includes/LoadBalancer.php

   1 <?php
   2 /**
   3  *
   4  * @package MediaWiki
   5  */
   6
   7 /**
   8  * Depends on the database object
   9  */
  10 require_once( 'Database.php' );
  11
  12 # Valid database indexes
  13 # Operation-based indexes
  14 define( 'DB_SLAVE', -1 );     # Read from the slave (or only server)
  15 define( 'DB_MASTER', -2 );    # Write to master (or only server)
  16 define( 'DB_LAST', -3 );     # Whatever database was used last
  17
  18 # Obsolete aliases
  19 define( 'DB_READ', -1 );
  20 define( 'DB_WRITE', -2 );
  21
  22
  23 # Scale polling time so that under overload conditions, the database server
  24 # receives a SHOW STATUS query at an average interval of this many microseconds
  25 define( 'AVG_STATUS_POLL', 2000 );
  26
  27
  28 /**
  29  * Database load balancing object
  30  *
  31  * @todo document
  32  * @package MediaWiki
  33  */
  34 class LoadBalancer {
  35         /* private */ var $mServers, $mConnections, $mLoads, $mGroupLoads;
  36         /* private */ var $mFailFunction, $mErrorConnection;
  37         /* private */ var $mForce, $mReadIndex, $mLastIndex;
  38         /* private */ var $mWaitForFile, $mWaitForPos, $mWaitTimeout;
  39         /* private */ var $mLaggedSlaveMode, $mLastError = 'Unknown error';
  40
  41         function LoadBalancer()
  42         {
  43                 $this->mServers = array();
  44                 $this->mConnections = array();
  45                 $this->mFailFunction = false;
  46                 $this->mReadIndex = -1;
  47                 $this->mForce = -1;
  48                 $this->mLastIndex = -1;
  49                 $this->mErrorConnection = false;
  50         }
  51
  52         function newFromParams( $servers, $failFunction = false, $waitTimeout = 10 )
  53         {
  54                 $lb = new LoadBalancer;
  55                 $lb->initialise( $servers, $failFunction, $waitTimeout );
  56                 return $lb;
  57         }
  58
  59         function initialise( $servers, $failFunction = false, $waitTimeout = 10 )
  60         {
  61                 $this->mServers = $servers;
  62                 $this->mFailFunction = $failFunction;
  63                 $this->mReadIndex = -1;
  64                 $this->mWriteIndex = -1;
  65                 $this->mForce = -1;
  66                 $this->mConnections = array();
  67                 $this->mLastIndex = 1;
  68                 $this->mLoads = array();
  69                 $this->mWaitForFile = false;
  70                 $this->mWaitForPos = false;
  71                 $this->mWaitTimeout = $waitTimeout;
  72                 $this->mLaggedSlaveMode = false;
  73
  74                 foreach( $servers as $i => $server ) {
  75                         $this->mLoads[$i] = $server['load'];
  76                         if ( isset( $server['groupLoads'] ) ) {
  77                                 foreach ( $server['groupLoads'] as $group => $ratio ) {
  78                                         if ( !isset( $this->mGroupLoads[$group] ) ) {
  79                                                 $this->mGroupLoads[$group] = array();
  80                                         }
  81                                         $this->mGroupLoads[$group][$i] = $ratio;
  82                                 }
  83                         }
  84                 }
  85         }
  86
  87         /**
  88          * Given an array of non-normalised probabilities, this function will select
  89          * an element and return the appropriate key
  90          */
  91         function pickRandom( $weights )
  92         {
  93                 if ( !is_array( $weights ) || count( $weights ) == 0 ) {
  94                         return false;
  95                 }
  96
  97                 $sum = 0;
  98                 foreach ( $weights as $w ) {
  99                         $sum += $w;
 100                 }
 101
 102                 if ( $sum == 0 ) {
 103                         # No loads on any of them
 104                         # Just pick one at random
 105                         foreach ( $weights as $i => $w ) {
 106                                 $weights[$i] = 1;
 107                         }
 108                 }
 109                 $max = mt_getrandmax();
 110                 $rand = mt_rand(0, $max) / $max * $sum;
 111
 112                 $sum = 0;
 113                 foreach ( $weights as $i => $w ) {
 114                         $sum += $w;
 115                         if ( $sum >= $rand ) {
 116                                 break;
 117                         }
 118                 }
 119                 return $i;
 120         }
 121
 122         function getRandomNonLagged( $loads ) {
 123                 # Unset excessively lagged servers
