Completed
Branch master (e9bc15)
by
unknown
32:19
created

WANObjectCache::getMulti()   D

Complexity

Conditions 13
Paths 128

Size

Total Lines 80
Code Lines 50

Duplication

Lines 0
Ratio 0 %

Importance

Changes 0
Metric Value
cc 13
eloc 50
nc 128
nop 4
dl 0
loc 80
rs 4.7182
c 0
b 0
f 0

How to fix   Long Method    Complexity   

Long Method

Small methods make your code easier to understand, in particular if combined with a good name. Besides, if your method is small, finding a good name is usually much easier.

For example, if you find yourself adding comments to a method's body, this is usually a good sign to extract the commented part to a new method, and use the comment as a starting point when coming up with a good name for this new method.

Commonly applied refactorings include:

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
 * @ingroup Cache
20
 * @author Aaron Schulz
21
 */
22
23
use Psr\Log\LoggerAwareInterface;
24
use Psr\Log\LoggerInterface;
25
use Psr\Log\NullLogger;
26
27
/**
28
 * Multi-datacenter aware caching interface
29
 *
30
 * All operations go to the local datacenter cache, except for delete(),
31
 * touchCheckKey(), and resetCheckKey(), which broadcast to all datacenters.
32
 *
33
 * This class is intended for caching data from primary stores.
34
 * If the get() method does not return a value, then the caller
35
 * should query the new value and backfill the cache using set().
36
 * The preferred way to do this logic is through getWithSetCallback().
37
 * When querying the store on cache miss, the closest DB replica
38
 * should be used. Try to avoid heavyweight DB master or quorum reads.
39
 * When the source data changes, a purge method should be called.
40
 * Since purges are expensive, they should be avoided. One can do so if:
41
 *   - a) The object cached is immutable; or
42
 *   - b) Validity is checked against the source after get(); or
43
 *   - c) Using a modest TTL is reasonably correct and performant
44
 *
45
 * The simplest purge method is delete().
46
 *
47
 * There are two supported ways to handle broadcasted operations:
48
 *   - a) Configure the 'purge' EventRelayer to point to a valid PubSub endpoint
49
 *        that has subscribed listeners on the cache servers applying the cache updates.
50
 *   - b) Ignore the 'purge' EventRelayer configuration (default is NullEventRelayer)
51
 *        and set up mcrouter as the underlying cache backend, using one of the memcached
52
 *        BagOStuff classes as 'cache'. Use OperationSelectorRoute in the mcrouter settings
53
 *        to configure 'set' and 'delete' operations to go to all DCs via AllAsyncRoute and
54
 *        configure other operations to go to the local DC via PoolRoute (for reference,
55
 *        see https://github.com/facebook/mcrouter/wiki/List-of-Route-Handles).
56
 *
57
 * Broadcasted operations like delete() and touchCheckKey() are done asynchronously
58
 * in all datacenters this way, though the local one should likely be near immediate.
59
 *
60
 * This means that callers in all datacenters may see older values for however many
61
 * milliseconds that the purge took to reach that datacenter. As with any cache, this
62
 * should not be relied on for cases where reads are used to determine writes to source
63
 * (e.g. non-cache) data stores, except when reading immutable data.
64
 *
65
 * All values are wrapped in metadata arrays. Keys use a "WANCache:" prefix
66
 * to avoid collisions with keys that are not wrapped as metadata arrays. The
67
 * prefixes are as follows:
68
 *   - a) "WANCache:v" : used for regular value keys
69
 *   - b) "WANCache:i" : used for temporarily storing values of tombstoned keys
70
 *   - c) "WANCache:t" : used for storing timestamp "check" keys
71
 *   - d) "WANCache:m" : used for temporary mutex keys to avoid cache stampedes
72
 *
73
 * @ingroup Cache
74
 * @since 1.26
75
 */
76
class WANObjectCache implements IExpiringStore, LoggerAwareInterface {
77
	/** @var BagOStuff The local datacenter cache */
78
	protected $cache;
79
	/** @var HashBagOStuff[] Map of group PHP instance caches */
80
	protected $processCaches = [];
81
	/** @var string Purge channel name */
82
	protected $purgeChannel;
83
	/** @var EventRelayer Bus that handles purge broadcasts */
84
	protected $purgeRelayer;
85
	/** @var LoggerInterface */
86
	protected $logger;
87
88
	/** @var int ERR_* constant for the "last error" registry */
89
	protected $lastRelayError = self::ERR_NONE;
90
91
	/** @var mixed[] Temporary warm-up cache */
92
	private $warmupCache = [];
93
94
	/** Max time expected to pass between delete() and DB commit finishing */
95
	const MAX_COMMIT_DELAY = 3;
96
	/** Max replication+snapshot lag before applying TTL_LAGGED or disallowing set() */
97
	const MAX_READ_LAG = 7;
98
	/** Seconds to tombstone keys on delete() */
99
	const HOLDOFF_TTL = 11; // MAX_COMMIT_DELAY + MAX_READ_LAG + 1
100
101
	/** Seconds to keep dependency purge keys around */
102
	const CHECK_KEY_TTL = self::TTL_YEAR;
103
	/** Seconds to keep lock keys around */
104
	const LOCK_TTL = 10;
105
	/** Default remaining TTL at which to consider pre-emptive regeneration */
106
	const LOW_TTL = 30;
107
	/** Default time-since-expiry on a miss that makes a key "hot" */
108
	const LOCK_TSE = 1;
109
110
	/** Never consider performing "popularity" refreshes until a key reaches this age */
111
	const AGE_NEW = 60;
112
	/** The time length of the "popularity" refresh window for hot keys */
113
	const HOT_TTR = 900;
114
	/** Hits/second for a refresh to be expected within the "popularity" window */
115
	const HIT_RATE_HIGH = 1;
116
	/** Seconds to ramp up to the "popularity" refresh chance after a key is no longer new */
117
	const RAMPUP_TTL = 30;
118
119
	/** Idiom for getWithSetCallback() callbacks to avoid calling set() */
120
	const TTL_UNCACHEABLE = -1;
121
	/** Idiom for getWithSetCallback() callbacks to 'lockTSE' logic */
122
	const TSE_NONE = -1;
123
	/** Max TTL to store keys when a data sourced is lagged */
124
	const TTL_LAGGED = 30;
125
	/** Idiom for delete() for "no hold-off" */
126
	const HOLDOFF_NONE = 0;
127
	/** Idiom for getWithSetCallback() for "no minimum required as-of timestamp" */
128
	const MIN_TIMESTAMP_NONE = 0.0;
129
130
	/** Tiny negative float to use when CTL comes up >= 0 due to clock skew */
131
	const TINY_NEGATIVE = -0.000001;
132
133
	/** Cache format version number */
134
	const VERSION = 1;
135
136
	const FLD_VERSION = 0; // key to cache version number
137
	const FLD_VALUE = 1; // key to the cached value
138
	const FLD_TTL = 2; // key to the original TTL
139
	const FLD_TIME = 3; // key to the cache time
140
	const FLD_FLAGS = 4; // key to the flags bitfield
141
	const FLD_HOLDOFF = 5; // key to any hold-off TTL
142
143
	/** @var integer Treat this value as expired-on-arrival */
144
	const FLG_STALE = 1;
145
146
	const ERR_NONE = 0; // no error
147
	const ERR_NO_RESPONSE = 1; // no response
148
	const ERR_UNREACHABLE = 2; // can't connect
149
	const ERR_UNEXPECTED = 3; // response gave some error
150
	const ERR_RELAY = 4; // relay broadcast failed
151
152
	const VALUE_KEY_PREFIX = 'WANCache:v:';
153
	const INTERIM_KEY_PREFIX = 'WANCache:i:';
154
	const TIME_KEY_PREFIX = 'WANCache:t:';
155
	const MUTEX_KEY_PREFIX = 'WANCache:m:';
156
157
	const PURGE_VAL_PREFIX = 'PURGED:';
158
159
	const VFLD_DATA = 'WOC:d'; // key to the value of versioned data
160
	const VFLD_VERSION = 'WOC:v'; // key to the version of the value present
161
162
	const PC_PRIMARY = 'primary:1000'; // process cache name and max key count
163
164
	const DEFAULT_PURGE_CHANNEL = 'wancache-purge';
165
166
	/**
167
	 * @param array $params
168
	 *   - cache    : BagOStuff object for a persistent cache
169
	 *   - channels : Map of (action => channel string). Actions include "purge".
170
	 *   - relayers : Map of (action => EventRelayer object). Actions include "purge".
171
	 *   - logger   : LoggerInterface object
172
	 */
173
	public function __construct( array $params ) {
174
		$this->cache = $params['cache'];
175
		$this->purgeChannel = isset( $params['channels']['purge'] )
176
			? $params['channels']['purge']
177
			: self::DEFAULT_PURGE_CHANNEL;
178
		$this->purgeRelayer = isset( $params['relayers']['purge'] )
179
			? $params['relayers']['purge']
180
