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, $mAllowLagged;
  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                 $this->mAllowLag = false;
  51         }
  52
  53         static function newFromParams( $servers, $failFunction = false, $waitTimeout = 10 )
  54         {
  55                 $lb = new LoadBalancer;
  56                 $lb->initialise( $servers, $failFunction, $waitTimeout );
  57                 return $lb;
  58         }
  59
  60         function initialise( $servers, $failFunction = false, $waitTimeout = 10 )
  61         {
  62                 $this->mServers = $servers;
  63                 $this->mFailFunction = $failFunction;
  64                 $this->mReadIndex = -1;
  65                 $this->mWriteIndex = -1;
  66                 $this->mForce = -1;
  67                 $this->mConnections = array();
  68                 $this->mLastIndex = 1;
  69                 $this->mLoads = array();
  70                 $this->mWaitForFile = false;
  71                 $this->mWaitForPos = false;
  72                 $this->mWaitTimeout = $waitTimeout;
  73                 $this->mLaggedSlaveMode = false;
  74
  75                 foreach( $servers as $i => $server ) {
  76                         $this->mLoads[$i] = $server['load'];
  77                         if ( isset( $server['groupLoads'] ) ) {
  78                                 foreach ( $server['groupLoads'] as $group => $ratio ) {
  79                                         if ( !isset( $this->mGroupLoads[$group] ) ) {
  80                                                 $this->mGroupLoads[$group] = array();
  81                                         }
  82                                         $this->mGroupLoads[$group][$i] = $ratio;
  83                                 }
  84                         }
  85                 }
  86         }
  87
  88         /**
  89          * Given an array of non-normalised probabilities, this function will select
  90          * an element and return the appropriate key
  91          */
  92         function pickRandom( $weights )
  93         {
  94                 if ( !is_array( $weights ) || count( $weights ) == 0 ) {
  95                         return false;
  96                 }
  97
  98                 $sum = array_sum( $weights );
  99                 if ( $sum == 0 ) {
 100                         # No loads on any of them
 101                         # In previous versions, this triggered an unweighted random selection,
 102                         # but this feature has been removed as of April 2006 to allow for strict
 103                         # separation of query groups.
 104                         return false;
 105                 }
 106                 $max = mt_getrandmax();
 107                 $rand = mt_rand(0, $max) / $max * $sum;
 108
 109                 $sum = 0;
 110                 foreach ( $weights as $i => $w ) {
 111                         $sum += $w;
 112                         if ( $sum >= $rand ) {
 113                                 break;
 114                         }
 115                 }
 116                 return $i;
 117         }
 118
 119         function getRandomNonLagged( $loads ) {
 120                 # Unset excessively lagged servers
 121                 $lags = $this->getLagTimes();
 122                 foreach ( $lags as $i => $lag ) {
 123                         if ( isset( $this->mServers[$i]['max lag'] ) && $lag > $this->mServers[$i]['max lag'] ) {
 124                                 unset( $loads[$i] );
 125                         }
 126                 }
 127
 128                 # Find out if all the slaves with non-zero load are lagged
 129                 $sum = 0;
 130                 foreach ( $loads as $load ) {
 131                         $sum += $load;
 132                 }
 133                 if ( $sum == 0 ) {
 134                         # No appropriate DB servers except maybe the master and some slaves with zero load
 135                         # Do NOT use the master
 136                         # Instead, this function will return false, triggering read-only mode,
 137                         # and a lagged slave will be used instead.
