includes/LoadBalancer.php

   1 <?php
   2 /**
   3  *
   4  * @package MediaWiki
   5  */
   6
   7 /**
   8  * Depends on the database object
   9  */
  10
  11 # Valid database indexes
  12 # Operation-based indexes
  13 define( 'DB_SLAVE', -1 );     # Read from the slave (or only server)
  14 define( 'DB_MASTER', -2 );    # Write to master (or only server)
  15 define( 'DB_LAST', -3 );     # Whatever database was used last
  16
  17 # Obsolete aliases
  18 define( 'DB_READ', -1 );
  19 define( 'DB_WRITE', -2 );
  20
  21
  22 # Scale polling time so that under overload conditions, the database server
  23 # receives a SHOW STATUS query at an average interval of this many microseconds
  24 define( 'AVG_STATUS_POLL', 2000 );
  25
  26
  27 /**
  28  * Database load balancing object
  29  *
  30  * @todo document
  31  * @package MediaWiki
  32  */
  33 class LoadBalancer {
  34         /* private */ var $mServers, $mConnections, $mLoads, $mGroupLoads;
  35         /* private */ var $mFailFunction, $mErrorConnection;
  36         /* private */ var $mForce, $mReadIndex, $mLastIndex, $mAllowLagged;
  37         /* private */ var $mWaitForFile, $mWaitForPos, $mWaitTimeout;
  38         /* private */ var $mLaggedSlaveMode, $mLastError = 'Unknown error';
  39
  40         function LoadBalancer()
  41         {
  42                 $this->mServers = array();
  43                 $this->mConnections = array();
  44                 $this->mFailFunction = false;
  45                 $this->mReadIndex = -1;
  46                 $this->mForce = -1;
  47                 $this->mLastIndex = -1;
  48                 $this->mErrorConnection = false;
  49                 $this->mAllowLag = false;
  50         }
  51
  52         function newFromParams( $servers, $failFunction = false, $waitTimeout = 10 )
  53         {
  54                 $lb = new LoadBalancer;
  55                 $lb->initialise( $servers, $failFunction, $waitTimeout );
  56                 return $lb;
  57         }
  58
  59         function initialise( $servers, $failFunction = false, $waitTimeout = 10 )
  60         {
  61                 $this->mServers = $servers;
  62                 $this->mFailFunction = $failFunction;
  63                 $this->mReadIndex = -1;
  64                 $this->mWriteIndex = -1;
  65                 $this->mForce = -1;
  66                 $this->mConnections = array();
  67                 $this->mLastIndex = 1;
  68                 $this->mLoads = array();
  69                 $this->mWaitForFile = false;
  70                 $this->mWaitForPos = false;
  71                 $this->mWaitTimeout = $waitTimeout;
  72                 $this->mLaggedSlaveMode = false;
  73
  74                 foreach( $servers as $i => $server ) {
  75                         $this->mLoads[$i] = $server['load'];
  76                         if ( isset( $server['groupLoads'] ) ) {
  77                                 foreach ( $server['groupLoads'] as $group => $ratio ) {
  78                                         if ( !isset( $this->mGroupLoads[$group] ) ) {
  79                                                 $this->mGroupLoads[$group] = array();
  80                                         }
  81                                         $this->mGroupLoads[$group][$i] = $ratio;
  82                                 }
  83                         }
  84                 }
  85         }
  86
  87         /**
  88          * Given an array of non-normalised probabilities, this function will select
  89          * an element and return the appropriate key
  90          */
  91         function pickRandom( $weights )
  92         {
  93                 if ( !is_array( $weights ) || count( $weights ) == 0 ) {
  94                         return false;
  95                 }
  96
  97                 $sum = array_sum( $weights );
  98                 if ( $sum == 0 ) {
  99                         # No loads on any of them
 100                         # In previous versions, this triggered an unweighted random selection,
 101                         # but this feature has been removed as of April 2006 to allow for strict
 102                         # separation of query groups.
