includes/poolcounter/PoolCounterRedis.php

   1 <?php
   2 /**
   3  * This program is free software; you can redistribute it and/or modify
   4  * it under the terms of the GNU General Public License as published by
   5  * the Free Software Foundation; either version 2 of the License, or
   6  * (at your option) any later version.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along
  14  * with this program; if not, write to the Free Software Foundation, Inc.,
  15  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
  16  * http://www.gnu.org/copyleft/gpl.html
  17  *
  18  * @file
  19  * @author Aaron Schulz
  20  */
  21
  22 /**
  23  * Version of PoolCounter that uses Redis
  24  *
  25  * There are four main redis keys used to track each pool counter key:
  26  *   - poolcounter:l-slots-*     : A list of available slot IDs for a pool.
  27  *   - poolcounter:z-renewtime-* : A sorted set of (slot ID, UNIX timestamp as score)
  28  *                                 used for tracking the next time a slot should be
  29  *                                 released. This is -1 when a slot is created, and is
  30  *                                 set when released (expired), locked, and unlocked.
  31  *   - poolcounter:z-wait-*      : A sorted set of (slot ID, UNIX timestamp as score)
  32  *                                 used for tracking waiting processes (and wait time).
  33  *   - poolcounter:l-wakeup-*    : A list pushed to for the sake of waking up processes
  34  *                                 when a any process in the pool finishes (lasts for 1ms).
  35  * For a given pool key, all the redis keys start off non-existing and are deleted if not
  36  * used for a while to prevent garbage from building up on the server. They are atomically
  37  * re-initialized as needed. The "z-renewtime" key is used for detecting sessions which got
  38  * slots but then disappeared. Stale entries from there have their timestamp updated and the
  39  * corresponding slots freed up. The "z-wait" key is used for detecting processes registered
  40  * as waiting but that disappeared. Stale entries from there are deleted and the corresponding
  41  * slots are freed up. The worker count is included in all the redis key names as it does not
  42  * vary within each $wgPoolCounterConf type and doing so handles configuration changes.
  43  *
  44  * This class requires Redis 2.6 as it makes use Lua scripts for fast atomic operations.
  45  * Also this should be on a server plenty of RAM for the working set to avoid evictions.
  46  * Evictions could temporarily allow wait queues to double in size or temporarily cause
  47  * pools to appear as full when they are not. Using volatile-ttl and bumping memory-samples
  48  * in redis.conf can be helpful otherwise.
  49  *
  50  * @ingroup Redis
  51  * @since 1.23
  52  */
  53 class PoolCounterRedis extends PoolCounter {
  54         /** @var HashRing */
  55         protected $ring;
  56         /** @var RedisConnectionPool */
  57         protected $pool;
  58         /** @var array (server label => host) map */
  59         protected $serversByLabel;
  60         /** @var string SHA-1 of the key */
  61         protected $keySha1;
  62         /** @var int TTL for locks to expire (work should finish in this time) */
  63         protected $lockTTL;
  64
  65         /** @var RedisConnRef */
  66         protected $conn;
  67         /** @var string Pool slot value */
  68         protected $slot;
  69         /** @var int AWAKE_* constant */
  70         protected $onRelease;
  71         /** @var string Unique string to identify this process */
  72         protected $session;
  73         /** @var int UNIX timestamp */
  74         protected $slotTime;
  75
  76         const AWAKE_ONE = 1; // wake-up if when a slot can be taken from an existing process
  77         const AWAKE_ALL = 2; // wake-up if an existing process finishes and wake up such others
  78
  79         /** @var array List of active PoolCounterRedis objects in this script */
  80         protected static $active = null;
  81
  82         function __construct( $conf, $type, $key ) {
  83                 parent::__construct( $conf, $type, $key );
  84
  85                 $this->serversByLabel = $conf['servers'];
  86                 $this->ring = new HashRing( array_fill_keys( array_keys( $conf['servers'] ), 100 ) );
  87
  88                 $conf['redisConfig']['serializer'] = 'none'; // for use with Lua
  89                 $this->pool = RedisConnectionPool::singleton( $conf['redisConfig'] );
  90
  91                 $this->keySha1 = sha1( $this->key );
  92                 $met = ini_get( 'max_execution_time' ); // usually 0 in CLI mode
  93                 $this->lockTTL = $met ? 2 * $met : 3600;
  94
  95                 if ( self::$active === null ) {
