blob: 524dc0a6e3f197b66d9b3b7b8ef21c522c285cb1 [file] [log] [blame]
Scott Bakere7144bc2019-10-01 14:16:47 -07001// Diskv (disk-vee) is a simple, persistent, key-value store.
2// It stores all data flatly on the filesystem.
3
4package diskv
5
6import (
7 "bytes"
8 "errors"
9 "fmt"
10 "io"
11 "io/ioutil"
12 "os"
13 "path/filepath"
14 "strings"
15 "sync"
16 "syscall"
17)
18
19const (
20 defaultBasePath = "diskv"
21 defaultFilePerm os.FileMode = 0666
22 defaultPathPerm os.FileMode = 0777
23)
24
25var (
26 defaultTransform = func(s string) []string { return []string{} }
27 errCanceled = errors.New("canceled")
28 errEmptyKey = errors.New("empty key")
29 errBadKey = errors.New("bad key")
30 errImportDirectory = errors.New("can't import a directory")
31)
32
33// TransformFunction transforms a key into a slice of strings, with each
34// element in the slice representing a directory in the file path where the
35// key's entry will eventually be stored.
36//
37// For example, if TransformFunc transforms "abcdef" to ["ab", "cde", "f"],
38// the final location of the data file will be <basedir>/ab/cde/f/abcdef
39type TransformFunction func(s string) []string
40
41// Options define a set of properties that dictate Diskv behavior.
42// All values are optional.
43type Options struct {
44 BasePath string
45 Transform TransformFunction
46 CacheSizeMax uint64 // bytes
47 PathPerm os.FileMode
48 FilePerm os.FileMode
49 // If TempDir is set, it will enable filesystem atomic writes by
50 // writing temporary files to that location before being moved
51 // to BasePath.
52 // Note that TempDir MUST be on the same device/partition as
53 // BasePath.
54 TempDir string
55
56 Index Index
57 IndexLess LessFunction
58
59 Compression Compression
60}
61
62// Diskv implements the Diskv interface. You shouldn't construct Diskv
63// structures directly; instead, use the New constructor.
64type Diskv struct {
65 Options
66 mu sync.RWMutex
67 cache map[string][]byte
68 cacheSize uint64
69}
70
71// New returns an initialized Diskv structure, ready to use.
72// If the path identified by baseDir already contains data,
73// it will be accessible, but not yet cached.
74func New(o Options) *Diskv {
75 if o.BasePath == "" {
76 o.BasePath = defaultBasePath
77 }
78 if o.Transform == nil {
79 o.Transform = defaultTransform
80 }
81 if o.PathPerm == 0 {
82 o.PathPerm = defaultPathPerm
83 }
84 if o.FilePerm == 0 {
85 o.FilePerm = defaultFilePerm
86 }
87
88 d := &Diskv{
89 Options: o,
90 cache: map[string][]byte{},
91 cacheSize: 0,
92 }
93
94 if d.Index != nil && d.IndexLess != nil {
95 d.Index.Initialize(d.IndexLess, d.Keys(nil))
96 }
97
98 return d
99}
100
101// Write synchronously writes the key-value pair to disk, making it immediately
102// available for reads. Write relies on the filesystem to perform an eventual
103// sync to physical media. If you need stronger guarantees, see WriteStream.
104func (d *Diskv) Write(key string, val []byte) error {
105 return d.WriteStream(key, bytes.NewBuffer(val), false)
106}
107
108// WriteStream writes the data represented by the io.Reader to the disk, under
109// the provided key. If sync is true, WriteStream performs an explicit sync on
110// the file as soon as it's written.
111//
112// bytes.Buffer provides io.Reader semantics for basic data types.
113func (d *Diskv) WriteStream(key string, r io.Reader, sync bool) error {
114 if len(key) <= 0 {
115 return errEmptyKey
116 }
117
118 d.mu.Lock()
119 defer d.mu.Unlock()
120
121 return d.writeStreamWithLock(key, r, sync)
122}
123
124// createKeyFileWithLock either creates the key file directly, or
125// creates a temporary file in TempDir if it is set.
