3 * Proxy backend that mirrors writes to several internal backends.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
18 * http://www.gnu.org/copyleft/gpl.html
21 * @ingroup FileBackend
22 * @author Aaron Schulz
26 * @brief Proxy backend that mirrors writes to several internal backends.
28 * This class defines a multi-write backend. Multiple backends can be
29 * registered to this proxy backend and it will act as a single backend.
30 * Use this when all access to those backends is through this proxy backend.
31 * At least one of the backends must be declared the "master" backend.
33 * Only use this class when transitioning from one storage system to another.
35 * Read operations are only done on the 'master' backend for consistency.
36 * Write operations are performed on all backends, in the order defined.
37 * If an operation fails on one backend it will be rolled back from the others.
39 * @ingroup FileBackend
42 class FileBackendMultiWrite
extends FileBackend
{
43 /** @var array Prioritized list of FileBackendStore objects.
44 * array of (backend index => backends)
46 protected $backends = array();
48 /** @var int Index of master backend */
49 protected $masterIndex = -1;
51 /** @var int Bitfield */
52 protected $syncChecks = 0;
54 /** @var string|bool */
55 protected $autoResync = false;
58 protected $noPushDirConts = array();
61 protected $noPushQuickOps = false;
63 /* Possible internal backend consistency checks */
69 * Construct a proxy backend that consists of several internal backends.
70 * Locking, journaling, and read-only checks are handled by the proxy backend.
72 * Additional $config params include:
73 * - backends : Array of backend config and multi-backend settings.
74 * Each value is the config used in the constructor of a
75 * FileBackendStore class, but with these additional settings:
76 * - class : The name of the backend class
77 * - isMultiMaster : This must be set for one backend.
78 * - template: : If given a backend name, this will use
79 * the config of that backend as a template.
80 * Values specified here take precedence.
81 * - syncChecks : Integer bitfield of internal backend sync checks to perform.
82 * Possible bits include the FileBackendMultiWrite::CHECK_* constants.
83 * There are constants for SIZE, TIME, and SHA1.
84 * The checks are done before allowing any file operations.
85 * - autoResync : Automatically resync the clone backends to the master backend
86 * when pre-operation sync checks fail. This should only be used
87 * if the master backend is stable and not missing any files.
88 * Use "conservative" to limit resyncing to copying newer master
89 * backend files over older (or non-existing) clone backend files.
90 * Cases that cannot be handled will result in operation abortion.
91 * - noPushQuickOps : (hack) Only apply doQuickOperations() to the master backend.
92 * - noPushDirConts : (hack) Only apply directory functions to the master backend.
94 * @param array $config
95 * @throws FileBackendError
97 public function __construct( array $config ) {
98 parent
::__construct( $config );
99 $this->syncChecks
= isset( $config['syncChecks'] )
100 ?
$config['syncChecks']
102 $this->autoResync
= isset( $config['autoResync'] )
103 ?
$config['autoResync']
105 $this->noPushQuickOps
= isset( $config['noPushQuickOps'] )
106 ?
$config['noPushQuickOps']
108 $this->noPushDirConts
= isset( $config['noPushDirConts'] )
109 ?
$config['noPushDirConts']
111 // Construct backends here rather than via registration
112 // to keep these backends hidden from outside the proxy.
113 $namesUsed = array();
114 foreach ( $config['backends'] as $index => $config ) {
115 if ( isset( $config['template'] ) ) {
116 // Config is just a modified version of a registered backend's.
117 // This should only be used when that config is used only by this backend.