  96                         self::$active = array();
  97                         register_shutdown_function( array( __CLASS__, 'releaseAll' ) );
  98                 }
  99         }
 100
 101         /**
 102          * @return Status Uses RediConnRef as value on success
 103          */
 104         protected function getConnection() {
 105                 if ( !isset( $this->conn ) ) {
 106                         $conn = false;
 107                         $servers = $this->ring->getLocations( $this->key, 3 );
 108                         ArrayUtils::consistentHashSort( $servers, $this->key );
 109                         foreach ( $servers as $server ) {
 110                                 $conn = $this->pool->getConnection( $this->serversByLabel[$server] );
 111                                 if ( $conn ) {
 112                                         break;
 113                                 }
 114                         }
 115                         if ( !$conn ) {
 116                                 return Status::newFatal( 'pool-servererror', implode( ', ', $servers ) );
 117                         }
 118                         $this->conn = $conn;
 119                 }
 120                 return Status::newGood( $this->conn );
 121         }
 122
 123         function acquireForMe() {
 124                 $section = new ProfileSection( __METHOD__ );
 125
 126                 $status = $this->precheckAcquire();
 127                 if ( !$status->isGood() ) {
 128                         return $status;
 129                 }
 130
 131                 return $this->waitForSlotOrNotif( self::AWAKE_ONE );
 132         }
 133
 134         function acquireForAnyone() {
 135                 $section = new ProfileSection( __METHOD__ );
 136
 137                 $status = $this->precheckAcquire();
 138                 if ( !$status->isGood() ) {
 139                         return $status;
 140                 }
 141
 142                 return $this->waitForSlotOrNotif( self::AWAKE_ALL );
 143         }
 144
 145         function release() {
 146                 $section = new ProfileSection( __METHOD__ );
 147
 148                 if ( $this->slot === null ) {
 149                         return Status::newGood( PoolCounter::NOT_LOCKED ); // not locked
 150                 }
 151
 152                 $status = $this->getConnection();
 153                 if ( !$status->isOK() ) {
 154                         return $status;
 155                 }
 156                 $conn = $status->value;
 157
 158                 static $script =
 159 <<<LUA
 160                 local kSlots,kSlotsNextRelease,kWakeup,kWaiting = unpack(KEYS)
 161                 local rMaxWorkers,rExpiry,rSlot,rSlotTime,rAwakeAll,rTime = unpack(ARGV)
 162                 -- Add the slots back to the list (if rSlot is "w" then it is not a slot).
 163                 -- Treat the list as expired if the "next release" time sorted-set is missing.
 164                 if rSlot ~= 'w' and redis.call('exists',kSlotsNextRelease) == 1 then
 165                         if 1*redis.call('zScore',kSlotsNextRelease,rSlot) ~= (rSlotTime + rExpiry) then
 166                                 -- Slot lock expired and was released already
 167                         elseif redis.call('lLen',kSlots) >= 1*rMaxWorkers then
 168                                 -- Slots somehow got out of sync; reset the list for sanity
 169                                 redis.call('del',kSlots,kSlotsNextRelease)
 170                         elseif redis.call('lLen',kSlots) == (1*rMaxWorkers - 1) and redis.call('zCard',kWaiting) == 0 then
 171                                 -- Slot list will be made full; clear it to save space (it re-inits as needed)
 172                                 -- since nothing is waiting on being unblocked by a push to the list
 173                                 redis.call('del',kSlots,kSlotsNextRelease)
 174                         else
 175                                 -- Add slot back to pool and update the "next release" time
 176                                 redis.call('rPush',kSlots,rSlot)
 177                                 redis.call('zAdd',kSlotsNextRelease,rTime + 30,rSlot)
 178                                 -- Always keep renewing the expiry on use
 179                                 redis.call('expireAt',kSlots,math.ceil(rTime + rExpiry))
 180                                 redis.call('expireAt',kSlotsNextRelease,math.ceil(rTime + rExpiry))
 181                         end
 182                 end
 183                 -- Update an ephemeral list to wake up other clients that can
 184                 -- reuse any cached work from this process. Only do this if no
 185                 -- slots are currently free (e.g. clients could be waiting).
