08424c9c8adb5e08ed0721982725ca3e6942a6fb
[lhc/web/wiklou.git] / includes / libs / objectcache / WANObjectCache.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 * @ingroup Cache
20 */
21
22 use Liuggio\StatsdClient\Factory\StatsdDataFactoryInterface;
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 three 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 * - c) Ignore the 'purge' EventRelayer configuration (default is NullEventRelayer)
57 * and set up dynomite as cache middleware between the web servers and either
58 * memcached or redis. This will also broadcast all key setting operations, not just purges,
59 * which can be useful for cache warming. Writes are eventually consistent via the
60 * Dynamo replication model (see https://github.com/Netflix/dynomite).
61 *
62 * Broadcasted operations like delete() and touchCheckKey() are done asynchronously
63 * in all datacenters this way, though the local one should likely be near immediate.
64 *
65 * This means that callers in all datacenters may see older values for however many
66 * milliseconds that the purge took to reach that datacenter. As with any cache, this
67 * should not be relied on for cases where reads are used to determine writes to source
68 * (e.g. non-cache) data stores, except when reading immutable data.
69 *
70 * All values are wrapped in metadata arrays. Keys use a "WANCache:" prefix
71 * to avoid collisions with keys that are not wrapped as metadata arrays. The
72 * prefixes are as follows:
73 * - a) "WANCache:v" : used for regular value keys
74 * - b) "WANCache:i" : used for temporarily storing values of tombstoned keys
75 * - c) "WANCache:t" : used for storing timestamp "check" keys
76 * - d) "WANCache:m" : used for temporary mutex keys to avoid cache stampedes
77 *
78 * @ingroup Cache
79 * @since 1.26
80 */
81 class WANObjectCache implements IExpiringStore, LoggerAwareInterface {
82 /** @var BagOStuff The local datacenter cache */
83 protected $cache;
84 /** @var HashBagOStuff[] Map of group PHP instance caches */
85 protected $processCaches = [];
86 /** @var string Purge channel name */
87 protected $purgeChannel;
88 /** @var EventRelayer Bus that handles purge broadcasts */
89 protected $purgeRelayer;
90 /** @var LoggerInterface */
91 protected $logger;
92 /** @var StatsdDataFactoryInterface */
93 protected $stats;
94 /** @var bool Whether to use "interim" caching while keys are tombstoned */
95 protected $useInterimHoldOffCaching = true;
96 /** @var callable|null Function that takes a WAN cache callback and runs it later */
97 protected $asyncHandler;
98
99 /** @var int ERR_* constant for the "last error" registry */
100 protected $lastRelayError = self::ERR_NONE;
101
102 /** @var int Callback stack depth for getWithSetCallback() */
103 private $callbackDepth = 0;
104 /** @var mixed[] Temporary warm-up cache */
105 private $warmupCache = [];
106 /** @var int Key fetched */
107 private $warmupKeyMisses = 0;
108
109 /** Max time expected to pass between delete() and DB commit finishing */
110 const MAX_COMMIT_DELAY = 3;
111 /** Max replication+snapshot lag before applying TTL_LAGGED or disallowing set() */
112 const MAX_READ_LAG = 7;
113 /** Seconds to tombstone keys on delete() */
114 const HOLDOFF_TTL = 11; // MAX_COMMIT_DELAY + MAX_READ_LAG + 1
115
116 /** Seconds to keep dependency purge keys around */
117 const CHECK_KEY_TTL = self::TTL_YEAR;
118 /** Seconds to keep interim value keys for tombstoned keys around */
119 const INTERIM_KEY_TTL = 1;
120
121 /** Seconds to keep lock keys around */
122 const LOCK_TTL = 10;
123 /** Default remaining TTL at which to consider pre-emptive regeneration */
124 const LOW_TTL = 30;
125 /** Default time-since-expiry on a miss that makes a key "hot" */
126 const LOCK_TSE = 1;
127
128 /** Never consider performing "popularity" refreshes until a key reaches this age */
129 const AGE_NEW = 60;
130 /** The time length of the "popularity" refresh window for hot keys */
131 const HOT_TTR = 900;
132 /** Hits/second for a refresh to be expected within the "popularity" window */
133 const HIT_RATE_HIGH = 1;
134 /** Seconds to ramp up to the "popularity" refresh chance after a key is no longer new */
135 const RAMPUP_TTL = 30;
136
137 /** Idiom for getWithSetCallback() callbacks to avoid calling set() */
138 const TTL_UNCACHEABLE = -1;
139 /** Idiom for getWithSetCallback() callbacks to 'lockTSE' logic */
140 const TSE_NONE = -1;
141 /** Max TTL to store keys when a data sourced is lagged */
142 const TTL_LAGGED = 30;
143 /** Idiom for delete() for "no hold-off" */
144 const HOLDOFF_NONE = 0;
145 /** Idiom for set()/getWithSetCallback() for "do not augment the storage medium TTL" */
146 const STALE_TTL_NONE = 0;
147 /** Idiom for set()/getWithSetCallback() for "no post-expired grace period" */
148 const GRACE_TTL_NONE = 0;
149
150 /** Idiom for getWithSetCallback() for "no minimum required as-of timestamp" */
151 const MIN_TIMESTAMP_NONE = 0.0;
152
153 /** Tiny negative float to use when CTL comes up >= 0 due to clock skew */
154 const TINY_NEGATIVE = -0.000001;
155
156 /** Cache format version number */
157 const VERSION = 1;
158
159 const FLD_VERSION = 0; // key to cache version number
160 const FLD_VALUE = 1; // key to the cached value
161 const FLD_TTL = 2; // key to the original TTL
162 const FLD_TIME = 3; // key to the cache time
163 const FLD_FLAGS = 4; // key to the flags bitfield
164 const FLD_HOLDOFF = 5; // key to any hold-off TTL
165
166 /** @var int Treat this value as expired-on-arrival */
167 const FLG_STALE = 1;
168
169 const ERR_NONE = 0; // no error
170 const ERR_NO_RESPONSE = 1; // no response
171 const ERR_UNREACHABLE = 2; // can't connect
172 const ERR_UNEXPECTED = 3; // response gave some error
173 const ERR_RELAY = 4; // relay broadcast failed
174
175 const VALUE_KEY_PREFIX = 'WANCache:v:';
176 const INTERIM_KEY_PREFIX = 'WANCache:i:';
177 const TIME_KEY_PREFIX = 'WANCache:t:';
178 const MUTEX_KEY_PREFIX = 'WANCache:m:';
179
180 const PURGE_VAL_PREFIX = 'PURGED:';
181
182 const VFLD_DATA = 'WOC:d'; // key to the value of versioned data
183 const VFLD_VERSION = 'WOC:v'; // key to the version of the value present
184
185 const PC_PRIMARY = 'primary:1000'; // process cache name and max key count
186
187 const DEFAULT_PURGE_CHANNEL = 'wancache-purge';
188
189 /**
190 * @param array $params
191 * - cache : BagOStuff object for a persistent cache
192 * - channels : Map of (action => channel string). Actions include "purge".
