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 use Psr\Log\LoggerInterface;
  22
  23 /**
  24  * Version of PoolCounter that uses Redis
  25  *
  26  * There are four main redis keys used to track each pool counter key:
  27  *   - poolcounter:l-slots-*     : A list of available slot IDs for a pool.
  28  *   - poolcounter:z-renewtime-* : A sorted set of (slot ID, UNIX timestamp as score)
  29  *                                 used for tracking the next time a slot should be
  30  *                                 released. This is -1 when a slot is created, and is
  31  *                                 set when released (expired), locked, and unlocked.
  32  *   - poolcounter:z-wait-*      : A sorted set of (slot ID, UNIX timestamp as score)
  33  *                                 used for tracking waiting processes (and wait time).
  34  *   - poolcounter:l-wakeup-*    : A list pushed to for the sake of waking up processes
  35  *                                 when a any process in the pool finishes (lasts for 1ms).
  36  * For a given pool key, all the redis keys start off non-existing and are deleted if not
  37  * used for a while to prevent garbage from building up on the server. They are atomically
  38  * re-initialized as needed. The "z-renewtime" key is used for detecting sessions which got
  39  * slots but then disappeared. Stale entries from there have their timestamp updated and the
  40  * corresponding slots freed up. The "z-wait" key is used for detecting processes registered
  41  * as waiting but that disappeared. Stale entries from there are deleted and the corresponding
  42  * slots are freed up. The worker count is included in all the redis key names as it does not
  43  * vary within each $wgPoolCounterConf type and doing so handles configuration changes.
  44  *
  45  * This class requires Redis 2.6 as it makes use Lua scripts for fast atomic operations.
  46  * Also this should be on a server plenty of RAM for the working set to avoid evictions.
  47  * Evictions could temporarily allow wait queues to double in size or temporarily cause
  48  * pools to appear as full when they are not. Using volatile-ttl and bumping memory-samples
  49  * in redis.conf can be helpful otherwise.
  50  *
  51  * @ingroup Redis
  52  * @since 1.23
  53  */
  54 class PoolCounterRedis extends PoolCounter {
  55         /** @var HashRing */
  56         protected $ring;
  57         /** @var RedisConnectionPool */
  58         protected $pool;
  59         /** @var LoggerInterface */
  60         protected $logger;
  61         /** @var array (server label => host) map */
  62         protected $serversByLabel;
  63         /** @var string SHA-1 of the key */
  64         protected $keySha1;
  65         /** @var int TTL for locks to expire (work should finish in this time) */
  66         protected $lockTTL;
  67
  68         /** @var RedisConnRef */
  69         protected $conn;
  70         /** @var string Pool slot value */
  71         protected $slot;
  72         /** @var int AWAKE_* constant */
  73         protected $onRelease;
  74         /** @var string Unique string to identify this process */
  75         protected $session;
  76         /** @var int UNIX timestamp */
  77         protected $slotTime;
  78
  79         const AWAKE_ONE = 1; // wake-up if when a slot can be taken from an existing process
  80         const AWAKE_ALL = 2; // wake-up if an existing process finishes and wake up such others
  81
  82         /** @var PoolCounterRedis[] List of active PoolCounterRedis objects in this script */
  83         protected static $active = null;
  84
  85         function __construct( $conf, $type, $key ) {
  86                 parent::__construct( $conf, $type, $key );
  87
  88                 $this->serversByLabel = $conf['servers'];
  89                 $this->ring = new HashRing( array_fill_keys( array_keys( $conf['servers'] ), 100 ) );
  90
  91                 $conf['redisConfig']['serializer'] = 'none'; // for use with Lua
  92                 $this->pool = RedisConnectionPool::singleton( $conf['redisConfig'] );
  93                 $this->logger = \MediaWiki\Logger\LoggerFactory::getInstance( 'redis' );
  94
  95                 $this->keySha1 = sha1( $this->key );
  96                 $met = ini_get( 'max_execution_time' ); // usually 0 in CLI mode
  97                 $this->lockTTL = $met ? 2 * $met : 3600;
  98
  99                 if ( self::$active === null ) {
