flatfs.go 22.6 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
31
	diskUsageCheckpointTimeout = 2.0 * time.Second
Tommi Virtanen's avatar
Tommi Virtanen committed
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 57
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
)

58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78
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 ""
}

79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
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
95 96
type Datastore struct {
	path string
97

98 99
	shardStr string
	getDir   ShardFunc
Jeromy's avatar
Jeromy committed
100 101 102

	// sychronize all writes and directory changes for added safety
	sync bool
103

104 105 106 107 108 109
	// atmoic operations should always be used with diskUsage
	diskUsage int64

	// these values should only be used during internalization or
	// inside the checkpoint loop
	dirty       bool
110
	storedValue diskUsageValue
111 112 113

	checkpointCh chan bool
	done         chan bool
114 115 116 117

	// opMap handles concurrent write operations (put/delete)
	// to the same key
	opMap *opMap
Tommi Virtanen's avatar
Tommi Virtanen committed
118 119
}

120 121 122 123 124
type diskUsageValue struct {
	diskUsage int64
	accuracy  initAccuracy
}

125 126
type ShardFunc func(string) string

127
type opT int
128

129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176
// 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()
}
177

178
func Create(path string, fun *ShardIdV1) error {
179

kpcyrd's avatar
kpcyrd committed
180
	err := os.Mkdir(path, 0755)
181 182
	if err != nil && !os.IsExist(err) {
		return err
Tommi Virtanen's avatar
Tommi Virtanen committed
183 184
	}

185 186
	dsFun, err := ReadShardFunc(path)
	switch err {
Kevin Atkinson's avatar
Kevin Atkinson committed
187
	case ErrShardingFileMissing:
188 189 190 191 192
		isEmpty, err := DirIsEmpty(path)
		if err != nil {
			return err
		}
		if !isEmpty {
193
			return fmt.Errorf("directory missing %s file: %s", SHARDING_FN, path)
194 195 196
		}

		err = WriteShardFunc(path, fun)
197 198 199
		if err != nil {
			return err
		}
200 201
		err = WriteReadme(path, fun)
		return err
202
	case nil:
203
		if fun.String() != dsFun.String() {
204
			return fmt.Errorf("specified shard func '%s' does not match repo shard func '%s'",
205
				fun.String(), dsFun.String())
206
		}
Kevin Atkinson's avatar
Kevin Atkinson committed
207
		return ErrDatastoreExists
208
	default:
209
		return err
210
	}
211 212
}

213
func Open(path string, syncFiles bool) (*Datastore, error) {
214 215
	_, err := os.Stat(path)
	if os.IsNotExist(err) {
Kevin Atkinson's avatar
Kevin Atkinson committed
216
		return nil, ErrDatastoreDoesNotExist
217 218 219 220
	} else if err != nil {
		return nil, err
	}

221
	shardId, err := ReadShardFunc(path)
222
	if err != nil {
223
		return nil, err
224 225
	}

226
	fs := &Datastore{
227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243
		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
244
	}
245 246 247 248 249 250

	fs.checkpointCh = make(chan bool, 1)
	fs.done = make(chan bool)
	go func() {
		fs.checkpointLoop()
	}()
251 252 253
	return fs, nil
}

254
// convenience method
255
func CreateOrOpen(path string, fun *ShardIdV1, sync bool) (*Datastore, error) {
256
	err := Create(path, fun)
Kevin Atkinson's avatar
Kevin Atkinson committed
257
	if err != nil && err != ErrDatastoreExists {
258 259 260 261 262
		return nil, err
	}
	return Open(path, sync)
}

263 264
func (fs *Datastore) ShardStr() string {
	return fs.shardStr
265 266
}

Tommi Virtanen's avatar
Tommi Virtanen committed
267
func (fs *Datastore) encode(key datastore.Key) (dir, file string) {
268
	noslash := key.String()[1:]
Jeromy's avatar
Jeromy committed
269 270
	dir = filepath.Join(fs.path, fs.getDir(noslash))
	file = filepath.Join(dir, noslash+extension)
Tommi Virtanen's avatar
Tommi Virtanen committed
271 272 273
	return dir, file
}