193 * - relayers : Map of (action => EventRelayer object). Actions include "purge".
194 * - logger : LoggerInterface object
195 * - stats : LoggerInterface object
196 * - asyncHandler : A function that takes a callback and runs it later. If supplied,
197 * whenever a preemptive refresh would be triggered in getWithSetCallback(), the
198 * current cache value is still used instead. However, the async-handler function
199 * receives a WAN cache callback that, when run, will execute the value generation
200 * callback supplied by the getWithSetCallback() caller. The result will be saved
201 * as normal. The handler is expected to call the WAN cache callback at an opportune
202 * time (e.g. HTTP post-send), though generally within a few 100ms. [optional]
203 */
204 public function __construct( array $params ) {
205 $this->cache = $params['cache'];
206 $this->purgeChannel = isset( $params['channels']['purge'] )
207 ? $params['channels']['purge']
208 : self::DEFAULT_PURGE_CHANNEL;
209 $this->purgeRelayer = isset( $params['relayers']['purge'] )
210 ? $params['relayers']['purge']
211 : new EventRelayerNull( [] );
212 $this->setLogger( isset( $params['logger'] ) ? $params['logger'] : new NullLogger() );
213 $this->stats = isset( $params['stats'] ) ? $params['stats'] : new NullStatsdDataFactory();
214 $this->asyncHandler = isset( $params['asyncHandler'] ) ? $params['asyncHandler'] : null;
215 }
216
217 /**
218 * @param LoggerInterface $logger
219 */
220 public function setLogger( LoggerInterface $logger ) {
221 $this->logger = $logger;
222 }
223
224 /**
225 * Get an instance that wraps EmptyBagOStuff
226 *
227 * @return WANObjectCache
228 */
229 public static function newEmpty() {
230 return new static( [
231 'cache' => new EmptyBagOStuff()
232 ] );
233 }
234
235 /**
236 * Fetch the value of a key from cache
237 *
238 * If supplied, $curTTL is set to the remaining TTL (current time left):
239 * - a) INF; if $key exists, has no TTL, and is not expired by $checkKeys
240 * - b) float (>=0); if $key exists, has a TTL, and is not expired by $checkKeys
241 * - c) float (<0); if $key is tombstoned, stale, or existing but expired by $checkKeys
242 * - d) null; if $key does not exist and is not tombstoned
243 *
244 * If a key is tombstoned, $curTTL will reflect the time since delete().
245 *
246 * The timestamp of $key will be checked against the last-purge timestamp
247 * of each of $checkKeys. Those $checkKeys not in cache will have the last-purge
248 * initialized to the current timestamp. If any of $checkKeys have a timestamp
249 * greater than that of $key, then $curTTL will reflect how long ago $key
250 * became invalid. Callers can use $curTTL to know when the value is stale.
251 * The $checkKeys parameter allow mass invalidations by updating a single key:
252 * - a) Each "check" key represents "last purged" of some source data
253 * - b) Callers pass in relevant "check" keys as $checkKeys in get()
254 * - c) When the source data that "check" keys represent changes,
255 * the touchCheckKey() method is called on them
256 *
257 * Source data entities might exists in a DB that uses snapshot isolation
258 * (e.g. the default REPEATABLE-READ in innoDB). Even for mutable data, that
259 * isolation can largely be maintained by doing the following:
260 * - a) Calling delete() on entity change *and* creation, before DB commit
261 * - b) Keeping transaction duration shorter than delete() hold-off TTL
262 *
263 * However, pre-snapshot values might still be seen if an update was made
264 * in a remote datacenter but the purge from delete() didn't relay yet.
265 *
266 * Consider using getWithSetCallback() instead of get() and set() cycles.
267 * That method has cache slam avoiding features for hot/expensive keys.
268 *
269 * @param string $key Cache key made from makeKey() or makeGlobalKey()
270 * @param mixed &$curTTL Approximate TTL left on the key if present/tombstoned [returned]
271 * @param array $checkKeys List of "check" keys
272 * @param float &$asOf UNIX timestamp of cached value; null on failure [returned]
273 * @return mixed Value of cache key or false on failure
274 */
275 final public function get( $key, &$curTTL = null, array $checkKeys = [], &$asOf = null ) {
276 $curTTLs = [];
277 $asOfs = [];
278 $values = $this->getMulti( [ $key ], $curTTLs, $checkKeys, $asOfs );
279 $curTTL = isset( $curTTLs[$key] ) ? $curTTLs[$key] : null;
280 $asOf = isset( $asOfs[$key] ) ? $asOfs[$key] : null;
281
282 return isset( $values[$key] ) ? $values[$key] : false;
283 }
284
285 /**
286 * Fetch the value of several keys from cache
287 *
288 * @see WANObjectCache::get()
289 *
290 * @param array $keys List of cache keys made from makeKey() or makeGlobalKey()
291 * @param array &$curTTLs Map of (key => approximate TTL left) for existing keys [returned]
292 * @param array $checkKeys List of check keys to apply to all $keys. May also apply "check"
293 * keys to specific cache keys only by using cache keys as keys in the $checkKeys array.
294 * @param float[] &$asOfs Map of (key => UNIX timestamp of cached value; null on failure)
295 * @return array Map of (key => value) for keys that exist and are not tombstoned
296 */
297 final public function getMulti(
298 array $keys, &$curTTLs = [], array $checkKeys = [], array &$asOfs = []
299 ) {
300 $result = [];
301 $curTTLs = [];
302 $asOfs = [];
303
304 $vPrefixLen = strlen( self::VALUE_KEY_PREFIX );
305 $valueKeys = self::prefixCacheKeys( $keys, self::VALUE_KEY_PREFIX );
306
307 $checkKeysForAll = [];
308 $checkKeysByKey = [];
309 $checkKeysFlat = [];
310 foreach ( $checkKeys as $i => $checkKeyGroup ) {
311 $prefixed = self::prefixCacheKeys( (array)$checkKeyGroup, self::TIME_KEY_PREFIX );
312 $checkKeysFlat = array_merge( $checkKeysFlat, $prefixed );
313 // Is this check keys for a specific cache key, or for all keys being fetched?
