More performance improvements to metadata lazyloading.

Comment and term meta lazyloading for `WP_Query` loops, introduced in 4.4,
depended on filter callback methods belonging to `WP_Query` objects. This meant
storing `WP_Query` objects in the `$wp_filter` global (via `add_filter()`),
requiring that PHP retain the objects in memory, even when the local variables
would typically be expunged during normal garbage collection. In cases where a
large number of `WP_Query` objects were instantiated on a single pageload,
and/or where the contents of the `WP_Query` objects were quite large, serious
performance issues could result.

We skirt this problem by moving metadata lazyloading out of `WP_Query`. The
new `WP_Metadata_Lazyloader` class acts as a lazyload queue. Query instances
register items whose metadata should be lazyloaded - such as post terms, or
comments - and a `WP_Metadata_Lazyloader` method will intercept comment and
term meta requests to perform the cache priming. Since `WP_Metadata_Lazyloader`
instances are far smaller than `WP_Query` (containing only object IDs), and
clean up after themselves far better than the previous `WP_Query` methods (bp
only running their callbacks a single time for a given set of queued objects),
the resource use is decreased dramatically.

See [36525] for an earlier step in this direction.

Props lpawlik, stevegrunwell, boonebgorges.
Fixes #35816.
Built from https://develop.svn.wordpress.org/trunk@36566


git-svn-id: http://core.svn.wordpress.org/trunk@36533 1a063a9b-81f0-0310-95a4-ce76da25c4cd

