[FileBackend] Added container stat caching to reduce RTTs to high latency backends.
authorAaron <aschulz@wikimedia.org>
Mon, 23 Apr 2012 20:27:58 +0000 (13:27 -0700)
committerAaron <aschulz@wikimedia.org>
Thu, 26 Apr 2012 01:07:11 +0000 (18:07 -0700)
Change-Id: I73575fd65ca06a238803b30f8de6873801a224b3

includes/filerepo/backend/FileBackendStore.php
includes/filerepo/backend/SwiftFileBackend.php

index 55dedc1..f4dd4c0 100644 (file)
@@ -19,6 +19,9 @@
  * @since 1.19
  */
 abstract class FileBackendStore extends FileBackend {
+       /** @var BagOStuff */
+       protected $memCache;
+
        /** @var Array Map of paths to small (RAM/disk) cache items */
        protected $cache = array(); // (storage path => key => value)
        protected $maxCacheSize = 100; // integer; max paths with entries
@@ -31,6 +34,16 @@ abstract class FileBackendStore extends FileBackend {
 
        protected $maxFileSize = 4294967296; // integer bytes (4GiB)
 
+       /**
+        * @see FileBackend::__construct()
+        *
+        * @param $config Array
+        */
+       public function __construct( array $config ) {
+               parent::__construct( $config );
+               $this->memCache = new EmptyBagOStuff(); // disabled by default
+       }
+
        /**
         * Get the maximum allowable file size given backend
         * medium restrictions and basic performance constraints.
@@ -390,11 +403,13 @@ abstract class FileBackendStore extends FileBackend {
 
                if ( $shard !== null ) { // confined to a single container/shard
                        $status->merge( $this->doCleanInternal( $fullCont, $dir, $params ) );
+                       $this->deleteContainerCache( $fullCont ); // purge cache
                } else { // directory is on several shards
                        wfDebug( __METHOD__ . ": iterating over all container shards.\n" );
                        list( $b, $shortCont, $r ) = self::splitStoragePath( $params['dir'] );
                        foreach ( $this->getContainerSuffixes( $shortCont ) as $suffix ) {
                                $status->merge( $this->doCleanInternal( "{$fullCont}{$suffix}", $dir, $params ) );
+                               $this->deleteContainerCache( "{$fullCont}{$suffix}" ); // purge cache
                        }
                }
 
@@ -760,9 +775,12 @@ abstract class FileBackendStore extends FileBackend {
                        }
                }
 
-               // Clear any cache entries (after locks acquired)
+               // Clear any file cache entries (after locks acquired)
                $this->clearCache();
 
+               // Load from the persistent container cache
+               $this->primeContainerCache( $performOps );
+
                // Actually attempt the operation batch...
                $subStatus = FileOp::attemptBatch( $performOps, $opts, $this->fileJournal );
 
@@ -1041,6 +1059,87 @@ abstract class FileBackendStore extends FileBackend {
        protected function resolveContainerPath( $container, $relStoragePath ) {
                return $relStoragePath;
        }
+
+       /**
+        * Get the cache key for a container
+        *
+        * @param $container Resolved container name
+        * @return string
+        */
+       private function containerCacheKey( $container ) {
+               return wfMemcKey( 'backend', $this->getName(), 'container', $container );
+       }
+
+       /**
+        * Set the cached info for a container
+        *
+        * @param $container Resolved container name
+        * @param $val mixed Information to cache
+        * @return void
+        */
+       final protected function setContainerCache( $container, $val ) {
+               $this->memCache->set( $this->containerCacheKey( $container ), $val, 7*86400 );
+       }
+
+       /**
+        * Delete the cached info for a container
+        *
+        * @param $container Resolved container name
+        * @return void
+        */
+       final protected function deleteContainerCache( $container ) {
+               $this->memCache->delete( $this->containerCacheKey( $container ) );
+       }
+
+       /**
+        * Do a batch lookup from cache for container stats for all containers
+        * used in a list of container names, storage paths, or FileOp objects.
+        *
+        * @param $items Array List of storage paths or FileOps
+        * @return void
+        */
+       final protected function primeContainerCache( array $items ) {
+               $paths = array(); // list of storage paths
+               $contNames = array(); // (cache key => resolved container name)
+               // Get all the paths/containers from the items...
+               foreach ( $items as $item ) {
+                       if ( $item instanceof FileOp ) {
+                               $paths = array_merge( $paths, $item->storagePathsRead() );
+                               $paths = array_merge( $paths, $item->storagePathsChanged() );
+                       } elseif ( self::isStoragePath( $item ) ) {
+                               $paths[] = $item;
+                       } elseif ( is_string( $item ) ) { // full container name
+                               $contNames[$this->containerCacheKey( $item )] = $item;
+                       }
+               }
+               // Get all the corresponding cache keys for paths...
+               foreach ( $paths as $path ) {
+                       list( $fullCont, $r, $s ) = $this->resolveStoragePath( $path );
+                       if ( $fullCont !== null ) { // valid path for this backend
+                               $contNames[$this->containerCacheKey( $fullCont )] = $fullCont;
+                       }
+               }
+
+               $contInfo = array(); // (resolved container name => cache value)
+               // Get all cache entries for these container cache keys...
+               $values = $this->memCache->getBatch( array_keys( $contNames ) );
+               foreach ( $values as $cacheKey => $val ) {
+                       $contInfo[$contNames[$cacheKey]] = $val;
+               }
+
+               // Populate the container process cache for the backend...
+               $this->doPrimeContainerCache( array_filter( $contInfo, 'is_array' ) );
+       }
+
+       /**
+        * Fill the backend-specific process cache given an array of
+        * resolved container names and their corresponding cached info.
+        * Only containers that actually exist should appear in the map.
+        *
+        * @param $containerInfo Array Map of resolved container names to cached info
+        * @return void
+        */
+       protected function doPrimeContainerCache( array $containerInfo ) {}
 }
 