274
func (fs *Datastore) decode(file string) (key datastore.Key, ok bool) {
Jeromy's avatar
Jeromy committed
275
	if filepath.Ext(file) != extension {
276 277 278
		return datastore.Key{}, false
	}
	name := file[:len(file)-len(extension)]
Jeromy's avatar
Jeromy committed
279
	return datastore.NewKey(name), true
280 281
}

282 283
func (fs *Datastore) makeDir(dir string) error {
	if err := fs.makeDirNoSync(dir); err != nil {
Jeromy's avatar
Jeromy committed
284
		return err
285 286 287 288 289 290
	}

	// 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
291 292 293 294
	if fs.sync {
		if err := syncDir(fs.path); err != nil {
			return err
		}
295 296 297 298
	}
	return nil
}

299
func (fs *Datastore) makeDirNoSync(dir string) error {
kpcyrd's avatar
kpcyrd committed
300
	if err := os.Mkdir(dir, 0755); err != nil {
Jeromy's avatar
Jeromy committed
301 302 303 304 305
		// EEXIST is safe to ignore here, that just means the prefix
		// directory already existed.
		if !os.IsExist(err) {
			return err
		}
306
		return nil
Jeromy's avatar
Jeromy committed
307
	}
308 309 310

	// Track DiskUsage of this NEW folder
	fs.updateDiskUsage(dir, true)
Jeromy's avatar
Jeromy committed
311 312 313
	return nil
}

314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
// 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.
329
	err = os.Rename(tmpPath, path)
330 331 332 333
	fs.updateDiskUsage(path, true)
	return err
}

Jeromy's avatar
Jeromy committed
334
var putMaxRetries = 6
335

336 337 338 339 340 341 342 343 344 345
// 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
346 347 348 349 350 351
func (fs *Datastore) Put(key datastore.Key, value interface{}) error {
	val, ok := value.([]byte)
	if !ok {
		return datastore.ErrInvalidType
	}

352
	var err error
Jeromy's avatar
Jeromy committed
353
	for i := 1; i <= putMaxRetries; i++ {
354 355 356 357 358
		err = fs.doWriteOp(&op{
			typ: opPut,
			key: key,
			v:   val,
		})
359
		if err == nil {
Jeromy's avatar
Jeromy committed
360
			break
361 362 363
		}

		if !strings.Contains(err.Error(), "too many open files") {
Jeromy's avatar
Jeromy committed
364
			break
365 366
		}

Or Rikon's avatar
Or Rikon committed
367
		log.Errorf("too many open files, retrying in %dms", 100*i)
Jeromy's avatar
Jeromy committed
368
		time.Sleep(time.Millisecond * 100 * time.Duration(i))
369 370 371 372
	}
	return err
}

373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409
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
}

410
func (fs *Datastore) doPut(key datastore.Key, val []byte) error {
411

Tommi Virtanen's avatar
Tommi Virtanen committed
412
	dir, path := fs.encode(key)
413
	if err := fs.makeDir(dir); err != nil {
414
		return err
Tommi Virtanen's avatar
Tommi Virtanen committed
415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436
	}

	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
437
	if fs.sync {
438
		if err := syncFile(tmp); err != nil {
Jeromy's avatar
Jeromy committed
439 440
			return err
		}
441
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
442 443 444 445 446
	if err := tmp.Close(); err != nil {
		return err
	}
	closed = true

447
	err = fs.renameAndUpdateDiskUsage(tmp.Name(), path)
Tommi Virtanen's avatar
Tommi Virtanen committed
448 449 450 451 452
	if err != nil {
		return err
	}
	removed = true

Jeromy's avatar
Jeromy committed
453 454 455 456
	if fs.sync {
		if err := syncDir(dir); err != nil {
			return err
		}
457
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
458 459 460
	return nil
}