			: new EventRelayerNull( [] );
181
		$this->setLogger( isset( $params['logger'] ) ? $params['logger'] : new NullLogger() );
182
	}
183
184
	public function setLogger( LoggerInterface $logger ) {
185
		$this->logger = $logger;
186
	}
187
188
	/**
189
	 * Get an instance that wraps EmptyBagOStuff
190
	 *
191
	 * @return WANObjectCache
192
	 */
193
	public static function newEmpty() {
194
		return new self( [
195
			'cache'   => new EmptyBagOStuff(),
196
			'pool'    => 'empty',
197
			'relayer' => new EventRelayerNull( [] )
198
		] );
199
	}
200
201
	/**
202
	 * Fetch the value of a key from cache
203
	 *
204
	 * If supplied, $curTTL is set to the remaining TTL (current time left):
205
	 *   - a) INF; if $key exists, has no TTL, and is not expired by $checkKeys
206
	 *   - b) float (>=0); if $key exists, has a TTL, and is not expired by $checkKeys
207
	 *   - c) float (<0); if $key is tombstoned, stale, or existing but expired by $checkKeys
208
	 *   - d) null; if $key does not exist and is not tombstoned
209
	 *
210
	 * If a key is tombstoned, $curTTL will reflect the time since delete().
211
	 *
212
	 * The timestamp of $key will be checked against the last-purge timestamp
213
	 * of each of $checkKeys. Those $checkKeys not in cache will have the last-purge
214
	 * initialized to the current timestamp. If any of $checkKeys have a timestamp
215
	 * greater than that of $key, then $curTTL will reflect how long ago $key
216
	 * became invalid. Callers can use $curTTL to know when the value is stale.
217
	 * The $checkKeys parameter allow mass invalidations by updating a single key:
218
	 *   - a) Each "check" key represents "last purged" of some source data
219
	 *   - b) Callers pass in relevant "check" keys as $checkKeys in get()
220
	 *   - c) When the source data that "check" keys represent changes,
221
	 *        the touchCheckKey() method is called on them
222
	 *
223
	 * Source data entities might exists in a DB that uses snapshot isolation
224
	 * (e.g. the default REPEATABLE-READ in innoDB). Even for mutable data, that
225
	 * isolation can largely be maintained by doing the following:
226
	 *   - a) Calling delete() on entity change *and* creation, before DB commit
227
	 *   - b) Keeping transaction duration shorter than delete() hold-off TTL
228
	 *
229
	 * However, pre-snapshot values might still be seen if an update was made
230
	 * in a remote datacenter but the purge from delete() didn't relay yet.
231
	 *
232
	 * Consider using getWithSetCallback() instead of get() and set() cycles.
233
	 * That method has cache slam avoiding features for hot/expensive keys.
234
	 *
235
	 * @param string $key Cache key
236
	 * @param mixed $curTTL Approximate TTL left on the key if present/tombstoned [returned]
237
	 * @param array $checkKeys List of "check" keys
238
	 * @param float &$asOf UNIX timestamp of cached value; null on failure [returned]
239
	 * @return mixed Value of cache key or false on failure
240
	 */
241
	final public function get( $key, &$curTTL = null, array $checkKeys = [], &$asOf = null ) {
242
		$curTTLs = [];
243
		$asOfs = [];
244
		$values = $this->getMulti( [ $key ], $curTTLs, $checkKeys, $asOfs );
245
		$curTTL = isset( $curTTLs[$key] ) ? $curTTLs[$key] : null;
246
		$asOf = isset( $asOfs[$key] ) ? $asOfs[$key] : null;
247
248
		return isset( $values[$key] ) ? $values[$key] : false;
249
	}
250
251
	/**
252
	 * Fetch the value of several keys from cache
253
	 *
254
	 * @see WANObjectCache::get()
255
	 *
256
	 * @param array $keys List of cache keys
257
	 * @param array $curTTLs Map of (key => approximate TTL left) for existing keys [returned]
258
	 * @param array $checkKeys List of check keys to apply to all $keys. May also apply "check"
259
	 *  keys to specific cache keys only by using cache keys as keys in the $checkKeys array.
260
	 * @param float[] &$asOfs Map of (key =>  UNIX timestamp of cached value; null on failure)
261
	 * @return array Map of (key => value) for keys that exist
262
	 */
263
	final public function getMulti(
264
		array $keys, &$curTTLs = [], array $checkKeys = [], array &$asOfs = []
265
	) {
266
		$result = [];
267
		$curTTLs = [];
268
		$asOfs = [];
269
270
		$vPrefixLen = strlen( self::VALUE_KEY_PREFIX );
271
		$valueKeys = self::prefixCacheKeys( $keys, self::VALUE_KEY_PREFIX );
272
273
		$checkKeysForAll = [];
274
		$checkKeysByKey = [];
275
		$checkKeysFlat = [];
276
		foreach ( $checkKeys as $i => $keys ) {
277
			$prefixed = self::prefixCacheKeys( (array)$keys, self::TIME_KEY_PREFIX );
278
			$checkKeysFlat = array_merge( $checkKeysFlat, $prefixed );
279
			// Is this check keys for a specific cache key, or for all keys being fetched?
280
			if ( is_int( $i ) ) {
281
				$checkKeysForAll = array_merge( $checkKeysForAll, $prefixed );
282
			} else {
283
				$checkKeysByKey[$i] = isset( $checkKeysByKey[$i] )
284
					? array_merge( $checkKeysByKey[$i], $prefixed )
285
					: $prefixed;
286
			}
287
		}
288
289
		// Fetch all of the raw values
290
		$keysGet = array_merge( $valueKeys, $checkKeysFlat );
291
		if ( $this->warmupCache ) {
292
			$wrappedValues = array_intersect_key( $this->warmupCache, array_flip( $keysGet ) );
293
			$keysGet = array_diff( $keysGet, array_keys( $wrappedValues ) ); // keys left to fetch
294
		} else {
295
			$wrappedValues = [];
296
		}
297
		$wrappedValues += $this->cache->getMulti( $keysGet );
298
		// Time used to compare/init "check" keys (derived after getMulti() to be pessimistic)
299
		$now = microtime( true );
300
301
		// Collect timestamps from all "check" keys
302
		$purgeValuesForAll = $this->processCheckKeys( $checkKeysForAll, $wrappedValues, $now );
303
		$purgeValuesByKey = [];
304
		foreach ( $checkKeysByKey as $cacheKey => $checks ) {
305
			$purgeValuesByKey[$cacheKey] =
306
				$this->processCheckKeys( $checks, $wrappedValues, $now );
307
		}
308
309
		// Get the main cache value for each key and validate them
310
		foreach ( $valueKeys as $vKey ) {
311
			if ( !isset( $wrappedValues[$vKey] ) ) {
312
				continue; // not found
313
			}
314
315
			$key = substr( $vKey, $vPrefixLen ); // unprefix
316
317
			list( $value, $curTTL ) = $this->unwrap( $wrappedValues[$vKey], $now );
318
			if ( $value !== false ) {
319
				$result[$key] = $value;
320
321
				// Force dependant keys to be invalid for a while after purging
322
				// to reduce race conditions involving stale data getting cached
323
				$purgeValues = $purgeValuesForAll;
324
				if ( isset( $purgeValuesByKey[$key] ) ) {
325
					$purgeValues = array_merge( $purgeValues, $purgeValuesByKey[$key] );
326
				}
327
				foreach ( $purgeValues as $purge ) {
328
					$safeTimestamp = $purge[self::FLD_TIME] + $purge[self::FLD_HOLDOFF];
329
					if ( $safeTimestamp >= $wrappedValues[$vKey][self::FLD_TIME] ) {
330
						// How long ago this value was expired by *this* check key
331
						$ago = min( $purge[self::FLD_TIME] - $now, self::TINY_NEGATIVE );
332
						// How long ago this value was expired by *any* known check key
333
						$curTTL = min( $curTTL, $ago );
334
					}
335
				}
336
			}
337
			$curTTLs[$key] = $curTTL;
338
			$asOfs[$key] = ( $value !== false ) ? $wrappedValues[$vKey][self::FLD_TIME] : null;
339
		}
340
341
		return $result;
342
	}
343
344
	/**
345
	 * @since 1.27
346
	 * @param array $timeKeys List of prefixed time check keys
347
	 * @param array $wrappedValues
348
	 * @param float $now
349
	 * @return array List of purge value arrays
350
	 */
351
	private function processCheckKeys( array $timeKeys, array $wrappedValues, $now ) {
352
		$purgeValues = [];
353
		foreach ( $timeKeys as $timeKey ) {
354
			$purge = isset( $wrappedValues[$timeKey] )
355
				? self::parsePurgeValue( $wrappedValues[$timeKey] )
356
				: false;
357
			if ( $purge === false ) {
358
				// Key is not set or invalid; regenerate
359
				$newVal = $this->makePurgeValue( $now, self::HOLDOFF_TTL );
360
				$this->cache->add( $timeKey, $newVal, self::CHECK_KEY_TTL );
361
				$purge = self::parsePurgeValue( $newVal );
362
			}
363
			$purgeValues[] = $purge;
364
		}
365
		return $purgeValues;
366
	}
367
368
	/**
369
	 * Set the value of a key in cache
370
	 *
371
	 * Simply calling this method when source data changes is not valid because
372
	 * the changes do not replicate to the other WAN sites. In that case, delete()
