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  * Database load balancing object
  24  *
  25  * @todo document
  26  * @package MediaWiki
  27  */
  28 class LoadBalancer {
  29         /* private */ var $mServers, $mConnections, $mLoads, $mGroupLoads;
  30         /* private */ var $mFailFunction, $mErrorConnection;
  31         /* private */ var $mForce, $mReadIndex, $mLastIndex;
  32         /* private */ var $mWaitForFile, $mWaitForPos, $mWaitTimeout;
  33         /* private */ var $mLaggedSlaveMode;
  34
  35         function LoadBalancer()
  36         {
  37                 $this->mServers = array();
  38                 $this->mConnections = array();
  39                 $this->mFailFunction = false;
  40                 $this->mReadIndex = -1;
  41                 $this->mForce = -1;
  42                 $this->mLastIndex = -1;
  43                 $this->mErrorConnection = false;
  44         }
  45
  46         function newFromParams( $servers, $failFunction = false, $waitTimeout = 10 )
  47         {
  48                 $lb = new LoadBalancer;
  49                 $lb->initialise( $servers, $failFunction, $waitTimeout );
  50                 return $lb;
  51         }
  52
  53         function initialise( $servers, $failFunction = false, $waitTimeout = 10 )
  54         {
  55                 $this->mServers = $servers;
  56                 $this->mFailFunction = $failFunction;
  57                 $this->mReadIndex = -1;
  58                 $this->mWriteIndex = -1;
  59                 $this->mForce = -1;
  60                 $this->mConnections = array();
  61                 $this->mLastIndex = 1;
  62                 $this->mLoads = array();
  63                 $this->mWaitForFile = false;
  64                 $this->mWaitForPos = false;
  65                 $this->mWaitTimeout = $waitTimeout;
  66                 $this->mLaggedSlaveMode = false;
  67
  68                 foreach( $servers as $i => $server ) {
  69                         $this->mLoads[$i] = $server['load'];
  70                         if ( isset( $server['groupLoads'] ) ) {
  71                                 foreach ( $server['groupLoads'] as $group => $ratio ) {
  72                                         if ( !isset( $this->mGroupLoads[$group] ) ) {
  73                                                 $this->mGroupLoads[$group] = array();
  74                                         }
  75                                         $this->mGroupLoads[$group][$i] = $ratio;
  76                                 }
  77                         }
  78                 }
  79         }
  80
  81         /**
  82          * Given an array of non-normalised probabilities, this function will select
  83          * an element and return the appropriate key
  84          */
  85         function pickRandom( $weights )
  86         {
  87                 if ( !is_array( $weights ) || count( $weights ) == 0 ) {
  88                         return false;
  89                 }
  90
  91                 $sum = 0;
  92                 foreach ( $weights as $w ) {
  93                         $sum += $w;
  94                 }
  95
  96                 if ( $sum == 0 ) {
  97                         # No loads on any of them
  98                         # Just pick one at random
  99                         foreach ( $weights as $i => $w ) {
 100                                 $weights[$i] = 1;
 101                         }
 102                 }
 103                 $max = mt_getrandmax();
 104                 $rand = mt_rand(0, $max) / $max * $sum;
 105
 106                 $sum = 0;
 107                 foreach ( $weights as $i => $w ) {
 108                         $sum += $w;
 109                         if ( $sum >= $rand ) {
 110                                 break;
 111                         }
 112                 }
 113                 return $i;
 114         }
 115
 116         function getRandomNonLagged( $loads ) {
 117                 # Unset excessively lagged servers
 118                 $lags = $this->getLagTimes();
 119                 foreach ( $lags as $i => $lag ) {
 120                         if ( isset( $this->mServers[$i]['max lag'] ) && $lag > $this->mServers[$i]['max lag'] ) {
 121                                 unset( $loads[$i] );
 122                         }
 123                 }
 124
 125
 126                 # Find out if all the slaves with non-zero load are lagged
 127                 $sum = 0;
 128                 foreach ( $loads as $load ) {
 129                         $sum += $load;
 130                 }
 131                 if ( $sum == 0 ) {
 132                         # No appropriate DB servers except maybe the master and some slaves with zero load
 133                         # Do NOT use the master
 134                         # Instead, this function will return false, triggering read-only mode,
 135                         # and a lagged slave will be used instead.
