flatfs.go 20 KB
Newer Older
Tommi Virtanen's avatar
Tommi Virtanen committed
1 2 3 4 5 6 7
// 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 (
	"errors"
8
	"fmt"
Tommi Virtanen's avatar
Tommi Virtanen committed
9
	"io/ioutil"
10 11
	"math"
	"math/rand"
Tommi Virtanen's avatar
Tommi Virtanen committed
12
	"os"
Jeromy's avatar
Jeromy committed
13
	"path/filepath"
14
	"strconv"
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"
Jakub Sztandera's avatar
Jakub Sztandera committed
22
	"github.com/jbenet/go-os-rename"
23

Jakub Sztandera's avatar
Jakub Sztandera committed
24
	logging "github.com/ipfs/go-log"
Tommi Virtanen's avatar
Tommi Virtanen committed
25 26
)

27 28
var log = logging.Logger("flatfs")

Tommi Virtanen's avatar
Tommi Virtanen committed
29
const (
30 31
	extension                  = ".data"
	diskUsageCheckpointPercent = 1.0
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 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
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
)

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
74 75
type Datastore struct {
	path string
76

77 78
	shardStr string
	getDir   ShardFunc
Jeromy's avatar
Jeromy committed
79 80 81

	// sychronize all writes and directory changes for added safety
	sync bool
82 83 84 85 86 87 88

	diskUsage           int64
	diskUsageCheckpoint int64

	// opMap handles concurrent write operations (put/delete)
	// to the same key
	opMap *opMap
Tommi Virtanen's avatar
Tommi Virtanen committed
89 90
}

91 92
type ShardFunc func(string) string

93
type opT int
94

95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142
// 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()
}
143

144
func Create(path string, fun *ShardIdV1) error {
145

kpcyrd's avatar
kpcyrd committed
146
	err := os.Mkdir(path, 0755)
147 148
	if err != nil && !os.IsExist(err) {
		return err
Tommi Virtanen's avatar
Tommi Virtanen committed
149 150
	}

151 152
	dsFun, err := ReadShardFunc(path)
	switch err {
Kevin Atkinson's avatar
Kevin Atkinson committed
153
	case ErrShardingFileMissing:
154 155 156 157 158
		isEmpty, err := DirIsEmpty(path)
		if err != nil {
			return err
		}
		if !isEmpty {
159
			return fmt.Errorf("directory missing %s file: %s", SHARDING_FN, path)
160 161 162
		}

		err = WriteShardFunc(path, fun)
163 164 165
		if err != nil {
			return err
		}
166 167
		err = WriteReadme(path, fun)
		return err
168
	case nil:
169
		if fun.String() != dsFun.String() {
170
			return fmt.Errorf("specified shard func '%s' does not match repo shard func '%s'",
171
				fun.String(), dsFun.String())
172
		}
Kevin Atkinson's avatar
Kevin Atkinson committed
173
		return ErrDatastoreExists
174
	default:
175
		return err
176
	}
177 178
}

179
func Open(path string, syncFiles bool) (*Datastore, error) {
180 181
	_, err := os.Stat(path)
	if os.IsNotExist(err) {
Kevin Atkinson's avatar
Kevin Atkinson committed
182
		return nil, ErrDatastoreDoesNotExist
183 184 185 186
	} else if err != nil {
		return nil, err
	}

187
	shardId, err := ReadShardFunc(path)
188
	if err != nil {
189
		return nil, err
190 191
	}

192
	fs := &Datastore{
193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209
		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
210
	}
211 212 213
	return fs, nil
}

214
// convenience method
215
func CreateOrOpen(path string, fun *ShardIdV1, sync bool) (*Datastore, error) {
216
	err := Create(path, fun)
Kevin Atkinson's avatar
Kevin Atkinson committed
217
	if err != nil && err != ErrDatastoreExists {
218 219 220 221 222
		return nil, err
	}
	return Open(path, sync)
}

223 224
func (fs *Datastore) ShardStr() string {
	return fs.shardStr
225 226
}

Tommi Virtanen's avatar
Tommi Virtanen committed
227
func (fs *Datastore) encode(key datastore.Key) (dir, file string) {
228
	noslash := key.String()[1:]
Jeromy's avatar
Jeromy committed
229 230
	dir = filepath.Join(fs.path, fs.getDir(noslash))
	file = filepath.Join(dir, noslash+extension)
Tommi Virtanen's avatar
Tommi Virtanen committed
231 232 233
	return dir, file
}