Jeromy's avatar
Jeromy committed
461 462
func (fs *Datastore) putMany(data map[datastore.Key]interface{}) error {
	var dirsToSync []string
463
	files := make(map[*os.File]*op)
Jeromy's avatar
Jeromy committed
464 465 466 467 468 469 470

	for key, value := range data {
		val, ok := value.([]byte)
		if !ok {
			return datastore.ErrInvalidType
		}
		dir, path := fs.encode(key)
471
		if err := fs.makeDirNoSync(dir); err != nil {
Jeromy's avatar
Jeromy committed
472 473 474 475 476 477 478 479 480 481 482 483 484
			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
		}

485 486 487 488 489 490
		files[tmp] = &op{
			typ:  opRename,
			path: path,
			tmp:  tmp.Name(),
			key:  key,
		}
Jeromy's avatar
Jeromy committed
491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510
	}

	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
511
		if fs.sync {
512
			if err := syncFile(fi); err != nil {
Jeromy's avatar
Jeromy committed
513 514
				return err
			}
Jeromy's avatar
Jeromy committed
515 516 517 518 519 520 521 522 523 524 525
		}

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

		// signify closed
		ops[fi] = 1
	}

	// move files to their proper places
526 527
	for fi, op := range files {
		fs.doWriteOp(op)
Jeromy's avatar
Jeromy committed
528 529 530 531 532
		// signify removed
		ops[fi] = 2
	}

	// now sync the dirs for those files
Jeromy's avatar
Jeromy committed
533 534 535 536 537
	if fs.sync {
		for _, dir := range dirsToSync {
			if err := syncDir(dir); err != nil {
				return err
			}
Jeromy's avatar
Jeromy committed
538 539
		}

Jeromy's avatar
Jeromy committed
540 541 542 543
		// sync top flatfs dir
		if err := syncDir(fs.path); err != nil {
			return err
		}
Jeromy's avatar
Jeromy committed
544 545 546 547 548
	}

	return nil
}

Tommi Virtanen's avatar
Tommi Virtanen committed
549 550 551 552 553 554 555 556 557 558 559 560 561 562
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
563 564 565 566 567 568 569 570 571
	_, 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
572 573
}

574 575 576
// 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
577
func (fs *Datastore) Delete(key datastore.Key) error {
578 579 580 581 582 583 584 585 586 587
	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
588
	_, path := fs.encode(key)
589 590 591

	fSize := fileSize(path)

Tommi Virtanen's avatar
Tommi Virtanen committed
592 593
	switch err := os.Remove(path); {
	case err == nil:
594 595
		atomic.AddInt64(&fs.diskUsage, -fSize)
		fs.checkpointDiskUsage()
Tommi Virtanen's avatar
Tommi Virtanen committed
596 597 598 599 600 601
		return nil
	case os.IsNotExist(err):
		return datastore.ErrNotFound
	default:
		return err
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
602 603 604
}

func (fs *Datastore) Query(q query.Query) (query.Results, error) {
605 606 607 608 609 610 611 612 613 614 615
	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")
	}

616
	reschan := make(chan query.Result, query.KeysOnlyBufSize)
Jeromy's avatar
Jeromy committed
617 618
	go func() {
		defer close(reschan)
619
		err := fs.walkTopLevel(fs.path, reschan)
Jeromy's avatar
Jeromy committed
620
		if err != nil {
621
			reschan <- query.Result{Error: errors.New("walk failed: " + err.Error())}
622
		}
Jeromy's avatar
Jeromy committed
623 624
	}()
	return query.ResultsWithChan(q, reschan), nil
Tommi Virtanen's avatar
Tommi Virtanen committed
625 626
}

627 628 629 630 631
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
632
	defer dir.Close()
633 634 635 636 637
	names, err := dir.Readdirnames(-1)
	if err != nil {
		return err
	}
	for _, dir := range names {
Kevin Atkinson's avatar
Kevin Atkinson committed
638 639 640 641 642

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

643 644 645 646
		err = fs.walk(filepath.Join(path, dir), reschan)
		if err != nil {
			return err
		}
Kevin Atkinson's avatar
Kevin Atkinson committed
647