 103                         return false;
 104                 }
 105                 $max = mt_getrandmax();
 106                 $rand = mt_rand(0, $max) / $max * $sum;
 107
 108                 $sum = 0;
 109                 foreach ( $weights as $i => $w ) {
 110                         $sum += $w;
 111                         if ( $sum >= $rand ) {
 112                                 break;
 113                         }
 114                 }
 115                 return $i;
 116         }
 117
 118         function getRandomNonLagged( $loads ) {
 119                 # Unset excessively lagged servers
 120                 $lags = $this->getLagTimes();
 121                 foreach ( $lags as $i => $lag ) {
 122                         if ( isset( $this->mServers[$i]['max lag'] ) && $lag > $this->mServers[$i]['max lag'] ) {
 123                                 unset( $loads[$i] );
 124                         }
 125                 }
 126
 127                 # Find out if all the slaves with non-zero load are lagged
 128                 $sum = 0;
 129                 foreach ( $loads as $load ) {
 130                         $sum += $load;
 131                 }
 132                 if ( $sum == 0 ) {
 133                         # No appropriate DB servers except maybe the master and some slaves with zero load
 134                         # Do NOT use the master
 135                         # Instead, this function will return false, triggering read-only mode,
 136                         # and a lagged slave will be used instead.
 137                         return false;
 138                 }
 139
 140                 if ( count( $loads ) == 0 ) {
 141                         return false;
 142                 }
 143
 144                 #wfDebugLog( 'connect', var_export( $loads, true ) );
 145
 146                 # Return a random representative of the remainder
 147                 return $this->pickRandom( $loads );
 148         }
 149
 150         /**
 151          * Get the index of the reader connection, which may be a slave
 152          * This takes into account load ratios and lag times. It should
 153          * always return a consistent index during a given invocation
 154          *
 155          * Side effect: opens connections to databases
 156          */
 157         function getReaderIndex() {
 158                 global $wgReadOnly, $wgDBClusterTimeout;
 159
 160                 $fname = 'LoadBalancer::getReaderIndex';
 161                 wfProfileIn( $fname );
 162
 163                 $i = false;
 164                 if ( $this->mForce >= 0 ) {
 165                         $i = $this->mForce;
 166                 } else {
 167                         if ( $this->mReadIndex >= 0 ) {
 168                                 $i = $this->mReadIndex;
 169                         } else {
 170                                 # $loads is $this->mLoads except with elements knocked out if they
 171                                 # don't work
 172                                 $loads = $this->mLoads;
 173                                 $done = false;
 174                                 $totalElapsed = 0;
 175                                 do {
 176                                         if ( $wgReadOnly or $this->mAllowLagged ) {
 177                                                 $i = $this->pickRandom( $loads );
 178                                         } else {
 179                                                 $i = $this->getRandomNonLagged( $loads );
 180                                                 if ( $i === false && count( $loads ) != 0 )  {
 181                                                         # All slaves lagged. Switch to read-only mode
 182                                                         $wgReadOnly = wfMsgNoDB( 'readonly_lag' );
 183                                                         $i = $this->pickRandom( $loads );
 184                                                 }
 185                                         }
 186                                         $serverIndex = $i;
 187                                         if ( $i !== false ) {
 188                                                 wfDebugLog( 'connect', "$fname: Using reader #$i: {$this->mServers[$i]['host']}...\n" );
 189                                                 $this->openConnection( $i );
 190
 191                                                 if ( !$this->isOpen( $i ) ) {
 192                                                         wfDebug( "$fname: Failed\n" );
 193                                                         unset( $loads[$i] );
 194                                                         $sleepTime = 0;
 195                                                 } else {
 196                                                         $status = $this->mConnections[$i]->getStatus("Thread%");
 197                                                         if ( isset( $this->mServers[$i]['max threads'] ) &&
 198                                                           $status['Threads_running'] > $this->mServers[$i]['max threads'] )
 199                                                         {
 200                                                                 # Too much load, back off and wait for a while.
 201                                                                 # The sleep time is scaled by the number of threads connected,
 202                                                                 # to produce a roughly constant global poll rate.