118 $config = $config + FileBackendGroup
::singleton()->config( $config['template'] );
120 $name = $config['name'];
121 if ( isset( $namesUsed[$name] ) ) { // don't break FileOp predicates
122 throw new FileBackendError( "Two or more backends defined with the name $name." );
124 $namesUsed[$name] = 1;
125 // Alter certain sub-backend settings for sanity
126 unset( $config['readOnly'] ); // use proxy backend setting
127 unset( $config['fileJournal'] ); // use proxy backend journal
128 unset( $config['lockManager'] ); // lock under proxy backend
129 $config['wikiId'] = $this->wikiId
; // use the proxy backend wiki ID
130 if ( !empty( $config['isMultiMaster'] ) ) {
131 if ( $this->masterIndex
>= 0 ) {
132 throw new FileBackendError( 'More than one master backend defined.' );
134 $this->masterIndex
= $index; // this is the "master"
135 $config['fileJournal'] = $this->fileJournal
; // log under proxy backend
137 // Create sub-backend object
138 if ( !isset( $config['class'] ) ) {
139 throw new FileBackendError( 'No class given for a backend config.' );
141 $class = $config['class'];
142 $this->backends
[$index] = new $class( $config );
144 if ( $this->masterIndex
< 0 ) { // need backends and must have a master
145 throw new FileBackendError( 'No master backend defined.' );
149 final protected function doOperationsInternal( array $ops, array $opts ) {
150 $status = Status
::newGood();
152 $mbe = $this->backends
[$this->masterIndex
]; // convenience
154 // Try to lock those files for the scope of this function...
155 if ( empty( $opts['nonLocking'] ) ) {
156 // Try to lock those files for the scope of this function...
157 $scopeLock = $this->getScopedLocksForOps( $ops, $status );
158 if ( !$status->isOK() ) {
159 return $status; // abort
162 // Clear any cache entries (after locks acquired)
164 $opts['preserveCache'] = true; // only locked files are cached
165 // Get the list of paths to read/write...
166 $relevantPaths = $this->fileStoragePathsForOps( $ops );
167 // Check if the paths are valid and accessible on all backends...
168 $status->merge( $this->accessibilityCheck( $relevantPaths ) );
169 if ( !$status->isOK() ) {
170 return $status; // abort
172 // Do a consistency check to see if the backends are consistent...
173 $syncStatus = $this->consistencyCheck( $relevantPaths );
174 if ( !$syncStatus->isOK() ) {
175 wfDebugLog( 'FileOperation', get_class( $this ) .
176 " failed sync check: " . FormatJson
::encode( $relevantPaths ) );
177 // Try to resync the clone backends to the master on the spot...
178 if ( !$this->autoResync ||
!$this->resyncFiles( $relevantPaths )->isOK() ) {
179 $status->merge( $syncStatus );
181 return $status; // abort
184 // Actually attempt the operation batch on the master backend...
185 $realOps = $this->substOpBatchPaths( $ops, $mbe );
186 $masterStatus = $mbe->doOperations( $realOps, $opts );
187 $status->merge( $masterStatus );
188 // Propagate the operations to the clone backends if there were no unexpected errors
189 // and if there were either no expected errors or if the 'force' option was used.
190 // However, if nothing succeeded at all, then don't replicate any of the operations.
191 // If $ops only had one operation, this might avoid backend sync inconsistencies.
192 if ( $masterStatus->isOK() && $masterStatus->successCount
> 0 ) {
193 foreach ( $this->backends
as $index => $backend ) {
194 if ( $index !== $this->masterIndex
) { // not done already
195 $realOps = $this->substOpBatchPaths( $ops, $backend );
196 $status->merge( $backend->doOperations( $realOps, $opts ) );
200 // Make 'success', 'successCount', and 'failCount' fields reflect
201 // the overall operation, rather than all the batches for each backend.
202 // Do this by only using success values from the master backend's batch.
