) Get(key []byte) ([]byte, error) { 192 dat, err := db.db.Get(key, nil) 193 if err != nil { 194 return nil, err 195 } 196 return dat, nil 197 } 198 199 // Put inserts the given value into the key-value store. 200 func (db *Database) Put(key []byte, value []byte) error { 201 return db.db.Put(key, value, nil) 202 } 203 204 // Delete removes the key from the key-value store. 205 func (db *Database) Delete(key []byte) error { 206 return db.db.Delete(key, nil) 207 } 208 209 // NewBatch creates a write-only key-value store that buffers changes to its host 210 // database until a final write is called. 211 func (db *Database) NewBatch() ethdb.Batch { 212 return &batch{ 213 db: db.db, 214 b: new(leveldb.Batch), 215 } 216 } 217 218 // NewBatchWithSize creates a write-only database batch with pre-allocated buffer. 219 func (db *Database) NewBatchWithSize(size int) ethdb.Batch { 220 return &batch{ 221 db: db.db, 222 b: leveldb.MakeBatch(size), 223 } 224 } 225 226 // NewIterator creates a binary-alphabetical iterator over a subset 227 // of database content with a particular key prefix, starting at a particular 228 // initial key (or after, if it does not exist). 229 func (db *Database) NewIterator(prefix []byte, start []byte) ethdb.Iterator { 230 return db.db.NewIterator(bytesPrefixRange(prefix, start), nil) 231 } 232 233 // NewSnapshot creates a database snapshot based on the current state. 234 // The created snapshot will not be affected by all following mutations 235 // happened on the database. 236 // Note don't forget to release the snapshot once it's used up, otherwise 237 // the stale data will never be cleaned up by the underlying compactor. 238 func (db *Database) NewSnapshot() (ethdb.Snapshot, error) { 239 snap, err := db.db.GetSnapshot() 240 if err != nil { 241 return nil, err 242 } 243 return &snapshot{db: snap}, nil 244 } 245 246 // Stat returns a particular internal stat of the database. 247 func (db *Database) Stat(property string) (string, error) { 248 return db.db.GetProperty(property) 249 } 250 251 // Compact flattens the underlying data store for the given key range. In essence, 252 // deleted and overwritten versions are discarded, and the data is rearranged to 253 // reduce the cost of operations needed to access them. 254 // 255 // A nil start is treated as a key before all keys in the data store; a nil limit 256 // is treated as a key after all keys in the data store. If both is nil then it 257 // will compact entire data store. 258 func (db *Database) Compact(start []byte, limit []byte) error { 259 return db.db.CompactRange(util.Range{Start: start, Limit: limit}) 260 } 261 262 // Path returns the path to the database directory. 263 func (db *Database) Path() string { 264 return db.fn 265 } 266 267 // meter periodically retrieves internal leveldb counters and reports them to 268 // the metrics subsystem. 269 // 270 // This is how a LevelDB stats table looks like (currently): 271 // 272 // Compactions 273 // Level | Tables | Size(MB) | Time(sec) | Read(MB) | Write(MB) 274 // -------+------------+---------------+---------------+---------------+--------------- 275 // 0 | 0 | 0.00000 | 1.27969 | 0.00000 | 12.31098 276 // 1 | 85 | 109.27913 | 28.09293 | 213.92493 | 214.26294 277 // 2 | 523 | 1000.37159 | 7.26059 | 66.86342 | 66.77884 278 // 3 | 570 | 1113.18458 | 0.00000 | 0.00000 | 0.00000 279 // 280 // This is how the write delay look like (currently): 281 // DelayN:5 Delay:406.604657ms Paused: false 282 // 283 // This is how the iostats look like (currently): 284 // Read(MB):3895.04860 Write(MB):3654.64712 285 func (db *Database) meter(refresh time.Duration) { 286 // Create the counters to store current and previous compaction values 287 compactions := make([][]float64, 2) 288 for i := 0; i < 2; i++ { 289 compactions[i] = make([]float64, 4) 290 } 291 // Create storage for iostats. 292 var iostats [2]float64 293 294 // Create storage and warning log tracer for write delay. 