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