203 $status->success
= $masterStatus->success
;
204 $status->successCount
= $masterStatus->successCount
;
205 $status->failCount
= $masterStatus->failCount
;
211 * Check that a set of files are consistent across all internal backends
213 * @param array $paths List of storage paths
216 public function consistencyCheck( array $paths ) {
217 $status = Status
::newGood();
218 if ( $this->syncChecks
== 0 ||
count( $this->backends
) <= 1 ) {
219 return $status; // skip checks
222 $mBackend = $this->backends
[$this->masterIndex
];
223 foreach ( $paths as $path ) {
224 $params = array( 'src' => $path, 'latest' => true );
225 $mParams = $this->substOpPaths( $params, $mBackend );
226 // Stat the file on the 'master' backend
227 $mStat = $mBackend->getFileStat( $mParams );
228 if ( $this->syncChecks
& self
::CHECK_SHA1
) {
229 $mSha1 = $mBackend->getFileSha1Base36( $mParams );
233 // Check if all clone backends agree with the master...
234 foreach ( $this->backends
as $index => $cBackend ) {
235 if ( $index === $this->masterIndex
) {
238 $cParams = $this->substOpPaths( $params, $cBackend );
239 $cStat = $cBackend->getFileStat( $cParams );
240 if ( $mStat ) { // file is in master
241 if ( !$cStat ) { // file should exist
242 $status->fatal( 'backend-fail-synced', $path );
245 if ( $this->syncChecks
& self
::CHECK_SIZE
) {
246 if ( $cStat['size'] != $mStat['size'] ) { // wrong size
247 $status->fatal( 'backend-fail-synced', $path );
251 if ( $this->syncChecks
& self
::CHECK_TIME
) {
252 $mTs = wfTimestamp( TS_UNIX
, $mStat['mtime'] );
253 $cTs = wfTimestamp( TS_UNIX
, $cStat['mtime'] );
254 if ( abs( $mTs - $cTs ) > 30 ) { // outdated file somewhere
255 $status->fatal( 'backend-fail-synced', $path );
259 if ( $this->syncChecks
& self
::CHECK_SHA1
) {
260 if ( $cBackend->getFileSha1Base36( $cParams ) !== $mSha1 ) { // wrong SHA1
261 $status->fatal( 'backend-fail-synced', $path );
265 } else { // file is not in master
266 if ( $cStat ) { // file should not exist
267 $status->fatal( 'backend-fail-synced', $path );
277 * Check that a set of file paths are usable across all internal backends
279 * @param array $paths List of storage paths
282 public function accessibilityCheck( array $paths ) {
283 $status = Status
::newGood();
284 if ( count( $this->backends
) <= 1 ) {
285 return $status; // skip checks
288 foreach ( $paths as $path ) {
289 foreach ( $this->backends
as $backend ) {
290 $realPath = $this->substPaths( $path, $backend );
291 if ( !$backend->isPathUsableInternal( $realPath ) ) {
292 $status->fatal( 'backend-fail-usable', $path );
301 * Check that a set of files are consistent across all internal backends
302 * and re-synchronize those files againt the "multi master" if needed.
304 * @param array $paths List of storage paths
307 public function resyncFiles( array $paths ) {
308 $status = Status
::newGood();
310 $mBackend = $this->backends
[$this->masterIndex
];
311 foreach ( $paths as $path ) {
312 $mPath = $this->substPaths( $path, $mBackend );
313 $mSha1 = $mBackend->getFileSha1Base36( array( 'src' => $mPath, 'latest' => true ) );
314 $mStat = $mBackend->getFileStat( array( 'src' => $mPath, 'latest' => true ) );
315 if ( $mStat === null ||
( $mSha1 !== false && !$mStat ) ) { // sanity
316 $status->fatal( 'backend-fail-internal', $this->name
);
317 continue; // file is not available on the master backend...
319 // Check of all clone backends agree with the master...
320 foreach ( $this->backends
as $index => $cBackend ) {
321 if ( $index === $this->masterIndex
) {
324 $cPath = $this->substPaths( $path, $cBackend );
325 $cSha1 = $cBackend->getFileSha1Base36( array( 'src' => $cPath, 'latest' => true ) );
326 $cStat = $cBackend->getFileStat( array( 'src' => $cPath, 'latest' => true ) );
327 if ( $cStat === null ||
( $cSha1 !== false && !$cStat ) ) { // sanity
328 $status->fatal( 'backend-fail-internal', $cBackend->getName() );
329 continue; // file is not available on the clone backend...