 138                         return false;
 139                 }
 140
 141                 if ( count( $loads ) == 0 ) {
 142                         return false;
 143                 }
 144
 145                 #wfDebugLog( 'connect', var_export( $loads, true ) );
 146
 147                 # Return a random representative of the remainder
 148                 return $this->pickRandom( $loads );
 149         }
 150
 151         /**
 152          * Get the index of the reader connection, which may be a slave
 153          * This takes into account load ratios and lag times. It should
 154          * always return a consistent index during a given invocation
 155          *
 156          * Side effect: opens connections to databases
 157          */
 158         function getReaderIndex() {
 159                 global $wgReadOnly, $wgDBClusterTimeout;
 160
 161                 $fname = 'LoadBalancer::getReaderIndex';
 162                 wfProfileIn( $fname );
 163
 164                 $i = false;
 165                 if ( $this->mForce >= 0 ) {
 166                         $i = $this->mForce;
 167                 } else {
 168                         if ( $this->mReadIndex >= 0 ) {
 169                                 $i = $this->mReadIndex;
 170                         } else {
 171                                 # $loads is $this->mLoads except with elements knocked out if they
 172                                 # don't work
 173                                 $loads = $this->mLoads;
 174                                 $done = false;
 175                                 $totalElapsed = 0;
 176                                 do {
 177                                         if ( $wgReadOnly or $this->mAllowLagged ) {
 178                                                 $i = $this->pickRandom( $loads );
 179                                         } else {
 180                                                 $i = $this->getRandomNonLagged( $loads );
 181                                                 if ( $i === false && count( $loads ) != 0 )  {
 182                                                         # All slaves lagged. Switch to read-only mode
 183                                                         $wgReadOnly = wfMsgNoDB( 'readonly_lag' );
 184                                                         $i = $this->pickRandom( $loads );
 185                                                 }
 186                                         }
 187                                         $serverIndex = $i;
 188                                         if ( $i !== false ) {
 189                                                 wfDebugLog( 'connect', "$fname: Using reader #$i: {$this->mServers[$i]['host']}...\n" );
 190                                                 $this->openConnection( $i );
 191
 192                                                 if ( !$this->isOpen( $i ) ) {
 193                                                         wfDebug( "$fname: Failed\n" );
 194                                                         unset( $loads[$i] );
 195                                                         $sleepTime = 0;
 196                                                 } else {
 197                                                         $status = $this->mConnections[$i]->getStatus("Thread%");
 198                                                         if ( isset( $this->mServers[$i]['max threads'] ) &&
 199                                                           $status['Threads_running'] > $this->mServers[$i]['max threads'] )
 200                                                         {
 201                                                                 # Too much load, back off and wait for a while.
 202                                                                 # The sleep time is scaled by the number of threads connected,
 203                                                                 # to produce a roughly constant global poll rate.
 204                                                                 $sleepTime = AVG_STATUS_POLL * $status['Threads_connected'];
 205
 206                                                                 # If we reach the timeout and exit the loop, don't use it
 207                                                                 $i = false;
 208                                                         } else {
 209                                                                 $done = true;
 210                                                                 $sleepTime = 0;
 211                                                         }
 212                                                 }
 213                                         } else {
 214                                                 $sleepTime = 500000;
 215                                         }
 216                                         if ( $sleepTime ) {