373
	 * should be used instead. This method is intended for use on cache misses.
374
	 *
375
	 * If the data was read from a snapshot-isolated transactions (e.g. the default
376
	 * REPEATABLE-READ in innoDB), use 'since' to avoid the following race condition:
377
	 *   - a) T1 starts
378
	 *   - b) T2 updates a row, calls delete(), and commits
379
	 *   - c) The HOLDOFF_TTL passes, expiring the delete() tombstone
380
	 *   - d) T1 reads the row and calls set() due to a cache miss
381
	 *   - e) Stale value is stuck in cache
382
	 *
383
	 * Setting 'lag' and 'since' help avoids keys getting stuck in stale states.
384
	 *
385
	 * Example usage:
386
	 * @code
387
	 *     $dbr = wfGetDB( DB_REPLICA );
388
	 *     $setOpts = Database::getCacheSetOptions( $dbr );
389
	 *     // Fetch the row from the DB
390
	 *     $row = $dbr->selectRow( ... );
391
	 *     $key = $cache->makeKey( 'building', $buildingId );
392
	 *     $cache->set( $key, $row, $cache::TTL_DAY, $setOpts );
393
	 * @endcode
394
	 *
395
	 * @param string $key Cache key
396
	 * @param mixed $value
397
	 * @param integer $ttl Seconds to live. Special values are:
398
	 *   - WANObjectCache::TTL_INDEFINITE: Cache forever
399
	 * @param array $opts Options map:
400
	 *   - lag : Seconds of replica DB lag. Typically, this is either the replica DB lag
401
	 *      before the data was read or, if applicable, the replica DB lag before
402
	 *      the snapshot-isolated transaction the data was read from started.
403
	 *      Use false to indicate that replication is not running.
404
	 *      Default: 0 seconds
405
	 *   - since : UNIX timestamp of the data in $value. Typically, this is either
406
	 *      the current time the data was read or (if applicable) the time when
407
	 *      the snapshot-isolated transaction the data was read from started.
408
	 *      Default: 0 seconds
409
	 *   - pending : Whether this data is possibly from an uncommitted write transaction.
410
	 *      Generally, other threads should not see values from the future and
411
	 *      they certainly should not see ones that ended up getting rolled back.
412
	 *      Default: false
413
	 *   - lockTSE : if excessive replication/snapshot lag is detected, then store the value
414
	 *      with this TTL and flag it as stale. This is only useful if the reads for
415
	 *      this key use getWithSetCallback() with "lockTSE" set.
416
	 *      Default: WANObjectCache::TSE_NONE
417
	 *   - staleTTL : Seconds to keep the key around if it is stale. The get()/getMulti()
418
	 *      methods return such stale values with a $curTTL of 0, and getWithSetCallback()
419
	 *      will call the regeneration callback in such cases, passing in the old value
420
	 *      and its as-of time to the callback. This is useful if adaptiveTTL() is used
421
	 *      on the old value's as-of time when it is verified as still being correct.
422
	 *      Default: 0.
423
	 * @note Options added in 1.28: staleTTL
424
	 * @return bool Success
425
	 */
426
	final public function set( $key, $value, $ttl = 0, array $opts = [] ) {
427
		$now = microtime( true );
428
		$lockTSE = isset( $opts['lockTSE'] ) ? $opts['lockTSE'] : self::TSE_NONE;
429
		$age = isset( $opts['since'] ) ? max( 0, $now - $opts['since'] ) : 0;
430
		$lag = isset( $opts['lag'] ) ? $opts['lag'] : 0;
431
		$staleTTL = isset( $opts['staleTTL'] ) ? $opts['staleTTL'] : 0;
432
433
		// Do not cache potentially uncommitted data as it might get rolled back
434
		if ( !empty( $opts['pending'] ) ) {
435
			$this->logger->info( "Rejected set() for $key due to pending writes." );
436
437
			return true; // no-op the write for being unsafe
438
		}
439
440
		$wrapExtra = []; // additional wrapped value fields
441
		// Check if there's a risk of writing stale data after the purge tombstone expired
442
		if ( $lag === false || ( $lag + $age ) > self::MAX_READ_LAG ) {
443
			// Case A: read lag with "lockTSE"; save but record value as stale
444
			if ( $lockTSE >= 0 ) {
445
				$ttl = max( 1, (int)$lockTSE ); // set() expects seconds
446
				$wrapExtra[self::FLD_FLAGS] = self::FLG_STALE; // mark as stale
447
			// Case B: any long-running transaction; ignore this set()
448
			} elseif ( $age > self::MAX_READ_LAG ) {
449
				$this->logger->warning( "Rejected set() for $key due to snapshot lag." );
450
451
				return true; // no-op the write for being unsafe
452
			// Case C: high replication lag; lower TTL instead of ignoring all set()s
453
			} elseif ( $lag === false || $lag > self::MAX_READ_LAG ) {
454
				$ttl = $ttl ? min( $ttl, self::TTL_LAGGED ) : self::TTL_LAGGED;
455
				$this->logger->warning( "Lowered set() TTL for $key due to replication lag." );
456
			// Case D: medium length request with medium replication lag; ignore this set()
457
			} else {
458
				$this->logger->warning( "Rejected set() for $key due to high read lag." );
459
460
				return true; // no-op the write for being unsafe
461
			}
462
		}
463
464
		// Wrap that value with time/TTL/version metadata
465
		$wrapped = $this->wrap( $value, $ttl, $now ) + $wrapExtra;
466
467
		$func = function ( $cache, $key, $cWrapped ) use ( $wrapped ) {
468
			return ( is_string( $cWrapped ) )
469
				? false // key is tombstoned; do nothing
470
				: $wrapped;
471
		};
472
473
		return $this->cache->merge( self::VALUE_KEY_PREFIX . $key, $func, $ttl + $staleTTL, 1 );
474
	}
475
476
	/**
477
	 * Purge a key from all datacenters
478
	 *
479
	 * This should only be called when the underlying data (being cached)
480
	 * changes in a significant way. This deletes the key and starts a hold-off
481
	 * period where the key cannot be written to for a few seconds (HOLDOFF_TTL).
482
	 * This is done to avoid the following race condition:
483
	 *   - a) Some DB data changes and delete() is called on a corresponding key
484
	 *   - b) A request refills the key with a stale value from a lagged DB
485
	 *   - c) The stale value is stuck there until the key is expired/evicted
486
	 *
487
	 * This is implemented by storing a special "tombstone" value at the cache
488
	 * key that this class recognizes; get() calls will return false for the key
489
	 * and any set() calls will refuse to replace tombstone values at the key.
490
	 * For this to always avoid stale value writes, the following must hold:
491
	 *   - a) Replication lag is bounded to being less than HOLDOFF_TTL; or
492
	 *   - b) If lag is higher, the DB will have gone into read-only mode already
493
	 *
494
	 * Note that set() can also be lag-aware and lower the TTL if it's high.
495
	 *
496
	 * When using potentially long-running ACID transactions, a good pattern is
497
	 * to use a pre-commit hook to issue the delete. This means that immediately
498
	 * after commit, callers will see the tombstone in cache upon purge relay.
499
	 * It also avoids the following race condition:
500
	 *   - a) T1 begins, changes a row, and calls delete()
501
	 *   - b) The HOLDOFF_TTL passes, expiring the delete() tombstone
502
	 *   - c) T2 starts, reads the row and calls set() due to a cache miss
503
	 *   - d) T1 finally commits
504
	 *   - e) Stale value is stuck in cache
505
	 *
506
	 * Example usage:
507
	 * @code
508
	 *     $dbw->startAtomic( __METHOD__ ); // start of request
509
	 *     ... <execute some stuff> ...
510
	 *     // Update the row in the DB
511
	 *     $dbw->update( ... );
512
	 *     $key = $cache->makeKey( 'homes', $homeId );
513
	 *     // Purge the corresponding cache entry just before committing
514
	 *     $dbw->onTransactionPreCommitOrIdle( function() use ( $cache, $key ) {
515
	 *         $cache->delete( $key );
516
	 *     } );
517
	 *     ... <execute some stuff> ...
518
	 *     $dbw->endAtomic( __METHOD__ ); // end of request
519
	 * @endcode
520
	 *
521
	 * The $ttl parameter can be used when purging values that have not actually changed
522
	 * recently. For example, a cleanup script to purge cache entries does not really need
523
	 * a hold-off period, so it can use HOLDOFF_NONE. Likewise for user-requested purge.
524
	 * Note that $ttl limits the effective range of 'lockTSE' for getWithSetCallback().
525
	 *
526
	 * If called twice on the same key, then the last hold-off TTL takes precedence. For
527
	 * idempotence, the $ttl should not vary for different delete() calls on the same key.