314 if ( is_int( $i ) ) {
315 $checkKeysForAll = array_merge( $checkKeysForAll, $prefixed );
316 } else {
317 $checkKeysByKey[$i] = isset( $checkKeysByKey[$i] )
318 ? array_merge( $checkKeysByKey[$i], $prefixed )
319 : $prefixed;
320 }
321 }
322
323 // Fetch all of the raw values
324 $keysGet = array_merge( $valueKeys, $checkKeysFlat );
325 if ( $this->warmupCache ) {
326 $wrappedValues = array_intersect_key( $this->warmupCache, array_flip( $keysGet ) );
327 $keysGet = array_diff( $keysGet, array_keys( $wrappedValues ) ); // keys left to fetch
328 $this->warmupKeyMisses += count( $keysGet );
329 } else {
330 $wrappedValues = [];
331 }
332 if ( $keysGet ) {
333 $wrappedValues += $this->cache->getMulti( $keysGet );
334 }
335 // Time used to compare/init "check" keys (derived after getMulti() to be pessimistic)
336 $now = $this->getCurrentTime();
337
338 // Collect timestamps from all "check" keys
339 $purgeValuesForAll = $this->processCheckKeys( $checkKeysForAll, $wrappedValues, $now );
340 $purgeValuesByKey = [];
341 foreach ( $checkKeysByKey as $cacheKey => $checks ) {
342 $purgeValuesByKey[$cacheKey] =
343 $this->processCheckKeys( $checks, $wrappedValues, $now );
344 }
345
346 // Get the main cache value for each key and validate them
347 foreach ( $valueKeys as $vKey ) {
348 if ( !isset( $wrappedValues[$vKey] ) ) {
349 continue; // not found
350 }
351
352 $key = substr( $vKey, $vPrefixLen ); // unprefix
353
354 list( $value, $curTTL ) = $this->unwrap( $wrappedValues[$vKey], $now );
355 if ( $value !== false ) {
356 $result[$key] = $value;
357
358 // Force dependant keys to be invalid for a while after purging
359 // to reduce race conditions involving stale data getting cached
360 $purgeValues = $purgeValuesForAll;
361 if ( isset( $purgeValuesByKey[$key] ) ) {
362 $purgeValues = array_merge( $purgeValues, $purgeValuesByKey[$key] );
363 }
364 foreach ( $purgeValues as $purge ) {
365 $safeTimestamp = $purge[self::FLD_TIME] + $purge[self::FLD_HOLDOFF];
366 if ( $safeTimestamp >= $wrappedValues[$vKey][self::FLD_TIME] ) {
367 // How long ago this value was expired by *this* check key
368 $ago = min( $purge[self::FLD_TIME] - $now, self::TINY_NEGATIVE );
369 // How long ago this value was expired by *any* known check key
370 $curTTL = min( $curTTL, $ago );
371 }
372 }
373 }
374 $curTTLs[$key] = $curTTL;
375 $asOfs[$key] = ( $value !== false ) ? $wrappedValues[$vKey][self::FLD_TIME] : null;
376 }
377
378 return $result;
379 }
380
381 /**
382 * @since 1.27
383 * @param array $timeKeys List of prefixed time check keys
384 * @param array $wrappedValues
385 * @param float $now
386 * @return array List of purge value arrays
387 */
388 private function processCheckKeys( array $timeKeys, array $wrappedValues, $now ) {
389 $purgeValues = [];
390 foreach ( $timeKeys as $timeKey ) {
391 $purge = isset( $wrappedValues[$timeKey] )
392 ? $this->parsePurgeValue( $wrappedValues[$timeKey] )
393 : false;
394 if ( $purge === false ) {
395 // Key is not set or invalid; regenerate
396 $newVal = $this->makePurgeValue( $now, self::HOLDOFF_TTL );
397 $this->cache->add( $timeKey, $newVal, self::CHECK_KEY_TTL );
398 $purge = $this->parsePurgeValue( $newVal );
399 }
400 $purgeValues[] = $purge;
401 }
402 return $purgeValues;
403 }
404
405 /**
406 * Set the value of a key in cache
407 *
408 * Simply calling this method when source data changes is not valid because
409 * the changes do not replicate to the other WAN sites. In that case, delete()
410 * should be used instead. This method is intended for use on cache misses.
411 *
412 * If the data was read from a snapshot-isolated transactions (e.g. the default
413 * REPEATABLE-READ in innoDB), use 'since' to avoid the following race condition:
414 * - a) T1 starts
415 * - b) T2 updates a row, calls delete(), and commits
416 * - c) The HOLDOFF_TTL passes, expiring the delete() tombstone
417 * - d) T1 reads the row and calls set() due to a cache miss
418 * - e) Stale value is stuck in cache
419 *
420 * Setting 'lag' and 'since' help avoids keys getting stuck in stale states.
421 *
422 * Be aware that this does not update the process cache for getWithSetCallback()
423 * callers. Keys accessed via that method are not generally meant to also be set
424 * using this primitive method.
425 *
426 * Do not use this method on versioned keys accessed via getWithSetCallback().
427 *
428 * Example usage:
429 * @code
430 * $dbr = wfGetDB( DB_REPLICA );
431 * $setOpts = Database::getCacheSetOptions( $dbr );
432 * // Fetch the row from the DB
433 * $row = $dbr->selectRow( ... );
434 * $key = $cache->makeKey( 'building', $buildingId );
435 * $cache->set( $key, $row, $cache::TTL_DAY, $setOpts );
436 * @endcode
437 *
438 * @param string $key Cache key
439 * @param mixed $value
440 * @param int $ttl Seconds to live. Special values are:
441 * - WANObjectCache::TTL_INDEFINITE: Cache forever
442 * @param array $opts Options map:
443 * - lag : Seconds of replica DB lag. Typically, this is either the replica DB lag
444 * before the data was read or, if applicable, the replica DB lag before
445 * the snapshot-isolated transaction the data was read from started.
446 * Use false to indicate that replication is not running.
447 * Default: 0 seconds
448 * - since : UNIX timestamp of the data in $value. Typically, this is either
449 * the current time the data was read or (if applicable) the time when
450 * the snapshot-isolated transaction the data was read from started.
