Code Coverage |
||||||||||
Lines |
Functions and Methods |
Classes and Traits |
||||||||
| Total | |
0.00% |
0 / 87 |
|
0.00% |
0 / 6 |
CRAP | |
0.00% |
0 / 1 |
| ElasticaWrite | |
0.00% |
0 / 87 |
|
0.00% |
0 / 6 |
272 | |
0.00% |
0 / 1 |
| build | |
0.00% |
0 / 9 |
|
0.00% |
0 / 1 |
2 | |||
| serde | |
0.00% |
0 / 8 |
|
0.00% |
0 / 1 |
42 | |||
| __construct | |
0.00% |
0 / 9 |
|
0.00% |
0 / 1 |
2 | |||
| allowRetries | |
0.00% |
0 / 1 |
|
0.00% |
0 / 1 |
2 | |||
| doJob | |
0.00% |
0 / 35 |
|
0.00% |
0 / 1 |
30 | |||
| requeueError | |
0.00% |
0 / 25 |
|
0.00% |
0 / 1 |
6 | |||
| 1 | <?php |
| 2 | |
| 3 | namespace CirrusSearch\Job; |
| 4 | |
| 5 | use CirrusSearch\Connection; |
| 6 | use CirrusSearch\DataSender; |
| 7 | use CirrusSearch\UpdateGroup; |
| 8 | use MediaWiki\Logger\LoggerFactory; |
| 9 | use MediaWiki\MediaWikiServices; |
| 10 | use MediaWiki\Status\Status; |
| 11 | use Wikimedia\Assert\Assert; |
| 12 | |
| 13 | /** |
| 14 | * Performs writes to elasticsearch indexes with requeuing and an |
| 15 | * exponential backoff (if supported by jobqueue) when the index |
| 16 | * writes fail. |
| 17 | * |
| 18 | * @license GPL-2.0-or-later |
| 19 | */ |
| 20 | class ElasticaWrite extends CirrusGenericJob { |
| 21 | |
| 22 | private const MAX_ERROR_RETRY = 4; |
| 23 | |
| 24 | /** |
| 25 | * @var array Map from method name to list of classes to |
| 26 | * handle serialization for each argument. |
| 27 | */ |
| 28 | private static $SERDE = [ |
| 29 | 'sendData' => [ null, ElasticaDocumentsJsonSerde::class ], |
| 30 | ]; |
| 31 | |
| 32 | /** |
| 33 | * @param string $cluster |
| 34 | * @param string $updateGroup UpdateGroup::* constant |
| 35 | * @param string $method |
| 36 | * @param array $arguments |
| 37 | * @param array $params |
| 38 | * @param string|null $updateKind the kind of update to perform (used for monitoring) |
| 39 | * @param int|null $rootEventTime the time of MW event that caused this update (used for monitoring) |
| 40 | * @return self |
| 41 | */ |
| 42 | public static function build( |
| 43 | string $cluster, |
| 44 | string $updateGroup, |
| 45 | string $method, |
| 46 | array $arguments, |
| 47 | array $params = [], |
| 48 | ?string $updateKind = null, |
| 49 | ?int $rootEventTime = null |
| 50 | ): self { |
| 51 | return new self( [ |
| 52 | 'method' => $method, |
| 53 | 'arguments' => self::serde( $method, $arguments ), |
| 54 | 'cluster' => $cluster, |
| 55 | 'update_group' => $updateGroup, |
| 56 | CirrusTitleJob::UPDATE_KIND => $updateKind, |
| 57 | CirrusTitleJob::ROOT_EVENT_TIME => $rootEventTime |
| 58 | ] + $params ); |
| 59 | } |
| 60 | |
| 61 | private static function serde( string $method, array $arguments, bool $serialize = true ): array { |
| 62 | if ( isset( self::$SERDE[$method] ) ) { |
| 63 | foreach ( self::$SERDE[$method] as $i => $serde ) { |
| 64 | if ( $serde !== null && array_key_exists( $i, $arguments ) ) { |
| 65 | $impl = new $serde(); |
| 66 | if ( $serialize ) { |
| 67 | $arguments[$i] = $impl->serialize( $arguments[$i] ); |
| 68 | } else { |
| 69 | $arguments[$i] = $impl->deserialize( $arguments[$i] ); |
| 70 | } |
| 71 | } |
| 72 | } |
| 73 | } |
| 74 | return $arguments; |
| 75 | } |
| 76 | |
| 77 | /** |
| 78 | * Entry point for jobs received from the job queue. Creating new |
| 79 | * jobs should be done via self::build. |
| 80 | */ |
| 81 | public function __construct( array $params ) { |
| 82 | parent::__construct( $params + [ |
| 83 | 'createdAt' => time(), |
| 84 | 'errorCount' => 0, |
| 85 | 'retryCount' => 0, |
| 86 | 'cluster' => null, |
| 87 | CirrusTitleJob::UPDATE_KIND => null, |
| 88 | CirrusTitleJob::ROOT_EVENT_TIME => null, |
| 89 | // BC for jobs created pre-1.42 |
| 90 | 'update_group' => UpdateGroup::PAGE, |
| 91 | ] ); |
| 92 | } |
| 93 | |
| 94 | /** |
| 95 | * This job handles all its own retries internally. These jobs are so |
| 96 | * numerous that if they were to start failing they would possibly |
| 97 | * overflow the job queue and bring down redis in production. |
| 98 | * |
| 99 | * Basically we just can't let these jobs hang out in the abandoned |
