HTMLCacheUpdateJob.php

Go to the documentation of this file.

<?php
use MediaWiki\MediaWikiServices;
 
class HTMLCacheUpdateJob extends Job {
    function __construct( Title $title, array $params ) {
        parent::__construct( 'htmlCacheUpdate', $title, $params );
        // Avoid the overhead of de-duplication when it would be pointless.
        // Note that these jobs always set page_touched to the current time,
        // so letting the older existing job "win" is still correct.
        $this->removeDuplicates = (
            // Ranges rarely will line up
            !isset( $params['range'] ) &&
            // Multiple pages per job make matches unlikely
            !( isset( $params['pages'] ) && count( $params['pages'] ) != 1 )
        );
    }
 
    public static function newForBacklinks( Title $title, $table ) {
        return new self(
            $title,
            [
                'table' => $table,
                'recursive' => true
            ] + Job::newRootJobParams( // "overall" refresh links job info
                "htmlCacheUpdate:{$table}:{$title->getPrefixedText()}"
            )
        );
    }
 
    function run() {
        global $wgUpdateRowsPerJob, $wgUpdateRowsPerQuery;
 
        if ( isset( $this->params['table'] ) && !isset( $this->params['pages'] ) ) {
            $this->params['recursive'] = true; // b/c; base job
        }
 
        // Job to purge all (or a range of) backlink pages for a page
        if ( !empty( $this->params['recursive'] ) ) {
            // Convert this into no more than $wgUpdateRowsPerJob HTMLCacheUpdateJob per-title
            // jobs and possibly a recursive HTMLCacheUpdateJob job for the rest of the backlinks
            $jobs = BacklinkJobUtils::partitionBacklinkJob(
                $this,
                $wgUpdateRowsPerJob,
                $wgUpdateRowsPerQuery, // jobs-per-title
                // Carry over information for de-duplication
                [ 'params' => $this->getRootJobParams() ]
            );
            JobQueueGroup::singleton()->push( $jobs );
        // Job to purge pages for a set of titles
        } elseif ( isset( $this->params['pages'] ) ) {
            $this->invalidateTitles( $this->params['pages'] );
        // Job to update a single title
        } else {
            $t = $this->title;
            $this->invalidateTitles( [
                $t->getArticleID() => [ $t->getNamespace(), $t->getDBkey() ]
            ] );
        }
 
        return true;
    }
 
    protected function invalidateTitles( array $pages ) {
        global $wgUpdateRowsPerQuery, $wgUseFileCache;
 
        // Get all page IDs in this query into an array
        $pageIds = array_keys( $pages );
        if ( !$pageIds ) {
            return;
        }
 
        // Bump page_touched to the current timestamp. This used to use the root job timestamp
        // (e.g. template/file edit time), which was a bit more efficient when template edits are
        // rare and don't effect the same pages much. However, this way allows for better
        // de-duplication, which is much more useful for wikis with high edit rates. Note that
        // RefreshLinksJob, which is enqueued alongside HTMLCacheUpdateJob, saves the parser output
        // since it has to parse anyway. We assume that vast majority of the cache jobs finish
        // before the link jobs, so using the current timestamp instead of the root timestamp is
        // not expected to invalidate these cache entries too often.
        $touchTimestamp = wfTimestampNow();
        // If page_touched is higher than this, then something else already bumped it after enqueue
        $condTimestamp = isset( $this->params['rootJobTimestamp'] )
            ? $this->params['rootJobTimestamp']
            : $touchTimestamp;
 
        $dbw = wfGetDB( DB_MASTER );
        $factory = MediaWikiServices::getInstance()->getDBLoadBalancerFactory();
        $ticket = $factory->getEmptyTransactionTicket( __METHOD__ );
        // Update page_touched (skipping pages already touched since the root job).
        // Check $wgUpdateRowsPerQuery for sanity; batch jobs are sized by that already.
        foreach ( array_chunk( $pageIds, $wgUpdateRowsPerQuery ) as $batch ) {
            $factory->commitAndWaitForReplication( __METHOD__, $ticket );
 
            $dbw->update( 'page',
                [ 'page_touched' => $dbw->timestamp( $touchTimestamp ) ],
                [ 'page_id' => $batch,
                    // don't invalidated pages that were already invalidated
                    "page_touched < " . $dbw->addQuotes( $dbw->timestamp( $condTimestamp ) )
                ],
                __METHOD__
            );
        }
        // Get the list of affected pages (races only mean something else did the purge)
        $titleArray = TitleArray::newFromResult( $dbw->select(
            'page',
            [ 'page_namespace', 'page_title' ],
            [ 'page_id' => $pageIds, 'page_touched' => $dbw->timestamp( $touchTimestamp ) ],
            __METHOD__
        ) );
 
        // Update CDN; call purge() directly so as to not bother with secondary purges
        $urls = [];
        foreach ( $titleArray as $title ) {
            $urls = array_merge( $urls, $title->getCdnUrls() );
        }
        CdnCacheUpdate::purge( $urls );
 
        // Update file cache
        if ( $wgUseFileCache ) {
            foreach ( $titleArray as $title ) {
                HTMLFileCache::clearFileCache( $title );
            }
        }
    }
 
    public function workItemCount() {
        if ( !empty( $this->params['recursive'] ) ) {
            return 0; // nothing actually purged
        } elseif ( isset( $this->params['pages'] ) ) {
            return count( $this->params['pages'] );
        }
 
        return 1; // one title
    }
}