 217                                                         $totalElapsed += $sleepTime;
 218                                                         $x = "{$this->mServers[$serverIndex]['host']} [$serverIndex]";
 219                                                         wfProfileIn( "$fname-sleep $x" );
 220                                                         usleep( $sleepTime );
 221                                                         wfProfileOut( "$fname-sleep $x" );
 222                                         }
 223                                 } while ( count( $loads ) && !$done && $totalElapsed / 1e6 < $wgDBClusterTimeout );
 224
 225                                 if ( $totalElapsed / 1e6 >= $wgDBClusterTimeout ) {
 226                                         $this->mErrorConnection = false;
 227                                         $this->mLastError = 'All servers busy';
 228                                 }
 229
 230                                 if ( $i !== false && $this->isOpen( $i ) ) {
 231                                         # Wait for the session master pos for a short time
 232                                         if ( $this->mWaitForFile ) {
 233                                                 if ( !$this->doWait( $i ) ) {
 234                                                         $this->mServers[$i]['slave pos'] = $this->mConnections[$i]->getSlavePos();
 235                                                 }
 236                                         }
 237                                         if ( $i !== false ) {
 238                                                 $this->mReadIndex = $i;
 239                                         }
 240                                 } else {
 241                                         $i = false;
 242                                 }
 243                         }
 244                 }
 245                 wfProfileOut( $fname );
 246                 return $i;
 247         }
 248
 249         /**
 250          * Get a random server to use in a query group
 251          */
 252         function getGroupIndex( $group ) {
 253                 if ( isset( $this->mGroupLoads[$group] ) ) {
 254                         $i = $this->pickRandom( $this->mGroupLoads[$group] );
 255                 } else {
 256                         $i = false;
 257                 }
 258                 wfDebug( "Query group $group => $i\n" );
 259                 return $i;
 260         }
 261
 262         /**
 263          * Set the master wait position
 264          * If a DB_SLAVE connection has been opened already, waits
 265          * Otherwise sets a variable telling it to wait if such a connection is opened
 266          */
 267         function waitFor( $file, $pos ) {
 268                 $fname = 'LoadBalancer::waitFor';
 269                 wfProfileIn( $fname );
 270
 271                 wfDebug( "User master pos: $file $pos\n" );
 272                 $this->mWaitForFile = false;
 273                 $this->mWaitForPos = false;
 274
 275                 if ( count( $this->mServers ) > 1 ) {
 276                         $this->mWaitForFile = $file;
 277                         $this->mWaitForPos = $pos;
 278                         $i = $this->mReadIndex;
 279
 280                         if ( $i > 0 ) {
 281                                 if ( !$this->doWait( $i ) ) {
 282                                         $this->mServers[$i]['slave pos'] = $this->mConnections[$i]->getSlavePos();
 283                                         $this->mLaggedSlaveMode = true;
 284                                 }
 285                         }
 286                 }
 287                 wfProfileOut( $fname );
 288         }
 289
 290         /**
 291          * Wait for a given slave to catch up to the master pos stored in $this
 292          */
 293         function doWait( $index ) {
 294                 global $wgMemc;
 295
 296                 $retVal = false;
 297
 298                 # Debugging hacks
 299                 if ( isset( $this->mServers[$index]['lagged slave'] ) ) {
 300                         return false;
 301                 } elseif ( isset( $this->mServers[$index]['fake slave'] ) ) {
 302                         return true;
 303                 }
 304
 305                 $key = 'masterpos:' . $index;
 306                 $memcPos = $wgMemc->get( $key );
 307                 if ( $memcPos ) {
 308                         list( $file, $pos ) = explode( ' ', $memcPos );
 309                         # If the saved position is later than the requested position, return now
 310                         if ( $file == $this->mWaitForFile && $this->mWaitForPos <= $pos ) {
 311                                 $retVal = true;
 312                         }
 313                 }
 314
 315                 if ( !$retVal && $this->isOpen( $index ) ) {
 316                         $conn =& $this->mConnections[$index];
