flatfs.go 22.3 KB
Newer Older
Tommi Virtanen's avatar
Tommi Virtanen committed
1 2 3 4 5 6
// Package flatfs is a Datastore implementation that stores all
// objects in a two-level directory structure in the local file
// system, regardless of the hierarchy of the keys.
package flatfs

import (
7
	"encoding/json"
Tommi Virtanen's avatar
Tommi Virtanen committed
8
	"errors"
9
	"fmt"
Tommi Virtanen's avatar
Tommi Virtanen committed
10
	"io/ioutil"
11 12
	"math"
	"math/rand"
Tommi Virtanen's avatar
Tommi Virtanen committed
13
	"os"
Jeromy's avatar
Jeromy committed
14
	"path/filepath"
Tommi Virtanen's avatar
Tommi Virtanen committed
15
	"strings"
16 17
	"sync"
	"sync/atomic"
18
	"time"
Tommi Virtanen's avatar
Tommi Virtanen committed
19

Jeromy's avatar
Jeromy committed
20 21
	"github.com/ipfs/go-datastore"
	"github.com/ipfs/go-datastore/query"
22

Jakub Sztandera's avatar
Jakub Sztandera committed
23
	logging "github.com/ipfs/go-log"
Tommi Virtanen's avatar
Tommi Virtanen committed
24 25
)

26 27
var log = logging.Logger("flatfs")

Tommi Virtanen's avatar
Tommi Virtanen committed
28
const (
29 30
	extension                  = ".data"
	diskUsageCheckpointPercent = 1.0
Tommi Virtanen's avatar
Tommi Virtanen committed
31 32
)

33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56
var (
	// DiskUsageFile is the name of the file to cache the size of the
	// datastore in disk
	DiskUsageFile = "diskUsage.cache"
	// DiskUsageFilesAverage is the maximum number of files per folder
	// to stat in order to calculate the size of the datastore.
	// The size of the rest of the files in a folder will be assumed
	// to be the average of the values obtained. This includes
	// regular files and directories.
	DiskUsageFilesAverage = 2000
	// DiskUsageCalcTimeout is the maximum time to spend
	// calculating the DiskUsage upon a start when no
	// DiskUsageFile is present.
	// If this period did not suffice to read the size of the datastore,
	// the remaining sizes will be stimated.
	DiskUsageCalcTimeout = 5 * time.Minute
)

const (
	opPut = iota
	opDelete
	opRename
)

57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77
type initAccuracy string

const (
	exactA    initAccuracy = "initial-exact"
	approxA   initAccuracy = "initial-approximate"
	timedoutA initAccuracy = "initial-timed-out"
)

func combineAccuracy(a, b initAccuracy) initAccuracy {
	if a == timedoutA || b == timedoutA {
		return timedoutA
	}
	if a == approxA || b == approxA {
		return approxA
	}
	if a == exactA && b == exactA {
		return exactA
	}
	return ""
}

78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93
var _ datastore.Datastore = (*Datastore)(nil)

var (
	ErrDatastoreExists       = errors.New("datastore already exists")
	ErrDatastoreDoesNotExist = errors.New("datastore directory does not exist")
	ErrShardingFileMissing   = fmt.Errorf("%s file not found in datastore", SHARDING_FN)
)

func init() {
	rand.Seed(time.Now().UTC().UnixNano())
}

// Datastore implements the go-datastore Interface.
// Note this datastore cannot guarantee order of concurrent
// write operations to the same key. See the explanation in
// Put().
Tommi Virtanen's avatar
Tommi Virtanen committed
94 95
type Datastore struct {
	path string
96

97 98
	shardStr string
	getDir   ShardFunc
Jeromy's avatar
Jeromy committed
99 100 101

	// sychronize all writes and directory changes for added safety
	sync bool
102

103 104
	diskUsage   int64
	storedValue diskUsageValue
105 106 107

	checkpointCh chan bool
	done         chan bool
108 109 110 111

	// opMap handles concurrent write operations (put/delete)
	// to the same key
	opMap *opMap
Tommi Virtanen's avatar
Tommi Virtanen committed
112 113
}

114 115 116 117 118
type diskUsageValue struct {
	diskUsage int64
	accuracy  initAccuracy
}

