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