wp-includes/class-wp-metadata-lazyloader.php

@@ -0,0 +1,164 @@
+<?php
+
+/**
+ * Lazyloader for object metadata.
+ *
+ * When loading many objects of a given type, such as posts in a WP_Query loop, it often makes
+ * sense to prime various metadata caches at the beginning of the loop. This means fetching all
+ * relevant metadata with a single database query, a technique that has the potential to improve
+ * performance dramatically in some cases.
+ *
+ * In cases where the given metadata may not even be used in the loop, we can improve performance
+ * even more by only priming the metadata cache for affected items the first time a piece of metadata
+ * is requested - ie, by lazyloading it. So, for example, comment meta may not be loaded into the
+ * cache in the comments section of a post until the first time get_comment_meta() is called in the
+ * context of the comment loop.
+ *
+ * WP uses the WP_Metadata_Lazyloader class to queue objects for metadata cache priming. The class
+ * then detects the relevant get_*_meta() function call, and queries the metadata of all queued objects.
+ *
+ * Do not access this class directly. Use the wp_metadata_lazyloader() function.
+ *
+ * @since 4.5.0
+ */
+class WP_Metadata_Lazyloader {
+	/**
+	 * Pending objects queue.
+	 *
+	 * @since 4.5.0
+	 * @var array
+	 */
+	protected $pending_objects;
+
+	/**
+	 * Settings for supported object types.
+	 *
+	 * @since 4.5.0
+	 * @var array
+	 */
+	protected $settings = array();
+
+	/**
+	 * Constructor.
+	 *
+	 * @since 4.5.0
+	 */
+	public function __construct() {
+		$this->settings = array(
+			'term' => array(
+				'filter'   => 'get_term_metadata',
+				'callback' => array( $this, 'lazyload_term_meta' ),
+			),
+			'comment' => array(
+				'filter'   => 'get_comment_metadata',
+				'callback' => array( $this, 'lazyload_comment_meta' ),
+			),
+		);
+	}
+
+	/**
+	 * Add objects to the metadata lazyload queue.
+	 *
+	 * @since 4.5.0
+	 *
+	 * @param string $object_type Type of object whose meta is to be lazyloaded. Accepts 'term' or 'comment'.
+	 * @param array  $object_ids  Array of object IDs.
+	 * @return bool|WP_Error True on success, WP_Error on failure.
+	 */
+	public function queue_objects( $object_type, $object_ids ) {
+		if ( ! isset( $this->settings[ $object_type ] ) ) {
+			return new WP_Error( 'invalid_object_type', __( 'Invalid object type' ) );
+		}
+
+		$type_settings = $this->settings[ $object_type ];
+
+		if ( ! isset( $this->pending_objects[ $object_type ] ) ) {
+			$this->pending_objects[ $object_type ] = array();
+		}
+
+		foreach ( $object_ids as $object_id ) {
+			// Keyed by ID for faster lookup.
+			if ( ! isset( $this->pending_objects[ $object_type ][ $object_id ] ) ) {
+				$this->pending_objects[ $object_type ][ $object_id ] = 1;
+			}
+		}
+
+		add_filter( $type_settings['filter'], $type_settings['callback'] );
+
+		/**
+		 * Fires after objects are added to the metadata lazyload queue.
+		 *
+		 * @since 4.5.0
+		 *
+		 * @param array                  $object_ids  Object IDs.
+		 * @param string                 $object_type Type of object being queued.
+		 * @param WP_Metadata_Lazyloader $lazyloader  The lazyloader object.
+		 */
+		do_action( 'metadata_lazyloader_queued_objects', $object_ids, $object_type, $this );
+	}
+
+	/**
+	 * Reset lazyload queue for a given object type.
+	 *
+	 * @since 4.5.0
+	 *
+	 * @param string $object_type Object type. Accepts 'comment' or 'term'.
+	 * @return bool|WP_Error True on success, WP_Error on failure.
+	 */
+	public function reset_queue( $object_type ) {
+		if ( ! isset( $this->settings[ $object_type ] ) ) {
+			return new WP_Error( 'invalid_object_type', __( 'Invalid object type' ) );
+		}
+
+		$type_settings = $this->settings[ $object_type ];
+
+		$this->pending_objects[ $object_type ] = array();
+		remove_filter( $type_settings['filter'], $type_settings['callback'] );
+	}
+
+	/**
+	 * Lazyloads term meta for queued terms.
+	 *
+	 * This method is public so that it can be used as a filter callback. As a rule, there
+	 * is no need to invoke it directly.
+	 *
+	 * @since 4.5.0
+	 * @access public
+	 *
+	 * @param mixed $check The `$check` param passed from the 'get_term_metadata' hook.
+	 * @return mixed In order not to short-circuit `get_metadata()`. Generally, this is `null`, but it could be
+	 *               another value if filtered by a plugin.
+	 */
+	public function lazyload_term_meta( $check ) {
+		if ( ! empty( $this->pending_objects['term'] ) ) {
+			update_termmeta_cache( array_keys( $this->pending_objects['term'] ) );
+
+			// No need to run again for this set of terms.
+			$this->reset_queue( 'term' );
+		}
+
+		return $check;
+	}
+
+	/**
+	 * Lazyload comment meta for queued comments.
+	 *
+	 * This method is public so that it can be used as a filter callback. As a rule, there is no need to invoke it
+	 * directly, from either inside or outside the `WP_Query` object.
+	 *
+	 * @since 4.5.0
+	 *
+	 * @param mixed $check The `$check` param passed from the 'get_comment_metadata' hook.
+	 * @return mixed The original value of `$check`, so as not to short-circuit `get_comment_metadata()`.
+	 */
+	public function lazyload_comment_meta( $check ) {
+		if ( ! empty( $this->pending_objects['comment'] ) ) {
+			update_meta_cache( 'comment', array_keys( $this->pending_objects['comment'] ) );
+
+			// No need to run again for this set of comments.
+			$this->reset_queue( 'comment' );
+		}
+
+		return $check;
+	}
+}

wp-includes/comment-template.php

@@ -1393,9 +1393,6 @@ function comments_template( $file = '/comments.php', $separate_comments = false
 	 */
 	$wp_query->comments = apply_filters( 'comments_array', $comments_flat, $post->ID );
 
