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