126func (d *Diskv) createKeyFileWithLock(key string) (*os.File, error) {
127 if d.TempDir != "" {
128 if err := os.MkdirAll(d.TempDir, d.PathPerm); err != nil {
129 return nil, fmt.Errorf("temp mkdir: %s", err)
130 }
131 f, err := ioutil.TempFile(d.TempDir, "")
132 if err != nil {
133 return nil, fmt.Errorf("temp file: %s", err)
134 }
135
136 if err := f.Chmod(d.FilePerm); err != nil {
137 f.Close() // error deliberately ignored
138 os.Remove(f.Name()) // error deliberately ignored
139 return nil, fmt.Errorf("chmod: %s", err)
140 }
141 return f, nil
142 }
143
144 mode := os.O_WRONLY | os.O_CREATE | os.O_TRUNC // overwrite if exists
145 f, err := os.OpenFile(d.completeFilename(key), mode, d.FilePerm)
146 if err != nil {
147 return nil, fmt.Errorf("open file: %s", err)
148 }
149 return f, nil
150}
151
152// writeStream does no input validation checking.
153func (d *Diskv) writeStreamWithLock(key string, r io.Reader, sync bool) error {
154 if err := d.ensurePathWithLock(key); err != nil {
155 return fmt.Errorf("ensure path: %s", err)
156 }
157
158 f, err := d.createKeyFileWithLock(key)
159 if err != nil {
160 return fmt.Errorf("create key file: %s", err)
161 }
162
163 wc := io.WriteCloser(&nopWriteCloser{f})
164 if d.Compression != nil {
165 wc, err = d.Compression.Writer(f)
166 if err != nil {
167 f.Close() // error deliberately ignored
168 os.Remove(f.Name()) // error deliberately ignored
169 return fmt.Errorf("compression writer: %s", err)
170 }
171 }
172
173 if _, err := io.Copy(wc, r); err != nil {
174 f.Close() // error deliberately ignored
175 os.Remove(f.Name()) // error deliberately ignored
176 return fmt.Errorf("i/o copy: %s", err)
177 }
178
179 if err := wc.Close(); err != nil {
180 f.Close() // error deliberately ignored
181 os.Remove(f.Name()) // error deliberately ignored
182 return fmt.Errorf("compression close: %s", err)
183 }
184
185 if sync {
186 if err := f.Sync(); err != nil {
187 f.Close() // error deliberately ignored
188 os.Remove(f.Name()) // error deliberately ignored
189 return fmt.Errorf("file sync: %s", err)
190 }
191 }
192
193 if err := f.Close(); err != nil {
194 return fmt.Errorf("file close: %s", err)
195 }
196
197 if f.Name() != d.completeFilename(key) {
198 if err := os.Rename(f.Name(), d.completeFilename(key)); err != nil {
199 os.Remove(f.Name()) // error deliberately ignored
200 return fmt.Errorf("rename: %s", err)
201 }
202 }
203
204 if d.Index != nil {
205 d.Index.Insert(key)
206 }
207
208 d.bustCacheWithLock(key) // cache only on read
209
210 return nil
211}
212
213// Import imports the source file into diskv under the destination key. If the
214// destination key already exists, it's overwritten. If move is true, the
215// source file is removed after a successful import.
216func (d *Diskv) Import(srcFilename, dstKey string, move bool) (err error) {
217 if dstKey == "" {
218 return errEmptyKey
219 }
220
221 if fi, err := os.Stat(srcFilename); err != nil {
222 return err
223 } else if fi.IsDir() {
224 return errImportDirectory
225 }
226
227 d.mu.Lock()
228 defer d.mu.Unlock()
229
230 if err := d.ensurePathWithLock(dstKey); err != nil {
231 return fmt.Errorf("ensure path: %s", err)
232 }
233
234 if move {
235 if err := syscall.Rename(srcFilename, d.completeFilename(dstKey)); err == nil {
236 d.bustCacheWithLock(dstKey)
237 return nil
238 } else if err != syscall.EXDEV {
239 // If it failed due to being on a different device, fall back to copying
240 return err
241 }
242 }
243
244 f, err := os.Open(srcFilename)
245 if err != nil {
246 return err
247 }
248 defer f.Close()
249 err = d.writeStreamWithLock(dstKey, f, false)
250 if err == nil && move {
251 err = os.Remove(srcFilename)
252 }
253 return err
254}
255
256// Read reads the key and returns the value.