331 if ( $mSha1 === $cSha1 ) {
332 // already synced; nothing to do
333 } elseif ( $mSha1 !== false ) { // file is in master
334 if ( $this->autoResync
=== 'conservative'
335 && $cStat && $cStat['mtime'] > $mStat['mtime']
337 $status->fatal( 'backend-fail-synced', $path );
338 continue; // don't rollback data
340 $fsFile = $mBackend->getLocalReference(
341 array( 'src' => $mPath, 'latest' => true ) );
342 $status->merge( $cBackend->quickStore(
343 array( 'src' => $fsFile->getPath(), 'dst' => $cPath )
345 } elseif ( $mStat === false ) { // file is not in master
346 if ( $this->autoResync
=== 'conservative' ) {
347 $status->fatal( 'backend-fail-synced', $path );
348 continue; // don't delete data
350 $status->merge( $cBackend->quickDelete( array( 'src' => $cPath ) ) );
359 * Get a list of file storage paths to read or write for a list of operations
361 * @param array $ops Same format as doOperations()
362 * @return array List of storage paths to files (does not include directories)
364 protected function fileStoragePathsForOps( array $ops ) {
366 foreach ( $ops as $op ) {
367 if ( isset( $op['src'] ) ) {
368 // For things like copy/move/delete with "ignoreMissingSource" and there
369 // is no source file, nothing should happen and there should be no errors.
370 if ( empty( $op['ignoreMissingSource'] )
371 ||
$this->fileExists( array( 'src' => $op['src'] ) )
373 $paths[] = $op['src'];
376 if ( isset( $op['srcs'] ) ) {
377 $paths = array_merge( $paths, $op['srcs'] );
379 if ( isset( $op['dst'] ) ) {
380 $paths[] = $op['dst'];
384 return array_values( array_unique( array_filter( $paths, 'FileBackend::isStoragePath' ) ) );
388 * Substitute the backend name in storage path parameters
389 * for a set of operations with that of a given internal backend.
391 * @param array $ops List of file operation arrays
392 * @param FileBackendStore $backend
395 protected function substOpBatchPaths( array $ops, FileBackendStore
$backend ) {
396 $newOps = array(); // operations
397 foreach ( $ops as $op ) {
398 $newOp = $op; // operation
399 foreach ( array( 'src', 'srcs', 'dst', 'dir' ) as $par ) {
400 if ( isset( $newOp[$par] ) ) { // string or array
401 $newOp[$par] = $this->substPaths( $newOp[$par], $backend );
411 * Same as substOpBatchPaths() but for a single operation
413 * @param array $ops File operation array
414 * @param FileBackendStore $backend
417 protected function substOpPaths( array $ops, FileBackendStore
$backend ) {
418 $newOps = $this->substOpBatchPaths( array( $ops ), $backend );
424 * Substitute the backend of storage paths with an internal backend's name
426 * @param array|string $paths List of paths or single string path
427 * @param FileBackendStore $backend
428 * @return array|string
430 protected function substPaths( $paths, FileBackendStore
$backend ) {
432 '!^mwstore://' . preg_quote( $this->name
) . '/!',
433 StringUtils
::escapeRegexReplacement( "mwstore://{$backend->getName()}/" ),
434 $paths // string or array
439 * Substitute the backend of internal storage paths with the proxy backend's name
441 * @param array|string $paths List of paths or single string path
442 * @return array|string
444 protected function unsubstPaths( $paths ) {
446 '!^mwstore://([^/]+)!',
447 StringUtils
::escapeRegexReplacement( "mwstore://{$this->name}" ),
448 $paths // string or array
452 protected function doQuickOperationsInternal( array $ops ) {
453 $status = Status
::newGood();
454 // Do the operations on the master backend; setting Status fields...