451 * Default: 0 seconds
452 * - pending : Whether this data is possibly from an uncommitted write transaction.
453 * Generally, other threads should not see values from the future and
454 * they certainly should not see ones that ended up getting rolled back.
455 * Default: false
456 * - lockTSE : if excessive replication/snapshot lag is detected, then store the value
457 * with this TTL and flag it as stale. This is only useful if the reads for this key
458 * use getWithSetCallback() with "lockTSE" set. Note that if "staleTTL" is set
459 * then it will still add on to this TTL in the excessive lag scenario.
460 * Default: WANObjectCache::TSE_NONE
461 * - staleTTL : Seconds to keep the key around if it is stale. The get()/getMulti()
462 * methods return such stale values with a $curTTL of 0, and getWithSetCallback()
463 * will call the regeneration callback in such cases, passing in the old value
464 * and its as-of time to the callback. This is useful if adaptiveTTL() is used
465 * on the old value's as-of time when it is verified as still being correct.
466 * Default: WANObjectCache::STALE_TTL_NONE.
467 * @note Options added in 1.28: staleTTL
468 * @return bool Success
469 */
470 final public function set( $key, $value, $ttl = 0, array $opts = [] ) {
471 $now = $this->getCurrentTime();
472 $lockTSE = isset( $opts['lockTSE'] ) ? $opts['lockTSE'] : self::TSE_NONE;
473 $staleTTL = isset( $opts['staleTTL'] ) ? $opts['staleTTL'] : self::STALE_TTL_NONE;
474 $age = isset( $opts['since'] ) ? max( 0, $now - $opts['since'] ) : 0;
475 $lag = isset( $opts['lag'] ) ? $opts['lag'] : 0;
476
477 // Do not cache potentially uncommitted data as it might get rolled back
478 if ( !empty( $opts['pending'] ) ) {
479 $this->logger->info( 'Rejected set() for {cachekey} due to pending writes.',
480 [ 'cachekey' => $key ] );
481
482 return true; // no-op the write for being unsafe
483 }
484
485 $wrapExtra = []; // additional wrapped value fields
486 // Check if there's a risk of writing stale data after the purge tombstone expired
487 if ( $lag === false || ( $lag + $age ) > self::MAX_READ_LAG ) {
488 // Case A: read lag with "lockTSE"; save but record value as stale
489 if ( $lockTSE >= 0 ) {
490 $ttl = max( 1, (int)$lockTSE ); // set() expects seconds
491 $wrapExtra[self::FLD_FLAGS] = self::FLG_STALE; // mark as stale
492 // Case B: any long-running transaction; ignore this set()
493 } elseif ( $age > self::MAX_READ_LAG ) {
494 $this->logger->info( 'Rejected set() for {cachekey} due to snapshot lag.',
495 [ 'cachekey' => $key ] );
496
497 return true; // no-op the write for being unsafe
498 // Case C: high replication lag; lower TTL instead of ignoring all set()s
499 } elseif ( $lag === false || $lag > self::MAX_READ_LAG ) {
500 $ttl = $ttl ? min( $ttl, self::TTL_LAGGED ) : self::TTL_LAGGED;
501 $this->logger->warning( 'Lowered set() TTL for {cachekey} due to replication lag.',
502 [ 'cachekey' => $key ] );
503 // Case D: medium length request with medium replication lag; ignore this set()
504 } else {
505 $this->logger->info( 'Rejected set() for {cachekey} due to high read lag.',
506 [ 'cachekey' => $key ] );
507
508 return true; // no-op the write for being unsafe
509 }
510 }
511
512 // Wrap that value with time/TTL/version metadata
513 $wrapped = $this->wrap( $value, $ttl, $now ) + $wrapExtra;
514
515 $func = function ( $cache, $key, $cWrapped ) use ( $wrapped ) {
516 return ( is_string( $cWrapped ) )
517 ? false // key is tombstoned; do nothing
518 : $wrapped;
519 };
520
521 return $this->cache->merge( self::VALUE_KEY_PREFIX . $key, $func, $ttl + $staleTTL, 1 );
522 }
523
524 /**
525 * Purge a key from all datacenters
526 *
527 * This should only be called when the underlying data (being cached)
528 * changes in a significant way. This deletes the key and starts a hold-off
529 * period where the key cannot be written to for a few seconds (HOLDOFF_TTL).
530 * This is done to avoid the following race condition:
531 * - a) Some DB data changes and delete() is called on a corresponding key
532 * - b) A request refills the key with a stale value from a lagged DB
533 * - c) The stale value is stuck there until the key is expired/evicted
534 *
535 * This is implemented by storing a special "tombstone" value at the cache
536 * key that this class recognizes; get() calls will return false for the key
537 * and any set() calls will refuse to replace tombstone values at the key.
538 * For this to always avoid stale value writes, the following must hold:
539 * - a) Replication lag is bounded to being less than HOLDOFF_TTL; or
540 * - b) If lag is higher, the DB will have gone into read-only mode already
541 *
542 * Note that set() can also be lag-aware and lower the TTL if it's high.
543 *
544 * Be aware that this does not clear the process cache. Even if it did, callbacks
545 * used by getWithSetCallback() might still return stale data in the case of either
546 * uncommitted or not-yet-replicated changes (callback generally use replica DBs).
547 *
548 * When using potentially long-running ACID transactions, a good pattern is
549 * to use a pre-commit hook to issue the delete. This means that immediately
550 * after commit, callers will see the tombstone in cache upon purge relay.
551 * It also avoids the following race condition:
552 * - a) T1 begins, changes a row, and calls delete()
553 * - b) The HOLDOFF_TTL passes, expiring the delete() tombstone
554 * - c) T2 starts, reads the row and calls set() due to a cache miss
555 * - d) T1 finally commits
556 * - e) Stale value is stuck in cache
557 *
558 * Example usage:
559 * @code
560 * $dbw->startAtomic( __METHOD__ ); // start of request
561 * ... <execute some stuff> ...
562 * // Update the row in the DB
563 * $dbw->update( ... );
564 * $key = $cache->makeKey( 'homes', $homeId );
565 * // Purge the corresponding cache entry just before committing
566 * $dbw->onTransactionPreCommitOrIdle( function() use ( $cache, $key ) {
567 * $cache->delete( $key );
568 * } );
569 * ... <execute some stuff> ...
570 * $dbw->endAtomic( __METHOD__ ); // end of request
571 * @endcode
572 *
573 * The $ttl parameter can be used when purging values that have not actually changed
574 * recently. For example, a cleanup script to purge cache entries does not really need
575 * a hold-off period, so it can use HOLDOFF_NONE. Likewise for user-requested purge.