257// If the key is available in the cache, Read won't touch the disk.
258// If the key is not in the cache, Read will have the side-effect of
259// lazily caching the value.
260func (d *Diskv) Read(key string) ([]byte, error) {
261 rc, err := d.ReadStream(key, false)
262 if err != nil {
263 return []byte{}, err
264 }
265 defer rc.Close()
266 return ioutil.ReadAll(rc)
267}
268
269// ReadStream reads the key and returns the value (data) as an io.ReadCloser.
270// If the value is cached from a previous read, and direct is false,
271// ReadStream will use the cached value. Otherwise, it will return a handle to
272// the file on disk, and cache the data on read.
273//
274// If direct is true, ReadStream will lazily delete any cached value for the
275// key, and return a direct handle to the file on disk.
276//
277// If compression is enabled, ReadStream taps into the io.Reader stream prior
278// to decompression, and caches the compressed data.
279func (d *Diskv) ReadStream(key string, direct bool) (io.ReadCloser, error) {
280 d.mu.RLock()
281 defer d.mu.RUnlock()
282
283 if val, ok := d.cache[key]; ok {
284 if !direct {
285 buf := bytes.NewBuffer(val)
286 if d.Compression != nil {
287 return d.Compression.Reader(buf)
288 }
289 return ioutil.NopCloser(buf), nil
290 }
291
292 go func() {
293 d.mu.Lock()
294 defer d.mu.Unlock()
295 d.uncacheWithLock(key, uint64(len(val)))
296 }()
297 }
298
299 return d.readWithRLock(key)
300}
301
302// read ignores the cache, and returns an io.ReadCloser representing the
303// decompressed data for the given key, streamed from the disk. Clients should
304// acquire a read lock on the Diskv and check the cache themselves before
305// calling read.
306func (d *Diskv) readWithRLock(key string) (io.ReadCloser, error) {
307 filename := d.completeFilename(key)
308
309 fi, err := os.Stat(filename)
310 if err != nil {
311 return nil, err
312 }
313 if fi.IsDir() {
314 return nil, os.ErrNotExist
315 }
316
317 f, err := os.Open(filename)
318 if err != nil {
319 return nil, err
320 }
321
322 var r io.Reader
323 if d.CacheSizeMax > 0 {
324 r = newSiphon(f, d, key)
325 } else {
326 r = &closingReader{f}
327 }
328
329 var rc = io.ReadCloser(ioutil.NopCloser(r))
330 if d.Compression != nil {
331 rc, err = d.Compression.Reader(r)
332 if err != nil {
333 return nil, err
334 }
335 }
336
337 return rc, nil
338}
339
340// closingReader provides a Reader that automatically closes the
341// embedded ReadCloser when it reaches EOF
342type closingReader struct {
343 rc io.ReadCloser
344}
345
346func (cr closingReader) Read(p []byte) (int, error) {
347 n, err := cr.rc.Read(p)
348 if err == io.EOF {
349 if closeErr := cr.rc.Close(); closeErr != nil {
350 return n, closeErr // close must succeed for Read to succeed
351 }
352 }
353 return n, err
354}
355
356// siphon is like a TeeReader: it copies all data read through it to an
357// internal buffer, and moves that buffer to the cache at EOF.
358type siphon struct {
359 f *os.File
360 d *Diskv
361 key string
362 buf *bytes.Buffer
363}
364
365// newSiphon constructs a siphoning reader that represents the passed file.
366// When a successful series of reads ends in an EOF, the siphon will write
367// the buffered data to Diskv's cache under the given key.
368func newSiphon(f *os.File, d *Diskv, key string) io.Reader {
369 return &siphon{
370 f: f,
371 d: d,
372 key: key,
373 buf: &bytes.Buffer{},
374 }
375}
376
377// Read implements the io.Reader interface for siphon.
