includes/LoadBalancer.php

   1 <?php
   2 /**
   3  *
   4  * @package MediaWiki
   5  */
   6
   7
   8 /**
   9  * Database load balancing object
  10  *
  11  * @todo document
  12  * @package MediaWiki
  13  */
  14 class LoadBalancer {
  15         /* private */ var $mServers, $mConnections, $mLoads, $mGroupLoads;
  16         /* private */ var $mFailFunction, $mErrorConnection;
  17         /* private */ var $mForce, $mReadIndex, $mLastIndex, $mAllowLagged;
  18         /* private */ var $mWaitForFile, $mWaitForPos, $mWaitTimeout;
  19         /* private */ var $mLaggedSlaveMode, $mLastError = 'Unknown error';
  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          */
  25         const AVG_STATUS_POLL = 2000;
  26
  27         function LoadBalancer( $servers, $failFunction = false, $waitTimeout = 10, $waitForMasterNow = false )
  28         {
  29                 $this->mServers = $servers;
  30                 $this->mFailFunction = $failFunction;
  31                 $this->mReadIndex = -1;
  32                 $this->mWriteIndex = -1;
  33                 $this->mForce = -1;
  34                 $this->mConnections = array();
  35                 $this->mLastIndex = 1;
  36                 $this->mLoads = array();
  37                 $this->mWaitForFile = false;
  38                 $this->mWaitForPos = false;
  39                 $this->mWaitTimeout = $waitTimeout;
  40                 $this->mLaggedSlaveMode = false;
  41                 $this->mErrorConnection = false;
  42                 $this->mAllowLag = false;
  43
  44                 foreach( $servers as $i => $server ) {
  45                         $this->mLoads[$i] = $server['load'];
  46                         if ( isset( $server['groupLoads'] ) ) {
  47                                 foreach ( $server['groupLoads'] as $group => $ratio ) {
  48                                         if ( !isset( $this->mGroupLoads[$group] ) ) {
  49                                                 $this->mGroupLoads[$group] = array();
  50                                         }
  51                                         $this->mGroupLoads[$group][$i] = $ratio;
  52                                 }
  53                         }
  54                 }
  55                 if ( $waitForMasterNow ) {
  56                         $this->loadMasterPos();
  57                 }
  58         }
  59
  60         static function newFromParams( $servers, $failFunction = false, $waitTimeout = 10 )
  61         {
  62                 return new LoadBalancer( $servers, $failFunction, $waitTimeout );
  63         }
  64
  65         /**
  66          * Given an array of non-normalised probabilities, this function will select
  67          * an element and return the appropriate key
  68          */
  69         function pickRandom( $weights )
  70         {
  71                 if ( !is_array( $weights ) || count( $weights ) == 0 ) {
  72                         return false;
  73                 }
  74
  75                 $sum = array_sum( $weights );
  76                 if ( $sum == 0 ) {
  77                         # No loads on any of them
  78                         # In previous versions, this triggered an unweighted random selection,
  79                         # but this feature has been removed as of April 2006 to allow for strict
  80                         # separation of query groups.
  81                         return false;
  82                 }
  83                 $max = mt_getrandmax();
  84                 $rand = mt_rand(0, $max) / $max * $sum;
  85
  86                 $sum = 0;
  87                 foreach ( $weights as $i => $w ) {
  88                         $sum += $w;
  89                         if ( $sum >= $rand ) {
  90                                 break;
  91                         }
  92                 }
  93                 return $i;
  94         }
  95
  96         function getRandomNonLagged( $loads ) {
  97                 # Unset excessively lagged servers
  98                 $lags = $this->getLagTimes();
  99                 foreach ( $lags as $i => $lag ) {
 100                         if ( isset( $this->mServers[$i]['max lag'] ) && $lag > $this->mServers[$i]['max lag'] ) {
 101                                 unset( $loads[$i] );
 102                         }
 103                 }
 104
 105                 # Find out if all the slaves with non-zero load are lagged
 106                 $sum = 0;
 107                 foreach ( $loads as $load ) {
 108                         $sum += $load;
 109                 }
 110                 if ( $sum == 0 ) {
 111                         # No appropriate DB servers except maybe the master and some slaves with zero load
 112                         # Do NOT use the master
 113                         # Instead, this function will return false, triggering read-only mode,
 114                         # and a lagged slave will be used instead.