119 120
type ShardFunc func(string) string

121
type opT int
122

123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
// op wraps useful arguments of write operations
type op struct {
	typ  opT           // operation type
	key  datastore.Key // datastore key. Mandatory.
	tmp  string        // temp file path
	path string        // file path
	v    []byte        // value
}

type opMap struct {
	ops sync.Map
}

type opResult struct {
	mu      sync.RWMutex
	success bool

	opMap *opMap
	name  string
}

// Returns nil if there's nothing to do.
func (m *opMap) Begin(name string) *opResult {
	for {
		myOp := &opResult{opMap: m, name: name}
		myOp.mu.Lock()
		opIface, loaded := m.ops.LoadOrStore(name, myOp)
		if !loaded { // no one else doing ops with this key
			return myOp
		}

		op := opIface.(*opResult)
		// someone else doing ops with this key, wait for
		// the result
		op.mu.RLock()
		if op.success {
			return nil
		}

		// if we are here, we will retry the operation
	}
}

func (o *opResult) Finish(ok bool) {
	o.success = ok
	o.opMap.ops.Delete(o.name)
	o.mu.Unlock()
}
171

172
func Create(path string, fun *ShardIdV1) error {
173

kpcyrd's avatar
kpcyrd committed
174
	err := os.Mkdir(path, 0755)
175 176
	if err != nil && !os.IsExist(err) {
		return err
Tommi Virtanen's avatar
Tommi Virtanen committed
177 178
	}

179 180
	dsFun, err := ReadShardFunc(path)
	switch err {
Kevin Atkinson's avatar
Kevin Atkinson committed
181
	case ErrShardingFileMissing:
182 183 184 185 186
		isEmpty, err := DirIsEmpty(path)
		if err != nil {
			return err
		}
		if !isEmpty {
187
			return fmt.Errorf("directory missing %s file: %s", SHARDING_FN, path)
188 189 190
		}

		err = WriteShardFunc(path, fun)
191 192 193
		if err != nil {
			return err
		}
194 195
		err = WriteReadme(path, fun)
		return err
196
	case nil:
197
		if fun.String() != dsFun.String() {
198
			return fmt.Errorf("specified shard func '%s' does not match repo shard func '%s'",
199
				fun.String(), dsFun.String())
200
		}
Kevin Atkinson's avatar
Kevin Atkinson committed
201
		return ErrDatastoreExists
202
	default:
203
		return err
204
	}
205 206
}

207
func Open(path string, syncFiles bool) (*Datastore, error) {
208 209
	_, err := os.Stat(path)
	if os.IsNotExist(err) {
Kevin Atkinson's avatar
Kevin Atkinson committed
210
		return nil, ErrDatastoreDoesNotExist
211 212 213 214
	} else if err != nil {
		return nil, err
	}

215
	shardId, err := ReadShardFunc(path)
216
	if err != nil {
217
		return nil, err
218 219
	}

220
	fs := &Datastore{
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
		path:      path,
		shardStr:  shardId.String(),
		getDir:    shardId.Func(),
		sync:      syncFiles,
		diskUsage: 0,
		opMap:     new(opMap),
	}

	// This sets diskUsage to the correct value
	// It might be slow, but allowing it to happen
	// while the datastore is usable might
	// cause diskUsage to not be accurate.
	err = fs.calculateDiskUsage()
	if err != nil {
		// Cannot stat() all
		// elements in the datastore.
		return nil, err
238
	}
239
	fs.checkpointLoop()
240 241 242
	return fs, nil
}

243
// convenience method
244
func CreateOrOpen(path string, fun *ShardIdV1, sync bool) (*Datastore, error) {
245
	err := Create(path, fun)
Kevin Atkinson's avatar
Kevin Atkinson committed
246
	if err != nil && err != ErrDatastoreExists {
247 248 249 250 251
		return nil, err
	}
	return Open(path, sync)
}

252 253
func (fs *Datastore) ShardStr() string {
	return fs.shardStr
254 255
}

Tommi Virtanen's avatar
Tommi Virtanen committed
256
func (fs *Datastore) encode(key datastore.Key) (dir, file string) {
257
	noslash := key.String()[1:]
Jeromy's avatar
Jeromy committed
258 259
	dir = filepath.Join(fs.path, fs.getDir(noslash))
	file = filepath.Join(dir, noslash+extension)
Tommi Virtanen's avatar
Tommi Virtanen committed
260 261 262
	return dir, file
}