-	// Set up lazy-loading for comment metadata.
-	add_action( 'get_comment_metadata', array( $wp_query, 'lazyload_comment_meta' ), 10, 2 );
-
 	$comments = &$wp_query->comments;
 	$wp_query->comment_count = count($wp_query->comments);
 	$wp_query->max_num_comment_pages = $comment_query->max_num_pages;
@@ -2030,6 +2027,8 @@ function wp_list_comments( $args = array(), $comments = null ) {
 	if ( null === $r['reverse_top_level'] )
 		$r['reverse_top_level'] = ( 'desc' == get_option('comment_order') );
 
+	wp_queue_comments_for_comment_meta_lazyload( $_comments );
+
 	if ( empty( $r['walker'] ) ) {
 		$walker = new Walker_Comment;
 	} else {

wp-includes/comment.php

@@ -468,6 +468,30 @@ function update_comment_meta($comment_id, $meta_key, $meta_value, $prev_value =
 	return update_metadata('comment', $comment_id, $meta_key, $meta_value, $prev_value);
 }
 
+/**
+ * Queue comments for metadata lazyloading.
+ *
+ * @since 4.5.0
+ *
+ * @param array $comments Array of comment objects.
+ */
+function wp_queue_comments_for_comment_meta_lazyload( $comments ) {
+	// Don't use `wp_list_pluck()` to avoid by-reference manipulation.
+	$comment_ids = array();
+	if ( is_array( $comments ) ) {
+		foreach ( $comments as $comment ) {
+			if ( $comment instanceof WP_Comment ) {
+				$comment_ids[] = $comment->comment_ID;
+			}
+		}
+	}
+
+	if ( $comment_ids ) {
+		$lazyloader = wp_metadata_lazyloader();
+		$lazyloader->queue_objects( 'comment', $comment_ids );
+	}
+}
+
 /**
  * Sets the cookies used to store an unauthenticated commentator's identity. Typically used
  * to recall previous comments by this commentator that are still held in moderation.

wp-includes/meta.php

@@ -851,6 +851,23 @@ function update_meta_cache($meta_type, $object_ids) {
 	return $cache;
 }
 
+/**
+ * Get the metadata lazyloading queue.
+ *
+ * @since 4.5.0
+ *
+ * @return WP_Metadata_Lazyloader $lazyloader Metadata lazyloader queue.
+ */
+function wp_metadata_lazyloader() {
+	static $wp_metadata_lazyloader;
+
+	if ( null === $wp_metadata_lazyloader ) {
+		$wp_metadata_lazyloader = new WP_Metadata_Lazyloader();
+	}
+
+	return $wp_metadata_lazyloader;
+}
+
 /**
  * Given a meta query, generates SQL clauses to be appended to a main query.
  *

wp-includes/post.php

@@ -5947,6 +5947,43 @@ function wp_delete_auto_drafts() {
 	}
 }
 
+/**
+ * Queue posts for lazyloading of term meta.
+ *
+ * @since 4.5.0
+ *
+ * @param array $posts Array of WP_Post objects.
+ */
+function wp_queue_posts_for_term_meta_lazyload( $posts ) {
+	$post_type_taxonomies = $term_ids = array();
+	foreach ( $posts as $post ) {
+		if ( ! ( $post instanceof WP_Post ) ) {
+			continue;
+		}
+
+		if ( ! isset( $post_type_taxonomies[ $post->post_type ] ) ) {
+			$post_type_taxonomies[ $post->post_type ] = get_object_taxonomies( $post->post_type );
+		}
+
+		foreach ( $post_type_taxonomies[ $post->post_type ] as $taxonomy ) {
+			// Term cache should already be primed by `update_post_term_cache()`.
+			$terms = get_object_term_cache( $post->ID, $taxonomy );
+			if ( false !== $terms ) {
+				foreach ( $terms as $term ) {
+					if ( ! isset( $term_ids[ $term->term_id ] ) ) {
+						$term_ids[] = $term->term_id;
+					}
+				}
+			}
+		}
+	}
+
+	if ( $term_ids ) {
+		$lazyloader = wp_metadata_lazyloader();
+		$lazyloader->queue_objects( 'term', $term_ids );
+	}
+}
+
 /**
  * Update the custom taxonomies' term counts when a post's status is changed.
  *

wp-includes/query.php

@@ -3605,11 +3605,6 @@ class WP_Query {
 		if ( $this->posts )
 			$this->posts = array_map( 'get_post', $this->posts );
 