576 * Note that $ttl limits the effective range of 'lockTSE' for getWithSetCallback().
577 *
578 * If called twice on the same key, then the last hold-off TTL takes precedence. For
579 * idempotence, the $ttl should not vary for different delete() calls on the same key.
580 *
581 * @param string $key Cache key
582 * @param int $ttl Tombstone TTL; Default: WANObjectCache::HOLDOFF_TTL
583 * @return bool True if the item was purged or not found, false on failure
584 */
585 final public function delete( $key, $ttl = self::HOLDOFF_TTL ) {
586 $key = self::VALUE_KEY_PREFIX . $key;
587
588 if ( $ttl <= 0 ) {
589 // Publish the purge to all datacenters
590 $ok = $this->relayDelete( $key );
591 } else {
592 // Publish the purge to all datacenters
593 $ok = $this->relayPurge( $key, $ttl, self::HOLDOFF_NONE );
594 }
595
596 return $ok;
597 }
598
599 /**
600 * Fetch the value of a timestamp "check" key
601 *
602 * The key will be *initialized* to the current time if not set,
603 * so only call this method if this behavior is actually desired
604 *
605 * The timestamp can be used to check whether a cached value is valid.
606 * Callers should not assume that this returns the same timestamp in
607 * all datacenters due to relay delays.
608 *
609 * The level of staleness can roughly be estimated from this key, but
610 * if the key was evicted from cache, such calculations may show the
611 * time since expiry as ~0 seconds.
612 *
613 * Note that "check" keys won't collide with other regular keys.
614 *
615 * @param string $key
616 * @return float UNIX timestamp
617 */
618 final public function getCheckKeyTime( $key ) {
619 return $this->getMultiCheckKeyTime( [ $key ] )[$key];
620 }
621
622 /**
623 * Fetch the values of each timestamp "check" key
624 *
625 * This works like getCheckKeyTime() except it takes a list of keys
626 * and returns a map of timestamps instead of just that of one key
627 *
628 * This might be useful if both:
629 * - a) a class of entities each depend on hundreds of other entities
630 * - b) these other entities are depended upon by millions of entities
631 *
632 * The later entities can each use a "check" key to invalidate their dependee entities.
633 * However, it is expensive for the former entities to verify against all of the relevant
634 * "check" keys during each getWithSetCallback() call. A less expensive approach is to do
635 * these verifications only after a "time-till-verify" (TTV) has passed. This is a middle
636 * ground between using blind TTLs and using constant verification. The adaptiveTTL() method
637 * can be used to dynamically adjust the TTV. Also, the initial TTV can make use of the
638 * last-modified times of the dependant entities (either from the DB or the "check" keys).
639 *
640 * Example usage:
641 * @code
642 * $value = $cache->getWithSetCallback(
643 * $cache->makeGlobalKey( 'wikibase-item', $id ),
644 * self::INITIAL_TTV, // initial time-till-verify
645 * function ( $oldValue, &$ttv, &$setOpts, $oldAsOf ) use ( $checkKeys, $cache ) {
646 * $now = microtime( true );
647 * // Use $oldValue if it passes max ultimate age and "check" key comparisons
648 * if ( $oldValue &&
649 * $oldAsOf > max( $cache->getMultiCheckKeyTime( $checkKeys ) ) &&
650 * ( $now - $oldValue['ctime'] ) <= self::MAX_CACHE_AGE
651 * ) {
652 * // Increase time-till-verify by 50% of last time to reduce overhead
653 * $ttv = $cache->adaptiveTTL( $oldAsOf, self::MAX_TTV, self::MIN_TTV, 1.5 );
654 * // Unlike $oldAsOf, "ctime" is the ultimate age of the cached data
655 * return $oldValue;
656 * }
657 *
658 * $mtimes = []; // dependency last-modified times; passed by reference
659 * $value = [ 'data' => $this->fetchEntityData( $mtimes ), 'ctime' => $now ];
660 * // Guess time-till-change among the dependencies, e.g. 1/(total change rate)
661 * $ttc = 1 / array_sum( array_map(
662 * function ( $mtime ) use ( $now ) {
663 * return 1 / ( $mtime ? ( $now - $mtime ) : 900 );
664 * },
665 * $mtimes
666 * ) );
667 * // The time-to-verify should not be overly pessimistic nor optimistic
668 * $ttv = min( max( $ttc, self::MIN_TTV ), self::MAX_TTV );
669 *
670 * return $value;
671 * },
672 * [ 'staleTTL' => $cache::TTL_DAY ] // keep around to verify and re-save
673 * );
674 * @endcode
675 *
676 * @see WANObjectCache::getCheckKeyTime()
677 * @see WANObjectCache::getWithSetCallback()
678 *
679 * @param array $keys
680 * @return float[] Map of (key => UNIX timestamp)
681 * @since 1.31
682 */
683 final public function getMultiCheckKeyTime( array $keys ) {
684 $rawKeys = [];
685 foreach ( $keys as $key ) {
686 $rawKeys[$key] = self::TIME_KEY_PREFIX . $key;
687 }
688
689 $rawValues = $this->cache->getMulti( $rawKeys );
690 $rawValues += array_fill_keys( $rawKeys, false );
691
692 $times = [];
693 foreach ( $rawKeys as $key => $rawKey ) {
694 $purge = $this->parsePurgeValue( $rawValues[$rawKey] );
695 if ( $purge !== false ) {
696 $time = $purge[self::FLD_TIME];
697 } else {
698 // Casting assures identical floats for the next getCheckKeyTime() calls
699 $now = (string)$this->getCurrentTime();
700 $this->cache->add(
701 $rawKey,
702 $this->makePurgeValue( $now, self::HOLDOFF_TTL ),
703 self::CHECK_KEY_TTL
704 );
705 $time = (float)$now;
706 }
707
708 $times[$key] = $time;
709 }
710
711 return $times;
712 }
713
714 /**
715 * Purge a "check" key from all datacenters, invalidating keys that use it
716 *
717 * This should only be called when the underlying data (being cached)
718 * changes in a significant way, and it is impractical to call delete()
719 * on all keys that should be changed. When get() is called on those
720 * keys, the relevant "check" keys must be supplied for this to work.
721 *
722 * The "check" key essentially represents a last-modified time of an entity.
723 * When the key is touched, the timestamp will be updated to the current time.
