HTMLCacheUpdateJob.php

Go to the documentation of this file.

<?php
use MediaWiki\MediaWikiServices;
use MediaWiki\Page\PageReference;
 
class HTMLCacheUpdateJob extends Job {
    private const NORMAL_MAX_LAG = 10;
 
    public 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 )
        );
        $this->params += [ 'causeAction' => 'unknown', 'causeAgent' => 'unknown' ];
    }
 
    public static function newForBacklinks( PageReference $page, $table, $params = [] ) {
        $title = Title::castFromPageReference( $page );
        return new self(
            $title,
            [
                'table' => $table,
                'recursive' => true
            ] + Job::newRootJobParams( // "overall" refresh links job info
                "htmlCacheUpdate:{$table}:{$title->getPrefixedText()}"
            ) + $params
        );
    }
 
    public function run() {
        $updateRowsPerJob = MediaWikiServices::getInstance()->getMainConfig()->get( 'UpdateRowsPerJob' );
        $updateRowsPerQuery = MediaWikiServices::getInstance()->getMainConfig()->get( 'UpdateRowsPerQuery' );
        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'] ) ) {
            // Carry over information for de-duplication
            $extraParams = $this->getRootJobParams();
            // Carry over cause information for logging
            $extraParams['causeAction'] = $this->params['causeAction'];
            $extraParams['causeAgent'] = $this->params['causeAgent'];
            // 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,
                $updateRowsPerJob,
                $updateRowsPerQuery, // jobs-per-title
                // Carry over information for de-duplication
                [ 'params' => $extraParams ]
            );
            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 ) {
        // Get all page IDs in this query into an array
        $pageIds = array_keys( $pages );
        if ( !$pageIds ) {
            return;
        }
 
        $rootTsUnix = wfTimestampOrNull( TS_UNIX, $this->params['rootJobTimestamp'] ?? null );
        // Bump page_touched to the current timestamp. This previously used the root job timestamp
        // (e.g. template/file edit time), which is a bit more efficient when template edits are
        // rare and don't effect the same pages much. However, this way better de-duplicates jobs,
        // which is much more useful for wikis with high edit rates. Note that RefreshLinksJob,
        // 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.
        $newTouchedUnix = time();
        // Timestamp used to bypass pages already invalided since the triggering event
        $casTsUnix = $rootTsUnix ?? $newTouchedUnix;
 
        $services = MediaWikiServices::getInstance();
        $config = $services->getMainConfig();
 
        $lbFactory = $services->getDBLoadBalancerFactory();
        $dbw = $lbFactory->getMainLB()->getConnectionRef( DB_PRIMARY );
        $ticket = $lbFactory->getEmptyTransactionTicket( __METHOD__ );
        // Update page_touched (skipping pages already touched since the root job).
        // Check $wgUpdateRowsPerQuery; batch jobs are sized by that already.
        $batches = array_chunk( $pageIds, $config->get( 'UpdateRowsPerQuery' ) );
        foreach ( $batches as $batch ) {
            $dbw->update( 'page',
                [ 'page_touched' => $dbw->timestamp( $newTouchedUnix ) ],
                [
                    'page_id' => $batch,
                    "page_touched < " . $dbw->addQuotes( $dbw->timestamp( $casTsUnix ) )
                ],
                __METHOD__
            );
            if ( count( $batches ) > 1 ) {
                $lbFactory->commitAndWaitForReplication( __METHOD__, $ticket );
            }
        }
        // Get the list of affected pages (races only mean something else did the purge)
        $titleArray = TitleArray::newFromResult( $dbw->select(
            'page',
            array_merge(
                [ 'page_namespace', 'page_title' ],
                $config->get( 'PageLanguageUseDB' ) ? [ 'page_lang' ] : []
            ),
            [ 'page_id' => $pageIds, 'page_touched' => $dbw->timestamp( $newTouchedUnix ) ],
            __METHOD__
        ) );
 
        // Update CDN and file caches
        $htmlCache = MediaWikiServices::getInstance()->getHtmlCacheUpdater();
        $htmlCache->purgeTitleUrls(
            $titleArray,
            $htmlCache::PURGE_NAIVE | $htmlCache::PURGE_URLS_LINKSUPDATE_ONLY,
            [ $htmlCache::UNLESS_CACHE_MTIME_AFTER => $casTsUnix + self::NORMAL_MAX_LAG ]
        );
    }
 
    public function getDeduplicationInfo() {
        $info = parent::getDeduplicationInfo();
        if ( is_array( $info['params'] ) ) {
            // For per-pages jobs, the job title is that of the template that changed
            // (or similar), so remove that since it ruins duplicate detection
            if ( isset( $info['params']['pages'] ) ) {
                unset( $info['namespace'] );
                unset( $info['title'] );
            }
        }
 
        return $info;
    }
 
    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
    }
}