528
	 *
529
	 * @param string $key Cache key
530
	 * @param integer $ttl Tombstone TTL; Default: WANObjectCache::HOLDOFF_TTL
531
	 * @return bool True if the item was purged or not found, false on failure
532
	 */
533
	final public function delete( $key, $ttl = self::HOLDOFF_TTL ) {
534
		$key = self::VALUE_KEY_PREFIX . $key;
535
536
		if ( $ttl <= 0 ) {
537
			// Publish the purge to all datacenters
538
			$ok = $this->relayDelete( $key );
539
		} else {
540
			// Publish the purge to all datacenters
541
			$ok = $this->relayPurge( $key, $ttl, self::HOLDOFF_NONE );
542
		}
543
544
		return $ok;
545
	}
546
547
	/**
548
	 * Fetch the value of a timestamp "check" key
549
	 *
550
	 * The key will be *initialized* to the current time if not set,
551
	 * so only call this method if this behavior is actually desired
552
	 *
553
	 * The timestamp can be used to check whether a cached value is valid.
554
	 * Callers should not assume that this returns the same timestamp in
555
	 * all datacenters due to relay delays.
556
	 *
557
	 * The level of staleness can roughly be estimated from this key, but
558
	 * if the key was evicted from cache, such calculations may show the
559
	 * time since expiry as ~0 seconds.
560
	 *
561
	 * Note that "check" keys won't collide with other regular keys.
562
	 *
563
	 * @param string $key
564
	 * @return float UNIX timestamp of the check key
565
	 */
566
	final public function getCheckKeyTime( $key ) {
567
		$key = self::TIME_KEY_PREFIX . $key;
568
569
		$purge = self::parsePurgeValue( $this->cache->get( $key ) );
570
		if ( $purge !== false ) {
571
			$time = $purge[self::FLD_TIME];
572
		} else {
573
			// Casting assures identical floats for the next getCheckKeyTime() calls
574
			$now = (string)microtime( true );
575
			$this->cache->add( $key,
576
				$this->makePurgeValue( $now, self::HOLDOFF_TTL ),
577
				self::CHECK_KEY_TTL
578
			);
579
			$time = (float)$now;
580
		}
581
582
		return $time;
583
	}
584
585
	/**
586
	 * Purge a "check" key from all datacenters, invalidating keys that use it
587
	 *
588
	 * This should only be called when the underlying data (being cached)
589
	 * changes in a significant way, and it is impractical to call delete()
590
	 * on all keys that should be changed. When get() is called on those
591
	 * keys, the relevant "check" keys must be supplied for this to work.
592
	 *
593
	 * The "check" key essentially represents a last-modified field.
594
	 * When touched, the field will be updated on all cache servers.
595
	 * Keys using it via get(), getMulti(), or getWithSetCallback() will
596
	 * be invalidated. It is treated as being HOLDOFF_TTL seconds in the future
597
	 * by those methods to avoid race conditions where dependent keys get updated
598
	 * with stale values (e.g. from a DB replica DB).
599
	 *
600
	 * This is typically useful for keys with hardcoded names or in some cases
601
	 * dynamically generated names where a low number of combinations exist.
602
	 * When a few important keys get a large number of hits, a high cache
603
	 * time is usually desired as well as "lockTSE" logic. The resetCheckKey()
604
	 * method is less appropriate in such cases since the "time since expiry"
605
	 * cannot be inferred, causing any get() after the reset to treat the key
606
	 * as being "hot", resulting in more stale value usage.
607
	 *
608
	 * Note that "check" keys won't collide with other regular keys.
609
	 *
610
	 * @see WANObjectCache::get()
611
	 * @see WANObjectCache::getWithSetCallback()
612
	 * @see WANObjectCache::resetCheckKey()
613
	 *
614
	 * @param string $key Cache key
615
	 * @param int $holdoff HOLDOFF_TTL or HOLDOFF_NONE constant
616
	 * @return bool True if the item was purged or not found, false on failure
617
	 */
618
	final public function touchCheckKey( $key, $holdoff = self::HOLDOFF_TTL ) {
619
		// Publish the purge to all datacenters
620
		return $this->relayPurge( self::TIME_KEY_PREFIX . $key, self::CHECK_KEY_TTL, $holdoff );
621
	}
622
623
	/**
624
	 * Delete a "check" key from all datacenters, invalidating keys that use it
625
	 *
626
	 * This is similar to touchCheckKey() in that keys using it via get(), getMulti(),
627
	 * or getWithSetCallback() will be invalidated. The differences are:
628
	 *   - a) The "check" key will be deleted from all caches and lazily
629
	 *        re-initialized when accessed (rather than set everywhere)
630
	 *   - b) Thus, dependent keys will be known to be invalid, but not
631
	 *        for how long (they are treated as "just" purged), which
632
	 *        effects any lockTSE logic in getWithSetCallback()
633
	 *   - c) Since "check" keys are initialized only on the server the key hashes
634
	 *        to, any temporary ejection of that server will cause the value to be
635
	 *        seen as purged as a new server will initialize the "check" key.
636
	 *
637
	 * The advantage is that this does not place high TTL keys on every cache
638
	 * server, making it better for code that will cache many different keys
639
	 * and either does not use lockTSE or uses a low enough TTL anyway.
640
	 *
641
	 * This is typically useful for keys with dynamically generated names
642
	 * where a high number of combinations exist.
643
	 *
644
	 * Note that "check" keys won't collide with other regular keys.
645
	 *
646
	 * @see WANObjectCache::get()
647
	 * @see WANObjectCache::getWithSetCallback()
648
	 * @see WANObjectCache::touchCheckKey()
649
	 *
650
	 * @param string $key Cache key
651
	 * @return bool True if the item was purged or not found, false on failure
652
	 */
653
	final public function resetCheckKey( $key ) {
654
		// Publish the purge to all datacenters
655
		return $this->relayDelete( self::TIME_KEY_PREFIX . $key );
656
	}
657
658
	/**
659
	 * Method to fetch/regenerate cache keys
660
	 *
661
	 * On cache miss, the key will be set to the callback result via set()
662
	 * (unless the callback returns false) and that result will be returned.
663
	 * The arguments supplied to the callback are:
664
	 *   - $oldValue : current cache value or false if not present
665
	 *   - &$ttl : a reference to the TTL which can be altered
666
	 *   - &$setOpts : a reference to options for set() which can be altered
667
	 *   - $oldAsOf : generation UNIX timestamp of $oldValue or null if not present (since 1.28)
668
	 *
669
	 * It is strongly recommended to set the 'lag' and 'since' fields to avoid race conditions
670
	 * that can cause stale values to get stuck at keys. Usually, callbacks ignore the current
671
	 * value, but it can be used to maintain "most recent X" values that come from time or
672
	 * sequence based source data, provided that the "as of" id/time is tracked. Note that
673
	 * preemptive regeneration and $checkKeys can result in a non-false current value.
674
	 *
675
	 * Usage of $checkKeys is similar to get() and getMulti(). However, rather than the caller
676
	 * having to inspect a "current time left" variable (e.g. $curTTL, $curTTLs), a cache
677
	 * regeneration will automatically be triggered using the callback.
678
	 *
679
	 * The simplest way to avoid stampedes for hot keys is to use
680
	 * the 'lockTSE' option in $opts. If cache purges are needed, also:
681
	 *   - a) Pass $key into $checkKeys
682
	 *   - b) Use touchCheckKey( $key ) instead of delete( $key )
683
	 *
684
	 * Example usage (typical key):
685
	 * @code
686
	 *     $catInfo = $cache->getWithSetCallback(
687
	 *         // Key to store the cached value under
688
	 *         $cache->makeKey( 'cat-attributes', $catId ),
689
	 *         // Time-to-live (in seconds)
690
	 *         $cache::TTL_MINUTE,
691
	 *         // Function that derives the new key value
692
	 *         function ( $oldValue, &$ttl, array &$setOpts ) {
693
	 *             $dbr = wfGetDB( DB_REPLICA );
694
	 *             // Account for any snapshot/replica DB lag
695
	 *             $setOpts += Database::getCacheSetOptions( $dbr );
696
	 *
697
	 *             return $dbr->selectRow( ... );
698
	 *        }
699
	 *     );
700
	 * @endcode
701
	 *
702
	 * Example usage (key that is expensive and hot):
703
	 * @code
704
	 *     $catConfig = $cache->getWithSetCallback(
705
	 *         // Key to store the cached value under
706
	 *         $cache->makeKey( 'site-cat-config' ),
707