-
-		if ( $q['update_post_term_cache'] ) {
-			add_filter( 'get_term_metadata', array( $this, 'lazyload_term_meta' ), 10, 2 );
-		}
-
 		if ( ! $q['suppress_filters'] ) {
 			/**
 			 * Filter the raw post results array, prior to status checks.
@@ -3738,7 +3733,7 @@ class WP_Query {
 
 		// If comments have been fetched as part of the query, make sure comment meta lazy-loading is set up.
 		if ( ! empty( $this->comments ) ) {
-			add_filter( 'get_comment_metadata', array( $this, 'lazyload_comment_meta' ), 10, 2 );
+			wp_queue_comments_for_comment_meta_lazyload( $this->comments );
 		}
 
 		if ( ! $q['suppress_filters'] ) {
@@ -3770,6 +3765,10 @@ class WP_Query {
 			$this->posts = array();
 		}
 
+		if ( $q['update_post_term_cache'] ) {
+			wp_queue_posts_for_term_meta_lazyload( $this->posts );
+		}
+
 		return $this->posts;
 	}
 
@@ -4834,106 +4833,32 @@ class WP_Query {
 	}
 
 	/**
-	 * Lazy-loads termmeta for located posts.
+	 * Lazyload term meta for posts in the loop.
-	 *
-	 * As a rule, term queries (`get_terms()` and `wp_get_object_terms()`) prime the metadata cache for matched
-	 * terms by default. However, this can cause a slight performance penalty, especially when that metadata is
-	 * not actually used. In the context of a `WP_Query` instance, we're able to avoid this potential penalty.
-	 * `update_object_term_cache()`, called from `update_post_caches()`, does not 'update_term_meta_cache'.
-	 * Instead, the first time `get_term_meta()` is called from within a `WP_Query` loop, the current method
-	 * detects the fact, and then primes the metadata cache for all terms attached to all posts in the loop,
-	 * with a single database query.
-	 *
-	 * This method is public so that it can be used as a filter callback. As a rule, there is no need to invoke it
-	 * directly, from either inside or outside the `WP_Query` object.
 	 *
 	 * @since 4.4.0
-	 * @access public
+	 * @deprecated 4.5.0 See wp_queue_posts_for_term_meta_lazyload().
 	 *
-	 * @param mixed $check  The `$check` param passed from the 'get_term_metadata' hook.
+	 * @param mixed $check
-	 * @param int  $term_id ID of the term whose metadata is being cached.
+	 * @param int   $term_id
-	 * @return mixed In order not to short-circuit `get_metadata()`. Generally, this is `null`, but it could be
+	 * @return mixed
-	 *               another value if filtered by a plugin.
 	 */
 	public function lazyload_term_meta( $check, $term_id ) {
-		// We can only lazyload if the entire post object is present.
+		_deprecated_function( __METHOD__, '4.5.0' );
-		$posts = array();
-		foreach ( $this->posts as $post ) {
-			if ( $post instanceof WP_Post ) {
-				$posts[] = $post;
-			}
-		}
-
-		if ( ! empty( $posts ) ) {
-			// Fetch cached term_ids for each post. Keyed by term_id for faster lookup.
-			$term_ids = array();
-			foreach ( $posts as $post ) {
-				$taxonomies = get_object_taxonomies( $post->post_type );
-				foreach ( $taxonomies as $taxonomy ) {
-					// Term cache should already be primed by 'update_post_term_cache'.
-					$terms = get_object_term_cache( $post->ID, $taxonomy );
-					if ( false !== $terms ) {
-						foreach ( $terms as $term ) {
-							if ( ! isset( $term_ids[ $term->term_id ] ) ) {
-								$term_ids[ $term->term_id ] = 1;
-							}
-						}
-					}
-				}
-			}
-
-			/*
-			 * Only update the metadata cache for terms belonging to these posts if the term_id passed
-			 * to `get_term_meta()` matches one of those terms. This prevents a single call to
-			 * `get_term_meta()` from priming metadata for all `WP_Query` objects.
-			 */
-			if ( isset( $term_ids[ $term_id ] ) ) {
-				update_termmeta_cache( array_keys( $term_ids ) );
-				remove_filter( 'get_term_metadata', array( $this, 'lazyload_term_meta' ), 10, 2 );
-			}
-		}
-
-		// If no terms were found, there's no need to run this again.
-		if ( empty( $term_ids ) ) {
-			remove_filter( 'get_term_metadata', array( $this, 'lazyload_term_meta' ), 10, 2 );
-		}
-
 		return $check;
 	}
 