263
func (fs *Datastore) decode(file string) (key datastore.Key, ok bool) {
Jeromy's avatar
Jeromy committed
264
	if filepath.Ext(file) != extension {
265 266 267
		return datastore.Key{}, false
	}
	name := file[:len(file)-len(extension)]
Jeromy's avatar
Jeromy committed
268
	return datastore.NewKey(name), true
269 270
}

271 272
func (fs *Datastore) makeDir(dir string) error {
	if err := fs.makeDirNoSync(dir); err != nil {
Jeromy's avatar
Jeromy committed
273
		return err
274 275 276 277 278 279
	}

	// In theory, if we create a new prefix dir and add a file to
	// it, the creation of the prefix dir itself might not be
	// durable yet. Sync the root dir after a successful mkdir of
	// a prefix dir, just to be paranoid.
Jeromy's avatar
Jeromy committed
280 281 282 283
	if fs.sync {
		if err := syncDir(fs.path); err != nil {
			return err
		}
284 285 286 287
	}
	return nil
}

288
func (fs *Datastore) makeDirNoSync(dir string) error {
kpcyrd's avatar
kpcyrd committed
289
	if err := os.Mkdir(dir, 0755); err != nil {
Jeromy's avatar
Jeromy committed
290 291 292 293 294
		// EEXIST is safe to ignore here, that just means the prefix
		// directory already existed.
		if !os.IsExist(err) {
			return err
		}
295
		return nil
Jeromy's avatar
Jeromy committed
296
	}
297 298 299

	// Track DiskUsage of this NEW folder
	fs.updateDiskUsage(dir, true)
Jeromy's avatar
Jeromy committed
300 301 302
	return nil
}

303 304 305 306 307 308 309 310 311 312 313 314 315 316 317
// This function always runs under an opLock. Therefore, only one thread is
// touching the affected files.
func (fs *Datastore) renameAndUpdateDiskUsage(tmpPath, path string) error {
	fi, err := os.Stat(path)

	// Destination exists, we need to discount it from diskUsage
	if fs != nil && err == nil {
		atomic.AddInt64(&fs.diskUsage, -fi.Size())
	} else if !os.IsNotExist(err) {
		return err
	}

	// Rename and add new file's diskUsage. If the rename fails,
	// it will either a) Re-add the size of an existing file, which
	// was sustracted before b) Add 0 if there is no existing file.
318
	err = os.Rename(tmpPath, path)
319 320 321 322
	fs.updateDiskUsage(path, true)
	return err
}

Jeromy's avatar
Jeromy committed
323
var putMaxRetries = 6
324

325 326 327 328 329 330 331 332 333 334
// Put stores a key/value in the datastore.
//
// Note, that we do not guarantee order of write operations (Put or Delete)
// to the same key in this datastore.
//
// For example. i.e. in the case of two concurrent Put, we only guarantee
// that one of them will come through, but cannot assure which one even if
// one arrived slightly later than the other. In the case of a
// concurrent Put and a Delete operation, we cannot guarantee which one
// will win.
Tommi Virtanen's avatar
Tommi Virtanen committed
335 336 337 338 339 340
func (fs *Datastore) Put(key datastore.Key, value interface{}) error {
	val, ok := value.([]byte)
	if !ok {
		return datastore.ErrInvalidType
	}

341
	var err error
Jeromy's avatar
Jeromy committed
342
	for i := 1; i <= putMaxRetries; i++ {
343 344 345 346 347
		err = fs.doWriteOp(&op{
			typ: opPut,
			key: key,
			v:   val,
		})
348
		if err == nil {
Jeromy's avatar
Jeromy committed
349
			break
350 351 352
		}

		if !strings.Contains(err.Error(), "too many open files") {
Jeromy's avatar
Jeromy committed
353
			break
354 355
		}

Or Rikon's avatar
Or Rikon committed
356
		log.Errorf("too many open files, retrying in %dms", 100*i)
Jeromy's avatar
Jeromy committed
357
		time.Sleep(time.Millisecond * 100 * time.Duration(i))
358 359 360 361
	}
	return err
}