 136                         unset ( $loads[0] );
 137                 }
 138
 139                 if ( count( $loads ) == 0 ) {
 140                         return false;
 141                 }
 142
 143                 #wfDebugLog( 'connect', var_export( $loads, true ) );
 144
 145                 # Return a random representative of the remainder
 146                 return $this->pickRandom( $loads );
 147         }
 148
 149         /**
 150          * Get the index of the reader connection, which may be a slave
 151          * This takes into account load ratios and lag times. It should
 152          * always return a consistent index during a given invocation
 153          *
 154          * Side effect: opens connections to databases
 155          */
 156         function getReaderIndex()
 157         {
 158                 global $wgMaxLag, $wgReadOnly, $wgDBClusterTimeout;
 159
 160                 $fname = 'LoadBalancer::getReaderIndex';
 161                 wfProfileIn( $fname );
 162
 163                 $i = false;
 164                 if ( $this->mForce >= 0 ) {
 165                         $i = $this->mForce;
 166                 } else {
 167                         if ( $this->mReadIndex >= 0 ) {
 168                                 $i = $this->mReadIndex;
 169                         } else {
 170                                 # $loads is $this->mLoads except with elements knocked out if they
 171                                 # don't work
 172                                 $loads = $this->mLoads;
 173                                 $done = false;
 174                                 $totalElapsed = 0;
 175                                 do {
 176                                         if ( $wgReadOnly ) {
 177                                                 $i = $this->pickRandom( $loads );
 178                                         } else {
 179                                                 $i = $this->getRandomNonLagged( $loads );
 180                                                 if ( $i === false && count( $loads ) != 0 )  {
 181                                                         # All slaves lagged. Switch to read-only mode
 182                                                         $wgReadOnly = wfMsgNoDB( 'readonly_lag' );
 183                                                         $i = $this->pickRandom( $loads );
 184                                                 }
 185                                         }
 186                                         $serverIndex = $i;
 187                                         if ( $i !== false ) {
 188                                                 wfDebugLog( 'connect', "Using reader #$i: {$this->mServers[$i]['host']}...\n" );
 189                                                 $this->openConnection( $i );
 190
 191                                                 if ( !$this->isOpen( $i ) ) {
 192                                                         wfDebug( "Failed\n" );
 193                                                         unset( $loads[$i] );
 194                                                         $sleepTime = 0;
 195                                                 } else {
 196                                                         $status = $this->mConnections[$i]->getStatus();
 197                                                         if ( isset( $this->mServers[$i]['max threads'] ) &&
 198                                                           $status['Threads_running'] > $this->mServers[$i]['max threads'] )
 199                                                         {
 200                                                                 # Slave is lagged, wait for a while
 201                                                                 $sleepTime = 5000 * $status['Threads_connected'];
 202
 203                                                                 # If we reach the timeout and exit the loop, don't use it
 204                                                                 $i = false;
 205                                                         } else {
 206                                                                 $done = true;
 207                                                                 $sleepTime = 0;
 208                                                         }
 209                                                 }
 210                                         } else {
 211                                                 $sleepTime = 500000;
 212                                         }
 213                                         if ( $sleepTime ) {
 214                                                         $totalElapsed += $sleepTime;
 215                                                         $x = "{$this->mServers[$serverIndex]['host']} [$serverIndex]";
 216                                                         wfProfileIn( "$fname-sleep $x" );
 217                                                         usleep( $sleepTime );
 218                                                         wfProfileOut( "$fname-sleep $x" );
 219                                         }
 220                                 } while ( count( $loads ) && !$done && $totalElapsed / 1e6 < $wgDBClusterTimeout );
 221
 222                                 if ( $i !== false && $this->isOpen( $i ) ) {
 223                                         # Wait for the session master pos for a short time
 224                                         if ( $this->mWaitForFile ) {
 225                                                 if ( !$this->doWait( $i ) ) {
 226                                                         $this->mServers[$i]['slave pos'] = $this->mConnections[$i]->getSlavePos();
 227                                                 }
 228                                         }