648 649 650 651
	}
	return nil
}

652 653 654
// 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.
655
func folderSize(path string, deadline time.Time) (int64, initAccuracy, error) {
656 657 658 659
	var du int64

	folder, err := os.Open(path)
	if err != nil {
660
		return 0, "", err
661 662 663 664 665
	}
	defer folder.Close()

	stat, err := folder.Stat()
	if err != nil {
666
		return 0, "", err
667 668 669 670
	}

	files, err := folder.Readdirnames(-1)
	if err != nil {
671
		return 0, "", err
672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688
	}

	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]
	}

689
	accuracy := exactA
690
	for {
691 692 693
		// Do not process any files after deadline is over
		if time.Now().After(deadline) {
			accuracy = timedoutA
694 695 696
			break
		}

697 698 699 700
		if i >= totalFiles || filesProcessed >= maxFiles {
			if filesProcessed >= maxFiles {
				accuracy = approxA
			}
701 702 703 704 705 706 707 708
			break
		}

		// Stat the file
		fname := files[i]
		subpath := filepath.Join(path, fname)
		st, err := os.Stat(subpath)
		if err != nil {
709
			return 0, "", err
710 711 712 713
		}

		// Find folder size recursively
		if st.IsDir() {
714
			du2, acc, err := folderSize(filepath.Join(subpath), deadline)
715
			if err != nil {
716
				return 0, "", err
717
			}
718
			accuracy = combineAccuracy(acc, accuracy)
719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740
			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)
741
	return du, accuracy, nil
742 743 744 745
}

// calculateDiskUsage tries to read the DiskUsageFile for a cached
// diskUsage value, otherwise walks the datastore files.
746
// it is only safe to call in Open()
747 748 749
func (fs *Datastore) calculateDiskUsage() error {
	// Try to obtain a previously stored value from disk
	if persDu := fs.readDiskUsageFile(); persDu > 0 {
750
		fs.diskUsage = persDu
751 752 753 754 755
		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)
756
	du, accuracy, err := folderSize(fs.path, deadline)
757 758 759
	if err != nil {
		return err
	}
760
	if accuracy == timedoutA {
761 762 763 764 765 766 767
		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")
	}

768 769
	fs.storedValue.accuracy = accuracy
	fs.diskUsage = du
770
	fs.writeDiskUsageFile(du)
771

772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792
	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 {
793 794
		atomic.AddInt64(&fs.diskUsage, fsize)
		fs.checkpointDiskUsage()
795 796 797
	}
}

798
func (fs *Datastore) checkpointDiskUsage() {
799 800 801 802 803
	select {
	case fs.checkpointCh <- true:
		// msg send
	default:
		// checkpoint request already pending
804 805 806
	}
}

807
func (fs *Datastore) checkpointLoop() {
808 809
	timerActive := true
	timer := time.NewTimer(0)
810
	for {
811 812 813 814 815
		select {
		case _, more := <-fs.checkpointCh:
			du := atomic.LoadInt64(&fs.diskUsage)
			if !more { // shutting down
				fs.writeDiskUsageFile(du)
816 817 818
				if fs.dirty {
					log.Errorf("could not right final value of disk usage to file, future estimates may be inaccurate")
				}
819 820 821 822 823 824 825
				fs.done <- true
				return
			}
			fs.dirty = true
			// If the difference between the checkpointed disk usage and
			// current one is larger than than `diskUsageCheckpointPercent`
			// of the checkpointed: store it.
826 827 828 829 830
			newDu := float64(du)
			lastCheckpointDu := float64(fs.storedValue.diskUsage)
			diff := math.Abs(newDu - lastCheckpointDu)
			if (lastCheckpointDu * diskUsageCheckpointPercent / 100.0) < diff {
				fs.writeDiskUsageFile(du)
831
			}
832 833 834 835 836 837 838 839 840 841 842
			// Otherwise insure the value will be written to disk after
			// `diskUsageCheckpointTimeout`
			if fs.dirty && !timerActive {
				timer.Reset(diskUsageCheckpointTimeout)
			}
		case <-timer.C:
			timerActive = false
			if fs.dirty {
				du := atomic.LoadInt64(&fs.diskUsage)
				fs.writeDiskUsageFile(du)
			}
843
		}
844
	}
845 846
}