 100                         self::$active = [];
 101                         register_shutdown_function( [ __CLASS__, 'releaseAll' ] );
 102                 }
 103         }
 104
 105         /**
 106          * @return Status Uses RediConnRef as value on success
 107          */
 108         protected function getConnection() {
 109                 if ( !isset( $this->conn ) ) {
 110                         $conn = false;
 111                         $servers = $this->ring->getLocations( $this->key, 3 );
 112                         ArrayUtils::consistentHashSort( $servers, $this->key );
 113                         foreach ( $servers as $server ) {
 114                                 $conn = $this->pool->getConnection( $this->serversByLabel[$server], $this->logger );
 115                                 if ( $conn ) {
 116                                         break;
 117                                 }
 118                         }
 119                         if ( !$conn ) {
 120                                 return Status::newFatal( 'pool-servererror', implode( ', ', $servers ) );
 121                         }
 122                         $this->conn = $conn;
 123                 }
 124                 return Status::newGood( $this->conn );
 125         }
 126
 127         function acquireForMe() {
 128                 $status = $this->precheckAcquire();
 129                 if ( !$status->isGood() ) {
 130                         return $status;
 131                 }
 132
 133                 return $this->waitForSlotOrNotif( self::AWAKE_ONE );
 134         }
 135
 136         function acquireForAnyone() {
 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                 if ( $this->slot === null ) {
 147                         return Status::newGood( PoolCounter::NOT_LOCKED ); // not locked
 148                 }
 149
 150                 $status = $this->getConnection();
 151                 if ( !$status->isOK() ) {
 152                         return $status;
 153                 }
 154                 $conn = $status->value;
 155
 156                 // @codingStandardsIgnoreStart Generic.Files.LineLength
 157                 static $script =
 158 <<<LUA
 159                 local kSlots,kSlotsNextRelease,kWakeup,kWaiting = unpack(KEYS)
 160                 local rMaxWorkers,rExpiry,rSlot,rSlotTime,rAwakeAll,rTime = unpack(ARGV)
 161                 -- Add the slots back to the list (if rSlot is "w" then it is not a slot).
 162                 -- Treat the list as expired if the "next release" time sorted-set is missing.
 163                 if rSlot ~= 'w' and redis.call('exists',kSlotsNextRelease) == 1 then
 164                         if 1*redis.call('zScore',kSlotsNextRelease,rSlot) ~= (rSlotTime + rExpiry) then
 165                                 -- Slot lock expired and was released already
 166                         elseif redis.call('lLen',kSlots) >= 1*rMaxWorkers then
 167                                 -- Slots somehow got out of sync; reset the list for sanity
 168                                 redis.call('del',kSlots,kSlotsNextRelease)
 169                         elseif redis.call('lLen',kSlots) == (1*rMaxWorkers - 1) and redis.call('zCard',kWaiting) == 0 then
 170                                 -- Slot list will be made full; clear it to save space (it re-inits as needed)
 171                                 -- since nothing is waiting on being unblocked by a push to the list
 172                                 redis.call('del',kSlots,kSlotsNextRelease)
 173                         else
 174                                 -- Add slot back to pool and update the "next release" time
 175                                 redis.call('rPush',kSlots,rSlot)
 176                                 redis.call('zAdd',kSlotsNextRelease,rTime + 30,rSlot)
 177                                 -- Always keep renewing the expiry on use
 178                                 redis.call('expireAt',kSlots,math.ceil(rTime + rExpiry))
 179                                 redis.call('expireAt',kSlotsNextRelease,math.ceil(rTime + rExpiry))
 180                         end
 181                 end
 182                 -- Update an ephemeral list to wake up other clients that can
 183                 -- reuse any cached work from this process. Only do this if no
 184                 -- slots are currently free (e.g. clients could be waiting).