| 100 | * queue for a week like retries typically do. If these jobs get |
| 101 | * failed they will log to CirrusSearchChangeFailed which is a signal |
| 102 | * that some point in time around the failure needs to be reindexed |
| 103 | * manually. See https://wikitech.wikimedia.org/wiki/Search for more |
| 104 | * details. |
| 105 | * @return bool |
| 106 | */ |
| 107 | public function allowRetries() { |
| 108 | return false; |
| 109 | } |
| 110 | |
| 111 | /** |
| 112 | * @return bool |
| 113 | */ |
| 114 | protected function doJob() { |
| 115 | // While we can only have a single connection per job, we still |
| 116 | // use decideClusters() which includes a variety of safeguards. |
| 117 | $connections = $this->decideClusters( $this->params['update_group'] ); |
| 118 | if ( !$connections ) { |
| 119 | // Chosen cluster no longer exists in configuration. |
| 120 | return true; |
| 121 | } |
| 122 | Assert::precondition( count( $connections ) == 1, |
| 123 | 'per self::build() we must have a single connection' ); |
| 124 | |
| 125 | $conn = reset( $connections ); |
| 126 | $arguments = self::serde( $this->params['method'], $this->params['arguments'], false ); |
| 127 | |
| 128 | LoggerFactory::getInstance( 'CirrusSearch' )->debug( |
| 129 | "Running {method} on cluster {cluster} {diff}s after insertion", |
| 130 | [ |
| 131 | 'method' => $this->params['method'], |
| 132 | 'arguments' => $arguments, |
| 133 | 'diff' => time() - $this->params['createdAt'], |
| 134 | 'cluster' => $conn->getClusterName(), |
| 135 | ] |
| 136 | ); |
| 137 | |
| 138 | $sender = new DataSender( $conn, $this->searchConfig ); |
| 139 | try { |
| 140 | $status = $sender->{$this->params['method']}( ...$arguments ); |
| 141 | } catch ( \Exception $e ) { |
| 142 | LoggerFactory::getInstance( 'CirrusSearch' )->warning( |
| 143 | "Exception thrown while running DataSender::{method} in cluster {cluster}: {errorMessage}", |
| 144 | [ |
| 145 | 'method' => $this->params['method'], |
| 146 | 'cluster' => $conn->getClusterName(), |
| 147 | 'errorMessage' => $e->getMessage(), |
| 148 | 'exception' => $e, |
| 149 | ] |
| 150 | ); |
| 151 | $status = Status::newFatal( 'cirrussearch-send-failure' ); |
| 152 | } |
| 153 | |
| 154 | $ok = true; |
| 155 | if ( !$status->isOK() ) { |
| 156 | $action = $this->requeueError( $conn ) ? "Requeued" : "Dropped"; |
| 157 | $this->setLastError( "ElasticaWrite job failed: {$action}" ); |
| 158 | $ok = false; |
| 159 | } |
| 160 | |
| 161 | return $ok; |
| 162 | } |
| 163 | |
| 164 | /** |
| 165 | * Re-queue job that failed, or drop the job if it has failed |
| 166 | * too many times |
| 167 | * |
| 168 | * @param Connection $conn |
| 169 | * @return bool True when the job has been queued |
| 170 | */ |
| 171 | private function requeueError( Connection $conn ) { |
| 172 | if ( $this->params['errorCount'] >= self::MAX_ERROR_RETRY ) { |
| 173 | LoggerFactory::getInstance( 'CirrusSearchChangeFailed' )->warning( |
| 174 | "Dropping failing ElasticaWrite job for DataSender::{method} in cluster {cluster} after repeated failure", |
| 175 | [ |
| 176 | 'method' => $this->params['method'], |
| 177 | 'cluster' => $conn->getClusterName(), |
| 178 | ] |
| 179 | ); |
| 180 | return false; |
| 181 | } else { |
| 182 | $delay = $this->backoffDelay( $this->params['retryCount'] ); |
| 183 | $params = $this->params; |
| 184 | $params['errorCount']++; |
| 185 | unset( $params['jobReleaseTimestamp'] ); |
| 186 | $jobQueue = MediaWikiServices::getInstance()->getJobQueueGroup(); |
| 187 | $params += self::buildJobDelayOptions( self::class, $delay, $jobQueue ); |
| 188 | $job = new self( $params ); |
| 189 | // Individual failures should have already logged specific errors, |
| 190 | LoggerFactory::getInstance( 'CirrusSearch' )->info( |
| 191 | "ElasticaWrite job reported failure on cluster {cluster}. Requeueing job with delay of {delay}.", |
| 192 | [ |
| 193 | 'cluster' => $conn->getClusterName(), |
| 194 | 'delay' => $delay |
| 195 | ] |
| 196 | ); |
| 197 | $jobQueue->push( $job ); |
| 198 | return true; |
| 199 | } |
| 200 | } |
| 201 | } |