724 * Keys using the "check" key via get(), getMulti(), or getWithSetCallback() will
725 * be invalidated. This approach is useful if many keys depend on a single entity.
726 *
727 * The timestamp of the "check" key is treated as being HOLDOFF_TTL seconds in the
728 * future by get*() methods in order to avoid race conditions where keys are updated
729 * with stale values (e.g. from a lagged replica DB). A high TTL is set on the "check"
730 * key, making it possible to know the timestamp of the last change to the corresponding
731 * entities in most cases. This might use more cache space than resetCheckKey().
732 *
733 * When a few important keys get a large number of hits, a high cache time is usually
734 * desired as well as "lockTSE" logic. The resetCheckKey() method is less appropriate
735 * in such cases since the "time since expiry" cannot be inferred, causing any get()
736 * after the reset to treat the key as being "hot", resulting in more stale value usage.
737 *
738 * Note that "check" keys won't collide with other regular keys.
739 *
740 * @see WANObjectCache::get()
741 * @see WANObjectCache::getWithSetCallback()
742 * @see WANObjectCache::resetCheckKey()
743 *
744 * @param string $key Cache key
745 * @param int $holdoff HOLDOFF_TTL or HOLDOFF_NONE constant
746 * @return bool True if the item was purged or not found, false on failure
747 */
748 final public function touchCheckKey( $key, $holdoff = self::HOLDOFF_TTL ) {
749 // Publish the purge to all datacenters
750 return $this->relayPurge( self::TIME_KEY_PREFIX . $key, self::CHECK_KEY_TTL, $holdoff );
751 }
752
753 /**
754 * Delete a "check" key from all datacenters, invalidating keys that use it
755 *
756 * This is similar to touchCheckKey() in that keys using it via get(), getMulti(),
757 * or getWithSetCallback() will be invalidated. The differences are:
758 * - a) The "check" key will be deleted from all caches and lazily
759 * re-initialized when accessed (rather than set everywhere)
760 * - b) Thus, dependent keys will be known to be invalid, but not
761 * for how long (they are treated as "just" purged), which
762 * effects any lockTSE logic in getWithSetCallback()
763 * - c) Since "check" keys are initialized only on the server the key hashes
764 * to, any temporary ejection of that server will cause the value to be
765 * seen as purged as a new server will initialize the "check" key.
766 *
767 * The advantage here is that the "check" keys, which have high TTLs, will only
768 * be created when a get*() method actually uses that key. This is better when
769 * a large number of "check" keys are invalided in a short period of time.
770 *
771 * Note that "check" keys won't collide with other regular keys.
772 *
773 * @see WANObjectCache::get()
774 * @see WANObjectCache::getWithSetCallback()
775 * @see WANObjectCache::touchCheckKey()
776 *
777 * @param string $key Cache key
778 * @return bool True if the item was purged or not found, false on failure
779 */
780 final public function resetCheckKey( $key ) {
781 // Publish the purge to all datacenters
782 return $this->relayDelete( self::TIME_KEY_PREFIX . $key );
783 }
784
785 /**
786 * Method to fetch/regenerate cache keys
787 *
788 * On cache miss, the key will be set to the callback result via set()
789 * (unless the callback returns false) and that result will be returned.
790 * The arguments supplied to the callback are:
791 * - $oldValue : current cache value or false if not present
792 * - &$ttl : a reference to the TTL which can be altered
793 * - &$setOpts : a reference to options for set() which can be altered
794 * - $oldAsOf : generation UNIX timestamp of $oldValue or null if not present (since 1.28)
795 *
796 * It is strongly recommended to set the 'lag' and 'since' fields to avoid race conditions
797 * that can cause stale values to get stuck at keys. Usually, callbacks ignore the current
798 * value, but it can be used to maintain "most recent X" values that come from time or
799 * sequence based source data, provided that the "as of" id/time is tracked. Note that
800 * preemptive regeneration and $checkKeys can result in a non-false current value.
801 *
802 * Usage of $checkKeys is similar to get() and getMulti(). However, rather than the caller
803 * having to inspect a "current time left" variable (e.g. $curTTL, $curTTLs), a cache
804 * regeneration will automatically be triggered using the callback.
805 *
806 * The $ttl argument and "hotTTR" option (in $opts) use time-dependant randomization
807 * to avoid stampedes. Keys that are slow to regenerate and either heavily used
808 * or subject to explicit (unpredictable) purges, may need additional mechanisms.
809 * The simplest way to avoid stampedes for such keys is to use 'lockTSE' (in $opts).
810 * If explicit purges are needed, also:
811 * - a) Pass $key into $checkKeys
812 * - b) Use touchCheckKey( $key ) instead of delete( $key )
813 *
814 * Example usage (typical key):
815 * @code
816 * $catInfo = $cache->getWithSetCallback(
817 * // Key to store the cached value under
818 * $cache->makeKey( 'cat-attributes', $catId ),
819 * // Time-to-live (in seconds)
820 * $cache::TTL_MINUTE,
821 * // Function that derives the new key value
822 * function ( $oldValue, &$ttl, array &$setOpts ) {
823 * $dbr = wfGetDB( DB_REPLICA );
824 * // Account for any snapshot/replica DB lag
825 * $setOpts += Database::getCacheSetOptions( $dbr );
826 *
827 * return $dbr->selectRow( ... );
828 * }
829 * );
830 * @endcode
831 *
832 * Example usage (key that is expensive and hot):
833 * @code
834 * $catConfig = $cache->getWithSetCallback(
835 * // Key to store the cached value under
836 * $cache->makeKey( 'site-cat-config' ),
837 * // Time-to-live (in seconds)
838 * $cache::TTL_DAY,
839 * // Function that derives the new key value
840 * function ( $oldValue, &$ttl, array &$setOpts ) {
841 * $dbr = wfGetDB( DB_REPLICA );
842 * // Account for any snapshot/replica DB lag
843 * $setOpts += Database::getCacheSetOptions( $dbr );
844 *
845 * return CatConfig::newFromRow( $dbr->selectRow( ... ) );
846 * },
847 * [
848 * // Calling touchCheckKey() on this key invalidates the cache
849 * 'checkKeys' => [ $cache->makeKey( 'site-cat-config' ) ],