 229                                         if ( $i !== false ) {
 230                                                 $this->mReadIndex = $i;
 231                                         }
 232                                 } else {
 233                                         $i = false;
 234                                 }
 235                         }
 236                 }
 237                 wfProfileOut( $fname );
 238                 return $i;
 239         }
 240
 241         /**
 242          * Get a random server to use in a query group
 243          */
 244         function getGroupIndex( $group ) {
 245                 if ( isset( $this->mGroupLoads[$group] ) ) {
 246                         $i = $this->pickRandom( $this->mGroupLoads[$group] );
 247                 } else {
 248                         $i = false;
 249                 }
 250                 wfDebug( "Query group $group => $i\n" );
 251                 return $i;
 252         }
 253
 254         /**
 255          * Set the master wait position
 256          * If a DB_SLAVE connection has been opened already, waits
 257          * Otherwise sets a variable telling it to wait if such a connection is opened
 258          */
 259         function waitFor( $file, $pos ) {
 260                 $fname = 'LoadBalancer::waitFor';
 261                 wfProfileIn( $fname );
 262
 263                 wfDebug( "User master pos: $file $pos\n" );
 264                 $this->mWaitForFile = false;
 265                 $this->mWaitForPos = false;
 266
 267                 if ( count( $this->mServers ) > 1 ) {
 268                         $this->mWaitForFile = $file;
 269                         $this->mWaitForPos = $pos;
 270                         $i = $this->mReadIndex;
 271
 272                         if ( $i > 0 ) {
 273                                 if ( !$this->doWait( $i ) ) {
 274                                         $this->mServers[$i]['slave pos'] = $this->mConnections[$i]->getSlavePos();
 275                                         $this->mLaggedSlaveMode = true;
 276                                 }
 277                         }
 278                 }
 279                 wfProfileOut( $fname );
 280         }
 281
 282         /**
 283          * Wait for a given slave to catch up to the master pos stored in $this
 284          */
 285         function doWait( $index ) {
 286                 global $wgMemc;
 287
 288                 $retVal = false;
 289
 290                 # Debugging hacks
 291                 if ( isset( $this->mServers[$index]['lagged slave'] ) ) {
 292                         return false;
 293                 } elseif ( isset( $this->mServers[$index]['fake slave'] ) ) {
 294                         return true;
 295                 }
 296
 297                 $key = 'masterpos:' . $index;
 298                 $memcPos = $wgMemc->get( $key );
 299                 if ( $memcPos ) {
 300                         list( $file, $pos ) = explode( ' ', $memcPos );
 301                         # If the saved position is later than the requested position, return now
 302                         if ( $file == $this->mWaitForFile && $this->mWaitForPos <= $pos ) {
 303                                 $retVal = true;
 304                         }
 305                 }
 306
 307                 if ( !$retVal && $this->isOpen( $index ) ) {
 308                         $conn =& $this->mConnections[$index];
 309                         wfDebug( "Waiting for slave #$index to catch up...\n" );
 310                         $result = $conn->masterPosWait( $this->mWaitForFile, $this->mWaitForPos, $this->mWaitTimeout );
 311
 312                         if ( $result == -1 || is_null( $result ) ) {
 313                                 # Timed out waiting for slave, use master instead
 314                                 wfDebug( "Timed out waiting for slave #$index pos {$this->mWaitForFile} {$this->mWaitForPos}\n" );
 315                                 $retVal = false;
 316                         } else {
 317                                 $retVal = true;
 318                                 wfDebug( "Done\n" );
 319                         }
 320                 }
 321                 return $retVal;
 322         }
 323
 324         /**
 325          * Get a connection by index
 326          */
 327         function &getConnection( $i, $fail = true, $groups = array() )
 328         {
 329                 $fname = 'LoadBalancer::getConnection';
 330                 wfProfileIn( $fname );
 331
 332                 # Query groups
 333                 $groupIndex = false;
 334                 foreach ( $groups as $group ) {
 335                         $groupIndex = $this->getGroupIndex( $group );
 336                         if ( $groupIndex !== false ) {
 337                                 $i = $groupIndex;
 338                                 break;
 339                         }
 340                 }
 341
 342                 # Operation-based index
 343                 if ( $i == DB_SLAVE ) {
 344                         $i = $this->getReaderIndex();
 345                 } elseif ( $i == DB_MASTER ) {
 346                         $i = $this->getWriterIndex();
 347                 } elseif ( $i == DB_LAST ) {
 348                         # Just use $this->mLastIndex, which should already be set
 349                         $i = $this->mLastIndex;
 350                         if ( $i === -1 ) {
 351                                 # Oh dear, not set, best to use the writer for safety
 352                                 wfDebug( "Warning: DB_LAST used when there was no previous index\n" );
 353                                 $i = $this->getWriterIndex();
 354                         }
 355                 }
 356                 # Couldn't find a working server in getReaderIndex()?