455 $realOps = $this->substOpBatchPaths( $ops, $this->backends
[$this->masterIndex
] );
456 $masterStatus = $this->backends
[$this->masterIndex
]->doQuickOperations( $realOps );
457 $status->merge( $masterStatus );
458 // Propagate the operations to the clone backends...
459 if ( !$this->noPushQuickOps
) {
460 foreach ( $this->backends
as $index => $backend ) {
461 if ( $index !== $this->masterIndex
) { // not done already
462 $realOps = $this->substOpBatchPaths( $ops, $backend );
463 $status->merge( $backend->doQuickOperations( $realOps ) );
467 // Make 'success', 'successCount', and 'failCount' fields reflect
468 // the overall operation, rather than all the batches for each backend.
469 // Do this by only using success values from the master backend's batch.
470 $status->success
= $masterStatus->success
;
471 $status->successCount
= $masterStatus->successCount
;
472 $status->failCount
= $masterStatus->failCount
;
478 * @param string $path Storage path
479 * @return bool Path container should have dir changes pushed to all backends
481 protected function replicateContainerDirChanges( $path ) {
482 list( , $shortCont, ) = self
::splitStoragePath( $path );
484 return !in_array( $shortCont, $this->noPushDirConts
);
487 protected function doPrepare( array $params ) {
488 $status = Status
::newGood();
489 $replicate = $this->replicateContainerDirChanges( $params['dir'] );
490 foreach ( $this->backends
as $index => $backend ) {
491 if ( $replicate ||
$index == $this->masterIndex
) {
492 $realParams = $this->substOpPaths( $params, $backend );
493 $status->merge( $backend->doPrepare( $realParams ) );
500 protected function doSecure( array $params ) {
501 $status = Status
::newGood();
502 $replicate = $this->replicateContainerDirChanges( $params['dir'] );
503 foreach ( $this->backends
as $index => $backend ) {
504 if ( $replicate ||
$index == $this->masterIndex
) {
505 $realParams = $this->substOpPaths( $params, $backend );
506 $status->merge( $backend->doSecure( $realParams ) );
513 protected function doPublish( array $params ) {
514 $status = Status
::newGood();
515 $replicate = $this->replicateContainerDirChanges( $params['dir'] );
516 foreach ( $this->backends
as $index => $backend ) {
517 if ( $replicate ||
$index == $this->masterIndex
) {
518 $realParams = $this->substOpPaths( $params, $backend );
519 $status->merge( $backend->doPublish( $realParams ) );
526 protected function doClean( array $params ) {
527 $status = Status
::newGood();
528 $replicate = $this->replicateContainerDirChanges( $params['dir'] );
529 foreach ( $this->backends
as $index => $backend ) {
530 if ( $replicate ||
$index == $this->masterIndex
) {
531 $realParams = $this->substOpPaths( $params, $backend );
532 $status->merge( $backend->doClean( $realParams ) );
539 public function concatenate( array $params ) {
540 // We are writing to an FS file, so we don't need to do this per-backend
541 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
543 return $this->backends
[$this->masterIndex
]->concatenate( $realParams );
546 public function fileExists( array $params ) {
547 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
549 return $this->backends
[$this->masterIndex
]->fileExists( $realParams );
552 public function getFileTimestamp( array $params ) {
553 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
555 return $this->backends
[$this->masterIndex
]->getFileTimestamp( $realParams );
558 public function getFileSize( array $params ) {
559 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
561 return $this->backends
[$this->masterIndex
]->getFileSize( $realParams );
564 public function getFileStat( array $params ) {
565 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
567 return $this->backends
[$this->masterIndex
]->getFileStat( $realParams );
570 public function getFileXAttributes( array $params ) {
571 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
573 return $this->backends
[$this->masterIndex
]->getFileXAttributes( $realParams );