362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398
func (fs *Datastore) doOp(oper *op) error {
	switch oper.typ {
	case opPut:
		return fs.doPut(oper.key, oper.v)
	case opDelete:
		return fs.doDelete(oper.key)
	case opRename:
		return fs.renameAndUpdateDiskUsage(oper.tmp, oper.path)
	default:
		panic("bad operation, this is a bug")
	}
}

// doWrite optmizes out write operations (put/delete) to the same
// key by queueing them and suceeding all queued
// operations if one of them does. In such case,
// we assume that the first suceeding operation
// on that key was the last one to happen after
// all successful others.
func (fs *Datastore) doWriteOp(oper *op) error {
	keyStr := oper.key.String()

	opRes := fs.opMap.Begin(keyStr)
	if opRes == nil { // nothing to do, a concurrent op succeeded
		return nil
	}

	// Do the operation
	err := fs.doOp(oper)

	// Finish it. If no error, it will signal other operations
	// waiting on this result to succeed. Otherwise, they will
	// retry.
	opRes.Finish(err == nil)
	return err
}

399
func (fs *Datastore) doPut(key datastore.Key, val []byte) error {
400

Tommi Virtanen's avatar
Tommi Virtanen committed
401
	dir, path := fs.encode(key)
402
	if err := fs.makeDir(dir); err != nil {
403
		return err
Tommi Virtanen's avatar
Tommi Virtanen committed
404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
	}

	tmp, err := ioutil.TempFile(dir, "put-")
	if err != nil {
		return err
	}
	closed := false
	removed := false
	defer func() {
		if !closed {
			// silence errcheck
			_ = tmp.Close()
		}
		if !removed {
			// silence errcheck
			_ = os.Remove(tmp.Name())
		}
	}()

	if _, err := tmp.Write(val); err != nil {
		return err
	}
Jeromy's avatar
Jeromy committed
426
	if fs.sync {
427
		if err := syncFile(tmp); err != nil {
Jeromy's avatar
Jeromy committed
428 429
			return err
		}
430
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
431 432 433 434 435
	if err := tmp.Close(); err != nil {
		return err
	}
	closed = true

436
	err = fs.renameAndUpdateDiskUsage(tmp.Name(), path)
Tommi Virtanen's avatar
Tommi Virtanen committed
437 438 439 440 441
	if err != nil {
		return err
	}
	removed = true

Jeromy's avatar
Jeromy committed
442 443 444 445
	if fs.sync {
		if err := syncDir(dir); err != nil {
			return err
		}
446
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
447 448 449
	return nil
}

Jeromy's avatar
Jeromy committed
450 451
func (fs *Datastore) putMany(data map[datastore.Key]interface{}) error {
	var dirsToSync []string
452
	files := make(map[*os.File]*op)
Jeromy's avatar
Jeromy committed
453 454 455 456 457 458 459

	for key, value := range data {
		val, ok := value.([]byte)
		if !ok {
			return datastore.ErrInvalidType
		}
		dir, path := fs.encode(key)
460
		if err := fs.makeDirNoSync(dir); err != nil {
Jeromy's avatar
Jeromy committed
461 462 463 464 465 466 467 468 469 470 471 472 473
			return err
		}
		dirsToSync = append(dirsToSync, dir)

		tmp, err := ioutil.TempFile(dir, "put-")
		if err != nil {
			return err
		}

		if _, err := tmp.Write(val); err != nil {
			return err
		}

474 475 476 477 478 479
		files[tmp] = &op{
			typ:  opRename,
			path: path,
			tmp:  tmp.Name(),
			key:  key,
		}
Jeromy's avatar
Jeromy committed
480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499
	}

	ops := make(map[*os.File]int)

	defer func() {
		for fi, _ := range files {
			val, _ := ops[fi]
			switch val {
			case 0:
				_ = fi.Close()
				fallthrough
			case 1:
				_ = os.Remove(fi.Name())
			}
		}
	}()

	// Now we sync everything
	// sync and close files
	for fi, _ := range files {
Jeromy's avatar
Jeromy committed
500
		if fs.sync {
501
			if err := syncFile(fi); err != nil {
Jeromy's avatar
Jeromy committed
502 503
				return err
			}
Jeromy's avatar
Jeromy committed
504 505 506 507 508 509 510 511 512 513 514
		}

		if err := fi.Close(); err != nil {
			return err
		}

		// signify closed
		ops[fi] = 1
	}

	// move files to their proper places
515 516
	for fi, op := range files {
		fs.doWriteOp(op)
Jeromy's avatar
Jeromy committed
517 518 519 520 521
		// signify removed
		ops[fi] = 2
	}