 124                 $lags = $this->getLagTimes();
 125                 foreach ( $lags as $i => $lag ) {
 126                         if ( isset( $this->mServers[$i]['max lag'] ) && $lag > $this->mServers[$i]['max lag'] ) {
 127                                 unset( $loads[$i] );
 128                         }
 129                 }
 130
 131
 132                 # Find out if all the slaves with non-zero load are lagged
 133                 $sum = 0;
 134                 foreach ( $loads as $load ) {
 135                         $sum += $load;
 136                 }
 137                 if ( $sum == 0 ) {
 138                         # No appropriate DB servers except maybe the master and some slaves with zero load
 139                         # Do NOT use the master
 140                         # Instead, this function will return false, triggering read-only mode,
 141                         # and a lagged slave will be used instead.
 142                         unset ( $loads[0] );
 143                 }
 144
 145                 if ( count( $loads ) == 0 ) {
 146                         return false;
 147                 }
 148
 149                 #wfDebugLog( 'connect', var_export( $loads, true ) );
 150
 151                 # Return a random representative of the remainder
 152                 return $this->pickRandom( $loads );
 153         }
 154
 155         /**
 156          * Get the index of the reader connection, which may be a slave
 157          * This takes into account load ratios and lag times. It should
 158          * always return a consistent index during a given invocation
 159          *
 160          * Side effect: opens connections to databases
 161          */
 162         function getReaderIndex() {
 163                 global $wgReadOnly, $wgDBClusterTimeout;
 164
 165                 $fname = 'LoadBalancer::getReaderIndex';
 166                 wfProfileIn( $fname );
 167
 168                 $i = false;
 169                 if ( $this->mForce >= 0 ) {
 170                         $i = $this->mForce;
 171                 } else {
 172                         if ( $this->mReadIndex >= 0 ) {
 173                                 $i = $this->mReadIndex;
 174                         } else {
 175                                 # $loads is $this->mLoads except with elements knocked out if they
 176                                 # don't work
 177                                 $loads = $this->mLoads;
 178                                 $done = false;
 179                                 $totalElapsed = 0;
 180                                 do {
 181                                         if ( $wgReadOnly ) {
 182                                                 $i = $this->pickRandom( $loads );
 183                                         } else {
 184                                                 $i = $this->getRandomNonLagged( $loads );
 185                                                 if ( $i === false && count( $loads ) != 0 )  {
 186                                                         # All slaves lagged. Switch to read-only mode
 187                                                         $wgReadOnly = wfMsgNoDB( 'readonly_lag' );
 188                                                         $i = $this->pickRandom( $loads );
 189                                                 }
 190                                         }
 191                                         $serverIndex = $i;
 192                                         if ( $i !== false ) {
 193                                                 wfDebugLog( 'connect', "Using reader #$i: {$this->mServers[$i]['host']}...\n" );
 194                                                 $this->openConnection( $i );
 195
 196                                                 if ( !$this->isOpen( $i ) ) {
 197                                                         wfDebug( "Failed\n" );
 198                                                         unset( $loads[$i] );
 199                                                         $sleepTime = 0;
 200                                                 } else {
 201                                                         $status = $this->mConnections[$i]->getStatus();
 202                                                         if ( isset( $this->mServers[$i]['max threads'] ) &&
 203                                                           $status['Threads_running'] > $this->mServers[$i]['max threads'] )
 204                                                         {
 205                                                                 # Slave is lagged, wait for a while
 206                                                                 $sleepTime = AVG_STATUS_POLL * $status['Threads_connected'];
