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