	 *         // Time-to-live (in seconds)
708
	 *         $cache::TTL_DAY,
709
	 *         // Function that derives the new key value
710
	 *         function ( $oldValue, &$ttl, array &$setOpts ) {
711
	 *             $dbr = wfGetDB( DB_REPLICA );
712
	 *             // Account for any snapshot/replica DB lag
713
	 *             $setOpts += Database::getCacheSetOptions( $dbr );
714
	 *
715
	 *             return CatConfig::newFromRow( $dbr->selectRow( ... ) );
716
	 *         },
717
	 *         [
718
	 *             // Calling touchCheckKey() on this key invalidates the cache
719
	 *             'checkKeys' => [ $cache->makeKey( 'site-cat-config' ) ],
720
	 *             // Try to only let one datacenter thread manage cache updates at a time
721
	 *             'lockTSE' => 30,
722
	 *             // Avoid querying cache servers multiple times in a web request
723
	 *             'pcTTL' => $cache::TTL_PROC_LONG
724
	 *         ]
725
	 *     );
726
	 * @endcode
727
	 *
728
	 * Example usage (key with dynamic dependencies):
729
	 * @code
730
	 *     $catState = $cache->getWithSetCallback(
731
	 *         // Key to store the cached value under
732
	 *         $cache->makeKey( 'cat-state', $cat->getId() ),
733
	 *         // Time-to-live (seconds)
734
	 *         $cache::TTL_HOUR,
735
	 *         // Function that derives the new key value
736
	 *         function ( $oldValue, &$ttl, array &$setOpts ) {
737
	 *             // Determine new value from the DB
738
	 *             $dbr = wfGetDB( DB_REPLICA );
739
	 *             // Account for any snapshot/replica DB lag
740
	 *             $setOpts += Database::getCacheSetOptions( $dbr );
741
	 *
742
	 *             return CatState::newFromResults( $dbr->select( ... ) );
743
	 *         },
744
	 *         [
745
	 *              // The "check" keys that represent things the value depends on;
746
	 *              // Calling touchCheckKey() on any of them invalidates the cache
747
	 *             'checkKeys' => [
748
	 *                 $cache->makeKey( 'sustenance-bowls', $cat->getRoomId() ),
749
	 *                 $cache->makeKey( 'people-present', $cat->getHouseId() ),
750
	 *                 $cache->makeKey( 'cat-laws', $cat->getCityId() ),
751
	 *             ]
752
	 *         ]
753
	 *     );
754
	 * @endcode
755
	 *
756
	 * Example usage (hot key holding most recent 100 events):
757
	 * @code
758
	 *     $lastCatActions = $cache->getWithSetCallback(
759
	 *         // Key to store the cached value under
760
	 *         $cache->makeKey( 'cat-last-actions', 100 ),
761
	 *         // Time-to-live (in seconds)
762
	 *         10,
763
	 *         // Function that derives the new key value
764
	 *         function ( $oldValue, &$ttl, array &$setOpts ) {
765
	 *             $dbr = wfGetDB( DB_REPLICA );
766
	 *             // Account for any snapshot/replica DB lag
767
	 *             $setOpts += Database::getCacheSetOptions( $dbr );
768
	 *
769
	 *             // Start off with the last cached list
770
	 *             $list = $oldValue ?: [];
771
	 *             // Fetch the last 100 relevant rows in descending order;
772
	 *             // only fetch rows newer than $list[0] to reduce scanning
773
	 *             $rows = iterator_to_array( $dbr->select( ... ) );
774
	 *             // Merge them and get the new "last 100" rows
775
	 *             return array_slice( array_merge( $new, $list ), 0, 100 );
776
	 *        },
777
	 *        [
778
	 *             // Try to only let one datacenter thread manage cache updates at a time
779
	 *             'lockTSE' => 30,
780
	 *             // Use a magic value when no cache value is ready rather than stampeding
781
	 *             'busyValue' => 'computing'
782
	 *        ]
783
	 *     );
784
	 * @endcode
785
	 *
786
	 * @see WANObjectCache::get()
787
	 * @see WANObjectCache::set()
788
	 *
789
	 * @param string $key Cache key
790
	 * @param integer $ttl Seconds to live for key updates. Special values are:
791
	 *   - WANObjectCache::TTL_INDEFINITE: Cache forever
792
	 *   - WANObjectCache::TTL_UNCACHEABLE: Do not cache at all
793
	 * @param callable $callback Value generation function
794
	 * @param array $opts Options map:
795
	 *   - checkKeys: List of "check" keys. The key at $key will be seen as invalid when either
796
	 *      touchCheckKey() or resetCheckKey() is called on any of these keys.
797
	 *      Default: [].
798
	 *   - lockTSE: If the key is tombstoned or expired (by checkKeys) less than this many seconds
799
	 *      ago, then try to have a single thread handle cache regeneration at any given time.
800
	 *      Other threads will try to use stale values if possible. If, on miss, the time since
801
	 *      expiration is low, the assumption is that the key is hot and that a stampede is worth
802
	 *      avoiding. Setting this above WANObjectCache::HOLDOFF_TTL makes no difference. The
803
	 *      higher this is set, the higher the worst-case staleness can be.
804
	 *      Use WANObjectCache::TSE_NONE to disable this logic.
805
	 *      Default: WANObjectCache::TSE_NONE.
806
	 *   - busyValue: If no value exists and another thread is currently regenerating it, use this
807
	 *      as a fallback value (or a callback to generate such a value). This assures that cache
808
	 *      stampedes cannot happen if the value falls out of cache. This can be used as insurance
809
	 *      against cache regeneration becoming very slow for some reason (greater than the TTL).
810
	 *      Default: null.
811
	 *   - pcTTL: Process cache the value in this PHP instance for this many seconds. This avoids
812
	 *      network I/O when a key is read several times. This will not cache when the callback
813
	 *      returns false, however. Note that any purges will not be seen while process cached;
814
	 *      since the callback should use replica DBs and they may be lagged or have snapshot
815
	 *      isolation anyway, this should not typically matter.
816
	 *      Default: WANObjectCache::TTL_UNCACHEABLE.
817
	 *   - pcGroup: Process cache group to use instead of the primary one. If set, this must be
818
	 *      of the format ALPHANUMERIC_NAME:MAX_KEY_SIZE, e.g. "mydata:10". Use this for storing
819
	 *      large values, small yet numerous values, or some values with a high cost of eviction.
820
	 *      It is generally preferable to use a class constant when setting this value.
821
	 *      This has no effect unless pcTTL is used.
822
	 *      Default: WANObjectCache::PC_PRIMARY.
823
	 *   - version: Integer version number. This allows for callers to make breaking changes to
824
	 *      how values are stored while maintaining compatability and correct cache purges. New
825
	 *      versions are stored alongside older versions concurrently. Avoid storing class objects
826
	 *      however, as this reduces compatibility (due to serialization).
827
	 *      Default: null.
828
	 *   - minAsOf: Reject values if they were generated before this UNIX timestamp.
829
	 *      This is useful if the source of a key is suspected of having possibly changed
830
	 *      recently, and the caller wants any such changes to be reflected.
831
	 *      Default: WANObjectCache::MIN_TIMESTAMP_NONE.
832
	 *   - hotTTR: Expected time-till-refresh for keys that average ~1 hit/second.
833
	 *      This should be greater than "ageNew". Keys with higher hit rates will regenerate
834
	 *      more often. This is useful when a popular key is changed but the cache purge was
835
	 *      delayed or lost. Seldom used keys are rarely affected by this setting, unless an
836
	 *      extremely low "hotTTR" value is passed in.
837
	 *      Default: WANObjectCache::HOT_TTR.
838
	 *   - lowTTL: Consider pre-emptive updates when the current TTL (seconds) of the key is less
839
	 *      than this. It becomes more likely over time, becoming certain once the key is expired.
840
	 *      Default: WANObjectCache::LOW_TTL.
841
	 *   - ageNew: Consider popularity refreshes only once a key reaches this age in seconds.
842
	 *      Default: WANObjectCache::AGE_NEW.
843
	 * @return mixed Value found or written to the key
844
	 * @note Options added in 1.28: version, busyValue, hotTTR, ageNew, pcGroup, minAsOf
845
	 * @note Callable type hints are not used to avoid class-autoloading
846
	 */
847