 186                 if 1*rAwakeAll == 1 then
 187                         local count = redis.call('zCard',kWaiting)
 188                         for i = 1,count do
 189                                 redis.call('rPush',kWakeup,'w')
 190                         end
 191                         redis.call('pexpire',kWakeup,1)
 192                 end
 193                 return 1
 194 LUA;
 195                 try {
 196                         $res = $conn->luaEval( $script,
 197                                 array(
 198                                         $this->getSlotListKey(),
 199                                         $this->getSlotRTimeSetKey(),
 200                                         $this->getWakeupListKey(),
 201                                         $this->getWaitSetKey(),
 202                                         $this->workers,
 203                                         $this->lockTTL,
 204                                         $this->slot,
 205                                         $this->slotTime, // used for CAS-style sanity check
 206                                         ( $this->onRelease === self::AWAKE_ALL ) ? 1 : 0,
 207                                         microtime( true )
 208                                 ),
 209                                 4 # number of first argument(s) that are keys
 210                         );
 211                 } catch ( RedisException $e ) {
 212                         return Status::newFatal( 'pool-error-unknown', $e->getMessage() );
 213                 }
 214
 215                 $this->slot = null;
 216                 $this->slotTime = null;
 217                 $this->onRelease = null;
 218                 unset( self::$active[$this->session] );
 219
 220                 $this->onRelease();
 221
 222                 return Status::newGood( PoolCounter::RELEASED );
 223         }
 224
 225         /**
 226          * @param int $doWakeup AWAKE_* constant
 227          * @return Status
 228          */
 229         protected function waitForSlotOrNotif( $doWakeup ) {
 230                 if ( $this->slot !== null ) {
 231                         return Status::newGood( PoolCounter::LOCK_HELD ); // already acquired
 232                 }
 233
 234                 $status = $this->getConnection();
 235                 if ( !$status->isOK() ) {
 236                         return $status;
 237                 }
 238                 $conn = $status->value;
 239
 240                 $now = microtime( true );
 241                 try {
 242                         $slot = $this->initAndPopPoolSlotList( $conn, $now );
 243                         if ( ctype_digit( $slot ) ) {
 244                                 // Pool slot acquired by this process
 245                                 $slotTime = $now;
 246                         } elseif ( $slot === 'QUEUE_FULL' ) {
 247                                 // Too many processes are waiting for pooled processes to finish
 248                                 return Status::newGood( PoolCounter::QUEUE_FULL );
 249                         } elseif ( $slot === 'QUEUE_WAIT' ) {
 250                                 // This process is now registered as waiting
 251                                 $keys = ( $doWakeup == self::AWAKE_ALL )
 252                                         // Wait for an open slot or wake-up signal (preferring the later)
 253                                         ? array( $this->getWakeupListKey(), $this->getSlotListKey() )
 254                                         // Just wait for an actual pool slot
 255                                         : array( $this->getSlotListKey() );
 256
 257                                 $res = $conn->blPop( $keys, $this->timeout );
 258                                 if ( $res === array() ) {
 259                                         $conn->zRem( $this->getWaitSetKey(), $this->session ); // no longer waiting
 260                                         return Status::newGood( PoolCounter::TIMEOUT );
 261                                 }
 262
 263                                 $slot = $res[1]; // pool slot or "w" for wake-up notifications
 264                                 $slotTime = microtime( true ); // last microtime() was a few RTTs ago
 265                                 // Unregister this process as waiting and bump slot "next release" time
 266                                 $this->registerAcquisitionTime( $conn, $slot, $slotTime );
 267                         } else {
 268                                 return Status::newFatal( 'pool-error-unknown', "Server gave slot '$slot'." );
 269                         }
 270                 } catch ( RedisException $e ) {
 271                         return Status::newFatal( 'pool-error-unknown', $e->getMessage() );
 272                 }
 273
 274                 if ( $slot !== 'w' ) {
 275                         $this->slot = $slot;
 276                         $this->slotTime = $slotTime;
 277                         $this->onRelease = $doWakeup;
 278                         self::$active[$this->session] = $this;
 279                 }
 280
 281                 $this->onAcquire();
 282
 283                 return Status::newGood( $slot === 'w' ? PoolCounter::DONE : PoolCounter::LOCKED );
 284         }
 285
 286         /**
 287          * @param RedisConnRef $conn
 288          * @param float $now UNIX timestamp
 289          * @return string|bool False on failure
 290          */
 291         protected function initAndPopPoolSlotList( RedisConnRef $conn, $now ) {
 292                 static $script =
 293 <<<LUA
 294                 local kSlots,kSlotsNextRelease,kSlotWaits = unpack(KEYS)
 295                 local rMaxWorkers,rMaxQueue,rTimeout,rExpiry,rSess,rTime = unpack(ARGV)
 296                 -- Initialize if the "next release" time sorted-set is empty. The slot key
 297                 -- itself is empty if all slots are busy or when nothing is initialized.
 298                 -- If the list is empty but the set is not, then it is the later case.
 299                 -- For sanity, if the list exists but not the set, then reset everything.