234
func (fs *Datastore) decode(file string) (key datastore.Key, ok bool) {
Jeromy's avatar
Jeromy committed
235
	if filepath.Ext(file) != extension {
236 237 238
		return datastore.Key{}, false
	}
	name := file[:len(file)-len(extension)]
Jeromy's avatar
Jeromy committed
239
	return datastore.NewKey(name), true
240 241
}

242 243
func (fs *Datastore) makeDir(dir string) error {
	if err := fs.makeDirNoSync(dir); err != nil {
Jeromy's avatar
Jeromy committed
244
		return err
245 246 247 248 249 250
	}

	// 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
251 252 253 254
	if fs.sync {
		if err := syncDir(fs.path); err != nil {
			return err
		}
255 256 257 258
	}
	return nil
}

259
func (fs *Datastore) makeDirNoSync(dir string) error {
kpcyrd's avatar
kpcyrd committed
260
	if err := os.Mkdir(dir, 0755); err != nil {
Jeromy's avatar
Jeromy committed
261 262 263 264 265
		// EEXIST is safe to ignore here, that just means the prefix
		// directory already existed.
		if !os.IsExist(err) {
			return err
		}
266
		return nil
Jeromy's avatar
Jeromy committed
267
	}
268 269 270

	// Track DiskUsage of this NEW folder
	fs.updateDiskUsage(dir, true)
Jeromy's avatar
Jeromy committed
271 272 273
	return nil
}

274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293
// 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.
	err = osrename.Rename(tmpPath, path)
	fs.updateDiskUsage(path, true)
	return err
}

Jeromy's avatar
Jeromy committed
294
var putMaxRetries = 6
295

296 297 298 299 300 301 302 303 304 305
// 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
306 307 308 309 310 311
func (fs *Datastore) Put(key datastore.Key, value interface{}) error {
	val, ok := value.([]byte)
	if !ok {
		return datastore.ErrInvalidType
	}

312
	var err error
Jeromy's avatar
Jeromy committed
313
	for i := 1; i <= putMaxRetries; i++ {
314 315 316 317 318
		err = fs.doWriteOp(&op{
			typ: opPut,
			key: key,
			v:   val,
		})
319
		if err == nil {
Jeromy's avatar
Jeromy committed
320
			break
321 322 323
		}

		if !strings.Contains(err.Error(), "too many open files") {
Jeromy's avatar
Jeromy committed
324
			break
325 326
		}

Or Rikon's avatar
Or Rikon committed
327
		log.Errorf("too many open files, retrying in %dms", 100*i)
Jeromy's avatar
Jeromy committed
328
		time.Sleep(time.Millisecond * 100 * time.Duration(i))
329 330 331 332
	}
	return err
}

333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369
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
}

370
func (fs *Datastore) doPut(key datastore.Key, val []byte) error {
371

Tommi Virtanen's avatar
Tommi Virtanen committed
372
	dir, path := fs.encode(key)
373
	if err := fs.makeDir(dir); err != nil {
374
		return err
Tommi Virtanen's avatar
Tommi Virtanen committed
375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396
	}

	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
397
	if fs.sync {
398
		if err := syncFile(tmp); err != nil {
Jeromy's avatar
Jeromy committed
399 400
			return err
		}
401
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
402 403 404 405 406
	if err := tmp.Close(); err != nil {
		return err
	}
	closed = true

407
	err = fs.renameAndUpdateDiskUsage(tmp.Name(), path)
Tommi Virtanen's avatar
Tommi Virtanen committed
408 409 410 411 412
	if err != nil {
		return err
	}
	removed = true

Jeromy's avatar
Jeromy committed
413 414 415 416
	if fs.sync {
		if err := syncDir(dir); err != nil {
			return err
		}
417
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
418 419 420
	return nil
}

Jeromy's avatar
Jeromy committed
421 422
func (fs *Datastore) putMany(data map[datastore.Key]interface{}) error {
	var dirsToSync []string
423
	files := make(map[*os.File]*op)
Jeromy's avatar
Jeromy committed
424 425 426 427 428 429 430

	for key, value := range data {
		val, ok := value.([]byte)
		if !ok {
			return datastore.ErrInvalidType
		}
		dir, path := fs.encode(key)
431
		if err := fs.makeDirNoSync(dir); err != nil {
Jeromy's avatar
Jeromy committed
432 433 434 435 436 437 438 439 440 441 442 443 444
			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
		}

