*/
/**
+ * @brief Proxy backend that mirrors writes to several internal backends.
+ *
* This class defines a multi-write backend. Multiple backends can be
* registered to this proxy backend and it will act as a single backend.
* Use this when all access to those backends is through this proxy backend.
* At least one of the backends must be declared the "master" backend.
*
* Only use this class when transitioning from one storage system to another.
- *
- * The order that the backends are defined sets the priority of which
- * backend is read from or written to first. Functions like fileExists()
- * and getFileProps() will return information based on the first backend
- * that has the file. Special cases are listed below:
- * a) getFileTimestamp() will always check only the master backend to
- * avoid confusing and inconsistent results.
- *
- * All write operations are performed on all backends.
+ *
+ * Read operations are only done on the 'master' backend for consistency.
+ * Write operations are performed on all backends, in the order defined.
* If an operation fails on one backend it will be rolled back from the others.
*
* @ingroup FileBackend
+ * @since 1.19
*/
-class FileBackendMultiWrite extends FileBackendBase {
- /** @var Array Prioritized list of FileBackend objects */
- protected $fileBackends = array(); // array of (backend index => backends)
- protected $masterIndex = -1; // index of master backend
+class FileBackendMultiWrite extends FileBackend {
+ /** @var Array Prioritized list of FileBackendStore objects */
+ protected $backends = array(); // array of (backend index => backends)
+ protected $masterIndex = -1; // integer; index of master backend
+ protected $syncChecks = 0; // integer bitfield
+
+ /* Possible internal backend consistency checks */
+ const CHECK_SIZE = 1;
+ const CHECK_TIME = 2;
/**
* Construct a proxy backend that consists of several internal backends.
- * $config contains:
- * 'name' : The name of the proxy backend
- * 'lockManager' : Registered name of the file lock manager to use
+ * Additional $config params include:
* 'backends' : Array of backend config and multi-backend settings.
* Each value is the config used in the constructor of a
- * FileBackend class, but with these additional settings:
+ * FileBackendStore class, but with these additional settings:
* 'class' : The name of the backend class
* 'isMultiMaster' : This must be set for one backend.
+ * 'syncChecks' : Integer bitfield of internal backend sync checks to perform.
+ * Possible bits include self::CHECK_SIZE and self::CHECK_TIME.
+ * The checks are done before allowing any file operations.
* @param $config Array
*/
public function __construct( array $config ) {
parent::__construct( $config );
+ $namesUsed = array();
// Construct backends here rather than via registration
// to keep these backends hidden from outside the proxy.
foreach ( $config['backends'] as $index => $config ) {
+ $name = $config['name'];
+ if ( isset( $namesUsed[$name] ) ) { // don't break FileOp predicates
+ throw new MWException( "Two or more backends defined with the name $name." );
+ }
+ $namesUsed[$name] = 1;
if ( !isset( $config['class'] ) ) {
throw new MWException( 'No class given for a backend config.' );
}
$class = $config['class'];
- $this->fileBackends[$index] = new $class( $config );
+ $this->backends[$index] = new $class( $config );
if ( !empty( $config['isMultiMaster'] ) ) {
if ( $this->masterIndex >= 0 ) {
throw new MWException( 'More than one master backend defined.' );
if ( $this->masterIndex < 0 ) { // need backends and must have a master
throw new MWException( 'No master backend defined.' );
}
+ $this->syncChecks = isset( $config['syncChecks'] )
+ ? $config['syncChecks']
+ : self::CHECK_SIZE;
}
/**
- * @see FileBackendBase::doOperationsInternal()
+ * @see FileBackend::doOperationsInternal()
+ * @return Status
*/
final protected function doOperationsInternal( array $ops, array $opts ) {
$status = Status::newGood();
$performOps = array(); // list of FileOp objects
- $filesLockEx = $filesLockSh = array(); // storage paths to lock
+ $filesRead = array(); // storage paths read from
+ $filesChanged = array(); // storage paths written to
// Build up a list of FileOps. The list will have all the ops
// for one backend, then all the ops for the next, and so on.