 203                                                                 $sleepTime = AVG_STATUS_POLL * $status['Threads_connected'];
 204
 205                                                                 # If we reach the timeout and exit the loop, don't use it
 206                                                                 $i = false;
 207                                                         } else {
 208                                                                 $done = true;
 209                                                                 $sleepTime = 0;
 210                                                         }
 211                                                 }
 212                                         } else {
 213                                                 $sleepTime = 500000;
 214                                         }
 215                                         if ( $sleepTime ) {
 216                                                         $totalElapsed += $sleepTime;
 217                                                         $x = "{$this->mServers[$serverIndex]['host']} [$serverIndex]";
 218                                                         wfProfileIn( "$fname-sleep $x" );
 219                                                         usleep( $sleepTime );
 220                                                         wfProfileOut( "$fname-sleep $x" );
 221                                         }
 222                                 } while ( count( $loads ) && !$done && $totalElapsed / 1e6 < $wgDBClusterTimeout );
 223
 224                                 if ( $totalElapsed / 1e6 >= $wgDBClusterTimeout ) {
 225                                         $this->mErrorConnection = false;
 226                                         $this->mLastError = 'All servers busy';
 227                                 }
 228
 229                                 if ( $i !== false && $this->isOpen( $i ) ) {
 230                                         # Wait for the session master pos for a short time
 231                                         if ( $this->mWaitForFile ) {
 232                                                 if ( !$this->doWait( $i ) ) {
 233                                                         $this->mServers[$i]['slave pos'] = $this->mConnections[$i]->getSlavePos();
 234                                                 }
 235                                         }
 236                                         if ( $i !== false ) {
 237                                                 $this->mReadIndex = $i;
 238                                         }
 239                                 } else {
 240                                         $i = false;
 241                                 }
 242                         }
 243                 }
 244                 wfProfileOut( $fname );
 245                 return $i;
 246         }
 247
 248         /**
 249          * Get a random server to use in a query group
 250          */
 251         function getGroupIndex( $group ) {
 252                 if ( isset( $this->mGroupLoads[$group] ) ) {
 253                         $i = $this->pickRandom( $this->mGroupLoads[$group] );
 254                 } else {
 255                         $i = false;
 256                 }
 257                 wfDebug( "Query group $group => $i\n" );
 258                 return $i;
 259         }
 260
 261         /**
 262          * Set the master wait position
 263          * If a DB_SLAVE connection has been opened already, waits
 264          * Otherwise sets a variable telling it to wait if such a connection is opened
 265          */
 266         function waitFor( $file, $pos ) {
 267                 $fname = 'LoadBalancer::waitFor';
 268                 wfProfileIn( $fname );
 269
 270                 wfDebug( "User master pos: $file $pos\n" );
 271                 $this->mWaitForFile = false;
 272                 $this->mWaitForPos = false;
 273
 274                 if ( count( $this->mServers ) > 1 ) {
 275                         $this->mWaitForFile = $file;
 276                         $this->mWaitForPos = $pos;
 277                         $i = $this->mReadIndex;
 278
 279                         if ( $i > 0 ) {
 280                                 if ( !$this->doWait( $i ) ) {
 281                                         $this->mServers[$i]['slave pos'] = $this->mConnections[$i]->getSlavePos();
 282                                         $this->mLaggedSlaveMode = true;
 283                                 }
 284                         }
 285                 }
 286                 wfProfileOut( $fname );
 287         }
 288
 289         /**
 290          * Wait for a given slave to catch up to the master pos stored in $this
 291          */
 292         function doWait( $index ) {
 293                 global $wgMemc;
 294
 295                 $retVal = false;
 296
 297                 # Debugging hacks
 298                 if ( isset( $this->mServers[$index]['lagged slave'] ) ) {