 185                 if 1*rAwakeAll == 1 then
 186                         local count = redis.call('zCard',kWaiting)
 187                         for i = 1,count do
 188                                 redis.call('rPush',kWakeup,'w')
 189                         end
 190                         redis.call('pexpire',kWakeup,1)
 191                 end
 192                 return 1
 193 LUA;
 194                 // @codingStandardsIgnoreEnd
 195
 196                 try {
 197                         $conn->luaEval( $script,
 198                                 [
 199                                         $this->getSlotListKey(),
 200                                         $this->getSlotRTimeSetKey(),
 201                                         $this->getWakeupListKey(),
 202                                         $this->getWaitSetKey(),
 203                                         $this->workers,
 204                                         $this->lockTTL,
 205                                         $this->slot,
 206                                         $this->slotTime, // used for CAS-style sanity check
 207                                         ( $this->onRelease === self::AWAKE_ALL ) ? 1 : 0,
 208                                         microtime( true )
 209                                 ],
 210                                 4 # number of first argument(s) that are keys
 211                         );
 212                 } catch ( RedisException $e ) {
 213                         return Status::newFatal( 'pool-error-unknown', $e->getMessage() );
 214                 }
 215
 216                 $this->slot = null;
 217                 $this->slotTime = null;
 218                 $this->onRelease = null;
 219                 unset( self::$active[$this->session] );
 220
 221                 $this->onRelease();
 222
 223                 return Status::newGood( PoolCounter::RELEASED );
 224         }
 225
 226         /**
 227          * @param int $doWakeup AWAKE_* constant
 228          * @return Status
 229          */
 230         protected function waitForSlotOrNotif( $doWakeup ) {
 231                 if ( $this->slot !== null ) {
 232                         return Status::newGood( PoolCounter::LOCK_HELD ); // already acquired
 233                 }
 234
 235                 $status = $this->getConnection();
 236                 if ( !$status->isOK() ) {
 237                         return $status;
 238                 }
 239                 $conn = $status->value;
 240
 241                 $now = microtime( true );
 242                 try {
 243                         $slot = $this->initAndPopPoolSlotList( $conn, $now );
 244                         if ( ctype_digit( $slot ) ) {
 245                                 // Pool slot acquired by this process
 246                                 $slotTime = $now;
 247                         } elseif ( $slot === 'QUEUE_FULL' ) {
 248                                 // Too many processes are waiting for pooled processes to finish
 249                                 return Status::newGood( PoolCounter::QUEUE_FULL );
 250                         } elseif ( $slot === 'QUEUE_WAIT' ) {
 251                                 // This process is now registered as waiting
 252                                 $keys = ( $doWakeup == self::AWAKE_ALL )
 253                                         // Wait for an open slot or wake-up signal (preferring the latter)
 254                                         ? [ $this->getWakeupListKey(), $this->getSlotListKey() ]
 255                                         // Just wait for an actual pool slot
 256                                         : [ $this->getSlotListKey() ];
 257
 258                                 $res = $conn->blPop( $keys, $this->timeout );
 259                                 if ( $res === [] ) {
 260                                         $conn->zRem( $this->getWaitSetKey(), $this->session ); // no longer waiting
 261                                         return Status::newGood( PoolCounter::TIMEOUT );
 262                                 }
 263
 264                                 $slot = $res[1]; // pool slot or "w" for wake-up notifications
 265                                 $slotTime = microtime( true ); // last microtime() was a few RTTs ago
 266                                 // Unregister this process as waiting and bump slot "next release" time
 267                                 $this->registerAcquisitionTime( $conn, $slot, $slotTime );
 268                         } else {
 269                                 return Status::newFatal( 'pool-error-unknown', "Server gave slot '$slot'." );
 270                         }
 271                 } catch ( RedisException $e ) {
 272                         return Status::newFatal( 'pool-error-unknown', $e->getMessage() );
 273                 }
 274
 275                 if ( $slot !== 'w' ) {
 276                         $this->slot = $slot;
 277                         $this->slotTime = $slotTime;
 278                         $this->onRelease = $doWakeup;
 279                         self::$active[$this->session] = $this;
 280                 }
 281
 282                 $this->onAcquire();
 283
 284                 return Status::newGood( $slot === 'w' ? PoolCounter::DONE : PoolCounter::LOCKED );
 285         }
 286
 287         /**
 288          * @param RedisConnRef $conn
 289          * @param float $now UNIX timestamp
 290          * @return string|bool False on failure
 291          */
 292         protected function initAndPopPoolSlotList( RedisConnRef $conn, $now ) {
 293                 static $script =
 294 <<<LUA
 295                 local kSlots,kSlotsNextRelease,kSlotWaits = unpack(KEYS)
 296                 local rMaxWorkers,rMaxQueue,rTimeout,rExpiry,rSess,rTime = unpack(ARGV)
 297                 -- Initialize if the "next release" time sorted-set is empty. The slot key
 298                 -- itself is empty if all slots are busy or when nothing is initialized.
 299                 -- If the list is empty but the set is not, then it is the latter case.
 300                 -- For sanity, if the list exists but not the set, then reset everything.