	// now sync the dirs for those files
Jeromy's avatar
Jeromy committed
522 523 524 525 526
	if fs.sync {
		for _, dir := range dirsToSync {
			if err := syncDir(dir); err != nil {
				return err
			}
Jeromy's avatar
Jeromy committed
527 528
		}

Jeromy's avatar
Jeromy committed
529 530 531 532
		// sync top flatfs dir
		if err := syncDir(fs.path); err != nil {
			return err
		}
Jeromy's avatar
Jeromy committed
533 534 535 536 537
	}

	return nil
}

Tommi Virtanen's avatar
Tommi Virtanen committed
538 539 540 541 542 543 544 545 546 547 548 549 550 551
func (fs *Datastore) Get(key datastore.Key) (value interface{}, err error) {
	_, path := fs.encode(key)
	data, err := ioutil.ReadFile(path)
	if err != nil {
		if os.IsNotExist(err) {
			return nil, datastore.ErrNotFound
		}
		// no specific error to return, so just pass it through
		return nil, err
	}
	return data, nil
}

func (fs *Datastore) Has(key datastore.Key) (exists bool, err error) {
Tommi Virtanen's avatar
Tommi Virtanen committed
552 553 554 555 556 557 558 559 560
	_, path := fs.encode(key)
	switch _, err := os.Stat(path); {
	case err == nil:
		return true, nil
	case os.IsNotExist(err):
		return false, nil
	default:
		return false, err
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
561 562
}

563 564 565
// Delete removes a key/value from the Datastore. Please read
// the Put() explanation about the handling of concurrent write
// operations to the same key.
Tommi Virtanen's avatar
Tommi Virtanen committed
566
func (fs *Datastore) Delete(key datastore.Key) error {
567 568 569 570 571 572 573 574 575 576
	return fs.doWriteOp(&op{
		typ: opDelete,
		key: key,
		v:   nil,
	})
}

// This function always runs within an opLock for the given
// key, and not concurrently.
func (fs *Datastore) doDelete(key datastore.Key) error {
Tommi Virtanen's avatar
Tommi Virtanen committed
577
	_, path := fs.encode(key)
578 579 580

	fSize := fileSize(path)

Tommi Virtanen's avatar
Tommi Virtanen committed
581 582
	switch err := os.Remove(path); {
	case err == nil:
583 584
		newDu := atomic.AddInt64(&fs.diskUsage, -fSize)
		fs.checkpointDiskUsage(newDu)
Tommi Virtanen's avatar
Tommi Virtanen committed
585 586 587 588 589 590
		return nil
	case os.IsNotExist(err):
		return datastore.ErrNotFound
	default:
		return err
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
591 592 593
}

func (fs *Datastore) Query(q query.Query) (query.Results, error) {
594 595 596 597 598 599 600 601 602 603 604
	if (q.Prefix != "" && q.Prefix != "/") ||
		len(q.Filters) > 0 ||
		len(q.Orders) > 0 ||
		q.Limit > 0 ||
		q.Offset > 0 ||
		!q.KeysOnly {
		// TODO this is overly simplistic, but the only caller is
		// `ipfs refs local` for now, and this gets us moving.
		return nil, errors.New("flatfs only supports listing all keys in random order")
	}

605
	reschan := make(chan query.Result, query.KeysOnlyBufSize)
Jeromy's avatar
Jeromy committed
606 607
	go func() {
		defer close(reschan)
608
		err := fs.walkTopLevel(fs.path, reschan)
Jeromy's avatar
Jeromy committed
609
		if err != nil {
610
			reschan <- query.Result{Error: errors.New("walk failed: " + err.Error())}
611
		}
Jeromy's avatar
Jeromy committed
612 613
	}()
	return query.ResultsWithChan(q, reschan), nil
Tommi Virtanen's avatar
Tommi Virtanen committed
614 615
}