 299                         return false;
 300                 } elseif ( isset( $this->mServers[$index]['fake slave'] ) ) {
 301                         return true;
 302                 }
 303
 304                 $key = 'masterpos:' . $index;
 305                 $memcPos = $wgMemc->get( $key );
 306                 if ( $memcPos ) {
 307                         list( $file, $pos ) = explode( ' ', $memcPos );
 308                         # If the saved position is later than the requested position, return now
 309                         if ( $file == $this->mWaitForFile && $this->mWaitForPos <= $pos ) {
 310                                 $retVal = true;
 311                         }
 312                 }
 313
 314                 if ( !$retVal && $this->isOpen( $index ) ) {
 315                         $conn =& $this->mConnections[$index];
 316                         wfDebug( "Waiting for slave #$index to catch up...\n" );
 317                         $result = $conn->masterPosWait( $this->mWaitForFile, $this->mWaitForPos, $this->mWaitTimeout );
 318
 319                         if ( $result == -1 || is_null( $result ) ) {
 320                                 # Timed out waiting for slave, use master instead
 321                                 wfDebug( "Timed out waiting for slave #$index pos {$this->mWaitForFile} {$this->mWaitForPos}\n" );
 322                                 $retVal = false;
 323                         } else {
 324                                 $retVal = true;
 325                                 wfDebug( "Done\n" );
 326                         }
 327                 }
 328                 return $retVal;
 329         }
 330
 331         /**
 332          * Get a connection by index
 333          */
 334         function &getConnection( $i, $fail = true, $groups = array() )
 335         {
 336                 global $wgDBtype;
 337                 $fname = 'LoadBalancer::getConnection';
 338                 wfProfileIn( $fname );
 339
 340
 341                 # Query groups
 342                 if ( !is_array( $groups ) ) {
 343                         $groupIndex = $this->getGroupIndex( $groups, $i );
 344                         if ( $groupIndex !== false ) {
 345                                 $i = $groupIndex;
 346                         }
 347                 } else {
 348                         foreach ( $groups as $group ) {
 349                                 $groupIndex = $this->getGroupIndex( $group, $i );
 350                                 if ( $groupIndex !== false ) {
 351                                         $i = $groupIndex;
 352                                         break;
 353                                 }
 354                         }
 355                 }
 356
 357                 # For now, only go through all this for mysql databases
 358                 if ($wgDBtype != 'mysql') {
 359                         $i = $this->getWriterIndex();
 360                 }
 361                 # Operation-based index
 362                 elseif ( $i == DB_SLAVE ) {
 363                         $i = $this->getReaderIndex();
 364                 } elseif ( $i == DB_MASTER ) {
 365                         $i = $this->getWriterIndex();
 366                 } elseif ( $i == DB_LAST ) {
 367                         # Just use $this->mLastIndex, which should already be set
 368                         $i = $this->mLastIndex;
 369                         if ( $i === -1 ) {
 370                                 # Oh dear, not set, best to use the writer for safety
 371                                 wfDebug( "Warning: DB_LAST used when there was no previous index\n" );
 372                                 $i = $this->getWriterIndex();
 373                         }
 374                 }
 375                 # Couldn't find a working server in getReaderIndex()?
 376                 if ( $i === false ) {
 377                         $this->reportConnectionError( $this->mErrorConnection );
 378                 }
 379                 # Now we have an explicit index into the servers array
 380                 $this->openConnection( $i, $fail );
 381
 382                 wfProfileOut( $fname );
 383                 return $this->mConnections[$i];
 384         }
 385
 386         /**
 387          * Open a connection to the server given by the specified index
 388          * Index must be an actual index into the array
 389          * Returns success
 390          * @access private
 391          */
 392         function openConnection( $i, $fail = false ) {
 393                 $fname = 'LoadBalancer::openConnection';
 394                 wfProfileIn( $fname );
 395                 $success = true;
 396
 397                 if ( !$this->isOpen( $i ) ) {
 398                         $this->mConnections[$i] = $this->reallyOpenConnection( $this->mServers[$i] );
 399                 }
 400
 401                 if ( !$this->isOpen( $i ) ) {
 402                         wfDebug( "Failed to connect to database $i at {$this->mServers[$i]['host']}\n" );