 115                         return false;
 116                 }
 117
 118                 if ( count( $loads ) == 0 ) {
 119                         return false;
 120                 }
 121
 122                 #wfDebugLog( 'connect', var_export( $loads, true ) );
 123
 124                 # Return a random representative of the remainder
 125                 return $this->pickRandom( $loads );
 126         }
 127
 128         /**
 129          * Get the index of the reader connection, which may be a slave
 130          * This takes into account load ratios and lag times. It should
 131          * always return a consistent index during a given invocation
 132          *
 133          * Side effect: opens connections to databases
 134          */
 135         function getReaderIndex() {
 136                 global $wgReadOnly, $wgDBClusterTimeout;
 137
 138                 $fname = 'LoadBalancer::getReaderIndex';
 139                 wfProfileIn( $fname );
 140
 141                 $i = false;
 142                 if ( $this->mForce >= 0 ) {
 143                         $i = $this->mForce;
 144                 } else {
 145                         if ( $this->mReadIndex >= 0 ) {
 146                                 $i = $this->mReadIndex;
 147                         } else {
 148                                 # $loads is $this->mLoads except with elements knocked out if they
 149                                 # don't work
 150                                 $loads = $this->mLoads;
 151                                 $done = false;
 152                                 $totalElapsed = 0;
 153                                 do {
 154                                         if ( $wgReadOnly or $this->mAllowLagged ) {
 155                                                 $i = $this->pickRandom( $loads );
 156                                         } else {
 157                                                 $i = $this->getRandomNonLagged( $loads );
 158                                                 if ( $i === false && count( $loads ) != 0 )  {
 159                                                         # All slaves lagged. Switch to read-only mode
 160                                                         $wgReadOnly = wfMsgNoDB( 'readonly_lag' );
 161                                                         $i = $this->pickRandom( $loads );
 162                                                 }
 163                                         }
 164                                         $serverIndex = $i;
 165                                         if ( $i !== false ) {
 166                                                 wfDebugLog( 'connect', "$fname: Using reader #$i: {$this->mServers[$i]['host']}...\n" );
 167                                                 $this->openConnection( $i );
 168
 169                                                 if ( !$this->isOpen( $i ) ) {
 170                                                         wfDebug( "$fname: Failed\n" );
 171                                                         unset( $loads[$i] );
 172                                                         $sleepTime = 0;
 173                                                 } else {
 174                                                         $status = $this->mConnections[$i]->getStatus("Thread%");
 175                                                         if ( isset( $this->mServers[$i]['max threads'] ) &&
 176                                                           $status['Threads_running'] > $this->mServers[$i]['max threads'] )
 177                                                         {
 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                                                 }
 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, $i );
 322                         if ( $groupIndex !== false ) {
 323                                 $i = $groupIndex;
 324                         }
 325                 } else {
 326                         foreach ( $groups as $group ) {
 327                                 $groupIndex = $this->getGroupIndex( $group, $i );
 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         {
 431                 $fname = 'LoadBalancer::reportConnectionError';
 432                 wfProfileIn( $fname );
 433                 # Prevent infinite recursion
 434
 435                 static $reporting = false;
 436                 if ( !$reporting ) {
 437                         $reporting = true;
 438                         if ( !is_object( $conn ) ) {
 439                                 // No last connection, probably due to all servers being too busy
 440                                 $conn = new Database;
 441                                 if ( $this->mFailFunction ) {
 442                                         $conn->failFunction( $this->mFailFunction );
 443                                         $conn->reportConnectionError( $this->mLastError );
 444                                 } else {
 445                                         // If all servers were busy, mLastError will contain something sensible
 446                                         throw new DBConnectionError( $conn, $this->mLastError );
 447                                 }
 448                         } else {
 449                                 if ( $this->mFailFunction ) {
 450                                         $conn->failFunction( $this->mFailFunction );
 451                                 } else {
 452                                         $conn->failFunction( false );
 453                                 }
 454                                 $server = $conn->getProperty( 'mServer' );
 455                                 $conn->reportConnectionError( "{$this->mLastError} ({$server})" );
 456                         }
 457                         $reporting = false;
 458                 }
 459                 wfProfileOut( $fname );
 460         }
 461
 462         function getWriterIndex() {
 463                 return 0;
 464         }
 465
 466         /**
 467          * Force subsequent calls to getConnection(DB_SLAVE) to return the
 468          * given index. Set to -1 to restore the original load balancing
 469          * behaviour. I thought this was a good idea when I originally
 470          * wrote this class, but it has never been used.
 471          */
 472         function force( $i ) {
 473                 $this->mForce = $i;
 474         }
 475
 476         /**
 477          * Returns true if the specified index is a valid server index
 478          */
 479         function haveIndex( $i ) {
 480                 return array_key_exists( $i, $this->mServers );
 481         }
 482
 483         /**
 484          * Returns true if the specified index is valid and has non-zero load
 485          */
 486         function isNonZeroLoad( $i ) {
 487                 return array_key_exists( $i, $this->mServers ) && $this->mLoads[$i] != 0;
 488         }
 489
 490         /**
 491          * Get the number of defined servers (not the number of open connections)
 492          */
 493         function getServerCount() {
 494                 return count( $this->mServers );
 495         }
 496
 497         /**
 498          * Save master pos to the session and to memcached, if the session exists
 499          */
 500         function saveMasterPos() {
 501                 global $wgSessionStarted;
 502                 if ( $wgSessionStarted && count( $this->mServers ) > 1 ) {
 503                         # If this entire request was served from a slave without opening a connection to the
 504                         # master (however unlikely that may be), then we can fetch the position from the slave.