847
func (fs *Datastore) writeDiskUsageFile(du int64) {
848 849
	tmp, err := ioutil.TempFile(fs.path, "du-")
	if err != nil {
850
		log.Warningf("cound not write disk usage: %v", err)
851 852
		return
	}
853 854 855

	encoder := json.NewEncoder(tmp)
	if err := encoder.Encode(&fs.storedValue); err != nil {
856
		log.Warningf("cound not write disk usage: %v", err)
857 858
		return
	}
859

860
	if err := tmp.Close(); err != nil {
861
		log.Warningf("cound not write disk usage: %v", err)
862 863 864
		return
	}

865
	if err := os.Rename(tmp.Name(), filepath.Join(fs.path, DiskUsageFile)); err != nil {
866
		log.Warningf("cound not write disk usage: %v", err)
867
		return
868
	}
869 870 871

	fs.storedValue.diskUsage = du
	fs.dirty = false
872 873
}

874
// readDiskUsageFile is only safe to call in Open()
875 876 877 878 879 880
func (fs *Datastore) readDiskUsageFile() int64 {
	fpath := filepath.Join(fs.path, DiskUsageFile)
	duB, err := ioutil.ReadFile(fpath)
	if err != nil {
		return 0
	}
881
	err = json.Unmarshal(duB, &fs.storedValue)
882 883 884
	if err != nil {
		return 0
	}
885
	return fs.storedValue.diskUsage
886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905
}

// 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
}

906 907 908 909 910 911 912
// 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)
}

913 914 915
func (fs *Datastore) walk(path string, reschan chan query.Result) error {
	dir, err := os.Open(path)
	if err != nil {
916 917 918 919
		if os.IsNotExist(err) {
			// not an error if the file disappeared
			return nil
		}
920 921
		return err
	}
Kevin Atkinson's avatar
Kevin Atkinson committed
922
	defer dir.Close()
923 924 925 926 927 928 929 930 931 932

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

933 934 935 936 937 938 939 940 941 942 943 944
	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 {
945
			log.Warningf("failed to decode flatfs entry: %s", fn)
946 947 948 949 950 951 952 953 954 955 956 957
			continue
		}

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

958 959 960 961 962 963 964 965 966 967
// 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
968 969 970
	return nil
}

971 972 973 974
func (fs *Datastore) Close() error {
	return fs.deactivate()
}

Jeromy's avatar
Jeromy committed
975
type flatfsBatch struct {
Jeromy's avatar
Jeromy committed
976 977 978 979 980 981
	puts    map[datastore.Key]interface{}
	deletes map[datastore.Key]struct{}

	ds *Datastore
}

Jeromy's avatar
Jeromy committed
982
func (fs *Datastore) Batch() (datastore.Batch, error) {
Jeromy's avatar
Jeromy committed
983
	return &flatfsBatch{
Jeromy's avatar
Jeromy committed
984 985 986
		puts:    make(map[datastore.Key]interface{}),
		deletes: make(map[datastore.Key]struct{}),
		ds:      fs,
Jeromy's avatar
Jeromy committed
987
	}, nil
Jeromy's avatar
Jeromy committed
988 989
}

Jeromy's avatar
Jeromy committed
990
func (bt *flatfsBatch) Put(key datastore.Key, val interface{}) error {
Jeromy's avatar
Jeromy committed
991 992 993 994
	bt.puts[key] = val
	return nil
}

Jeromy's avatar
Jeromy committed
995
func (bt *flatfsBatch) Delete(key datastore.Key) error {
Jeromy's avatar
Jeromy committed
996 997 998 999
	bt.deletes[key] = struct{}{}
	return nil
}

Jeromy's avatar
Jeromy committed
1000
func (bt *flatfsBatch) Commit() error {
Jeromy's avatar
Jeromy committed
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
	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
1014 1015 1016
var _ datastore.ThreadSafeDatastore = (*Datastore)(nil)

func (*Datastore) IsThreadSafe() {}