 403                         if ( $fail ) {
 404                                 $this->reportConnectionError( $this->mConnections[$i] );
 405                         }
 406                         $this->mErrorConnection = $this->mConnections[$i];
 407                         $this->mConnections[$i] = false;
 408                         $success = false;
 409                 }
 410                 $this->mLastIndex = $i;
 411                 wfProfileOut( $fname );
 412                 return $success;
 413         }
 414
 415         /**
 416          * Test if the specified index represents an open connection
 417          * @access private
 418          */
 419         function isOpen( $index ) {
 420                 if( !is_integer( $index ) ) {
 421                         return false;
 422                 }
 423                 if ( array_key_exists( $index, $this->mConnections ) && is_object( $this->mConnections[$index] ) &&
 424                   $this->mConnections[$index]->isOpen() )
 425                 {
 426                         return true;
 427                 } else {
 428                         return false;
 429                 }
 430         }
 431
 432         /**
 433          * Really opens a connection
 434          * @access private
 435          */
 436         function reallyOpenConnection( &$server ) {
 437                 if( !is_array( $server ) ) {
 438                         wfDebugDieBacktrace( 'You must update your load-balancing configuration. See DefaultSettings.php entry for $wgDBservers.' );
 439                 }
 440
 441                 extract( $server );
 442
 443                 # Get class for this database type
 444                 if ($type != 'mysql' ) {
 445                         $class = 'Database' . ucfirst( $type );
 446                 } else {
 447                         $class = 'Database';
 448                 }
 449
 450                 # Create object
 451                 $db = new $class( $host, $user, $password, $dbname, 1, $flags );
 452                 $db->setLBInfo( $server );
 453                 return $db;
 454         }
 455
 456         function reportConnectionError( &$conn )
 457         {
 458                 $fname = 'LoadBalancer::reportConnectionError';
 459                 wfProfileIn( $fname );
 460                 # Prevent infinite recursion
 461
 462                 static $reporting = false;
 463                 if ( !$reporting ) {
 464                         $reporting = true;
 465                         if ( !is_object( $conn ) ) {
 466                                 // No last connection, probably due to all servers being too busy
 467                                 $conn = new Database;
 468                                 if ( $this->mFailFunction ) {
 469                                         $conn->failFunction( $this->mFailFunction );
 470                                         $conn->reportConnectionError( $this->mLastError );
 471                                 } else {
 472                                         // If all servers were busy, mLastError will contain something sensible
 473                                         wfEmergencyAbort( $conn, $this->mLastError );
 474                                 }
 475                         } else {
 476                                 if ( $this->mFailFunction ) {
 477                                         $conn->failFunction( $this->mFailFunction );
 478                                 } else {
 479                                         $conn->failFunction( false );
 480                                 }
 481                                 $conn->reportConnectionError( "{$this->mLastError} ({$conn->mServer})" );
 482                         }
 483                         $reporting = false;
 484                 }
 485                 wfProfileOut( $fname );
 486         }
 487
 488         function getWriterIndex()
 489         {
 490                 return 0;
 491         }
 492
 493         function force( $i )
 494         {
 495                 $this->mForce = $i;
 496         }
 497
 498         function haveIndex( $i )
 499         {
 500                 return array_key_exists( $i, $this->mServers );
 501         }
 502
 503         /**
 504          * Get the number of defined servers (not the number of open connections)
 505          */
 506         function getServerCount() {
 507                 return count( $this->mServers );
 508         }
 509
 510         /**
 511          * Save master pos to the session and to memcached, if the session exists
 512          */
 513         function saveMasterPos() {
 514                 global $wgSessionStarted;
 515                 if ( $wgSessionStarted && count( $this->mServers ) > 1 ) {
 516                         # If this entire request was served from a slave without opening a connection to the
 517                         # master (however unlikely that may be), then we can fetch the position from the slave.