850 * // Try to only let one datacenter thread manage cache updates at a time
851 * 'lockTSE' => 30,
852 * // Avoid querying cache servers multiple times in a web request
853 * 'pcTTL' => $cache::TTL_PROC_LONG
854 * ]
855 * );
856 * @endcode
857 *
858 * Example usage (key with dynamic dependencies):
859 * @code
860 * $catState = $cache->getWithSetCallback(
861 * // Key to store the cached value under
862 * $cache->makeKey( 'cat-state', $cat->getId() ),
863 * // Time-to-live (seconds)
864 * $cache::TTL_HOUR,
865 * // Function that derives the new key value
866 * function ( $oldValue, &$ttl, array &$setOpts ) {
867 * // Determine new value from the DB
868 * $dbr = wfGetDB( DB_REPLICA );
869 * // Account for any snapshot/replica DB lag
870 * $setOpts += Database::getCacheSetOptions( $dbr );
871 *
872 * return CatState::newFromResults( $dbr->select( ... ) );
873 * },
874 * [
875 * // The "check" keys that represent things the value depends on;
876 * // Calling touchCheckKey() on any of them invalidates the cache
877 * 'checkKeys' => [
878 * $cache->makeKey( 'sustenance-bowls', $cat->getRoomId() ),
879 * $cache->makeKey( 'people-present', $cat->getHouseId() ),
880 * $cache->makeKey( 'cat-laws', $cat->getCityId() ),
881 * ]
882 * ]
883 * );
884 * @endcode
885 *
886 * Example usage (hot key holding most recent 100 events):
887 * @code
888 * $lastCatActions = $cache->getWithSetCallback(
889 * // Key to store the cached value under
890 * $cache->makeKey( 'cat-last-actions', 100 ),
891 * // Time-to-live (in seconds)
892 * 10,
893 * // Function that derives the new key value
894 * function ( $oldValue, &$ttl, array &$setOpts ) {
895 * $dbr = wfGetDB( DB_REPLICA );
896 * // Account for any snapshot/replica DB lag
897 * $setOpts += Database::getCacheSetOptions( $dbr );
898 *
899 * // Start off with the last cached list
900 * $list = $oldValue ?: [];
901 * // Fetch the last 100 relevant rows in descending order;
902 * // only fetch rows newer than $list[0] to reduce scanning
903 * $rows = iterator_to_array( $dbr->select( ... ) );
904 * // Merge them and get the new "last 100" rows
905 * return array_slice( array_merge( $new, $list ), 0, 100 );
906 * },
907 * [
908 * // Try to only let one datacenter thread manage cache updates at a time
909 * 'lockTSE' => 30,
910 * // Use a magic value when no cache value is ready rather than stampeding
911 * 'busyValue' => 'computing'
912 * ]
913 * );
914 * @endcode
915 *
916 * Example usage (key holding an LRU subkey:value map; this can avoid flooding cache with
917 * keys for an unlimited set of (constraint,situation) pairs, thereby avoiding elevated
918 * cache evictions and wasted memory):
919 * @code
920 * $catSituationTolerabilityCache = $this->cache->getWithSetCallback(
921 * // Group by constraint ID/hash, cat family ID/hash, or something else useful
922 * $this->cache->makeKey( 'cat-situation-tolerablity-checks', $groupKey ),
923 * WANObjectCache::TTL_DAY, // rarely used groups should fade away
924 * // The $scenarioKey format is $constraintId:<ID/hash of $situation>
925 * function ( $cacheMap ) use ( $scenarioKey, $constraintId, $situation ) {
926 * $lruCache = MapCacheLRU::newFromArray( $cacheMap ?: [], self::CACHE_SIZE );
927 * $result = $lruCache->get( $scenarioKey ); // triggers LRU bump if present
928 * if ( $result === null || $this->isScenarioResultExpired( $result ) ) {
929 * $result = $this->checkScenarioTolerability( $constraintId, $situation );
930 * $lruCache->set( $scenarioKey, $result, 3 / 8 );
931 * }
932 * // Save the new LRU cache map and reset the map's TTL
933 * return $lruCache->toArray();
934 * },
935 * [
936 * // Once map is > 1 sec old, consider refreshing
937 * 'ageNew' => 1,
938 * // Update within 5 seconds after "ageNew" given a 1hz cache check rate
939 * 'hotTTR' => 5,
940 * // Avoid querying cache servers multiple times in a request; this also means
941 * // that a request can only alter the value of any given constraint key once
942 * 'pcTTL' => WANObjectCache::TTL_PROC_LONG
943 * ]
944 * );
945 * $tolerability = isset( $catSituationTolerabilityCache[$scenarioKey] )
946 * ? $catSituationTolerabilityCache[$scenarioKey]
947 * : $this->checkScenarioTolerability( $constraintId, $situation );
948 * @endcode
949 *
950 * @see WANObjectCache::get()
951 * @see WANObjectCache::set()
952 *
953 * @param string $key Cache key made from makeKey() or makeGlobalKey()
954 * @param int $ttl Seconds to live for key updates. Special values are:
955 * - WANObjectCache::TTL_INDEFINITE: Cache forever (subject to LRU-style evictions)
956 * - WANObjectCache::TTL_UNCACHEABLE: Do not cache (if the key exists, it is not deleted)
957 * @param callable $callback Value generation function
958 * @param array $opts Options map:
959 * - checkKeys: List of "check" keys. The key at $key will be seen as invalid when either
960 * touchCheckKey() or resetCheckKey() is called on any of the keys in this list. This
961 * is useful if thousands or millions of keys depend on the same entity. The entity can
962 * simply have its "check" key updated whenever the entity is modified.
963 * Default: [].
964 * - graceTTL: Consider reusing expired values instead of refreshing them if they expired
965 * less than this many seconds ago. The odds of a refresh becomes more likely over time,
966 * becoming certain once the grace period is reached. This can reduce traffic spikes
967 * when millions of keys are compared to the same "check" key and touchCheckKey()
968 * or resetCheckKey() is called on that "check" key.
969 * Default: WANObjectCache::GRACE_TTL_NONE.
970 * - lockTSE: If the key is tombstoned or expired (by checkKeys) less than this many seconds
971 * ago, then try to have a single thread handle cache regeneration at any given time.
972 * Other threads will try to use stale values if possible. If, on miss, the time since
973 * expiration is low, the assumption is that the key is hot and that a stampede is worth
974 * avoiding. Setting this above WANObjectCache::HOLDOFF_TTL makes no difference. The
975 * higher this is set, the higher the worst-case staleness can be.
976 * Use WANObjectCache::TSE_NONE to disable this logic.
977 * Default: WANObjectCache::TSE_NONE.