// These batches of ops are all part of a continuous array.
- // Also build up a list of files to lock...
- foreach ( $this->fileBackends as $index => $backend ) {
- $backendOps = $this->substOpPaths( $ops, $backend );
+ // Also build up a list of files read/changed...
+ foreach ( $this->backends as $index => $backend ) {
+ $backendOps = $this->substOpBatchPaths( $ops, $backend );
+ // Add on the operation batch for this backend
$performOps = array_merge( $performOps, $backend->getOperations( $backendOps ) );
- if ( $index == 0 && empty( $opts['nonLocking'] ) ) {
- // Set "files to lock" from the first batch so we don't try to set all
- // locks two or three times over (depending on the number of backends).
- // A lock on one storage path is a lock on all the backends.
+ if ( $index == 0 ) { // first batch
+ // Get the files used for these operations. Each backend has a batch of
+ // the same operations, so we only need to get them from the first batch.
foreach ( $performOps as $fileOp ) {
- $filesLockSh = array_merge( $filesLockSh, $fileOp->storagePathsRead() );
- $filesLockEx = array_merge( $filesLockEx, $fileOp->storagePathsChanged() );
+ $filesRead = array_merge( $filesRead, $fileOp->storagePathsRead() );
+ $filesChanged = array_merge( $filesChanged, $fileOp->storagePathsChanged() );
}
- // Optimization: if doing an EX lock anyway, don't also set an SH one
- $filesLockSh = array_diff( $filesLockSh, $filesLockEx );
- // Lock the paths under the proxy backend's name
- $this->unsubstPaths( $filesLockSh );
- $this->unsubstPaths( $filesLockEx );
+ // Get the paths under the proxy backend's name
+ $filesRead = $this->unsubstPaths( $filesRead );
+ $filesChanged = $this->unsubstPaths( $filesChanged );
}
}
// Try to lock those files for the scope of this function...
- $scopeLockS = $this->getScopedFileLocks( $filesLockSh, LockManager::LOCK_UW, $status );
- $scopeLockE = $this->getScopedFileLocks( $filesLockEx, LockManager::LOCK_EX, $status );
- if ( !$status->isOK() ) {
- return $status; // abort
+ if ( empty( $opts['nonLocking'] ) ) {
+ $filesLockSh = array_diff( $filesRead, $filesChanged ); // optimization
+ $filesLockEx = $filesChanged;
+ // Get a shared lock on the parent directory of each path changed
+ $filesLockSh = array_merge( $filesLockSh, array_map( 'dirname', $filesLockEx ) );
+ // Try to lock those files for the scope of this function...
+ $scopeLockS = $this->getScopedFileLocks( $filesLockSh, LockManager::LOCK_UW, $status );
+ $scopeLockE = $this->getScopedFileLocks( $filesLockEx, LockManager::LOCK_EX, $status );
+ if ( !$status->isOK() ) {
+ return $status; // abort
+ }
}
// Clear any cache entries (after locks acquired)
- foreach ( $this->fileBackends as $backend ) {
- $backend->clearCache();
+ $this->clearCache();
+
+ // Do a consistency check to see if the backends agree
+ if ( count( $this->backends ) > 1 ) {
+ $status->merge( $this->consistencyCheck( array_merge( $filesRead, $filesChanged ) ) );
+ if ( !$status->isOK() ) {
+ return $status; // abort
+ }
}
+
// Actually attempt the operation batch...
- $status->merge( FileOp::attemptBatch( $performOps, $opts ) );
+ $subStatus = FileOp::attemptBatch( $performOps, $opts, $this->fileJournal );
+
+ $success = array();
+ $failCount = 0;
+ $successCount = 0;
+ // Make 'success', 'successCount', and 'failCount' fields reflect
+ // the overall operation, rather than all the batches for each backend.
+ // Do this by only using success values from the master backend's batch.