 518                         if ( empty( $this->mConnections[0] ) ) {
 519                                 $conn =& $this->getConnection( DB_SLAVE );
 520                                 list( $file, $pos ) = $conn->getSlavePos();
 521                                 wfDebug( "Saving master pos fetched from slave: $file $pos\n" );
 522                         } else {
 523                                 $conn =& $this->getConnection( 0 );
 524                                 list( $file, $pos ) = $conn->getMasterPos();
 525                                 wfDebug( "Saving master pos: $file $pos\n" );
 526                         }
 527                         if ( $file !== false ) {
 528                                 $_SESSION['master_log_file'] = $file;
 529                                 $_SESSION['master_pos'] = $pos;
 530                         }
 531                 }
 532         }
 533
 534         /**
 535          * Loads the master pos from the session, waits for it if necessary
 536          */
 537         function loadMasterPos() {
 538                 if ( isset( $_SESSION['master_log_file'] ) && isset( $_SESSION['master_pos'] ) ) {
 539                         $this->waitFor( $_SESSION['master_log_file'], $_SESSION['master_pos'] );
 540                 }
 541         }
 542
 543         /**
 544          * Close all open connections
 545          */
 546         function closeAll() {
 547                 foreach( $this->mConnections as $i => $conn ) {
 548                         if ( $this->isOpen( $i ) ) {
 549                                 // Need to use this syntax because $conn is a copy not a reference
 550                                 $this->mConnections[$i]->close();
 551                         }
 552                 }
 553         }
 554
 555         function commitAll() {
 556                 foreach( $this->mConnections as $i => $conn ) {
 557                         if ( $this->isOpen( $i ) ) {
 558                                 // Need to use this syntax because $conn is a copy not a reference
 559                                 $this->mConnections[$i]->immediateCommit();
 560                         }
 561                 }
 562         }
 563
 564         function waitTimeout( $value = NULL ) {
 565                 return wfSetVar( $this->mWaitTimeout, $value );
 566         }
 567
 568         function getLaggedSlaveMode() {
 569                 return $this->mLaggedSlaveMode;
 570         }
 571
 572         /* Disables/enables lag checks */
 573         function allowLagged($mode=null) {
 574                 if ($mode===null)
 575                         return $this->mAllowLagged;
 576                 $this->mAllowLagged=$mode;
 577         }
 578
 579         function pingAll() {
 580                 $success = true;
 581                 foreach ( $this->mConnections as $i => $conn ) {
 582                         if ( $this->isOpen( $i ) ) {
 583                                 if ( !$this->mConnections[$i]->ping() ) {
 584                                         $success = false;
 585                                 }
 586                         }
 587                 }
 588                 return $success;
 589         }
 590
 591         /**
 592          * Get the hostname and lag time of the most-lagged slave
 593          * This is useful for maintenance scripts that need to throttle their updates
 594          */
 595         function getMaxLag() {
 596                 $maxLag = -1;
 597                 $host = '';
 598                 foreach ( $this->mServers as $i => $conn ) {
 599                         if ( $this->openConnection( $i ) ) {
 600                                 $lag = $this->mConnections[$i]->getLag();
 601                                 if ( $lag > $maxLag ) {
 602                                         $maxLag = $lag;
 603                                         $host = $this->mServers[$i]['host'];
 604                                 }
 605                         }
 606                 }
 607                 return array( $host, $maxLag );
 608         }
 609
 610         /**
 611          * Get lag time for each DB
 612          * Results are cached for a short time in memcached
 613          */
 614         function getLagTimes() {
 615                 global $wgDBname;
 616
 617                 $expiry = 5;
 618                 $requestRate = 10;
 619
 620                 global $wgMemc;
 621                 $times = $wgMemc->get( "$wgDBname:lag_times" );
 622                 if ( $times ) {
 623                         # Randomly recache with probability rising over $expiry
 624                         $elapsed = time() - $times['timestamp'];
 625                         $chance = max( 0, ( $expiry - $elapsed ) * $requestRate );
 626                         if ( mt_rand( 0, $chance ) != 0 ) {
 627                                 unset( $times['timestamp'] );
 628                                 return $times;
 629                         }
 630                 }
 631
 632                 # Cache key missing or expired
 633
 634                 $times = array();
 635                 foreach ( $this->mServers as $i => $conn ) {
 636                         if ($i==0) { # Master
 637                                 $times[$i] = 0;
 638                         } elseif ( $this->openConnection( $i ) ) {
 639                                 $times[$i] = $this->mConnections[$i]->getLag();
 640                         }
 641                 }
 642
 643                 # Add a timestamp key so we know when it was cached
 644                 $times['timestamp'] = time();
 645                 $wgMemc->set( "$wgDBname:lag_times", $times, $expiry );
 646
 647                 # But don't give the timestamp to the caller
 648                 unset($times['timestamp']);
 649                 return $times;
 650         }
 651 }
 652
 653 ?>