 317                         wfDebug( "Waiting for slave #$index to catch up...\n" );
 318                         $result = $conn->masterPosWait( $this->mWaitForFile, $this->mWaitForPos, $this->mWaitTimeout );
 319
 320                         if ( $result == -1 || is_null( $result ) ) {
 321                                 # Timed out waiting for slave, use master instead
 322                                 wfDebug( "Timed out waiting for slave #$index pos {$this->mWaitForFile} {$this->mWaitForPos}\n" );
 323                                 $retVal = false;
 324                         } else {
 325                                 $retVal = true;
 326                                 wfDebug( "Done\n" );
 327                         }
 328                 }
 329                 return $retVal;
 330         }
 331
 332         /**
 333          * Get a connection by index
 334          */
 335         function &getConnection( $i, $fail = true, $groups = array() )
 336         {
 337                 global $wgDBtype;
 338                 $fname = 'LoadBalancer::getConnection';
 339                 wfProfileIn( $fname );
 340
 341
 342                 # Query groups
 343                 if ( !is_array( $groups ) ) {
 344                         $groupIndex = $this->getGroupIndex( $groups, $i );
 345                         if ( $groupIndex !== false ) {
 346                                 $i = $groupIndex;
 347                         }
 348                 } else {
 349                         foreach ( $groups as $group ) {
 350                                 $groupIndex = $this->getGroupIndex( $group, $i );
 351                                 if ( $groupIndex !== false ) {
 352                                         $i = $groupIndex;
 353                                         break;
 354                                 }
 355                         }
 356                 }
 357
 358                 # For now, only go through all this for mysql databases
 359                 if ($wgDBtype != 'mysql') {
 360                         $i = $this->getWriterIndex();
 361                 }
 362                 # Operation-based index
 363                 elseif ( $i == DB_SLAVE ) {
 364                         $i = $this->getReaderIndex();
 365                 } elseif ( $i == DB_MASTER ) {
 366                         $i = $this->getWriterIndex();
 367                 } elseif ( $i == DB_LAST ) {
 368                         # Just use $this->mLastIndex, which should already be set
 369                         $i = $this->mLastIndex;
 370                         if ( $i === -1 ) {
 371                                 # Oh dear, not set, best to use the writer for safety
 372                                 wfDebug( "Warning: DB_LAST used when there was no previous index\n" );
 373                                 $i = $this->getWriterIndex();
 374                         }
 375                 }
 376                 # Couldn't find a working server in getReaderIndex()?
 377                 if ( $i === false ) {
 378                         $this->reportConnectionError( $this->mErrorConnection );
 379                 }
 380                 # Now we have an explicit index into the servers array
 381                 $this->openConnection( $i, $fail );
 382
 383                 wfProfileOut( $fname );
 384                 return $this->mConnections[$i];
 385         }
 386
 387         /**
 388          * Open a connection to the server given by the specified index
 389          * Index must be an actual index into the array
 390          * Returns success
 391          * @access private
 392          */
 393         function openConnection( $i, $fail = false ) {
 394                 $fname = 'LoadBalancer::openConnection';
 395                 wfProfileIn( $fname );
 396                 $success = true;
 397
 398                 if ( !$this->isOpen( $i ) ) {
 399                         $this->mConnections[$i] = $this->reallyOpenConnection( $this->mServers[$i] );
 400                 }
 401
 402                 if ( !$this->isOpen( $i ) ) {
 403                         wfDebug( "Failed to connect to database $i at {$this->mServers[$i]['host']}\n" );
 404                         if ( $fail ) {
 405                                 $this->reportConnectionError( $this->mConnections[$i] );
 406                         }
 407                         $this->mErrorConnection = $this->mConnections[$i];
 408                         $this->mConnections[$i] = false;
 409                         $success = false;
 410                 }
 411                 $this->mLastIndex = $i;
 412                 wfProfileOut( $fname );
 413                 return $success;
 414         }
 415
 416         /**
 417          * Test if the specified index represents an open connection
 418          * @access private
 419          */
 420         function isOpen( $index ) {
 421                 if( !is_integer( $index ) ) {