+ $batchStart = $this->masterIndex * count( $ops );
+ $batchEnd = $batchStart + count( $ops ) - 1;
+ for ( $i = $batchStart; $i <= $batchEnd; $i++ ) {
+ if ( !isset( $subStatus->success[$i] ) ) {
+ break; // failed out before trying this op
+ } elseif ( $subStatus->success[$i] ) {
+ ++$successCount;
+ } else {
+ ++$failCount;
+ }
+ $success[] = $subStatus->success[$i];
+ }
+ $subStatus->success = $success;
+ $subStatus->successCount = $successCount;
+ $subStatus->failCount = $failCount;
+
+ // Merge errors into status fields
+ $status->merge( $subStatus );
+ $status->success = $subStatus->success; // not done in merge()
+
+ return $status;
+ }
+
+ /**
+ * Check that a set of files are consistent across all internal backends
+ *
+ * @param $paths Array
+ * @return Status
+ */
+ public function consistencyCheck( array $paths ) {
+ $status = Status::newGood();
+ if ( $this->syncChecks == 0 ) {
+ return $status; // skip checks
+ }
+
+ $mBackend = $this->backends[$this->masterIndex];
+ foreach ( array_unique( $paths ) as $path ) {
+ $params = array( 'src' => $path, 'latest' => true );
+ // Stat the file on the 'master' backend
+ $mStat = $mBackend->getFileStat( $this->substOpPaths( $params, $mBackend ) );
+ // Check of all clone backends agree with the master...
+ foreach ( $this->backends as $index => $cBackend ) {
+ if ( $index === $this->masterIndex ) {
+ continue; // master
+ }
+ $cStat = $cBackend->getFileStat( $this->substOpPaths( $params, $cBackend ) );
+ if ( $mStat ) { // file is in master
+ if ( !$cStat ) { // file should exist
+ $status->fatal( 'backend-fail-synced', $path );
+ continue;
+ }
+ if ( $this->syncChecks & self::CHECK_SIZE ) {
+ if ( $cStat['size'] != $mStat['size'] ) { // wrong size
+ $status->fatal( 'backend-fail-synced', $path );
+ continue;
+ }
+ }
+ if ( $this->syncChecks & self::CHECK_TIME ) {
+ $mTs = wfTimestamp( TS_UNIX, $mStat['mtime'] );
+ $cTs = wfTimestamp( TS_UNIX, $cStat['mtime'] );
+ if ( abs( $mTs - $cTs ) > 30 ) { // outdated file somewhere
+ $status->fatal( 'backend-fail-synced', $path );
+ continue;
+ }
+ }
+ } else { // file is not in master
+ if ( $cStat ) { // file should not exist
+ $status->fatal( 'backend-fail-synced', $path );
+ }
+ }
+ }
+ }
return $status;
}
/**
* Substitute the backend name in storage path parameters
- * for a set of operations with a that of a given backend.
+ * for a set of operations with that of a given internal backend.
*
* @param $ops Array List of file operation arrays
- * @param $backend FileBackend
+ * @param $backend FileBackendStore
* @return Array
*/
- protected function substOpPaths( array $ops, FileBackend $backend ) {
+ protected function substOpBatchPaths( array $ops, FileBackendStore $backend ) {
$newOps = array(); // operations
foreach ( $ops as $op ) {
$newOp = $op; // operation
- foreach ( array( 'src', 'srcs', 'dst' ) as $par ) {
- if ( isset( $newOp[$par] ) ) {
- $newOp[$par] = preg_replace(
- '!^mwstore://' . preg_quote( $this->name ) . '/!',
- 'mwstore://' . $backend->getName() . '/',
- $newOp[$par] // string or array
- );
+ foreach ( array( 'src', 'srcs', 'dst', 'dir' ) as $par ) {
+ if ( isset( $newOp[$par] ) ) { // string or array
+ $newOp[$par] = $this->substPaths( $newOp[$par], $backend );
}
}
$newOps[] = $newOp;
}
/**
- * Replace the backend part of storage paths with this backend's name
+ * Same as substOpBatchPaths() but for a single operation
*
- * @param &$paths Array
- * @return void
+ * @param $op File operation array
+ * @param $backend FileBackendStore
+ * @return Array
*/
- protected function unsubstPaths( array &$paths ) {
- foreach ( $paths as &$path ) {
- $path = preg_replace( '!^mwstore://([^/]+)!', "mwstore://{$this->name}", $path );
- }
+ protected function substOpPaths( array $ops, FileBackendStore $backend ) {
+ $newOps = $this->substOpBatchPaths( array( $ops ), $backend );