616 617 618 619 620
func (fs *Datastore) walkTopLevel(path string, reschan chan query.Result) error {
	dir, err := os.Open(path)
	if err != nil {
		return err
	}
Kevin Atkinson's avatar
Kevin Atkinson committed
621
	defer dir.Close()
622 623 624 625 626
	names, err := dir.Readdirnames(-1)
	if err != nil {
		return err
	}
	for _, dir := range names {
Kevin Atkinson's avatar
Kevin Atkinson committed
627 628 629 630 631

		if len(dir) == 0 || dir[0] == '.' {
			continue
		}

632 633 634 635
		err = fs.walk(filepath.Join(path, dir), reschan)
		if err != nil {
			return err
		}
Kevin Atkinson's avatar
Kevin Atkinson committed
636

637 638 639 640
	}
	return nil
}

641 642 643
// folderSize estimates the diskUsage of a folder by reading
// up to DiskUsageFilesAverage entries in it and assumming any
// other files will have an avereage size.
644
func folderSize(path string, deadline time.Time) (int64, initAccuracy, error) {
645 646 647 648
	var du int64

	folder, err := os.Open(path)
	if err != nil {
649
		return 0, "", err
650 651 652 653 654
	}
	defer folder.Close()

	stat, err := folder.Stat()
	if err != nil {
655
		return 0, "", err
656 657 658 659
	}

	files, err := folder.Readdirnames(-1)
	if err != nil {
660
		return 0, "", err
661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677
	}

	totalFiles := len(files)
	i := 0
	filesProcessed := 0
	maxFiles := DiskUsageFilesAverage
	if maxFiles <= 0 {
		maxFiles = totalFiles
	}

	// randomize file order
	// https://stackoverflow.com/a/42776696
	for i := len(files) - 1; i > 0; i-- {
		j := rand.Intn(i + 1)
		files[i], files[j] = files[j], files[i]
	}

678
	accuracy := exactA
679
	for {
680 681 682
		// Do not process any files after deadline is over
		if time.Now().After(deadline) {
			accuracy = timedoutA
683 684 685
			break
		}

686 687 688 689
		if i >= totalFiles || filesProcessed >= maxFiles {
			if filesProcessed >= maxFiles {
				accuracy = approxA
			}
690 691 692 693 694 695 696 697
			break
		}

		// Stat the file
		fname := files[i]
		subpath := filepath.Join(path, fname)
		st, err := os.Stat(subpath)
		if err != nil {
698
			return 0, "", err
699 700 701 702
		}

		// Find folder size recursively
		if st.IsDir() {
703
			du2, acc, err := folderSize(filepath.Join(subpath), deadline)
704
			if err != nil {
705
				return 0, "", err
706
			}
707
			accuracy = combineAccuracy(acc, accuracy)
708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729
			du += du2
			filesProcessed++
		} else { // in any other case, add the file size
			du += st.Size()
			filesProcessed++
		}

		i++
	}

	nonProcessed := totalFiles - filesProcessed

	// Avg is total size in this folder up to now / total files processed
	// it includes folders ant not folders
	avg := 0.0
	if filesProcessed > 0 {
		avg = float64(du) / float64(filesProcessed)
	}
	duEstimation := int64(avg * float64(nonProcessed))
	du += duEstimation
	du += stat.Size()
	//fmt.Println(path, "total:", totalFiles, "totalStat:", i, "totalFile:", filesProcessed, "left:", nonProcessed, "avg:", int(avg), "est:", int(duEstimation), "du:", du)
730
	return du, accuracy, nil
731 732 733 734
}

// calculateDiskUsage tries to read the DiskUsageFile for a cached
// diskUsage value, otherwise walks the datastore files.
735
// it is only safe to call in Open()
736 737 738
func (fs *Datastore) calculateDiskUsage() error {
	// Try to obtain a previously stored value from disk
	if persDu := fs.readDiskUsageFile(); persDu > 0 {
739
		fs.diskUsage = persDu
740 741 742 743 744
		return nil
	}

	fmt.Printf("Calculating datastore size. This might take %s at most and will happen only once\n", DiskUsageCalcTimeout.String())
	deadline := time.Now().Add(DiskUsageCalcTimeout)
745
	du, accuracy, err := folderSize(fs.path, deadline)
746 747 748
	if err != nil {
		return err
	}
749
	if accuracy == timedoutA {
750 751 752 753 754 755 756
		fmt.Println("WARN: It took to long to calculate the datastore size")
		fmt.Printf("WARN: The total size (%d) is an estimation. You can fix errors by\n", du)
		fmt.Printf("WARN: replacing the %s file with the right disk usage in bytes and\n",
			filepath.Join(fs.path, DiskUsageFile))
		fmt.Println("WARN: re-opening the datastore")
	}