 /**
index c7e40e8..18ce451 100644 (file)
@@ -64,6 +64,8 @@ class SwiftFileBackend extends FileBackendStore {
                $this->shardViaHashLevels = isset( $config['shardViaHashLevels'] )
                        ? $config['shardViaHashLevels']
                        : '';
+               // Cache container info to mask latency
+               $this->memCache = wfGetMainCache();
        }
 
        /**
@@ -750,23 +752,30 @@ class SwiftFileBackend extends FileBackendStore {
         * Use $reCache if the file count or byte count is needed.
         *
         * @param $container string Container name
-        * @param $reCache bool Refresh the process cache
+        * @param $bypassCache bool Bypass all caches and load from Swift
         * @return CF_Container
+        * @throws InvalidResponseException
         */
-       protected function getContainer( $container, $reCache = false ) {
+       protected function getContainer( $container, $bypassCache = false ) {
                $conn = $this->getConnection(); // Swift proxy connection
-               if ( $reCache ) {
-                       unset( $this->connContainers[$container] ); // purge cache
+               if ( $bypassCache ) { // purge cache
+                       unset( $this->connContainers[$container] );
+               } elseif ( !isset( $this->connContainers[$container] ) ) {
+                       $this->primeContainerCache( array( $container ) ); // check persistent cache
                }
                if ( !isset( $this->connContainers[$container] ) ) {
                        $contObj = $conn->get_container( $container );
                        // NoSuchContainerException not thrown: container must exist
                        if ( count( $this->connContainers ) >= $this->maxContCacheSize ) { // trim cache?
                                reset( $this->connContainers );
-                               $key = key( $this->connContainers );
-                               unset( $this->connContainers[$key] );
+                               unset( $this->connContainers[key( $this->connContainers )] );
                        }
                        $this->connContainers[$container] = $contObj; // cache it
+                       if ( !$bypassCache ) {
+                               $this->setContainerCache( $container, // update persistent cache
+                                       array( 'bytes' => $contObj->bytes_used, 'count' => $contObj->object_count )
+                               );
+                       }
                }
                return $this->connContainers[$container];
        }
@@ -776,6 +785,7 @@ class SwiftFileBackend extends FileBackendStore {
         *
         * @param $container string Container name
         * @return CF_Container
+        * @throws InvalidResponseException
         */
        protected function createContainer( $container ) {
                $conn = $this->getConnection(); // Swift proxy connection
@@ -789,6 +799,7 @@ class SwiftFileBackend extends FileBackendStore {
         *
         * @param $container string Container name
         * @return void
+        * @throws InvalidResponseException
         */
        protected function deleteContainer( $container ) {
                $conn = $this->getConnection(); // Swift proxy connection
@@ -796,6 +807,28 @@ class SwiftFileBackend extends FileBackendStore {
                unset( $this->connContainers[$container] ); // purge cache
        }
 
+       /**
+        * @see FileBackendStore::doPrimeContainerCache()
+        * @return void
+        */
+       protected function doPrimeContainerCache( array $containerInfo ) {
+               try {
+                       $conn = $this->getConnection(); // Swift proxy connection
+                       foreach ( $containerInfo as $container => $info ) {
+                               $this->connContainers[$container] = new CF_Container(
+                                       $conn->cfs_auth,
+                                       $conn->cfs_http,
+                                       $container,
+                                       $info['count'],
+                                       $info['bytes']
+                               );
+                       }
+               } catch ( InvalidResponseException $e ) {
+               } catch ( Exception $e ) { // some other exception?
+                       $this->logException( $e, __METHOD__, array() );
+               }
+       }
+
        /**
         * Log an unexpected exception for this backend
         *