	final public function getWithSetCallback( $key, $ttl, $callback, array $opts = [] ) {
848
		$pcTTL = isset( $opts['pcTTL'] ) ? $opts['pcTTL'] : self::TTL_UNCACHEABLE;
849
850
		// Try the process cache if enabled
851
		if ( $pcTTL >= 0 ) {
852
			$group = isset( $opts['pcGroup'] ) ? $opts['pcGroup'] : self::PC_PRIMARY;
853
			$procCache = $this->getProcessCache( $group );
854
			$value = $procCache->get( $key );
855
		} else {
856
			$procCache = false;
857
			$value = false;
858
		}
859
860
		if ( $value === false ) {
861
			// Fetch the value over the network
862
			if ( isset( $opts['version'] ) ) {
863
				$version = $opts['version'];
864
				$asOf = null;
865
				$cur = $this->doGetWithSetCallback(
866
					$key,
867
					$ttl,
868
					function ( $oldValue, &$ttl, &$setOpts, $oldAsOf )
869
					use ( $callback, $version ) {
870
						if ( is_array( $oldValue )
871
							&& array_key_exists( self::VFLD_DATA, $oldValue )
872
						) {
873
							$oldData = $oldValue[self::VFLD_DATA];
874
						} else {
875
							// VFLD_DATA is not set if an old, unversioned, key is present
876
							$oldData = false;
877
						}
878
879
						return [
880
							self::VFLD_DATA => $callback( $oldData, $ttl, $setOpts, $oldAsOf ),
881
							self::VFLD_VERSION => $version
882
						];
883
					},
884
					$opts,
885
					$asOf
886
				);
887
				if ( $cur[self::VFLD_VERSION] === $version ) {
888
					// Value created or existed before with version; use it
889
					$value = $cur[self::VFLD_DATA];
890
				} else {
891
					// Value existed before with a different version; use variant key.
892
					// Reflect purges to $key by requiring that this key value be newer.
893
					$value = $this->doGetWithSetCallback(
894
						'cache-variant:' . md5( $key ) . ":$version",
895
						$ttl,
896
						$callback,
897
						// Regenerate value if not newer than $key
898
						[ 'version' => null, 'minAsOf' => $asOf ] + $opts
899
					);
900
				}
901
			} else {
902
				$value = $this->doGetWithSetCallback( $key, $ttl, $callback, $opts );
903
			}
904
905
			// Update the process cache if enabled
906
			if ( $procCache && $value !== false ) {
907
				$procCache->set( $key, $value, $pcTTL );
908
			}
909
		}
910
911
		return $value;
912
	}
913
914
	/**
915
	 * Do the actual I/O for getWithSetCallback() when needed
916
	 *
917
	 * @see WANObjectCache::getWithSetCallback()
918
	 *
919
	 * @param string $key
920
	 * @param integer $ttl
921
	 * @param callback $callback
922
	 * @param array $opts Options map for getWithSetCallback()
923
	 * @param float &$asOf Cache generation timestamp of returned value [returned]
924
	 * @return mixed
925
	 * @note Callable type hints are not used to avoid class-autoloading
926
	 */
927
	protected function doGetWithSetCallback( $key, $ttl, $callback, array $opts, &$asOf = null ) {
928
		$lowTTL = isset( $opts['lowTTL'] ) ? $opts['lowTTL'] : min( self::LOW_TTL, $ttl );
929
		$lockTSE = isset( $opts['lockTSE'] ) ? $opts['lockTSE'] : self::TSE_NONE;
930
		$checkKeys = isset( $opts['checkKeys'] ) ? $opts['checkKeys'] : [];
931
		$busyValue = isset( $opts['busyValue'] ) ? $opts['busyValue'] : null;
932
		$popWindow = isset( $opts['hotTTR'] ) ? $opts['hotTTR'] : self::HOT_TTR;
933
		$ageNew = isset( $opts['ageNew'] ) ? $opts['ageNew'] : self::AGE_NEW;
934
		$minTime = isset( $opts['minAsOf'] ) ? $opts['minAsOf'] : self::MIN_TIMESTAMP_NONE;
935
		$versioned = isset( $opts['version'] );
936
937
		// Get the current key value
938
		$curTTL = null;
939
		$cValue = $this->get( $key, $curTTL, $checkKeys, $asOf ); // current value
940
		$value = $cValue; // return value
941
942
		// Determine if a regeneration is desired
943
		if ( $value !== false
944
			&& $curTTL > 0
945
			&& $this->isValid( $value, $versioned, $asOf, $minTime )
946
			&& !$this->worthRefreshExpiring( $curTTL, $lowTTL )
947
			&& !$this->worthRefreshPopular( $asOf, $ageNew, $popWindow )
948
		) {
949
			return $value;
950
		}
951
952
		// A deleted key with a negative TTL left must be tombstoned
953
		$isTombstone = ( $curTTL !== null && $value === false );
954
		// Assume a key is hot if requested soon after invalidation
955
		$isHot = ( $curTTL !== null && $curTTL <= 0 && abs( $curTTL ) <= $lockTSE );
956
		// Use the mutex if there is no value and a busy fallback is given
957
		$checkBusy = ( $busyValue !== null && $value === false );
958
		// Decide whether a single thread should handle regenerations.
959
		// This avoids stampedes when $checkKeys are bumped and when preemptive
960
		// renegerations take too long. It also reduces regenerations while $key
961
		// is tombstoned. This balances cache freshness with avoiding DB load.
962
		$useMutex = ( $isHot || ( $isTombstone && $lockTSE > 0 ) || $checkBusy );
963
964
		$lockAcquired = false;
965
		if ( $useMutex ) {
966
			// Acquire a datacenter-local non-blocking lock
967
			if ( $this->cache->add( self::MUTEX_KEY_PREFIX . $key, 1, self::LOCK_TTL ) ) {
968
				// Lock acquired; this thread should update the key
969
				$lockAcquired = true;
970
			} elseif ( $value !== false && $this->isValid( $value, $versioned, $asOf, $minTime ) ) {
971
				// If it cannot be acquired; then the stale value can be used
972
				return $value;
973
			} else {
974
				// Use the INTERIM value for tombstoned keys to reduce regeneration load.
975
				// For hot keys, either another thread has the lock or the lock failed;
976
				// use the INTERIM value from the last thread that regenerated it.
977
				$wrapped = $this->cache->get( self::INTERIM_KEY_PREFIX . $key );
978
				list( $value ) = $this->unwrap( $wrapped, microtime( true ) );
979
				if ( $value !== false && $this->isValid( $value, $versioned, $asOf, $minTime ) ) {
980
					$asOf = $wrapped[self::FLD_TIME];
981
982
					return $value;
983
				}
984
				// Use the busy fallback value if nothing else
985
				if ( $busyValue !== null ) {
986
					return is_callable( $busyValue ) ? $busyValue() : $busyValue;
987
				}
988
			}
989
		}
990
991
		if ( !is_callable( $callback ) ) {
992
			throw new InvalidArgumentException( "Invalid cache miss callback provided." );
993
		}
994
995
		// Generate the new value from the callback...
996
		$setOpts = [];
997
		$value = call_user_func_array( $callback, [ $cValue, &$ttl, &$setOpts, $asOf ] );
998
		// When delete() is called, writes are write-holed by the tombstone,
999
		// so use a special INTERIM key to pass the new value around threads.
1000
		if ( ( $isTombstone && $lockTSE > 0 ) && $value !== false && $ttl >= 0 ) {
1001
			$tempTTL = max( 1, (int)$lockTSE ); // set() expects seconds
1002
			$newAsOf = microtime( true );
1003
			$wrapped = $this->wrap( $value, $tempTTL, $newAsOf );
1004
			// Avoid using set() to avoid pointless mcrouter broadcasting
1005
			$this->cache->merge(
1006
				self::INTERIM_KEY_PREFIX . $key,
1007
				function () use ( $wrapped ) {
1008
					return $wrapped;
1009
				},
1010
				$tempTTL,
1011
				1
1012
			);
1013
		}
1014
1015
		if ( $value !== false && $ttl >= 0 ) {
1016
			// Update the cache; this will fail if the key is tombstoned
1017
			$setOpts['lockTSE'] = $lockTSE;
1018
			$this->set( $key, $value, $ttl, $setOpts );
1019
		}
1020
1021
		if ( $lockAcquired ) {
1022
			// Avoid using delete() to avoid pointless mcrouter broadcasting
1023
			$this->cache->changeTTL( self::MUTEX_KEY_PREFIX . $key, 1 );
1024
		}
1025
1026
		return $value;
1027
	}
1028
1029
	/**
1030
	 * Method to fetch/regenerate multiple cache keys at once
1031
	 *
1032
	 * This works the same as getWithSetCallback() except:
1033
	 *   - a) The $keys argument expects the result of WANObjectCache::makeMultiKeys()
1034
	 *   - b) The $callback argument expects a callback taking the following arguments:
1035
	 *         - $id: ID of an entity to query
1036
	 *         - $oldValue : the prior cache value or false if none was present
1037
	 *         - &$ttl : a reference to the new value TTL in seconds
1038
	 *         - &$setOpts : a reference to options for set() which can be altered
1039
	 *         - $oldAsOf : generation UNIX timestamp of $oldValue or null if not present
1040
	 *        Aside from the additional $id argument, the other arguments function the same
1041
	 *        way they do in getWithSetCallback().
1042