757 758 759 760
	fs.storedValue.accuracy = accuracy
	fs.diskUsage = du
	fs.persistDiskUsageFile()

761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788
	return nil
}

func fileSize(path string) int64 {
	fi, err := os.Stat(path)
	if err != nil {
		return 0
	}
	return fi.Size()
}

// updateDiskUsage reads the size of path and atomically
// increases or decreases the diskUsage variable.
// setting add to false will subtract from disk usage.
func (fs *Datastore) updateDiskUsage(path string, add bool) {
	fsize := fileSize(path)
	if !add {
		fsize = -fsize
	}

	if fsize != 0 {
		newDu := atomic.AddInt64(&fs.diskUsage, fsize)
		fs.checkpointDiskUsage(newDu)
	}
}

func (fs *Datastore) checkpointDiskUsage(newDuInt int64) {
	newDu := float64(newDuInt)
789
	lastCheckpointDu := float64(atomic.LoadInt64(&fs.storedValue.diskUsage))
790 791 792 793 794
	diff := math.Abs(newDu - lastCheckpointDu)

	// If the difference between the checkpointed disk usage and
	// current one is larger than than 1% of the checkpointed: store it.
	if (lastCheckpointDu * diskUsageCheckpointPercent / 100.0) < diff {
795 796 797 798 799 800 801 802 803 804
		fs.signalCheckpoint()
	}
}

func (fs *Datastore) signalCheckpoint() {
	select {
	case fs.checkpointCh <- true:
		// msg send
	default:
		// checkpoint request already pending
805 806 807
	}
}

808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826
func (fs *Datastore) checkpointLoop() {
	fs.checkpointCh = make(chan bool, 1)
	fs.done = make(chan bool)
	go func() {
		for {
			//println("waiting...")
			_, more := <-fs.checkpointCh
			if more {
				//println("checkpoint")
				fs.persistDiskUsageFile()
			} else {
				//println("shutdown")
				fs.done <- true
				return
			}
		}
	}()
}

827
// persistDiskUsageFile updates the diskusage file with the last known
828 829
// value, should only be called during initialization or within the
// checkpoint loop
830
func (fs *Datastore) persistDiskUsageFile() {
831 832 833 834 835 836 837
	du := atomic.LoadInt64(&fs.diskUsage)

	origVal := fs.storedValue.diskUsage
	// update the stored diskUsage value now to prevent unnecessary
	// calls to persistDiskUsageFile.  On error role back the value to
	// the original
	atomic.StoreInt64(&fs.storedValue.diskUsage, du)
838 839 840

	tmp, err := ioutil.TempFile(fs.path, "du-")
	if err != nil {
841
		atomic.StoreInt64(&fs.storedValue.diskUsage, origVal)
842 843
		return
	}
844 845 846 847

	encoder := json.NewEncoder(tmp)
	if err := encoder.Encode(&fs.storedValue); err != nil {
		atomic.StoreInt64(&fs.storedValue.diskUsage, origVal)
848 849
		return
	}
850

851
	if err := tmp.Close(); err != nil {
852
		atomic.StoreInt64(&fs.storedValue.diskUsage, origVal)
853 854 855
		return
	}

856 857 858
	if err := os.Rename(tmp.Name(), filepath.Join(fs.path, DiskUsageFile)); err != nil {
		atomic.StoreInt64(&fs.storedValue.diskUsage, origVal)
	}
859 860
}

861
// readDiskUsageFile is only safe to call in Open()
862 863 864 865 866 867
func (fs *Datastore) readDiskUsageFile() int64 {
	fpath := filepath.Join(fs.path, DiskUsageFile)
	duB, err := ioutil.ReadFile(fpath)
	if err != nil {
		return 0
	}
868
	err = json.Unmarshal(duB, &fs.storedValue)
869 870 871
	if err != nil {
		return 0
	}
872
	return fs.storedValue.diskUsage
873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892
}