 357                 if ( $i === false ) {
 358                         $this->reportConnectionError( $this->mErrorConnection );
 359                 }
 360                 # Now we have an explicit index into the servers array
 361                 $this->openConnection( $i, $fail );
 362
 363                 wfProfileOut( $fname );
 364                 return $this->mConnections[$i];
 365         }
 366
 367         /**
 368          * Open a connection to the server given by the specified index
 369          * Index must be an actual index into the array
 370          * Returns success
 371          * @private
 372          */
 373         function openConnection( $i, $fail = false ) {
 374                 $fname = 'LoadBalancer::openConnection';
 375                 wfProfileIn( $fname );
 376                 $success = true;
 377
 378                 if ( !$this->isOpen( $i ) ) {
 379                         $this->mConnections[$i] = $this->reallyOpenConnection( $this->mServers[$i] );
 380                 }
 381
 382                 if ( !$this->isOpen( $i ) ) {
 383                         wfDebug( "Failed to connect to database $i at {$this->mServers[$i]['host']}\n" );
 384                         if ( $fail ) {
 385                                 $this->reportConnectionError( $this->mConnections[$i] );
 386                         }
 387                         $this->mErrorConnection = $this->mConnections[$i];
 388                         $this->mConnections[$i] = false;
 389                         $success = false;
 390                 }
 391                 $this->mLastIndex = $i;
 392                 wfProfileOut( $fname );
 393                 return $success;
 394         }
 395
 396         /**
 397          * Test if the specified index represents an open connection
 398          * @private
 399          */
 400         function isOpen( $index ) {
 401                 if( !is_integer( $index ) ) {
 402                         return false;
 403                 }
 404                 if ( array_key_exists( $index, $this->mConnections ) && is_object( $this->mConnections[$index] ) &&
 405                   $this->mConnections[$index]->isOpen() )
 406                 {
 407                         return true;
 408                 } else {
 409                         return false;
 410                 }
 411         }
 412
 413         /**
 414          * Really opens a connection
 415          * @private
 416          */
 417         function reallyOpenConnection( &$server ) {
 418                 if( !is_array( $server ) ) {
 419                         wfDebugDieBacktrace( 'You must update your load-balancing configuration. See DefaultSettings.php entry for $wgDBservers.' );
 420                 }
 421
 422                 extract( $server );
 423                 # Get class for this database type
 424                 $class = 'Database' . ucfirst( $type );
 425                 if ( !class_exists( $class ) ) {
 426                         require_once( "$class.php" );
 427                 }
 428
 429                 # Create object
 430                 return new $class( $host, $user, $password, $dbname, 1, $flags );
 431         }
 432
 433         function reportConnectionError( &$conn )
 434         {
 435                 $fname = 'LoadBalancer::reportConnectionError';
 436                 wfProfileIn( $fname );
 437                 # Prevent infinite recursion
 438
 439                 static $reporting = false;
 440                 if ( !$reporting ) {
 441                         $reporting = true;
 442                         if ( !is_object( $conn ) ) {
 443                                 $conn = new Database;
 444                         }
 445                         if ( $this->mFailFunction ) {
 446                                 $conn->failFunction( $this->mFailFunction );
 447                         } else {
 448                                 $conn->failFunction( false );
 449                         }
 450                         $conn->reportConnectionError();
 451                         $reporting = false;
 452                 }
 453                 wfProfileOut( $fname );
 454         }
 455
 456         function getWriterIndex()
 457         {
 458                 return 0;
 459         }
 460
 461         function force( $i )
 462         {
 463                 $this->mForce = $i;
 464         }
 465
 466         function haveIndex( $i )
 467         {
 468                 return array_key_exists( $i, $this->mServers );
 469         }
 470
 471         /**
 472          * Get the number of defined servers (not the number of open connections)
 473          */
 474         function getServerCount() {
 475                 return count( $this->mServers );
 476         }
 477
 478         /**
 479          * Save master pos to the session and to memcached, if the session exists
 480          */
 481         function saveMasterPos() {
 482                 global $wgSessionStarted;
 483                 if ( $wgSessionStarted && count( $this->mServers ) > 1 ) {
 484                         # If this entire request was served from a slave without opening a connection to the
 485                         # master (however unlikely that may be), then we can fetch the position from the slave.