+ return $newOps[0];
+ }
+
+ /**
+ * Substitute the backend of storage paths with an internal backend's name
+ *
+ * @param $paths Array|string List of paths or single string path
+ * @param $backend FileBackendStore
+ * @return Array|string
+ */
+ protected function substPaths( $paths, FileBackendStore $backend ) {
+ return preg_replace(
+ '!^mwstore://' . preg_quote( $this->name ) . '/!',
+ StringUtils::escapeRegexReplacement( "mwstore://{$backend->getName()}/" ),
+ $paths // string or array
+ );
+ }
+
+ /**
+ * Substitute the backend of internal storage paths with the proxy backend's name
+ *
+ * @param $paths Array|string List of paths or single string path
+ * @return Array|string
+ */
+ protected function unsubstPaths( $paths ) {
+ return preg_replace(
+ '!^mwstore://([^/]+)!',
+ StringUtils::escapeRegexReplacement( "mwstore://{$this->name}" ),
+ $paths // string or array
+ );
}
/**
- * @see FileBackendBase::prepare()
+ * @see FileBackend::doPrepare()
+ * @return Status
*/
- function prepare( array $params ) {
+ protected function doPrepare( array $params ) {
$status = Status::newGood();
foreach ( $this->backends as $backend ) {
$realParams = $this->substOpPaths( $params, $backend );
- $status->merge( $backend->prepare( $realParams ) );
+ $status->merge( $backend->doPrepare( $realParams ) );
}
return $status;
}
/**
- * @see FileBackendBase::secure()
+ * @see FileBackend::doSecure()
+ * @return Status
*/
- function secure( array $params ) {
+ protected function doSecure( array $params ) {
$status = Status::newGood();
foreach ( $this->backends as $backend ) {
$realParams = $this->substOpPaths( $params, $backend );
- $status->merge( $backend->secure( $realParams ) );
+ $status->merge( $backend->doSecure( $realParams ) );
}
return $status;
}
/**
- * @see FileBackendBase::clean()
+ * @see FileBackend::doClean()
+ * @return Status
*/
- function clean( array $params ) {
+ protected function doClean( array $params ) {
$status = Status::newGood();
foreach ( $this->backends as $backend ) {
$realParams = $this->substOpPaths( $params, $backend );
- $status->merge( $backend->clean( $realParams ) );
+ $status->merge( $backend->doClean( $realParams ) );
}
return $status;
}
/**
- * @see FileBackendBase::fileExists()
+ * @see FileBackend::getFileList()
*/
- function fileExists( array $params ) {
- # Hit all backends in case of failed operations (out of sync)
- foreach ( $this->backends as $backend ) {
- $realParams = $this->substOpPaths( $params, $backend );
- if ( $backend->fileExists( $realParams ) ) {
- return true;
- }
- }
- return false;
+ public function concatenate( array $params ) {
+ // We are writing to an FS file, so we don't need to do this per-backend
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->concatenate( $realParams );
+ }
+
+ /**
+ * @see FileBackend::fileExists()
+ */
+ public function fileExists( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->fileExists( $realParams );
}
/**
- * @see FileBackendBase::getFileTimestamp()
+ * @see FileBackend::getFileTimestamp()
*/
- function getFileTimestamp( array $params ) {
- // Skip non-master for consistent timestamps
+ public function getFileTimestamp( array $params ) {
$realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
return $this->backends[$this->masterIndex]->getFileTimestamp( $realParams );
}
/**
- * @see FileBackendBase::getFileSha1Base36()
+ * @see FileBackend::getFileSize()
*/
- function getFileSha1Base36( array $params ) {
- # Hit all backends in case of failed operations (out of sync)
- foreach ( $this->backends as $backend ) {
- $realParams = $this->substOpPaths( $params, $backend );
- $hash = $backend->getFileSha1Base36( $realParams );
- if ( $hash !== false ) {
- return $hash;
- }
- }
- return false;