 505                         if ( empty( $this->mConnections[0] ) ) {
 506                                 $conn =& $this->getConnection( DB_SLAVE );
 507                                 list( $file, $pos ) = $conn->getSlavePos();
 508                                 wfDebug( "Saving master pos fetched from slave: $file $pos\n" );
 509                         } else {
 510                                 $conn =& $this->getConnection( 0 );
 511                                 list( $file, $pos ) = $conn->getMasterPos();
 512                                 wfDebug( "Saving master pos: $file $pos\n" );
 513                         }
 514                         if ( $file !== false ) {
 515                                 $_SESSION['master_log_file'] = $file;
 516                                 $_SESSION['master_pos'] = $pos;
 517                         }
 518                 }
 519         }
 520
 521         /**
 522          * Loads the master pos from the session, waits for it if necessary
 523          */
 524         function loadMasterPos() {
 525                 if ( isset( $_SESSION['master_log_file'] ) && isset( $_SESSION['master_pos'] ) ) {
 526                         $this->waitFor( $_SESSION['master_log_file'], $_SESSION['master_pos'] );
 527                 }
 528         }
 529
 530         /**
 531          * Close all open connections
 532          */
 533         function closeAll() {
 534                 foreach( $this->mConnections as $i => $conn ) {
 535                         if ( $this->isOpen( $i ) ) {
 536                                 // Need to use this syntax because $conn is a copy not a reference
 537                                 $this->mConnections[$i]->close();
 538                         }
 539                 }
 540         }
 541
 542         function commitAll() {
 543                 foreach( $this->mConnections as $i => $conn ) {
 544                         if ( $this->isOpen( $i ) ) {
 545                                 // Need to use this syntax because $conn is a copy not a reference
 546                                 $this->mConnections[$i]->immediateCommit();
 547                         }
 548                 }
 549         }
 550
 551         function waitTimeout( $value = NULL ) {
 552                 return wfSetVar( $this->mWaitTimeout, $value );
 553         }
 554
 555         function getLaggedSlaveMode() {
 556                 return $this->mLaggedSlaveMode;
 557         }
 558
 559         /* Disables/enables lag checks */
 560         function allowLagged($mode=null) {
 561                 if ($mode===null)
 562                         return $this->mAllowLagged;
 563                 $this->mAllowLagged=$mode;
 564         }
 565
 566         function pingAll() {
 567                 $success = true;
 568                 foreach ( $this->mConnections as $i => $conn ) {
 569                         if ( $this->isOpen( $i ) ) {
 570                                 if ( !$this->mConnections[$i]->ping() ) {
 571                                         $success = false;
 572                                 }
 573                         }
 574                 }
 575                 return $success;
 576         }
 577
 578         /**
 579          * Get the hostname and lag time of the most-lagged slave
 580          * This is useful for maintenance scripts that need to throttle their updates
 581          */
 582         function getMaxLag() {
 583                 $maxLag = -1;
 584                 $host = '';
 585                 foreach ( $this->mServers as $i => $conn ) {
 586                         if ( $this->openConnection( $i ) ) {
 587                                 $lag = $this->mConnections[$i]->getLag();
 588                                 if ( $lag > $maxLag ) {
 589                                         $maxLag = $lag;
 590                                         $host = $this->mServers[$i]['host'];
 591                                 }
 592                         }
 593                 }
 594                 return array( $host, $maxLag );
 595         }
 596
 597         /**
 598          * Get lag time for each DB
 599          * Results are cached for a short time in memcached
 600          */
 601         function getLagTimes() {
 602                 global $wgDBname;
 603
 604                 wfProfileIn( __METHOD__ );
 605                 $expiry = 5;
 606                 $requestRate = 10;
 607
 608                 global $wgMemc;
 609                 $times = $wgMemc->get( "$wgDBname:lag_times" );
 610                 if ( $times ) {
 611                         # Randomly recache with probability rising over $expiry
 612                         $elapsed = time() - $times['timestamp'];
 613                         $chance = max( 0, ( $expiry - $elapsed ) * $requestRate );
 614                         if ( mt_rand( 0, $chance ) != 0 ) {
 615                                 unset( $times['timestamp'] );
 616                                 wfProfileOut( __METHOD__ );
 617                                 return $times;
 618                         }
 619                         wfIncrStats( 'lag_cache_miss_expired' );
 620                 } else {
 621                         wfIncrStats( 'lag_cache_miss_absent' );
 622                 }
 623
 624                 # Cache key missing or expired
 625
 626                 $times = array();
 627                 foreach ( $this->mServers as $i => $conn ) {
 628                         if ($i==0) { # Master
 629                                 $times[$i] = 0;
 630                         } elseif ( $this->openConnection( $i ) ) {
 631                                 $times[$i] = $this->mConnections[$i]->getLag();
 632                         }
 633                 }
 634
 635                 # Add a timestamp key so we know when it was cached
 636                 $times['timestamp'] = time();
 637                 $wgMemc->set( "$wgDBname:lag_times", $times, $expiry );
 638
 639                 # But don't give the timestamp to the caller
 640                 unset($times['timestamp']);
 641                 wfProfileOut( __METHOD__ );
 642                 return $times;
 643         }
 644 }
 645
 646 ?>