createFile(self::WRITE_LOCK_FILE); if (!$lock->lock(LOCK_EX)) { require_once 'Zend/Search/Lucene/Exception.php'; throw new Zend_Search_Lucene_Exception('Can\'t obtain exclusive index lock'); } return $lock; } /** * Release exclusive write lock * * @param Zend_Search_Lucene_Storage_Directory $lockDirectory */ public static function releaseWriteLock(Zend_Search_Lucene_Storage_Directory $lockDirectory) { $lock = $lockDirectory->getFileObject(self::WRITE_LOCK_FILE); $lock->unlock(); } /** * Obtain the exclusive "read escalation/de-escalation" lock * * Required to protect the escalate/de-escalate read lock process * on GFS (and potentially other) mounted filesystems. * * Why we need this: * While GFS supports cluster-wide locking via flock(), it's * implementation isn't quite what it should be. The locking * semantics that work consistently on a local filesystem tend to * fail on GFS mounted filesystems. This appears to be a design defect * in the implementation of GFS. How this manifests itself is that * conditional promotion of a shared lock to exclusive will always * fail, lock release requests are honored but not immediately * processed (causing erratic failures of subsequent conditional * requests) and the releasing of the exclusive lock before the * shared lock is set when a lock is demoted (which can open a window * of opportunity for another process to gain an exclusive lock when * it shoudln't be allowed to). * * @param Zend_Search_Lucene_Storage_Directory $lockDirectory * @return Zend_Search_Lucene_Storage_File * @throws Zend_Search_Lucene_Exception */ private static function _startReadLockProcessing(Zend_Search_Lucene_Storage_Directory $lockDirectory) { $lock = $lockDirectory->createFile(self::READ_LOCK_PROCESSING_LOCK_FILE); if (!$lock->lock(LOCK_EX)) { require_once 'Zend/Search/Lucene/Exception.php'; throw new Zend_Search_Lucene_Exception('Can\'t obtain exclusive lock for the read lock processing file'); } return $lock; } /** * Release the exclusive "read escalation/de-escalation" lock * * Required to protect the escalate/de-escalate read lock process * on GFS (and potentially other) mounted filesystems. * * @param Zend_Search_Lucene_Storage_Directory $lockDirectory */ private static function _stopReadLockProcessing(Zend_Search_Lucene_Storage_Directory $lockDirectory) { $lock = $lockDirectory->getFileObject(self::READ_LOCK_PROCESSING_LOCK_FILE); $lock->unlock(); } /** * Obtain shared read lock on the index * * It doesn't block other read or update processes, but prevent index from the premature cleaning-up * * @param Zend_Search_Lucene_Storage_Directory $defaultLockDirectory * @return Zend_Search_Lucene_Storage_File * @throws Zend_Search_Lucene_Exception */ public static function obtainReadLock(Zend_Search_Lucene_Storage_Directory $lockDirectory) { $lock = $lockDirectory->createFile(self::READ_LOCK_FILE); if (!$lock->lock(LOCK_SH)) { require_once 'Zend/Search/Lucene/Exception.php'; throw new Zend_Search_Lucene_Exception('Can\'t obtain shared reading index lock'); } return $lock; } /** * Release shared read lock * * @param Zend_Search_Lucene_Storage_Directory $lockDirectory */ public static function releaseReadLock(Zend_Search_Lucene_Storage_Directory $lockDirectory) { $lock = $lockDirectory->getFileObject(self::READ_LOCK_FILE); $lock->unlock(); } /** * Escalate Read lock to exclusive level * * @param Zend_Search_Lucene_Storage_Directory $lockDirectory * @return boolean */ public static function escalateReadLock(Zend_Search_Lucene_Storage_Directory $lockDirectory) { self::_startReadLockProcessing($lockDirectory); $lock = $lockDirectory->getFileObject(self::READ_LOCK_FILE); // First, release the shared lock for the benefit of GFS since // it will fail the conditional request to promote the lock to // "exclusive" while the shared lock is held (even when we are // the only holder). $lock->unlock(); // GFS is really poor. While the above "unlock" returns, GFS // doesn't clean up it's tables right away (which will potentially // cause the conditional locking for the "exclusive" lock to fail. // We will retry the conditional lock request several times on a // failure to get past this. The performance hit is negligible // in the grand scheme of things and only will occur with GFS // filesystems or if another local process has the shared lock // on local filesystems. for ($retries = 0; $retries < 10; $retries++) { if ($lock->lock(LOCK_EX, true)) { // Exclusive lock is obtained! self::_stopReadLockProcessing($lockDirectory); return true; } // wait 1 microsecond usleep(1); } // Restore lock state $lock->lock(LOCK_SH); self::_stopReadLockProcessing($lockDirectory); return false; } /** * De-escalate Read lock to shared level * * @param Zend_Search_Lucene_Storage_Directory $lockDirectory */ public static function deEscalateReadLock(Zend_Search_Lucene_Storage_Directory $lockDirectory) { $lock = $lockDirectory->getFileObject(self::READ_LOCK_FILE); $lock->lock(LOCK_SH); } /** * Obtain exclusive optimization lock on the index * * Returns lock object on success and false otherwise (doesn't block execution) * * @param Zend_Search_Lucene_Storage_Directory $lockDirectory * @return mixed */ public static function obtainOptimizationLock(Zend_Search_Lucene_Storage_Directory $lockDirectory) { $lock = $lockDirectory->createFile(self::OPTIMIZATION_LOCK_FILE); if (!$lock->lock(LOCK_EX, true)) { return false; } return $lock; } /** * Release exclusive optimization lock * * @param Zend_Search_Lucene_Storage_Directory $lockDirectory */ public static function releaseOptimizationLock(Zend_Search_Lucene_Storage_Directory $lockDirectory) { $lock = $lockDirectory->getFileObject(self::OPTIMIZATION_LOCK_FILE); $lock->unlock(); } }