 422                         return false;
 423                 }
 424                 if ( array_key_exists( $index, $this->mConnections ) && is_object( $this->mConnections[$index] ) &&
 425                   $this->mConnections[$index]->isOpen() )
 426                 {
 427                         return true;
 428                 } else {
 429                         return false;
 430                 }
 431         }
 432
 433         /**
 434          * Really opens a connection
 435          * @access private
 436          */
 437         function reallyOpenConnection( &$server ) {
 438                 if( !is_array( $server ) ) {
 439                         throw new MWException( 'You must update your load-balancing configuration. See DefaultSettings.php entry for $wgDBservers.' );
 440                 }
 441
 442                 extract( $server );
 443                 # Get class for this database type
 444                 $class = 'Database' . ucfirst( $type );
 445                 if ( !class_exists( $class ) ) {
 446                         require_once( "$class.php" );
 447                 }
 448
 449                 # Create object
 450                 $db = new $class( $host, $user, $password, $dbname, 1, $flags );
 451                 $db->setLBInfo( $server );
 452                 return $db;
 453         }
 454
 455         function reportConnectionError( &$conn )
 456         {
 457                 $fname = 'LoadBalancer::reportConnectionError';
 458                 wfProfileIn( $fname );
 459                 # Prevent infinite recursion
 460
 461                 static $reporting = false;
 462                 if ( !$reporting ) {
 463                         $reporting = true;
 464                         if ( !is_object( $conn ) ) {
 465                                 // No last connection, probably due to all servers being too busy
 466                                 $conn = new Database;
 467                                 if ( $this->mFailFunction ) {
 468                                         $conn->failFunction( $this->mFailFunction );
 469                                         $conn->reportConnectionError( $this->mLastError );
 470                                 } else {
 471                                         // If all servers were busy, mLastError will contain something sensible
 472                                         throw new DBConnectionError( $conn, $this->mLastError );
 473                                 }
 474                         } else {
 475                                 if ( $this->mFailFunction ) {
 476                                         $conn->failFunction( $this->mFailFunction );
 477                                 } else {
 478                                         $conn->failFunction( false );
 479                                 }
 480                                 $server = $conn->getProperty( 'mServer' );
 481                                 $conn->reportConnectionError( "{$this->mLastError} ({$server})" );
 482                         }
 483                         $reporting = false;
 484                 }
 485                 wfProfileOut( $fname );
 486         }
 487
 488         function getWriterIndex() {
 489                 return 0;
 490         }
 491
 492         /**
 493          * Force subsequent calls to getConnection(DB_SLAVE) to return the
 494          * given index. Set to -1 to restore the original load balancing
 495          * behaviour. I thought this was a good idea when I originally
 496          * wrote this class, but it has never been used.
 497          */
 498         function force( $i ) {
 499                 $this->mForce = $i;
 500         }
 501
 502         /**
 503          * Returns true if the specified index is a valid server index
 504          */
 505         function haveIndex( $i ) {
 506                 return array_key_exists( $i, $this->mServers );
 507         }
 508
 509         /**
 510          * Returns true if the specified index is valid and has non-zero load
 511          */
 512         function isNonZeroLoad( $i ) {
 513                 return array_key_exists( $i, $this->mServers ) && $this->mLoads[$i] != 0;
 514         }
 515
 516         /**
 517          * Get the number of defined servers (not the number of open connections)
 518          */
 519         function getServerCount() {
 520                 return count( $this->mServers );
 521         }
 522
 523         /**
 524          * Save master pos to the session and to memcached, if the session exists
 525          */
 526         function saveMasterPos() {
 527                 global $wgSessionStarted;
 528                 if ( $wgSessionStarted && count( $this->mServers ) > 1 ) {
 529                         # If this entire request was served from a slave without opening a connection to the
 530                         # master (however unlikely that may be), then we can fetch the position from the slave.