	 *   - c) The return value is a map of (cache key => value) in the order of $keyedIds
1043
	 *
1044
	 * @see WANObjectCache::getWithSetCallback()
1045
	 *
1046
	 * Example usage:
1047
	 * @code
1048
	 *     $rows = $cache->getMultiWithSetCallback(
1049
	 *         // Map of cache keys to entitiy IDs
1050
	 *         $cache->makeMultiKeys(
1051
	 *             $this->fileVersionIds(),
1052
	 *             function ( $id, WANObjectCache $cache ) {
1053
	 *                 return $cache->makeKey( 'file-version', $id );
1054
	 *             }
1055
	 *         ),
1056
	 *         // Time-to-live (in seconds)
1057
	 *         $cache::TTL_DAY,
1058
	 *         // Function that derives the new key value
1059
	 *         return function ( $id, $oldValue, &$ttl, array &$setOpts ) {
1060
	 *             $dbr = wfGetDB( DB_REPLICA );
1061
	 *             // Account for any snapshot/replica DB lag
1062
	 *             $setOpts += Database::getCacheSetOptions( $dbr );
1063
	 *
1064
	 *             // Load the row for this file
1065
	 *             $row = $dbr->selectRow( 'file', '*', [ 'id' => $id ], __METHOD__ );
1066
	 *
1067
	 *             return $row ? (array)$row : false;
1068
	 *         },
1069
	 *         [
1070
	 *             // Process cache for 30 seconds
1071
	 *             'pcTTL' => 30,
1072
	 *             // Use a dedicated 500 item cache (initialized on-the-fly)
1073
	 *             'pcGroup' => 'file-versions:500'
1074
	 *         ]
1075
	 *     );
1076
	 *     $files = array_map( [ __CLASS__, 'newFromRow' ], $rows );
1077
	 * @endcode
1078
	 *
1079
	 * @param ArrayIterator $keyedIds Result of WANObjectCache::makeMultiKeys()
1080
	 * @param integer $ttl Seconds to live for key updates
1081
	 * @param callable $callback Callback the yields entity regeneration callbacks
1082
	 * @param array $opts Options map
1083
	 * @return array Map of (cache key => value) in the same order as $keyedIds
1084
	 * @since 1.28
1085
	 */
1086
	final public function getMultiWithSetCallback(
1087
		ArrayIterator $keyedIds, $ttl, callable $callback, array $opts = []
1088
	) {
1089
		$keysWarmUp = iterator_to_array( $keyedIds, true );
1090
		$checkKeys = isset( $opts['checkKeys'] ) ? $opts['checkKeys'] : [];
1091
		foreach ( $checkKeys as $i => $checkKeyOrKeys ) {
1092
			if ( is_int( $i ) ) {
1093
				$keysWarmUp[] = $checkKeyOrKeys;
1094
			} else {
1095
				$keysWarmUp = array_merge( $keysWarmUp, $checkKeyOrKeys );
1096
			}
1097
		}
1098
1099
		$this->warmupCache = $this->cache->getMulti( $keysWarmUp );
1100
		$this->warmupCache += array_fill_keys( $keysWarmUp, false );
1101
1102
		// Wrap $callback to match the getWithSetCallback() format while passing $id to $callback
1103
		$id = null;
1104
		$func = function ( $oldValue, &$ttl, array $setOpts, $oldAsOf ) use ( $callback, &$id ) {
1105
			return $callback( $id, $oldValue, $ttl, $setOpts, $oldAsOf );
1106
		};
1107
1108
		$values = [];
1109
		foreach ( $keyedIds as $key => $id ) {
1110
			$values[$key] = $this->getWithSetCallback( $key, $ttl, $func, $opts );
1111
		}
1112
1113
		$this->warmupCache = [];
1114
1115
		return $values;
1116
	}
1117
1118
	/**
1119
	 * @see BagOStuff::makeKey()
1120
	 * @param string ... Key component
1121
	 * @return string
1122
	 * @since 1.27
1123
	 */
1124
	public function makeKey() {
1125
		return call_user_func_array( [ $this->cache, __FUNCTION__ ], func_get_args() );
1126
	}
1127
1128
	/**
1129
	 * @see BagOStuff::makeGlobalKey()
1130
	 * @param string ... Key component
1131
	 * @return string
1132
	 * @since 1.27
1133
	 */
1134
	public function makeGlobalKey() {
1135
		return call_user_func_array( [ $this->cache, __FUNCTION__ ], func_get_args() );
1136
	}
1137
1138
	/**
1139
	 * @param array $entities List of entity IDs
1140
	 * @param callable $keyFunc Callback yielding a key from (entity ID, this WANObjectCache)
1141
	 * @return ArrayIterator Iterator yielding (cache key => entity ID) in $entities order
1142
	 * @since 1.28
1143
	 */
1144
	public function makeMultiKeys( array $entities, callable $keyFunc ) {
1145
		$map = [];
1146
		foreach ( $entities as $entity ) {
1147
			$map[$keyFunc( $entity, $this )] = $entity;
1148
		}
1149
1150
		return new ArrayIterator( $map );
1151
	}
1152
1153
	/**
1154
	 * Get the "last error" registered; clearLastError() should be called manually
1155
	 * @return int ERR_* class constant for the "last error" registry
1156
	 */
1157
	final public function getLastError() {
1158
		if ( $this->lastRelayError ) {
1159
			// If the cache and the relayer failed, focus on the latter.
1160
			// An update not making it to the relayer means it won't show up
1161
			// in other DCs (nor will consistent re-hashing see up-to-date values).
1162
			// On the other hand, if just the cache update failed, then it should
1163
			// eventually be applied by the relayer.
1164
			return $this->lastRelayError;
1165
		}
1166
1167
		$code = $this->cache->getLastError();
1168
		switch ( $code ) {
1169
			case BagOStuff::ERR_NONE:
1170
				return self::ERR_NONE;
1171
			case BagOStuff::ERR_NO_RESPONSE:
1172
				return self::ERR_NO_RESPONSE;
1173
			case BagOStuff::ERR_UNREACHABLE:
1174
				return self::ERR_UNREACHABLE;
1175
			default:
1176
				return self::ERR_UNEXPECTED;
1177
		}
1178
	}
1179
1180
	/**
1181
	 * Clear the "last error" registry
1182
	 */
1183
	final public function clearLastError() {
1184
		$this->cache->clearLastError();
1185
		$this->lastRelayError = self::ERR_NONE;
1186
	}
1187
1188
	/**
1189
	 * Clear the in-process caches; useful for testing
1190
	 *
1191
	 * @since 1.27
1192
	 */
1193
	public function clearProcessCache() {
1194
		$this->processCaches = [];
1195
	}
1196
1197
	/**
1198
	 * @param integer $flag ATTR_* class constant
1199
	 * @return integer QOS_* class constant
1200
	 * @since 1.28
1201
	 */
1202
	public function getQoS( $flag ) {
1203
		return $this->cache->getQoS( $flag );
1204
	}
1205
1206
	/**
1207
	 * Get a TTL that is higher for objects that have not changed recently
1208
	 *
1209
	 * This is useful for keys that get explicit purges and DB or purge relay
1210
	 * lag is a potential concern (especially how it interacts with CDN cache)
1211
	 *
1212
	 * Example usage:
1213
	 * @code
1214
	 *     // Last-modified time of page
1215
	 *     $mtime = wfTimestamp( TS_UNIX, $page->getTimestamp() );
1216
	 *     // Get adjusted TTL. If $mtime is 3600 seconds ago and $minTTL/$factor left at
1217
	 *     // defaults, then $ttl is 3600 * .2 = 720. If $minTTL was greater than 720, then
1218
	 *     // $ttl would be $minTTL. If $maxTTL was smaller than 720, $ttl would be $maxTTL.
1219
	 *     $ttl = $cache->adaptiveTTL( $mtime, $cache::TTL_DAY );
1220
	 * @endcode
1221
	 *
1222
	 * @param integer|float $mtime UNIX timestamp
1223
	 * @param integer $maxTTL Maximum TTL (seconds)
1224
	 * @param integer $minTTL Minimum TTL (seconds); Default: 30
1225
	 * @param float $factor Value in the range (0,1); Default: .2
1226
	 * @return integer Adaptive TTL
1227
	 * @since 1.28
1228
	 */
1229
	public function adaptiveTTL( $mtime, $maxTTL, $minTTL = 30, $factor = .2 ) {
1230
		if ( is_float( $mtime ) || ctype_digit( $mtime ) ) {
1231
			$mtime = (int)$mtime; // handle fractional seconds and string integers
1232
		}
1233
1234
		if ( !is_int( $mtime ) || $mtime <= 0 ) {
1235
			return $minTTL; // no last-modified time provided
1236
		}
1237
1238
		$age = time() - $mtime;
1239
1240
		return (int)min( $maxTTL, max( $minTTL, $factor * $age ) );
1241
	}
1242
1243
	/**
1244
	 * Do the actual async bus purge of a key
1245
	 *
1246
	 * This must set the key to "PURGED:<UNIX timestamp>:<holdoff>"
1247
	 *
1248
	 * @param string $key Cache key
1249
	 * @param integer $ttl How long to keep the tombstone [seconds]
1250
	 * @param integer $holdoff HOLDOFF_* constant controlling how long to ignore sets for this key
1251
	 * @return bool Success
1252
	 */
1253
	protected function relayPurge( $key, $ttl, $holdoff ) {
1254
		if ( $this->purgeRelayer instanceof EventRelayerNull ) {
1255
			// This handles the mcrouter and the single-DC case
1256
			$ok = $this->cache->set( $key,
1257
				$this->makePurgeValue( microtime( true ), self::HOLDOFF_NONE ),
1258
				$ttl
1259
			);
1260
		} else {
1261
			$event = $this->cache->modifySimpleRelayEvent( [
1262
				'cmd' => 'set',
1263
				'key' => $key,
1264