445 446 447 448 449 450
		files[tmp] = &op{
			typ:  opRename,
			path: path,
			tmp:  tmp.Name(),
			key:  key,
		}
Jeromy's avatar
Jeromy committed
451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470
	}

	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
471
		if fs.sync {
472
			if err := syncFile(fi); err != nil {
Jeromy's avatar
Jeromy committed
473 474
				return err
			}
Jeromy's avatar
Jeromy committed
475 476 477 478 479 480 481 482 483 484 485
		}

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

		// signify closed
		ops[fi] = 1
	}

	// move files to their proper places
486 487
	for fi, op := range files {
		fs.doWriteOp(op)
Jeromy's avatar
Jeromy committed
488 489 490 491 492
		// signify removed
		ops[fi] = 2
	}

	// now sync the dirs for those files
Jeromy's avatar
Jeromy committed
493 494 495 496 497
	if fs.sync {
		for _, dir := range dirsToSync {
			if err := syncDir(dir); err != nil {
				return err
			}
Jeromy's avatar
Jeromy committed
498 499
		}

Jeromy's avatar
Jeromy committed
500 501 502 503
		// sync top flatfs dir
		if err := syncDir(fs.path); err != nil {
			return err
		}
Jeromy's avatar
Jeromy committed
504 505 506 507 508
	}

	return nil
}

Tommi Virtanen's avatar
Tommi Virtanen committed
509 510 511 512 513 514 515 516 517 518 519 520 521 522
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
523 524 525 526 527 528 529 530 531
	_, 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
532 533
}

534 535 536
// 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
537
func (fs *Datastore) Delete(key datastore.Key) error {
538 539 540 541 542 543 544 545 546 547
	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
548
	_, path := fs.encode(key)
549 550 551

	fSize := fileSize(path)

Tommi Virtanen's avatar
Tommi Virtanen committed
552 553
	switch err := os.Remove(path); {
	case err == nil:
554 555
		newDu := atomic.AddInt64(&fs.diskUsage, -fSize)
		fs.checkpointDiskUsage(newDu)
Tommi Virtanen's avatar
Tommi Virtanen committed
556 557 558 559 560 561
		return nil
	case os.IsNotExist(err):
		return datastore.ErrNotFound
	default:
		return err
	}
Tommi Virtanen's avatar
Tommi Virtanen committed
562 563 564
}

func (fs *Datastore) Query(q query.Query) (query.Results, error) {
565 566 567 568 569 570 571 572 573 574 575
	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")
	}

576
	reschan := make(chan query.Result, query.KeysOnlyBufSize)
Jeromy's avatar
Jeromy committed
577 578
	go func() {
		defer close(reschan)
579
		err := fs.walkTopLevel(fs.path, reschan)
Jeromy's avatar
Jeromy committed
580
		if err != nil {
581
			reschan <- query.Result{Error: errors.New("walk failed: " + err.Error())}
582
		}
Jeromy's avatar
Jeromy committed
583 584
	}()
	return query.ResultsWithChan(q, reschan), nil
Tommi Virtanen's avatar
Tommi Virtanen committed
585 586
}

587 588 589 590 591
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
592
	defer dir.Close()
593 594 595 596 597
	names, err := dir.Readdirnames(-1)
	if err != nil {
		return err
	}
	for _, dir := range names {
Kevin Atkinson's avatar
Kevin Atkinson committed
598 599 600 601 602

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

603 604 605 606
		err = fs.walk(filepath.Join(path, dir), reschan)
		if err != nil {
			return err
		}
Kevin Atkinson's avatar
Kevin Atkinson committed
607

608 609 610 611
	}
	return nil
}

612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
// 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.
func folderSize(path string, deadline time.Time) (int64, error) {
	var du int64

	folder, err := os.Open(path)
	if err != nil {
		return 0, err
	}
	defer folder.Close()

	stat, err := folder.Stat()
	if err != nil {
		return 0, err
	}

	files, err := folder.Readdirnames(-1)
	if err != nil {
		return 0, err
	}

	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]
	}

	for {
		if i >= totalFiles || filesProcessed >= maxFiles {
			break
		}

		// Do not process any files after deadline is over
		if time.Now().After(deadline) {
			break
		}

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

		// Find folder size recursively
		if st.IsDir() {
			du2, err := folderSize(filepath.Join(subpath), deadline)
			if err != nil {
				return 0, err
			}
			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)
	return du, nil
}