// DiskUsage implements the PersistentDatastore interface
// and returns the current disk usage in bytes used by
// this datastore.
//
// The size is approximative and may slightly differ from
// the real disk values.
func (fs *Datastore) DiskUsage() (uint64, error) {
	// it may differ from real disk values if
	// the filesystem has allocated for blocks
	// for a directory because it has many files in it
	// we don't account for "resized" directories.
	// In a large datastore, the differences should be
	// are negligible though.

	du := atomic.LoadInt64(&fs.diskUsage)
	return uint64(du), nil
}

893 894 895 896 897 898 899
// Accuracy returns a string representing the accuracy of the
// DiskUsage() result, the value returned is implementation defined
// and for informational purposes only
func (fs *Datastore) Accuracy() string {
	return string(fs.storedValue.accuracy)
}

900 901 902
func (fs *Datastore) walk(path string, reschan chan query.Result) error {
	dir, err := os.Open(path)
	if err != nil {
903 904 905 906
		if os.IsNotExist(err) {
			// not an error if the file disappeared
			return nil
		}
907 908
		return err
	}
Kevin Atkinson's avatar
Kevin Atkinson committed
909
	defer dir.Close()
910 911 912 913 914 915 916 917 918 919

	// ignore non-directories
	fileInfo, err := dir.Stat()
	if err != nil {
		return err
	}
	if !fileInfo.IsDir() {
		return nil
	}

920 921 922 923 924 925 926 927 928 929 930 931
	names, err := dir.Readdirnames(-1)
	if err != nil {
		return err
	}
	for _, fn := range names {

		if len(fn) == 0 || fn[0] == '.' {
			continue
		}

		key, ok := fs.decode(fn)
		if !ok {
932
			log.Warningf("failed to decode flatfs entry: %s", fn)
933 934 935 936 937 938 939 940 941 942 943 944
			continue
		}

		reschan <- query.Result{
			Entry: query.Entry{
				Key: key.String(),
			},
		}
	}
	return nil
}

945 946 947 948 949 950 951 952 953 954
// Deactivate closes background maintenance threads, most write
// operations will fail but readonly operations will continue to
// function
func (fs *Datastore) deactivate() error {
	if fs.checkpointCh != nil {
		fs.checkpointCh <- true
		close(fs.checkpointCh)
		<-fs.done
		fs.checkpointCh = nil
	}
Jeromy's avatar
Jeromy committed
955 956 957
	return nil
}

958 959 960 961
func (fs *Datastore) Close() error {
	return fs.deactivate()
}

Jeromy's avatar
Jeromy committed
962
type flatfsBatch struct {
Jeromy's avatar
Jeromy committed
963 964 965 966 967 968
	puts    map[datastore.Key]interface{}
	deletes map[datastore.Key]struct{}

	ds *Datastore
}

Jeromy's avatar
Jeromy committed
969
func (fs *Datastore) Batch() (datastore.Batch, error) {
Jeromy's avatar
Jeromy committed
970
	return &flatfsBatch{
Jeromy's avatar
Jeromy committed
971 972 973
		puts:    make(map[datastore.Key]interface{}),
		deletes: make(map[datastore.Key]struct{}),
		ds:      fs,
Jeromy's avatar
Jeromy committed
974
	}, nil
Jeromy's avatar
Jeromy committed
975 976
}

Jeromy's avatar
Jeromy committed
977
func (bt *flatfsBatch) Put(key datastore.Key, val interface{}) error {
Jeromy's avatar
Jeromy committed
978 979 980 981
	bt.puts[key] = val
	return nil
}

Jeromy's avatar
Jeromy committed
982
func (bt *flatfsBatch) Delete(key datastore.Key) error {
Jeromy's avatar
Jeromy committed
983 984 985 986
	bt.deletes[key] = struct{}{}
	return nil
}

Jeromy's avatar
Jeromy committed
987
func (bt *flatfsBatch) Commit() error {
Jeromy's avatar
Jeromy committed
988 989 990 991 992 993 994 995 996 997 998 999 1000
	if err := bt.ds.putMany(bt.puts); err != nil {
		return err
	}

	for k, _ := range bt.deletes {
		if err := bt.ds.Delete(k); err != nil {
			return err
		}
	}

	return nil
}

Tommi Virtanen's avatar
Tommi Virtanen committed
1001 1002 1003
var _ datastore.ThreadSafeDatastore = (*Datastore)(nil)

func (*Datastore) IsThreadSafe() {}