 207
 208                                                                 # If we reach the timeout and exit the loop, don't use it
 209                                                                 $i = false;
 210                                                         } else {
 211                                                                 $done = true;
 212                                                                 $sleepTime = 0;
 213                                                         }
 214                                                 }
 215                                         } else {
 216                                                 $sleepTime = 500000;
 217                                         }
 218                                         if ( $sleepTime ) {
 219                                                         $totalElapsed += $sleepTime;
 220                                                         $x = "{$this->mServers[$serverIndex]['host']} [$serverIndex]";
 221                                                         wfProfileIn( "$fname-sleep $x" );
 222                                                         usleep( $sleepTime );
 223                                                         wfProfileOut( "$fname-sleep $x" );
 224                                         }
 225                                 } while ( count( $loads ) && !$done && $totalElapsed / 1e6 < $wgDBClusterTimeout );
 226
 227                                 if ( $totalElapsed / 1e6 >= $wgDBClusterTimeout ) {
 228                                         $this->mErrorConnection = false;
 229                                         $this->mLastError = 'All servers busy';
 230                                 }
 231
 232                                 if ( $i !== false && $this->isOpen( $i ) ) {
 233                                         # Wait for the session master pos for a short time
 234                                         if ( $this->mWaitForFile ) {
 235                                                 if ( !$this->doWait( $i ) ) {
 236                                                         $this->mServers[$i]['slave pos'] = $this->mConnections[$i]->getSlavePos();
 237                                                 }
 238                                         }
 239                                         if ( $i !== false ) {
 240                                                 $this->mReadIndex = $i;
 241                                         }
 242                                 } else {
 243                                         $i = false;
 244                                 }
 245                         }
 246                 }
 247                 wfProfileOut( $fname );
 248                 return $i;
 249         }
 250
 251         /**
 252          * Get a random server to use in a query group
 253          */
 254         function getGroupIndex( $group ) {
 255                 if ( isset( $this->mGroupLoads[$group] ) ) {
 256                         $i = $this->pickRandom( $this->mGroupLoads[$group] );
 257                 } else {
 258                         $i = false;
 259                 }
 260                 wfDebug( "Query group $group => $i\n" );
 261                 return $i;
 262         }
 263
 264         /**
 265          * Set the master wait position
 266          * If a DB_SLAVE connection has been opened already, waits
 267          * Otherwise sets a variable telling it to wait if such a connection is opened
 268          */
 269         function waitFor( $file, $pos ) {
 270                 $fname = 'LoadBalancer::waitFor';
 271                 wfProfileIn( $fname );
 272
 273                 wfDebug( "User master pos: $file $pos\n" );
 274                 $this->mWaitForFile = false;
 275                 $this->mWaitForPos = false;
 276
 277                 if ( count( $this->mServers ) > 1 ) {
 278                         $this->mWaitForFile = $file;
 279                         $this->mWaitForPos = $pos;
 280                         $i = $this->mReadIndex;
 281
 282                         if ( $i > 0 ) {
 283                                 if ( !$this->doWait( $i ) ) {
 284                                         $this->mServers[$i]['slave pos'] = $this->mConnections[$i]->getSlavePos();
 285                                         $this->mLaggedSlaveMode = true;
 286                                 }
 287                         }
 288                 }
 289                 wfProfileOut( $fname );
 290         }
 291
 292         /**
 293          * Wait for a given slave to catch up to the master pos stored in $this
 294          */
 295         function doWait( $index ) {
 296                 global $wgMemc;
 297