378func (s *siphon) Read(p []byte) (int, error) {
379 n, err := s.f.Read(p)
380
381 if err == nil {
382 return s.buf.Write(p[0:n]) // Write must succeed for Read to succeed
383 }
384
385 if err == io.EOF {
386 s.d.cacheWithoutLock(s.key, s.buf.Bytes()) // cache may fail
387 if closeErr := s.f.Close(); closeErr != nil {
388 return n, closeErr // close must succeed for Read to succeed
389 }
390 return n, err
391 }
392
393 return n, err
394}
395
396// Erase synchronously erases the given key from the disk and the cache.
397func (d *Diskv) Erase(key string) error {
398 d.mu.Lock()
399 defer d.mu.Unlock()
400
401 d.bustCacheWithLock(key)
402
403 // erase from index
404 if d.Index != nil {
405 d.Index.Delete(key)
406 }
407
408 // erase from disk
409 filename := d.completeFilename(key)
410 if s, err := os.Stat(filename); err == nil {
411 if s.IsDir() {
412 return errBadKey
413 }
414 if err = os.Remove(filename); err != nil {
415 return err
416 }
417 } else {
418 // Return err as-is so caller can do os.IsNotExist(err).
419 return err
420 }
421
422 // clean up and return
423 d.pruneDirsWithLock(key)
424 return nil
425}
426
427// EraseAll will delete all of the data from the store, both in the cache and on
428// the disk. Note that EraseAll doesn't distinguish diskv-related data from non-
429// diskv-related data. Care should be taken to always specify a diskv base
430// directory that is exclusively for diskv data.
431func (d *Diskv) EraseAll() error {
432 d.mu.Lock()
433 defer d.mu.Unlock()
434 d.cache = make(map[string][]byte)
435 d.cacheSize = 0
436 if d.TempDir != "" {
437 os.RemoveAll(d.TempDir) // errors ignored
438 }
439 return os.RemoveAll(d.BasePath)
440}
441
442// Has returns true if the given key exists.
443func (d *Diskv) Has(key string) bool {
444 d.mu.Lock()
445 defer d.mu.Unlock()
446
447 if _, ok := d.cache[key]; ok {
448 return true
449 }
450
451 filename := d.completeFilename(key)
452 s, err := os.Stat(filename)
453 if err != nil {
454 return false
455 }
456 if s.IsDir() {
457 return false
458 }
459
460 return true
461}
462
463// Keys returns a channel that will yield every key accessible by the store,
464// in undefined order. If a cancel channel is provided, closing it will
465// terminate and close the keys channel.
466func (d *Diskv) Keys(cancel <-chan struct{}) <-chan string {
467 return d.KeysPrefix("", cancel)
468}
469
470// KeysPrefix returns a channel that will yield every key accessible by the
471// store with the given prefix, in undefined order. If a cancel channel is
472// provided, closing it will terminate and close the keys channel. If the
473// provided prefix is the empty string, all keys will be yielded.
474func (d *Diskv) KeysPrefix(prefix string, cancel <-chan struct{}) <-chan string {
475 var prepath string
476 if prefix == "" {
477 prepath = d.BasePath
478 } else {
479 prepath = d.pathFor(prefix)
480 }
481 c := make(chan string)
482 go func() {
483 filepath.Walk(prepath, walker(c, prefix, cancel))
484 close(c)
485 }()
486 return c
487}
488
489// walker returns a function which satisfies the filepath.WalkFunc interface.
490// It sends every non-directory file entry down the channel c.
491func walker(c chan<- string, prefix string, cancel <-chan struct{}) filepath.WalkFunc {
492 return func(path string, info os.FileInfo, err error) error {
493 if err != nil {
494 return err
495 }
496
497 if info.IsDir() || !strings.HasPrefix(info.Name(), prefix) {
498 return nil // "pass"
499 }
500
501 select {
502 case c <- info.Name():
503 case <-cancel:
504 return errCanceled
505 }
506
507 return nil
508 }
509}
510
511// pathFor returns the absolute path for location on the filesystem where the
512// data for the given key will be stored.
513func (d *Diskv) pathFor(key string) string {
514 return filepath.Join(d.BasePath, filepath.Join(d.Transform(key)...))