576 public function getFileContentsMulti( array $params ) {
577 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
578 $contentsM = $this->backends
[$this->masterIndex
]->getFileContentsMulti( $realParams );
580 $contents = array(); // (path => FSFile) mapping using the proxy backend's name
581 foreach ( $contentsM as $path => $data ) {
582 $contents[$this->unsubstPaths( $path )] = $data;
588 public function getFileSha1Base36( array $params ) {
589 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
591 return $this->backends
[$this->masterIndex
]->getFileSha1Base36( $realParams );
594 public function getFileProps( array $params ) {
595 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
597 return $this->backends
[$this->masterIndex
]->getFileProps( $realParams );
600 public function streamFile( array $params ) {
601 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
603 return $this->backends
[$this->masterIndex
]->streamFile( $realParams );
606 public function getLocalReferenceMulti( array $params ) {
607 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
608 $fsFilesM = $this->backends
[$this->masterIndex
]->getLocalReferenceMulti( $realParams );
610 $fsFiles = array(); // (path => FSFile) mapping using the proxy backend's name
611 foreach ( $fsFilesM as $path => $fsFile ) {
612 $fsFiles[$this->unsubstPaths( $path )] = $fsFile;
618 public function getLocalCopyMulti( array $params ) {
619 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
620 $tempFilesM = $this->backends
[$this->masterIndex
]->getLocalCopyMulti( $realParams );
622 $tempFiles = array(); // (path => TempFSFile) mapping using the proxy backend's name
623 foreach ( $tempFilesM as $path => $tempFile ) {
624 $tempFiles[$this->unsubstPaths( $path )] = $tempFile;
630 public function getFileHttpUrl( array $params ) {
631 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
633 return $this->backends
[$this->masterIndex
]->getFileHttpUrl( $realParams );
636 public function directoryExists( array $params ) {
637 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
639 return $this->backends
[$this->masterIndex
]->directoryExists( $realParams );
642 public function getDirectoryList( array $params ) {
643 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
645 return $this->backends
[$this->masterIndex
]->getDirectoryList( $realParams );
648 public function getFileList( array $params ) {
649 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
651 return $this->backends
[$this->masterIndex
]->getFileList( $realParams );
654 public function getFeatures() {
655 return $this->backends
[$this->masterIndex
]->getFeatures();
658 public function clearCache( array $paths = null ) {
659 foreach ( $this->backends
as $backend ) {
660 $realPaths = is_array( $paths ) ?
$this->substPaths( $paths, $backend ) : null;
661 $backend->clearCache( $realPaths );
665 public function preloadCache( array $paths ) {
666 $realPaths = $this->substPaths( $paths, $this->backends
[$this->masterIndex
] );
667 $this->backends
[$this->masterIndex
]->preloadCache( $realPaths );
670 public function preloadFileStat( array $params ) {
671 $realParams = $this->substOpPaths( $params, $this->backends
[$this->masterIndex
] );
672 return $this->backends
[$this->masterIndex
]->preloadFileStat( $realParams );
675 public function getScopedLocksForOps( array $ops, Status
$status ) {
676 $realOps = $this->substOpBatchPaths( $ops, $this->backends
[$this->masterIndex
] );
677 $fileOps = $this->backends
[$this->masterIndex
]->getOperationsInternal( $realOps );
678 // Get the paths to lock from the master backend
679 $paths = $this->backends
[$this->masterIndex
]->getPathsToLockForOpsInternal( $fileOps );
680 // Get the paths under the proxy backend's name
682 LockManager
::LOCK_UW
=> $this->unsubstPaths( $paths[LockManager
::LOCK_UW
] ),
683 LockManager
::LOCK_EX
=> $this->unsubstPaths( $paths[LockManager
::LOCK_EX
] )
686 // Actually acquire the locks
687 return array( $this->getScopedFileLocks( $pbPaths, 'mixed', $status ) );