 298                 $retVal = false;
 299
 300                 # Debugging hacks
 301                 if ( isset( $this->mServers[$index]['lagged slave'] ) ) {
 302                         return false;
 303                 } elseif ( isset( $this->mServers[$index]['fake slave'] ) ) {
 304                         return true;
 305                 }
 306
 307                 $key = 'masterpos:' . $index;
 308                 $memcPos = $wgMemc->get( $key );
 309                 if ( $memcPos ) {
 310                         list( $file, $pos ) = explode( ' ', $memcPos );
 311                         # If the saved position is later than the requested position, return now
 312                         if ( $file == $this->mWaitForFile && $this->mWaitForPos <= $pos ) {
 313                                 $retVal = true;
 314                         }
 315                 }
 316
 317                 if ( !$retVal && $this->isOpen( $index ) ) {
 318                         $conn =& $this->mConnections[$index];
 319                         wfDebug( "Waiting for slave #$index to catch up...\n" );
 320                         $result = $conn->masterPosWait( $this->mWaitForFile, $this->mWaitForPos, $this->mWaitTimeout );
 321
 322                         if ( $result == -1 || is_null( $result ) ) {
 323                                 # Timed out waiting for slave, use master instead
 324                                 wfDebug( "Timed out waiting for slave #$index pos {$this->mWaitForFile} {$this->mWaitForPos}\n" );
 325                                 $retVal = false;
 326                         } else {
 327                                 $retVal = true;
 328                                 wfDebug( "Done\n" );
 329                         }
 330                 }
 331                 return $retVal;
 332         }
 333
 334         /**
 335          * Get a connection by index
 336          */
 337         function &getConnection( $i, $fail = true, $groups = array() )
 338         {
 339                 $fname = 'LoadBalancer::getConnection';
 340                 wfProfileIn( $fname );
 341
 342                 # Query groups
 343                 $groupIndex = false;
 344                 foreach ( $groups as $group ) {
 345                         $groupIndex = $this->getGroupIndex( $group );
 346                         if ( $groupIndex !== false ) {
 347                                 $i = $groupIndex;
 348                                 break;
 349                         }
 350                 }
 351
 352                 # Operation-based index
 353                 if ( $i == DB_SLAVE ) {
 354                         $i = $this->getReaderIndex();
 355                 } elseif ( $i == DB_MASTER ) {
 356                         $i = $this->getWriterIndex();
 357                 } elseif ( $i == DB_LAST ) {
 358                         # Just use $this->mLastIndex, which should already be set
 359                         $i = $this->mLastIndex;
 360                         if ( $i === -1 ) {
 361                                 # Oh dear, not set, best to use the writer for safety
 362                                 wfDebug( "Warning: DB_LAST used when there was no previous index\n" );
 363                                 $i = $this->getWriterIndex();
 364                         }
 365                 }
 366                 # Couldn't find a working server in getReaderIndex()?
 367                 if ( $i === false ) {
 368                         $this->reportConnectionError( $this->mErrorConnection );
 369                 }
 370                 # Now we have an explicit index into the servers array
 371                 $this->openConnection( $i, $fail );
 372
 373                 wfProfileOut( $fname );
 374                 return $this->mConnections[$i];
 375         }
 376
 377         /**
 378          * Open a connection to the server given by the specified index
 379          * Index must be an actual index into the array
 380          * Returns success
 381          * @access private
 382          */
 383         function openConnection( $i, $fail = false ) {
 384                 $fname = 'LoadBalancer::openConnection';
 385                 wfProfileIn( $fname );
 386                 $success = true;
 387
 388                 if ( !$this->isOpen( $i ) ) {
 389                         $this->mConnections[$i] = $this->reallyOpenConnection( $this->mServers[$i] );
 390                 }
 391
 392                 if ( !$this->isOpen( $i ) ) {
 393                         wfDebug( "Failed to connect to database $i at {$this->mServers[$i]['host']}\n" );
 394                         if ( $fail ) {
 395                                 $this->reportConnectionError( $this->mConnections[$i] );
 396                         }
