includes/LoadBalancer.php

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