515}
516
517// ensurePathWithLock is a helper function that generates all necessary
518// directories on the filesystem for the given key.
519func (d *Diskv) ensurePathWithLock(key string) error {
520 return os.MkdirAll(d.pathFor(key), d.PathPerm)
521}
522
523// completeFilename returns the absolute path to the file for the given key.
524func (d *Diskv) completeFilename(key string) string {
525 return filepath.Join(d.pathFor(key), key)
526}
527
528// cacheWithLock attempts to cache the given key-value pair in the store's
529// cache. It can fail if the value is larger than the cache's maximum size.
530func (d *Diskv) cacheWithLock(key string, val []byte) error {
531 valueSize := uint64(len(val))
532 if err := d.ensureCacheSpaceWithLock(valueSize); err != nil {
533 return fmt.Errorf("%s; not caching", err)
534 }
535
536 // be very strict about memory guarantees
537 if (d.cacheSize + valueSize) > d.CacheSizeMax {
538 panic(fmt.Sprintf("failed to make room for value (%d/%d)", valueSize, d.CacheSizeMax))
539 }
540
541 d.cache[key] = val
542 d.cacheSize += valueSize
543 return nil
544}
545
546// cacheWithoutLock acquires the store's (write) mutex and calls cacheWithLock.
547func (d *Diskv) cacheWithoutLock(key string, val []byte) error {
548 d.mu.Lock()
549 defer d.mu.Unlock()
550 return d.cacheWithLock(key, val)
551}
552
553func (d *Diskv) bustCacheWithLock(key string) {
554 if val, ok := d.cache[key]; ok {
555 d.uncacheWithLock(key, uint64(len(val)))
556 }
557}
558
559func (d *Diskv) uncacheWithLock(key string, sz uint64) {
560 d.cacheSize -= sz
561 delete(d.cache, key)
562}
563
564// pruneDirsWithLock deletes empty directories in the path walk leading to the
565// key k. Typically this function is called after an Erase is made.
566func (d *Diskv) pruneDirsWithLock(key string) error {
567 pathlist := d.Transform(key)
568 for i := range pathlist {
569 dir := filepath.Join(d.BasePath, filepath.Join(pathlist[:len(pathlist)-i]...))
570
571 // thanks to Steven Blenkinsop for this snippet
572 switch fi, err := os.Stat(dir); true {
573 case err != nil:
574 return err
575 case !fi.IsDir():
576 panic(fmt.Sprintf("corrupt dirstate at %s", dir))
577 }
578
579 nlinks, err := filepath.Glob(filepath.Join(dir, "*"))
580 if err != nil {
581 return err
582 } else if len(nlinks) > 0 {
583 return nil // has subdirs -- do not prune
584 }
585 if err = os.Remove(dir); err != nil {
586 return err
587 }
588 }
589
590 return nil
591}
592
593// ensureCacheSpaceWithLock deletes entries from the cache in arbitrary order
594// until the cache has at least valueSize bytes available.
595func (d *Diskv) ensureCacheSpaceWithLock(valueSize uint64) error {
596 if valueSize > d.CacheSizeMax {
597 return fmt.Errorf("value size (%d bytes) too large for cache (%d bytes)", valueSize, d.CacheSizeMax)
598 }
599
600 safe := func() bool { return (d.cacheSize + valueSize) <= d.CacheSizeMax }
601
602 for key, val := range d.cache {
603 if safe() {
604 break
605 }
606
607 d.uncacheWithLock(key, uint64(len(val)))
608 }
609
610 if !safe() {
611 panic(fmt.Sprintf("%d bytes still won't fit in the cache! (max %d bytes)", valueSize, d.CacheSizeMax))
612 }
613
614 return nil
615}
616
617// nopWriteCloser wraps an io.Writer and provides a no-op Close method to
618// satisfy the io.WriteCloser interface.
619type nopWriteCloser struct {
620 io.Writer
621}
622
623func (wc *nopWriteCloser) Write(p []byte) (int, error) { return wc.Writer.Write(p) }
624func (wc *nopWriteCloser) Close() error { return nil }