 397                         $this->mErrorConnection = $this->mConnections[$i];
 398                         $this->mConnections[$i] = false;
 399                         $success = false;
 400                 }
 401                 $this->mLastIndex = $i;
 402                 wfProfileOut( $fname );
 403                 return $success;
 404         }
 405
 406         /**
 407          * Test if the specified index represents an open connection
 408          * @access private
 409          */
 410         function isOpen( $index ) {
 411                 if( !is_integer( $index ) ) {
 412                         return false;
 413                 }
 414                 if ( array_key_exists( $index, $this->mConnections ) && is_object( $this->mConnections[$index] ) &&
 415                   $this->mConnections[$index]->isOpen() )
 416                 {
 417                         return true;
 418                 } else {
 419                         return false;
 420                 }
 421         }
 422
 423         /**
 424          * Really opens a connection
 425          * @access private
 426          */
 427         function reallyOpenConnection( &$server ) {
 428                 if( !is_array( $server ) ) {
 429                         wfDebugDieBacktrace( 'You must update your load-balancing configuration. See DefaultSettings.php entry for $wgDBservers.' );
 430                 }
 431
 432                 extract( $server );
 433                 # Get class for this database type
 434                 $class = 'Database' . ucfirst( $type );
 435                 if ( !class_exists( $class ) ) {
 436                         require_once( "$class.php" );
 437                 }
 438
 439                 # Create object
 440                 $db = new $class( $host, $user, $password, $dbname, 1, $flags );
 441                 $db->setLBInfo( $server );
 442                 return $db;
 443         }
 444
 445         function reportConnectionError( &$conn )
 446         {
 447                 $fname = 'LoadBalancer::reportConnectionError';
 448                 wfProfileIn( $fname );
 449                 # Prevent infinite recursion
 450
 451                 static $reporting = false;
 452                 if ( !$reporting ) {
 453                         $reporting = true;
 454                         if ( !is_object( $conn ) ) {
 455                                 // No last connection, probably due to all servers being too busy
 456                                 $conn = new Database;
 457                                 if ( $this->mFailFunction ) {
 458                                         $conn->failFunction( $this->mFailFunction );
 459                                         $conn->reportConnectionError( $this->mLastError );
 460                                 } else {
 461                                         // If all servers were busy, mLastError will contain something sensible
 462                                         wfEmergencyAbort( $conn, $this->mLastError );
 463                                 }
 464                         } else {
 465                                 if ( $this->mFailFunction ) {
 466                                         $conn->failFunction( $this->mFailFunction );
 467                                 } else {
 468                                         $conn->failFunction( false );
 469                                 }
 470                                 $conn->reportConnectionError( "{$this->mLastError} ({$conn->mServer})" );
 471                         }
 472                         $reporting = false;
 473                 }
 474                 wfProfileOut( $fname );
 475         }
 476
 477         function getWriterIndex()
 478         {
 479                 return 0;
 480         }
 481
 482         function force( $i )
 483         {
 484                 $this->mForce = $i;
 485         }
 486
 487         function haveIndex( $i )
 488         {
 489                 return array_key_exists( $i, $this->mServers );
 490         }
 491
 492         /**
 493          * Get the number of defined servers (not the number of open connections)
 494          */
 495         function getServerCount() {
 496                 return count( $this->mServers );
 497         }
 498
 499         /**
 500          * Save master pos to the session and to memcached, if the session exists
 501          */
 502         function saveMasterPos() {
 503                 global $wgSessionStarted;
 504                 if ( $wgSessionStarted && count( $this->mServers ) > 1 ) {
 505                         # If this entire request was served from a slave without opening a connection to the
 506                         # master (however unlikely that may be), then we can fetch the position from the slave.