295 var ( 296 delaystats [2]int64 297 lastWritePaused time.Time 298 ) 299 300 var ( 301 errc chan error 302 merr error 303 ) 304 305 timer := time.NewTimer(refresh) 306 defer timer.Stop() 307 308 // Iterate ad infinitum and collect the stats 309 for i := 1; errc == nil && merr == nil; i++ { 310 // Retrieve the database stats 311 stats, err := db.db.GetProperty("leveldb.stats") 312 if err != nil { 313 db.log.Error("Failed to read database stats", "err", err) 314 merr = err 315 continue 316 } 317 // Find the compaction table, skip the header 318 lines := strings.Split(stats, "\n") 319 for len(lines) > 0 && strings.TrimSpace(lines[0]) != "Compactions" { 320 lines = lines[1:] 321 } 322 if len(lines) <= 3 { 323 db.log.Error("Compaction leveldbTable not found") 324 merr = errors.New("compaction leveldbTable not found") 325 continue 326 } 327 lines = lines[3:] 328 329 // Iterate over all the leveldbTable rows, and accumulate the entries 330 for j := 0; j < len(compactions[i%2]); j++ { 331 compactions[i%2][j] = 0 332 } 333 for _, line := range lines { 334 parts := strings.Split(line, "|") 335 if len(parts) != 6 { 336 break 337 } 338 for idx, counter := range parts[2:] { 339 value, err := strconv.ParseFloat(strings.TrimSpace(counter), 64) 340 if err != nil { 341 db.log.Error("Compaction entry parsing failed", "err", err) 342 merr = err 343 continue 344 } 345 compactions[i%2][idx] += value 346 } 347 } 348 // Update all the requested meters 349 if db.diskSizeGauge != nil { 350 db.diskSizeGauge.Update(int64(compactions[i%2][0] * 1024 * 1024)) 351 } 352 if db.compTimeMeter != nil { 353 db.compTimeMeter.Mark(int64((compactions[i%2][1] - compactions[(i-1)%2][1]) * 1000 * 1000 * 1000)) 354 } 355 if db.compReadMeter != nil { 356 db.compReadMeter.Mark(int64((compactions[i%2][2] - compactions[(i-1)%2][2]) * 1024 * 1024)) 357 } 358 if db.compWriteMeter != nil { 359 db.compWriteMeter.Mark(int64((compactions[i%2][3] - compactions[(i-1)%2][3]) * 1024 * 1024)) 360 } 361 // Retrieve the write delay statistic 362 writedelay, err := db.db.GetProperty("leveldb.writedelay") 363 if err != nil { 364 db.log.Error("Failed to read database write delay statistic", "err", err) 365 merr = err 366 continue 367 } 368 var ( 369 delayN int64 370 delayDuration string 371 duration time.Duration 372 paused bool 373 ) 374 if n, err := fmt.Sscanf(writedelay, "DelayN:%d Delay:%s Paused:%t", &delayN, &delayDuration, &paused); n != 3 || err != nil { 375 db.log.Error("Write delay statistic not found") 376 merr = err 377 continue 378 } 379 duration, err = time.ParseDuration(delayDuration) 380 if err != nil { 381 db.log.Error("Failed to parse delay duration", "err", err) 382 merr = err 383 continue 384 } 385 if db.writeDelayNMeter != nil { 386 db.writeDelayNMeter.Mark(delayN - delaystats[0]) 387 } 388 if db.writeDelayMeter != nil { 389 db.writeDelayMeter.Mark(duration.Nanoseconds() - delaystats[1]) 390 } 391 // If a warning that db is performing compaction has been displayed, any subsequent 392 // warnings will be withheld for one minute not to overwhelm the user. 393 if paused && delayN-delaystats[0] == 0 && duration.Nanoseconds()-delaystats[1] == 0 && 394 time.Now().After(lastWritePaused.Add(degradationWarnInterval)) { 395 db.log.Warn("Database compacting, degraded performance") 396 lastWritePaused = time.Now() 397 } 398 delaystats[0], delaystats[1] = delayN, duration.Nanoseconds() 399 400 // Retrieve the database iostats. 401 ioStats, err := db.db.GetProperty("leveldb.iostats") 402 if err != nil { 403 db.log.Error("Failed to read database iostats", "err", err) 404 merr = err 405 continue 406 } 407 var nRead, nWrite float64 408 parts := strings.Split(ioStats, " ") 409 if len(parts) < 2 { 410 db.log.Error("Bad syntax of ioStats", "ioStats", ioStats) 411 merr = fmt.Errorf("bad syntax of ioStats %s", ioStats) 412 continue 413 } 414 if n, err := fmt.Sscanf(parts[0], "Read(MB):%f", &nRead); n != 1 || err != nil { 415 db.log.Error("Bad syntax of read entry", "entry", parts[0]) 416 merr = err 417 continue 418 } 419 if n, err := fmt.Sscanf(parts[1], "Write(MB):%f", &nWrite); n != 1 || err != nil { 420 db.log.Error("Bad syntax of write entry", "entry", parts[1]) 421 merr = err 422 continue 423 } 424 if db.diskReadMeter != nil { 425 db.diskReadMeter.Mark(int64((nRead - iostats[0]) * 1024 * 1024)) 426 } 427 if db.diskWriteMeter != nil { 428 db.diskWriteMeter.Mark(int64((nWrite - iostats[1]) * 1024 * 1024)) 429 } 430 iostats[0], iostats[1] = nRead, nWrite 431 432 compCount, err := db.db.GetProperty("leveldb.compcount") 433 if err != nil { 434 db.log.Error("Failed to read database iostats", "err", err) 435 merr = err 436 continue 437 } 438 439 var ( 440 memComp uint32 441 level0Comp uint32 442 nonLevel0Comp uint32 443 seekComp uint32 444 ) 445 if n, err := fmt.Sscanf(compCount, "MemComp:%d Level0Comp:%d NonLevel0Comp:%d SeekComp:%d", &memComp, &level0Comp, &nonLevel0Comp, &seekComp); n != 4 || err != nil { 446 db.log.Error("Compaction count statistic not found") 447 merr = err 448 continue 449 } 450 db.memCompGauge.Update(int64(memComp)) 451 db.level0CompGauge.Update(int64(level0Comp)) 452 db.nonlevel0CompGauge.Update(int64(nonLevel0Comp)) 453 db.seekCompGauge.Update(int64(seekComp)) 454 455 // Sleep a bit, then repeat the stats collection 456 select { 457 case errc = <-db.quitChan: 458 // Quit requesting, stop hammering the database 459 case <-timer.C: 460 timer.Reset(refresh) 461 // Timeout, gather a new set of stats 462 } 463 } 464 465 if errc == nil { 466 errc = <-db.quitChan 467 } 468 errc <- merr 469 } 470 471 // batch is a write-only leveldb batch that commits changes to its host database 472 // when Write is called. A batch cannot be used concurrently. 473 type batch struct { 474 db *leveldb.DB 475 b *leveldb.Batch 476 size int 477 } 478 479 // Put inserts the given value into the batch for later committing. 480 func (b *batch) Put(key, value []byte) error { 481 b.b.Put(key, value) 482 b.size += len(key) + len(value) 483 return nil 484 } 485 486 // Delete inserts the a key removal into the batch for later committing. 487 func (b *batch) Delete(key []byte) error { 488 b.b.Delete(key) 489 b.size += len(key) 490 return nil 491 } 492 493 // ValueSize retrieves the amount of data queued up for writing. 494 func (b *batch) ValueSize() int { 495 return b.size 496 } 497 498 // Write flushes any accumulated data to disk. 499 func (b *batch) Write() error { 500 return b.db.Write(b.b, nil) 501 } 502 503 // Reset resets the batch for reuse. 504 func (b *batch) Reset() { 505 b.b.Reset() 506 b.size = 0 507 } 508 509 // Replay replays the batch contents. 510 func (b *batch) Replay(w ethdb.KeyValueWriter) error { 511 return b.b.Replay(&replayer{writer: w}) 512 } 513 514 // replayer is a small wrapper to implement the correct replay methods. 515 type replayer struct { 516 writer ethdb.KeyValueWriter 517 failure error 518 } 519 520 // Put inserts the given value into the key-value data store. 521 func (r *replayer) Put(key, value []byte) { 522 // If the replay already failed, stop executing ops 523 if r.failure != nil { 524 return 525 } 526 r.failure = r.writer.Put(key, value) 527 } 528 529 // Delete removes the key from the key-value data store. 530 func (r *replayer) Delete(key []byte) { 531 // If the replay already failed, stop executing ops 532 if r.failure != nil { 533 return 534 } 535 r.failure = r.writer.Delete(key) 536 } 537 538 // bytesPrefixRange returns key range that satisfy 539 // - the given prefix, and 540 // - the given seek position 541 func bytesPrefixRange(prefix, start []byte) *util.Range { 542 r := util.BytesPrefix(prefix) 543 r.Start = append(r.Start, start...) 544 return r 545 } 546 547 // snapshot wraps a leveldb snapshot for implementing the Snapshot interface. 548 type snapshot struct { 549 db *leveldb.Snapshot 550 } 551 552 // Has retrieves if a key is present in the snapshot backing by a key-value 553 // data store. 554 func (snap *snapshot) Has(key []byte) (bool, error) { 555 return snap.db.Has(key, nil) 556 } 557 558 // Get retrieves the given key if it's present in the snapshot backing by 559 // key-value data store. 560 func (snap *snapshot) Get(key []byte) ([]byte, error) { 561 return snap.db.Get(key, nil) 562 } 563 564 // Release releases associated resources. Release should always succeed and can 565 // be called multiple times without causing error. 566 func (snap *snapshot) Release() { 567 snap.db.Release() 568 }