// calculateDiskUsage tries to read the DiskUsageFile for a cached
// diskUsage value, otherwise walks the datastore files.
func (fs *Datastore) calculateDiskUsage() error {
	// Try to obtain a previously stored value from disk
	if persDu := fs.readDiskUsageFile(); persDu > 0 {
		atomic.StoreInt64(&fs.diskUsage, persDu)
		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)
	du, err := folderSize(fs.path, deadline)
	if err != nil {
		return err
	}
	if time.Now().After(deadline) {
		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")
	}

	atomic.StoreInt64(&fs.diskUsage, du)
	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)
	lastCheckpointDu := float64(atomic.LoadInt64(&fs.diskUsageCheckpoint))
	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 {
		fs.persistDiskUsageFile()
	}
}

func (fs *Datastore) persistDiskUsageFile() {
	// Store DiskUsage on clean shutdowns.
	du, err := fs.DiskUsage()
	if err != nil {
		// do not store on errors. just ignore them
		return
	}

	atomic.StoreInt64(&fs.diskUsageCheckpoint, int64(du))

	duB := []byte(fmt.Sprintf("%d", du))
	tmp, err := ioutil.TempFile(fs.path, "du-")
	if err != nil {
		return
	}
	if _, err := tmp.Write(duB); err != nil {
		return
	}
	if err := tmp.Close(); err != nil {
		return
	}

	osrename.Rename(tmp.Name(), filepath.Join(fs.path, DiskUsageFile))
}

func (fs *Datastore) readDiskUsageFile() int64 {
	fpath := filepath.Join(fs.path, DiskUsageFile)
	duB, err := ioutil.ReadFile(fpath)
	if err != nil {
		return 0
	}
	i, err := strconv.ParseInt(string(duB), 10, 64)
	if err != nil {
		return 0
	}
	atomic.StoreInt64(&fs.diskUsageCheckpoint, i)
	return i
}

// 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
}

817 818 819 820 821
func (fs *Datastore) walk(path string, reschan chan query.Result) error {
	dir, err := os.Open(path)
	if err != nil {
		return err
	}
Kevin Atkinson's avatar
Kevin Atkinson committed
822
	defer dir.Close()
823 824 825 826 827 828 829 830 831 832

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

833 834 835 836 837 838 839 840 841 842 843 844
	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 {
845
			log.Warningf("failed to decode flatfs entry: %s", fn)
846 847 848 849 850 851 852 853 854 855 856 857
			continue
		}

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

Jeromy's avatar
Jeromy committed
858
func (fs *Datastore) Close() error {
859
	fs.persistDiskUsageFile()
Jeromy's avatar
Jeromy committed
860 861 862
	return nil
}

Jeromy's avatar
Jeromy committed
863
type flatfsBatch struct {
Jeromy's avatar
Jeromy committed
864 865 866 867 868 869
	puts    map[datastore.Key]interface{}
	deletes map[datastore.Key]struct{}

	ds *Datastore
}

Jeromy's avatar
Jeromy committed
870
func (fs *Datastore) Batch() (datastore.Batch, error) {
Jeromy's avatar
Jeromy committed
871
	return &flatfsBatch{
Jeromy's avatar
Jeromy committed
872 873 874
		puts:    make(map[datastore.Key]interface{}),
		deletes: make(map[datastore.Key]struct{}),
		ds:      fs,
Jeromy's avatar
Jeromy committed
875
	}, nil
Jeromy's avatar
Jeromy committed
876 877
}

Jeromy's avatar
Jeromy committed
878
func (bt *flatfsBatch) Put(key datastore.Key, val interface{}) error {
Jeromy's avatar
Jeromy committed
879 880 881 882
	bt.puts[key] = val
	return nil
}

Jeromy's avatar
Jeromy committed
883
func (bt *flatfsBatch) Delete(key datastore.Key) error {
Jeromy's avatar
Jeromy committed
884 885 886 887
	bt.deletes[key] = struct{}{}
	return nil
}

Jeromy's avatar
Jeromy committed
888
func (bt *flatfsBatch) Commit() error {
Jeromy's avatar
Jeromy committed
889 890 891 892 893 894 895 896 897 898 899 900 901
	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
902 903 904
var _ datastore.ThreadSafeDatastore = (*Datastore)(nil)

func (*Datastore) IsThreadSafe() {}