 486                         if ( empty( $this->mConnections[0] ) ) {
 487                                 $conn =& $this->getConnection( DB_SLAVE );
 488                                 list( $file, $pos ) = $conn->getSlavePos();
 489                                 wfDebug( "Saving master pos fetched from slave: $file $pos\n" );
 490                         } else {
 491                                 $conn =& $this->getConnection( 0 );
 492                                 list( $file, $pos ) = $conn->getMasterPos();
 493                                 wfDebug( "Saving master pos: $file $pos\n" );
 494                         }
 495                         if ( $file !== false ) {
 496                                 $_SESSION['master_log_file'] = $file;
 497                                 $_SESSION['master_pos'] = $pos;
 498                         }
 499                 }
 500         }
 501
 502         /**
 503          * Loads the master pos from the session, waits for it if necessary
 504          */
 505         function loadMasterPos() {
 506                 if ( isset( $_SESSION['master_log_file'] ) && isset( $_SESSION['master_pos'] ) ) {
 507                         $this->waitFor( $_SESSION['master_log_file'], $_SESSION['master_pos'] );
 508                 }
 509         }
 510
 511         /**
 512          * Close all open connections
 513          */
 514         function closeAll() {
 515                 foreach( $this->mConnections as $i => $conn ) {
 516                         if ( $this->isOpen( $i ) ) {
 517                                 // Need to use this syntax because $conn is a copy not a reference
 518                                 $this->mConnections[$i]->close();
 519                         }
 520                 }
 521         }
 522
 523         function commitAll() {
 524                 foreach( $this->mConnections as $i => $conn ) {
 525                         if ( $this->isOpen( $i ) ) {
 526                                 // Need to use this syntax because $conn is a copy not a reference
 527                                 $this->mConnections[$i]->immediateCommit();
 528                         }
 529                 }
 530         }
 531
 532         function waitTimeout( $value = NULL ) {
 533                 return wfSetVar( $this->mWaitTimeout, $value );
 534         }
 535
 536         function getLaggedSlaveMode() {
 537                 return $this->mLaggedSlaveMode;
 538         }
 539
 540         function pingAll() {
 541                 $success = true;
 542                 foreach ( $this->mConnections as $i => $conn ) {
 543                         if ( $this->isOpen( $i ) ) {
 544                                 if ( !$this->mConnections[$i]->ping() ) {
 545                                         $success = false;
 546                                 }
 547                         }
 548                 }
 549                 return $success;
 550         }
 551
 552         /**
 553          * Get the hostname and lag time of the most-lagged slave
 554          * This is useful for maintenance scripts that need to throttle their updates
 555          */
 556         function getMaxLag() {
 557                 $maxLag = -1;
 558                 $host = '';
 559                 foreach ( $this->mServers as $i => $conn ) {
 560                         if ( $this->openConnection( $i ) ) {
 561                                 $lag = $this->mConnections[$i]->getLag();
 562                                 if ( $lag > $maxLag ) {
 563                                         $maxLag = $lag;
 564                                         $host = $this->mServers[$i]['host'];
 565                                 }
 566                         }
 567                 }
 568                 return array( $host, $maxLag );
 569         }
 570
 571         /**
 572          * Get lag time for each DB
 573          * Results are cached for a short time in memcached
 574          */
 575         function getLagTimes() {
 576                 $expiry = 5;
 577                 $requestRate = 10;
 578
 579                 global $wgMemc;
 580                 $times = $wgMemc->get( 'lag_times' );
 581                 if ( $times ) {
 582                         # Randomly recache with probability rising over $expiry
 583                         $elapsed = time() - $times['timestamp'];
 584                         $chance = max( 0, ( $expiry - $elapsed ) * $requestRate );
 585                         if ( mt_rand( 0, $chance ) != 0 ) {
 586                                 unset( $times['timestamp'] );
 587                                 return $times;
 588                         }
 589                 }
 590
 591                 # Cache key missing or expired
 592
 593                 $times = array();
 594                 foreach ( $this->mServers as $i => $conn ) {
 595                         if ( $this->openConnection( $i ) ) {
 596                                 $times[$i] = $this->mConnections[$i]->getLag();
 597                         }
 598                 }
 599
 600                 # Add a timestamp key so we know when it was cached
 601                 $times['timestamp'] = time();
 602                 $wgMemc->set( 'lag_times', $times, $expiry );
 603
 604                 # But don't give the timestamp to the caller
 605                 unset($times['timestamp']);
 606                 return $times;
 607         }
 608 }
 609
 610 ?>