 300                 if redis.call('exists',kSlotsNextRelease) == 0 then
 301                         redis.call('del',kSlots)
 302                         for i = 1,1*rMaxWorkers do
 303                                 redis.call('rPush',kSlots,i)
 304                                 redis.call('zAdd',kSlotsNextRelease,-1,i)
 305                         end
 306                 -- Otherwise do maintenance to clean up after network partitions
 307                 else
 308                         -- Find stale slot locks and add free them (avoid duplicates for sanity)
 309                         local staleLocks = redis.call('zRangeByScore',kSlotsNextRelease,0,rTime)
 310                         for k,slot in ipairs(staleLocks) do
 311                                 redis.call('lRem',kSlots,0,slot)
 312                                 redis.call('rPush',kSlots,slot)
 313                                 redis.call('zAdd',kSlotsNextRelease,rTime + 30,slot)
 314                         end
 315                         -- Find stale wait slot entries and remove them
 316                         redis.call('zRemRangeByScore',kSlotWaits,0,rTime - 2*rTimeout)
 317                 end
 318                 local slot
 319                 -- Try to acquire a slot if possible now
 320                 if redis.call('lLen',kSlots) > 0 then
 321                         slot = redis.call('lPop',kSlots)
 322                         -- Update the slot "next release" time
 323                         redis.call('zAdd',kSlotsNextRelease,rTime + rExpiry,slot)
 324                 elseif redis.call('zCard',kSlotWaits) >= 1*rMaxQueue then
 325                         slot = 'QUEUE_FULL'
 326                 else
 327                         slot = 'QUEUE_WAIT'
 328                         -- Register this process as waiting
 329                         redis.call('zAdd',kSlotWaits,rTime,rSess)
 330                         redis.call('expireAt',kSlotWaits,math.ceil(rTime + 2*rTimeout))
 331                 end
 332                 -- Always keep renewing the expiry on use
 333                 redis.call('expireAt',kSlots,math.ceil(rTime + rExpiry))
 334                 redis.call('expireAt',kSlotsNextRelease,math.ceil(rTime + rExpiry))
 335                 return slot
 336 LUA;
 337                 return $conn->luaEval( $script,
 338                         array(
 339                                 $this->getSlotListKey(),
 340                                 $this->getSlotRTimeSetKey(),
 341                                 $this->getWaitSetKey(),
 342                                 $this->workers,
 343                                 $this->maxqueue,
 344                                 $this->timeout,
 345                                 $this->lockTTL,
 346                                 $this->session,
 347                                 $now
 348                         ),
 349                         3 # number of first argument(s) that are keys
 350                 );
 351         }
 352
 353         /**
 354          * @param RedisConnRef $conn
 355          * @param string $slot
 356          * @param float $now
 357          * @return int|bool False on failure
 358          */
 359         protected function registerAcquisitionTime( RedisConnRef $conn, $slot, $now ) {
 360                 static $script =
 361 <<<LUA
 362                 local kSlots,kSlotsNextRelease,kSlotWaits = unpack(KEYS)
 363                 local rSlot,rExpiry,rSess,rTime = unpack(ARGV)
 364                 -- If rSlot is 'w' then the client was told to wake up but got no slot
 365                 if rSlot ~= 'w' then
 366                         -- Update the slot "next release" time
 367                         redis.call('zAdd',kSlotsNextRelease,rTime + rExpiry,rSlot)
 368                         -- Always keep renewing the expiry on use
 369                         redis.call('expireAt',kSlots,math.ceil(rTime + rExpiry))
 370                         redis.call('expireAt',kSlotsNextRelease,math.ceil(rTime + rExpiry))
 371                 end
 372                 -- Unregister this process as waiting
 373                 redis.call('zRem',kSlotWaits,rSess)
 374                 return 1
 375 LUA;
 376                 return $conn->luaEval( $script,
 377                         array(
 378                                 $this->getSlotListKey(),
 379                                 $this->getSlotRTimeSetKey(),
 380                                 $this->getWaitSetKey(),
 381                                 $slot,
 382                                 $this->lockTTL,
 383                                 $this->session,
 384                                 $now
 385                         ),
 386                         3 # number of first argument(s) that are keys
 387                 );
 388         }
 389
 390         /**
 391          * @return string
 392          */
 393         protected function getSlotListKey() {
 394                 return "poolcounter:l-slots-{$this->keySha1}-{$this->workers}";
 395         }
 396
 397         /**
 398          * @return string
 399          */
 400         protected function getSlotRTimeSetKey() {
 401                 return "poolcounter:z-renewtime-{$this->keySha1}-{$this->workers}";
 402         }
 403
 404         /**
 405          * @return string
 406          */
 407         protected function getWaitSetKey() {
 408                 return "poolcounter:z-wait-{$this->keySha1}-{$this->workers}";
 409         }
 410
 411         /**
 412          * @return string
 413          */
 414         protected function getWakeupListKey() {
 415                 return "poolcounter:l-wakeup-{$this->keySha1}-{$this->workers}";
 416         }
 417
 418         /**
 419          * Try to make sure that locks get released (even with exceptions and fatals)
 420          */
 421         public static function releaseAll() {
 422                 foreach ( self::$active as $poolCounter ) {
 423                         try {
 424                                 if ( $poolCounter->slot !== null ) {
 425                                         $poolCounter->release();
 426                                 }
 427                         } catch ( Exception $e ) {
 428                         }
 429                 }
 430         }
 431 }