+ public function getFileSize( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->getFileSize( $realParams );
}
/**
- * @see FileBackendBase::getFileProps()
+ * @see FileBackend::getFileStat()
*/
- function getFileProps( array $params ) {
- # Hit all backends in case of failed operations (out of sync)
- foreach ( $this->backends as $backend ) {
- $realParams = $this->substOpPaths( $params, $backend );
- $props = $backend->getFileProps( $realParams );
- if ( $props !== null ) {
- return $props;
- }
- }
- return null;
+ public function getFileStat( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->getFileStat( $realParams );
}
/**
- * @see FileBackendBase::streamFile()
+ * @see FileBackend::getFileContents()
*/
- function streamFile( array $params ) {
- $status = Status::newGood();
- foreach ( $this->backends as $backend ) {
- $realParams = $this->substOpPaths( $params, $backend );
- $subStatus = $backend->streamFile( $realParams );
- $status->merge( $subStatus );
- if ( $subStatus->isOK() ) {
- // Pass isOK() despite fatals from other backends
- $status->setResult( true );
- return $status;
- } else { // failure
- if ( headers_sent() ) {
- return $status; // died mid-stream...so this is already fubar
- } elseif ( strval( ob_get_contents() ) !== '' ) {
- ob_clean(); // output was buffered but not sent; clear it
- }
- }
- }
- return $status;
+ public function getFileContents( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->getFileContents( $realParams );
}
/**
- * @see FileBackendBase::getLocalReference()
+ * @see FileBackend::getFileSha1Base36()
*/
- function getLocalReference( array $params ) {
- # Hit all backends in case of failed operations (out of sync)
- foreach ( $this->backends as $backend ) {
- $realParams = $this->substOpPaths( $params, $backend );
- $fsFile = $backend->getLocalReference( $realParams );
- if ( $fsFile ) {
- return $fsFile;
- }
- }
- return null;
+ public function getFileSha1Base36( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->getFileSha1Base36( $realParams );
}
/**
- * @see FileBackendBase::getLocalCopy()
+ * @see FileBackend::getFileProps()
*/
- function getLocalCopy( array $params ) {
- # Hit all backends in case of failed operations (out of sync)
- foreach ( $this->backends as $backend ) {
- $realParams = $this->substOpPaths( $params, $backend );
- $tmpFile = $backend->getLocalCopy( $realParams );
- if ( $tmpFile ) {
- return $tmpFile;
- }
- }
- return null;
+ public function getFileProps( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->getFileProps( $realParams );
+ }
+
+ /**
+ * @see FileBackend::streamFile()
+ */
+ public function streamFile( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->streamFile( $realParams );
+ }
+
+ /**
+ * @see FileBackend::getLocalReference()
+ */
+ public function getLocalReference( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->getLocalReference( $realParams );
+ }
+
+ /**
+ * @see FileBackend::getLocalCopy()
+ */
+ public function getLocalCopy( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->getLocalCopy( $realParams );
}
/**
- * @see FileBackendBase::getFileList()
+ * @see FileBackend::getFileList()
*/
- function getFileList( array $params ) {
+ public function getFileList( array $params ) {
+ $realParams = $this->substOpPaths( $params, $this->backends[$this->masterIndex] );
+ return $this->backends[$this->masterIndex]->getFileList( $realParams );
+ }
+
+ /**
+ * @see FileBackend::clearCache()
+ */
+ public function clearCache( array $paths = null ) {
foreach ( $this->backends as $backend ) {
- # Get results from the first backend
- $realParams = $this->substOpPaths( $params, $backend );
- return $backend->getFileList( $realParams );
+ $realPaths = is_array( $paths ) ? $this->substPaths( $paths, $backend ) : null;
+ $backend->clearCache( $realPaths );
}
- return array(); // sanity
}
}