 301                 if redis.call('exists',kSlotsNextRelease) == 0 then
 302                         redis.call('del',kSlots)
 303                         for i = 1,1*rMaxWorkers do
 304                                 redis.call('rPush',kSlots,i)
 305                                 redis.call('zAdd',kSlotsNextRelease,-1,i)
 306                         end
 307                 -- Otherwise do maintenance to clean up after network partitions
 308                 else
 309                         -- Find stale slot locks and add free them (avoid duplicates for sanity)
 310                         local staleLocks = redis.call('zRangeByScore',kSlotsNextRelease,0,rTime)
 311                         for k,slot in ipairs(staleLocks) do
 312                                 redis.call('lRem',kSlots,0,slot)
 313                                 redis.call('rPush',kSlots,slot)
 314                                 redis.call('zAdd',kSlotsNextRelease,rTime + 30,slot)
 315                         end
 316                         -- Find stale wait slot entries and remove them
 317                         redis.call('zRemRangeByScore',kSlotWaits,0,rTime - 2*rTimeout)
 318                 end
 319                 local slot
 320                 -- Try to acquire a slot if possible now
 321                 if redis.call('lLen',kSlots) > 0 then
 322                         slot = redis.call('lPop',kSlots)
 323                         -- Update the slot "next release" time
 324                         redis.call('zAdd',kSlotsNextRelease,rTime + rExpiry,slot)
 325                 elseif redis.call('zCard',kSlotWaits) >= 1*rMaxQueue then
 326                         slot = 'QUEUE_FULL'
 327                 else
 328                         slot = 'QUEUE_WAIT'
 329                         -- Register this process as waiting
 330                         redis.call('zAdd',kSlotWaits,rTime,rSess)
 331                         redis.call('expireAt',kSlotWaits,math.ceil(rTime + 2*rTimeout))
 332                 end
 333                 -- Always keep renewing the expiry on use
 334                 redis.call('expireAt',kSlots,math.ceil(rTime + rExpiry))
 335                 redis.call('expireAt',kSlotsNextRelease,math.ceil(rTime + rExpiry))
 336                 return slot
 337 LUA;
 338                 return $conn->luaEval( $script,
 339                         [
 340                                 $this->getSlotListKey(),
 341                                 $this->getSlotRTimeSetKey(),
 342                                 $this->getWaitSetKey(),
 343                                 $this->workers,
 344                                 $this->maxqueue,
 345                                 $this->timeout,
 346                                 $this->lockTTL,
 347                                 $this->session,
 348                                 $now
 349                         ],
 350                         3 # number of first argument(s) that are keys
 351                 );
 352         }
 353
 354         /**
 355          * @param RedisConnRef $conn
 356          * @param string $slot
 357          * @param float $now
 358          * @return int|bool False on failure
 359          */
 360         protected function registerAcquisitionTime( RedisConnRef $conn, $slot, $now ) {
 361                 static $script =
 362 <<<LUA
 363                 local kSlots,kSlotsNextRelease,kSlotWaits = unpack(KEYS)
 364                 local rSlot,rExpiry,rSess,rTime = unpack(ARGV)
 365                 -- If rSlot is 'w' then the client was told to wake up but got no slot
 366                 if rSlot ~= 'w' then
 367                         -- Update the slot "next release" time
 368                         redis.call('zAdd',kSlotsNextRelease,rTime + rExpiry,rSlot)
 369                         -- Always keep renewing the expiry on use
 370                         redis.call('expireAt',kSlots,math.ceil(rTime + rExpiry))
 371                         redis.call('expireAt',kSlotsNextRelease,math.ceil(rTime + rExpiry))
 372                 end
 373                 -- Unregister this process as waiting
 374                 redis.call('zRem',kSlotWaits,rSess)
 375                 return 1
 376 LUA;
 377                 return $conn->luaEval( $script,
 378                         [
 379                                 $this->getSlotListKey(),
 380                                 $this->getSlotRTimeSetKey(),
 381                                 $this->getWaitSetKey(),
 382                                 $slot,
 383                                 $this->lockTTL,
 384                                 $this->session,
 385                                 $now
 386                         ],
 387                         3 # number of first argument(s) that are keys
 388                 );
 389         }
 390
 391         /**
 392          * @return string
 393          */
 394         protected function getSlotListKey() {
 395                 return "poolcounter:l-slots-{$this->keySha1}-{$this->workers}";
 396         }
 397
 398         /**
 399          * @return string
 400          */
 401         protected function getSlotRTimeSetKey() {
 402                 return "poolcounter:z-renewtime-{$this->keySha1}-{$this->workers}";
 403         }
 404
 405         /**
 406          * @return string
 407          */
 408         protected function getWaitSetKey() {
 409                 return "poolcounter:z-wait-{$this->keySha1}-{$this->workers}";
 410         }
 411
 412         /**
 413          * @return string
 414          */
 415         protected function getWakeupListKey() {
 416                 return "poolcounter:l-wakeup-{$this->keySha1}-{$this->workers}";
 417         }
 418
 419         /**
 420          * Try to make sure that locks get released (even with exceptions and fatals)
 421          */
 422         public static function releaseAll() {
 423                 foreach ( self::$active as $poolCounter ) {
 424                         try {
 425                                 if ( $poolCounter->slot !== null ) {
 426                                         $poolCounter->release();
 427                                 }
 428                         } catch ( Exception $e ) {
 429                         }
 430                 }
 431         }
 432 }