 	/**
-	 * Lazy-load comment meta when inside of a `WP_Query` loop.
+	 * Lazyload comment meta for comments in the loop.
-	 *
-	 * This method is public so that it can be used as a filter callback. As a rule, there is no need to invoke it
-	 * directly, from either inside or outside the `WP_Query` object.
 	 *
 	 * @since 4.4.0
+	 * @deprecated 4.5.0 See wp_queue_comments_for_comment_meta_lazyload().
 	 *
-	 * @param mixed $check     The `$check` param passed from the 'get_comment_metadata' hook.
+	 * @param mixed $check
-	 * @param int  $comment_id ID of the comment whose metadata is being cached.
+	 * @param int   $comment_id
-	 * @return mixed The original value of `$check`, to not affect 'get_comment_metadata'.
+	 * @return mixed
 	 */
 	public function lazyload_comment_meta( $check, $comment_id ) {
-		// Don't use `wp_list_pluck()` to avoid by-reference manipulation.
+		_deprecated_function( __METHOD__, '4.5.0' );
-		$comment_ids = array();
-		if ( is_array( $this->comments ) ) {
-			foreach ( $this->comments as $comment ) {
-				$comment_ids[] = $comment->comment_ID;
-			}
-		}
-
-		/*
-		 * Only update the metadata cache for comments belonging to these posts if the comment_id passed
-		 * to `get_comment_meta()` matches one of those comments. This prevents a single call to
-		 * `get_comment_meta()` from priming metadata for all `WP_Query` objects.
-		 */
-		if ( in_array( $comment_id, $comment_ids ) ) {
-			update_meta_cache( 'comment', $comment_ids );
-			remove_filter( 'get_comment_metadata', array( $this, 'lazyload_comment_meta' ), 10, 2 );
-		} elseif ( empty( $comment_ids ) ) {
-			remove_filter( 'get_comment_metadata', array( $this, 'lazyload_comment_meta' ), 10, 2 );
-		}
-
 		return $check;
 	}
 }

wp-includes/version.php

@@ -4,7 +4,7 @@
  *
  * @global string $wp_version
  */
-$wp_version = '4.5-alpha-36565';
+$wp_version = '4.5-alpha-36566';
 
 /**
  * Holds the WordPress DB revision, increments when changes are made to the WordPress DB schema.

wp-settings.php

@@ -135,6 +135,7 @@ require( ABSPATH . WPINC . '/class-wp-user-query.php' );
 require( ABSPATH . WPINC . '/session.php' );
 require( ABSPATH . WPINC . '/meta.php' );
 require( ABSPATH . WPINC . '/class-wp-meta-query.php' );
+require( ABSPATH . WPINC . '/class-wp-metadata-lazyloader.php' );
 require( ABSPATH . WPINC . '/general-template.php' );
 require( ABSPATH . WPINC . '/link-template.php' );
 require( ABSPATH . WPINC . '/author-template.php' );