 531                         if ( empty( $this->mConnections[0] ) ) {
 532                                 $conn =& $this->getConnection( DB_SLAVE );
 533                                 list( $file, $pos ) = $conn->getSlavePos();
 534                                 wfDebug( "Saving master pos fetched from slave: $file $pos\n" );
 535                         } else {
 536                                 $conn =& $this->getConnection( 0 );
 537                                 list( $file, $pos ) = $conn->getMasterPos();
 538                                 wfDebug( "Saving master pos: $file $pos\n" );
 539                         }
 540                         if ( $file !== false ) {
 541                                 $_SESSION['master_log_file'] = $file;
 542                                 $_SESSION['master_pos'] = $pos;
 543                         }
 544                 }
 545         }
 546
 547         /**
 548          * Loads the master pos from the session, waits for it if necessary
 549          */
 550         function loadMasterPos() {
 551                 if ( isset( $_SESSION['master_log_file'] ) && isset( $_SESSION['master_pos'] ) ) {
 552                         $this->waitFor( $_SESSION['master_log_file'], $_SESSION['master_pos'] );
 553                 }
 554         }
 555
 556         /**
 557          * Close all open connections
 558          */
 559         function closeAll() {
 560                 foreach( $this->mConnections as $i => $conn ) {
 561                         if ( $this->isOpen( $i ) ) {
 562                                 // Need to use this syntax because $conn is a copy not a reference
 563                                 $this->mConnections[$i]->close();
 564                         }
 565                 }
 566         }
 567
 568         function commitAll() {
 569                 foreach( $this->mConnections as $i => $conn ) {
 570                         if ( $this->isOpen( $i ) ) {
 571                                 // Need to use this syntax because $conn is a copy not a reference
 572                                 $this->mConnections[$i]->immediateCommit();
 573                         }
 574                 }
 575         }
 576
 577         function waitTimeout( $value = NULL ) {
 578                 return wfSetVar( $this->mWaitTimeout, $value );
 579         }
 580
 581         function getLaggedSlaveMode() {
 582                 return $this->mLaggedSlaveMode;
 583         }
 584
 585         /* Disables/enables lag checks */
 586         function allowLagged($mode=null) {
 587                 if ($mode===null)
 588                         return $this->mAllowLagged;
 589                 $this->mAllowLagged=$mode;
 590         }
 591
 592         function pingAll() {
 593                 $success = true;
 594                 foreach ( $this->mConnections as $i => $conn ) {
 595                         if ( $this->isOpen( $i ) ) {
 596                                 if ( !$this->mConnections[$i]->ping() ) {
 597                                         $success = false;
 598                                 }
 599                         }
 600                 }
 601                 return $success;
 602         }
 603
 604         /**
 605          * Get the hostname and lag time of the most-lagged slave
 606          * This is useful for maintenance scripts that need to throttle their updates
 607          */
 608         function getMaxLag() {
 609                 $maxLag = -1;
 610                 $host = '';
 611                 foreach ( $this->mServers as $i => $conn ) {
 612                         if ( $this->openConnection( $i ) ) {
 613                                 $lag = $this->mConnections[$i]->getLag();
 614                                 if ( $lag > $maxLag ) {
 615                                         $maxLag = $lag;
 616                                         $host = $this->mServers[$i]['host'];
 617                                 }
 618                         }
 619                 }
 620                 return array( $host, $maxLag );
 621         }
 622
 623         /**
 624          * Get lag time for each DB
 625          * Results are cached for a short time in memcached
 626          */
 627         function getLagTimes() {
 628                 global $wgDBname;
 629
 630                 wfProfileIn( __METHOD__ );
 631                 $expiry = 5;
 632                 $requestRate = 10;
 633
 634                 global $wgMemc;
 635                 $times = $wgMemc->get( "$wgDBname:lag_times" );
 636                 if ( $times ) {
 637                         # Randomly recache with probability rising over $expiry
 638                         $elapsed = time() - $times['timestamp'];
 639                         $chance = max( 0, ( $expiry - $elapsed ) * $requestRate );
 640                         if ( mt_rand( 0, $chance ) != 0 ) {
 641                                 unset( $times['timestamp'] );
 642                                 wfProfileOut( __METHOD__ );
 643                                 return $times;
 644                         }
 645                 }
 646
 647                 # Cache key missing or expired
 648
 649                 $times = array();
 650                 foreach ( $this->mServers as $i => $conn ) {
 651                         if ($i==0) { # Master
 652                                 $times[$i] = 0;
 653                         } elseif ( $this->openConnection( $i ) ) {
 654                                 $times[$i] = $this->mConnections[$i]->getLag();
 655                         }
 656                 }
 657
 658                 # Add a timestamp key so we know when it was cached
 659                 $times['timestamp'] = time();
 660                 $wgMemc->set( "$wgDBname:lag_times", $times, $expiry );
 661
 662                 # But don't give the timestamp to the caller
 663                 unset($times['timestamp']);
 664                 wfIncrStats( 'lag_cache_miss' );
 665                 wfProfileOut( __METHOD__ );
 666                 return $times;
 667         }
 668 }
 669
 670 ?>