 507                         if ( empty( $this->mConnections[0] ) ) {
 508                                 $conn =& $this->getConnection( DB_SLAVE );
 509                                 list( $file, $pos ) = $conn->getSlavePos();
 510                                 wfDebug( "Saving master pos fetched from slave: $file $pos\n" );
 511                         } else {
 512                                 $conn =& $this->getConnection( 0 );
 513                                 list( $file, $pos ) = $conn->getMasterPos();
 514                                 wfDebug( "Saving master pos: $file $pos\n" );
 515                         }
 516                         if ( $file !== false ) {
 517                                 $_SESSION['master_log_file'] = $file;
 518                                 $_SESSION['master_pos'] = $pos;
 519                         }
 520                 }
 521         }
 522
 523         /**
 524          * Loads the master pos from the session, waits for it if necessary
 525          */
 526         function loadMasterPos() {
 527                 if ( isset( $_SESSION['master_log_file'] ) && isset( $_SESSION['master_pos'] ) ) {
 528                         $this->waitFor( $_SESSION['master_log_file'], $_SESSION['master_pos'] );
 529                 }
 530         }
 531
 532         /**
 533          * Close all open connections
 534          */
 535         function closeAll() {
 536                 foreach( $this->mConnections as $i => $conn ) {
 537                         if ( $this->isOpen( $i ) ) {
 538                                 // Need to use this syntax because $conn is a copy not a reference
 539                                 $this->mConnections[$i]->close();
 540                         }
 541                 }
 542         }
 543
 544         function commitAll() {
 545                 foreach( $this->mConnections as $i => $conn ) {
 546                         if ( $this->isOpen( $i ) ) {
 547                                 // Need to use this syntax because $conn is a copy not a reference
 548                                 $this->mConnections[$i]->immediateCommit();
 549                         }
 550                 }
 551         }
 552
 553         function waitTimeout( $value = NULL ) {
 554                 return wfSetVar( $this->mWaitTimeout, $value );
 555         }
 556
 557         function getLaggedSlaveMode() {
 558                 return $this->mLaggedSlaveMode;
 559         }
 560
 561         function pingAll() {
 562                 $success = true;
 563                 foreach ( $this->mConnections as $i => $conn ) {
 564                         if ( $this->isOpen( $i ) ) {
 565                                 if ( !$this->mConnections[$i]->ping() ) {
 566                                         $success = false;
 567                                 }
 568                         }
 569                 }
 570                 return $success;
 571         }
 572
 573         /**
 574          * Get the hostname and lag time of the most-lagged slave
 575          * This is useful for maintenance scripts that need to throttle their updates
 576          */
 577         function getMaxLag() {
 578                 $maxLag = -1;
 579                 $host = '';
 580                 foreach ( $this->mServers as $i => $conn ) {
 581                         if ( $this->openConnection( $i ) ) {
 582                                 $lag = $this->mConnections[$i]->getLag();
 583                                 if ( $lag > $maxLag ) {
 584                                         $maxLag = $lag;
 585                                         $host = $this->mServers[$i]['host'];
 586                                 }
 587                         }
 588                 }
 589                 return array( $host, $maxLag );
 590         }
 591
 592         /**
 593          * Get lag time for each DB
 594          * Results are cached for a short time in memcached
 595          */
 596         function getLagTimes() {
 597                 global $wgDBname;
 598
 599                 $expiry = 5;
 600                 $requestRate = 10;
 601
 602                 global $wgMemc;
 603                 $times = $wgMemc->get( "$wgDBname:lag_times" );
 604                 if ( $times ) {
 605                         # Randomly recache with probability rising over $expiry
 606                         $elapsed = time() - $times['timestamp'];
 607                         $chance = max( 0, ( $expiry - $elapsed ) * $requestRate );
 608                         if ( mt_rand( 0, $chance ) != 0 ) {
 609                                 unset( $times['timestamp'] );
 610                                 return $times;
 611                         }
 612                 }
 613
 614                 # Cache key missing or expired
 615
 616                 $times = array();
 617                 foreach ( $this->mServers as $i => $conn ) {
 618                         if ( $this->openConnection( $i ) ) {
 619                                 $times[$i] = $this->mConnections[$i]->getLag();
 620                         }
 621                 }
 622
 623                 # Add a timestamp key so we know when it was cached
 624                 $times['timestamp'] = time();
 625                 $wgMemc->set( "$wgDBname:lag_times", $times, $expiry );
 626
 627                 # But don't give the timestamp to the caller
 628                 unset($times['timestamp']);
 629                 return $times;
 630         }
 631 }
 632
 633 ?>