				'val' => 'PURGED:$UNIXTIME$:' . (int)$holdoff,
1265
				'ttl' => max( $ttl, 1 ),
1266
				'sbt' => true, // substitute $UNIXTIME$ with actual microtime
1267
			] );
1268
1269
			$ok = $this->purgeRelayer->notify( $this->purgeChannel, $event );
1270
			if ( !$ok ) {
1271
				$this->lastRelayError = self::ERR_RELAY;
1272
			}
1273
		}
1274
1275
		return $ok;
1276
	}
1277
1278
	/**
1279
	 * Do the actual async bus delete of a key
1280
	 *
1281
	 * @param string $key Cache key
1282
	 * @return bool Success
1283
	 */
1284
	protected function relayDelete( $key ) {
1285
		if ( $this->purgeRelayer instanceof EventRelayerNull ) {
1286
			// This handles the mcrouter and the single-DC case
1287
			$ok = $this->cache->delete( $key );
1288
		} else {
1289
			$event = $this->cache->modifySimpleRelayEvent( [
1290
				'cmd' => 'delete',
1291
				'key' => $key,
1292
			] );
1293
1294
			$ok = $this->purgeRelayer->notify( $this->purgeChannel, $event );
1295
			if ( !$ok ) {
1296
				$this->lastRelayError = self::ERR_RELAY;
1297
			}
1298
		}
1299
1300
		return $ok;
1301
	}
1302
1303
	/**
1304
	 * Check if a key should be regenerated (using random probability)
1305
	 *
1306
	 * This returns false if $curTTL >= $lowTTL. Otherwise, the chance
1307
	 * of returning true increases steadily from 0% to 100% as the $curTTL
1308
	 * moves from $lowTTL to 0 seconds. This handles widely varying
1309
	 * levels of cache access traffic.
1310
	 *
1311
	 * @param float $curTTL Approximate TTL left on the key if present
1312
	 * @param float $lowTTL Consider a refresh when $curTTL is less than this
1313
	 * @return bool
1314
	 */
1315
	protected function worthRefreshExpiring( $curTTL, $lowTTL ) {
1316
		if ( $curTTL >= $lowTTL ) {
1317
			return false;
1318
		} elseif ( $curTTL <= 0 ) {
1319
			return true;
1320
		}
1321
1322
		$chance = ( 1 - $curTTL / $lowTTL );
1323
1324
		return mt_rand( 1, 1e9 ) <= 1e9 * $chance;
1325
	}
1326
1327
	/**
1328
	 * Check if a key is due for randomized regeneration due to its popularity
1329
	 *
1330
	 * This is used so that popular keys can preemptively refresh themselves for higher
1331
	 * consistency (especially in the case of purge loss/delay). Unpopular keys can remain
1332
	 * in cache with their high nominal TTL. This means popular keys keep good consistency,
1333
	 * whether the data changes frequently or not, and long-tail keys get to stay in cache
1334
	 * and get hits too. Similar to worthRefreshExpiring(), randomization is used.
1335
	 *
1336
	 * @param float $asOf UNIX timestamp of the value
1337
	 * @param integer $ageNew Age of key when this might recommend refreshing (seconds)
1338
	 * @param integer $timeTillRefresh Age of key when it should be refreshed if popular (seconds)
1339
	 * @return bool
1340
	 */
1341
	protected function worthRefreshPopular( $asOf, $ageNew, $timeTillRefresh ) {
1342
		$age = microtime( true ) - $asOf;
1343
		$timeOld = $age - $ageNew;
1344
		if ( $timeOld <= 0 ) {
1345
			return false;
1346
		}
1347
1348
		// Lifecycle is: new, ramp-up refresh chance, full refresh chance
1349
		$refreshWindowSec = max( $timeTillRefresh - $ageNew - self::RAMPUP_TTL / 2, 1 );
1350
		// P(refresh) * (# hits in $refreshWindowSec) = (expected # of refreshes)
1351
		// P(refresh) * ($refreshWindowSec * $popularHitsPerSec) = 1
1352
		// P(refresh) = 1/($refreshWindowSec * $popularHitsPerSec)
1353
		$chance = 1 / ( self::HIT_RATE_HIGH * $refreshWindowSec );
1354
1355
		// Ramp up $chance from 0 to its nominal value over RAMPUP_TTL seconds to avoid stampedes
1356
		$chance *= ( $timeOld <= self::RAMPUP_TTL ) ? $timeOld / self::RAMPUP_TTL : 1;
1357
1358
		return mt_rand( 1, 1e9 ) <= 1e9 * $chance;
1359
	}
1360
1361
	/**
1362
	 * Check whether $value is appropriately versioned and not older than $minTime (if set)
1363
	 *
1364
	 * @param array $value
1365
	 * @param bool $versioned
1366
	 * @param float $asOf The time $value was generated
1367
	 * @param float $minTime The last time the main value was generated (0.0 if unknown)
1368
	 * @return bool
1369
	 */
1370
	protected function isValid( $value, $versioned, $asOf, $minTime ) {
1371
		if ( $versioned && !isset( $value[self::VFLD_VERSION] ) ) {
1372
			return false;
1373
		} elseif ( $minTime > 0 && $asOf < $minTime ) {
1374
			return false;
1375
		}
1376
1377
		return true;
1378
	}
1379
1380
	/**
1381
	 * Do not use this method outside WANObjectCache
1382
	 *
1383
	 * @param mixed $value
1384
	 * @param integer $ttl [0=forever]
1385
	 * @param float $now Unix Current timestamp just before calling set()
1386
	 * @return array
1387
	 */
1388
	protected function wrap( $value, $ttl, $now ) {
1389
		return [
1390
			self::FLD_VERSION => self::VERSION,
1391
			self::FLD_VALUE => $value,
1392
			self::FLD_TTL => $ttl,
1393
			self::FLD_TIME => $now
1394
		];
1395
	}
1396
1397
	/**
1398
	 * Do not use this method outside WANObjectCache
1399
	 *
1400
	 * @param array|string|bool $wrapped
1401
	 * @param float $now Unix Current timestamp (preferrably pre-query)
1402
	 * @return array (mixed; false if absent/invalid, current time left)
1403
	 */
1404
	protected function unwrap( $wrapped, $now ) {
1405
		// Check if the value is a tombstone
1406
		$purge = self::parsePurgeValue( $wrapped );
1407
		if ( $purge !== false ) {
1408
			// Purged values should always have a negative current $ttl
1409
			$curTTL = min( $purge[self::FLD_TIME] - $now, self::TINY_NEGATIVE );
1410
			return [ false, $curTTL ];
1411
		}
1412
1413
		if ( !is_array( $wrapped ) // not found
1414
			|| !isset( $wrapped[self::FLD_VERSION] ) // wrong format
1415
			|| $wrapped[self::FLD_VERSION] !== self::VERSION // wrong version
1416
		) {
1417
			return [ false, null ];
1418
		}
1419
1420
		$flags = isset( $wrapped[self::FLD_FLAGS] ) ? $wrapped[self::FLD_FLAGS] : 0;
1421
		if ( ( $flags & self::FLG_STALE ) == self::FLG_STALE ) {
1422
			// Treat as expired, with the cache time as the expiration
1423
			$age = $now - $wrapped[self::FLD_TIME];
1424
			$curTTL = min( -$age, self::TINY_NEGATIVE );
1425
		} elseif ( $wrapped[self::FLD_TTL] > 0 ) {
1426
			// Get the approximate time left on the key
1427
			$age = $now - $wrapped[self::FLD_TIME];
1428
			$curTTL = max( $wrapped[self::FLD_TTL] - $age, 0.0 );
1429
		} else {
1430
			// Key had no TTL, so the time left is unbounded
1431
			$curTTL = INF;
1432
		}
1433
1434
		return [ $wrapped[self::FLD_VALUE], $curTTL ];
1435
	}
1436
1437
	/**
1438
	 * @param array $keys
1439
	 * @param string $prefix
1440
	 * @return string[]
1441
	 */
1442
	protected static function prefixCacheKeys( array $keys, $prefix ) {
1443
		$res = [];
1444
		foreach ( $keys as $key ) {
1445
			$res[] = $prefix . $key;
1446
		}
1447
1448
		return $res;
1449
	}
1450
1451
	/**
1452
	 * @param string $value Wrapped value like "PURGED:<timestamp>:<holdoff>"
1453
	 * @return array|bool Array containing a UNIX timestamp (float) and holdoff period (integer),
1454
	 *  or false if value isn't a valid purge value
1455
	 */
1456
	protected static function parsePurgeValue( $value ) {
1457
		if ( !is_string( $value ) ) {
1458
			return false;
1459
		}
1460
		$segments = explode( ':', $value, 3 );
1461
		if ( !isset( $segments[0] ) || !isset( $segments[1] )
1462
			|| "{$segments[0]}:" !== self::PURGE_VAL_PREFIX
1463
		) {
1464
			return false;
1465
		}
1466
		if ( !isset( $segments[2] ) ) {
1467
			// Back-compat with old purge values without holdoff
1468
			$segments[2] = self::HOLDOFF_TTL;
1469
		}
1470
		return [
1471
			self::FLD_TIME => (float)$segments[1],
1472
			self::FLD_HOLDOFF => (int)$segments[2],
1473
		];
1474
	}
1475
1476
	/**
1477
	 * @param float $timestamp
1478
	 * @param int $holdoff In seconds
1479
	 * @return string Wrapped purge value
1480
	 */
1481
	protected function makePurgeValue( $timestamp, $holdoff ) {
1482
		return self::PURGE_VAL_PREFIX . (float)$timestamp . ':' . (int)$holdoff;
1483
	}
1484
1485
	/**
1486
	 * @param string $group
1487
	 * @return HashBagOStuff
1488
	 */
1489
	protected function getProcessCache( $group ) {
1490
		if ( !isset( $this->processCaches[$group] ) ) {
1491
			list( , $n ) = explode( ':', $group );
1492
			$this->processCaches[$group] = new HashBagOStuff( [ 'maxKeys' => (int)$n ] );
1493
		}
1494
1495
		return $this->processCaches[$group];
1496
	}
1497
}
1498