978 * - busyValue: If no value exists and another thread is currently regenerating it, use this
979 * as a fallback value (or a callback to generate such a value). This assures that cache
980 * stampedes cannot happen if the value falls out of cache. This can be used as insurance
981 * against cache regeneration becoming very slow for some reason (greater than the TTL).
982 * Default: null.
983 * - pcTTL: Process cache the value in this PHP instance for this many seconds. This avoids
984 * network I/O when a key is read several times. This will not cache when the callback
985 * returns false, however. Note that any purges will not be seen while process cached;
986 * since the callback should use replica DBs and they may be lagged or have snapshot
987 * isolation anyway, this should not typically matter.
988 * Default: WANObjectCache::TTL_UNCACHEABLE.
989 * - pcGroup: Process cache group to use instead of the primary one. If set, this must be
990 * of the format ALPHANUMERIC_NAME:MAX_KEY_SIZE, e.g. "mydata:10". Use this for storing
991 * large values, small yet numerous values, or some values with a high cost of eviction.
992 * It is generally preferable to use a class constant when setting this value.
993 * This has no effect unless pcTTL is used.
994 * Default: WANObjectCache::PC_PRIMARY.
995 * - version: Integer version number. This allows for callers to make breaking changes to
996 * how values are stored while maintaining compatability and correct cache purges. New
997 * versions are stored alongside older versions concurrently. Avoid storing class objects
998 * however, as this reduces compatibility (due to serialization).
999 * Default: null.
1000 * - minAsOf: Reject values if they were generated before this UNIX timestamp.
1001 * This is useful if the source of a key is suspected of having possibly changed
1002 * recently, and the caller wants any such changes to be reflected.
1003 * Default: WANObjectCache::MIN_TIMESTAMP_NONE.
1004 * - hotTTR: Expected time-till-refresh (TTR) in seconds for keys that average ~1 hit per
1005 * second (e.g. 1Hz). Keys with a hit rate higher than 1Hz will refresh sooner than this
1006 * TTR and vise versa. Such refreshes won't happen until keys are "ageNew" seconds old.
1007 * This uses randomization to avoid triggering cache stampedes. The TTR is useful at
1008 * reducing the impact of missed cache purges, since the effect of a heavily referenced
1009 * key being stale is worse than that of a rarely referenced key. Unlike simply lowering
1010 * $ttl, seldomly used keys are largely unaffected by this option, which makes it
1011 * possible to have a high hit rate for the "long-tail" of less-used keys.
1012 * Default: WANObjectCache::HOT_TTR.
1013 * - lowTTL: Consider pre-emptive updates when the current TTL (seconds) of the key is less
1014 * than this. It becomes more likely over time, becoming certain once the key is expired.
1015 * This helps avoid cache stampedes that might be triggered due to the key expiring.
1016 * Default: WANObjectCache::LOW_TTL.
1017 * - ageNew: Consider popularity refreshes only once a key reaches this age in seconds.
1018 * Default: WANObjectCache::AGE_NEW.
1019 * - staleTTL: Seconds to keep the key around if it is stale. This means that on cache
1020 * miss the callback may get $oldValue/$oldAsOf values for keys that have already been
1021 * expired for this specified time. This is useful if adaptiveTTL() is used on the old
1022 * value's as-of time when it is verified as still being correct.
1023 * Default: WANObjectCache::STALE_TTL_NONE
1024 * @return mixed Value found or written to the key
1025 * @note Options added in 1.28: version, busyValue, hotTTR, ageNew, pcGroup, minAsOf
1026 * @note Options added in 1.31: staleTTL, graceTTL
1027 * @note Callable type hints are not used to avoid class-autoloading
1028 */
1029 final public function getWithSetCallback( $key, $ttl, $callback, array $opts = [] ) {
1030 $pcTTL = isset( $opts['pcTTL'] ) ? $opts['pcTTL'] : self::TTL_UNCACHEABLE;
1031
1032 // Try the process cache if enabled and the cache callback is not within a cache callback.
1033 // Process cache use in nested callbacks is not lag-safe with regard to HOLDOFF_TTL since
1034 // the in-memory value is further lagged than the shared one since it uses a blind TTL.
1035 if ( $pcTTL >= 0 && $this->callbackDepth == 0 ) {
1036 $group = isset( $opts['pcGroup'] ) ? $opts['pcGroup'] : self::PC_PRIMARY;
1037 $procCache = $this->getProcessCache( $group );
1038 $value = $procCache->get( $key );
1039 } else {
1040 $procCache = false;
1041 $value = false;
1042 }
1043
1044 if ( $value === false ) {
1045 // Fetch the value over the network
1046 if ( isset( $opts['version'] ) ) {
1047 $version = $opts['version'];
1048 $asOf = null;
1049 $cur = $this->doGetWithSetCallback(
1050 $key,
1051 $ttl,
1052 function ( $oldValue, &$ttl, &$setOpts, $oldAsOf )
1053 use ( $callback, $version ) {
1054 if ( is_array( $oldValue )
1055 && array_key_exists( self::VFLD_DATA, $oldValue )
1056 && array_key_exists( self::VFLD_VERSION, $oldValue )
1057 && $oldValue[self::VFLD_VERSION] === $version
1058 ) {
1059 $oldData = $oldValue[self::VFLD_DATA];
1060 } else {
1061 // VFLD_DATA is not set if an old, unversioned, key is present
1062 $oldData = false;
1063 $oldAsOf = null;
1064 }
1065
1066 return [
1067 self::VFLD_DATA => $callback( $oldData, $ttl, $setOpts, $oldAsOf ),
1068 self::VFLD_VERSION => $version
1069 ];
1070 },
1071 $opts,
1072 $asOf
1073 );
1074 if ( $cur[self::VFLD_VERSION] === $version ) {
1075 // Value created or existed before with version; use it
1076 $value = $cur[self::VFLD_DATA];
1077 } else {
1078 // Value existed before with a different version; use variant key.
1079 // Reflect purges to $key by requiring that this key value be newer.
1080 $value = $this->doGetWithSetCallback(
1081 $this->makeGlobalKey( 'WANCache-key-variant', md5( $key ), $version ),
1082 $ttl,
1083 $callback,
1084 // Regenerate value if not newer than $key
1085 [ 'version' => null, 'minAsOf' => $asOf ] + $opts
1086 );
1087 }
1088 } else {
1089 $value = $this->doGetWithSetCallback( $key, $ttl, $callback, $opts );
1090 }
1091
1092 // Update the process cache if enabled
1093 if ( $procCache && $value !== false ) {
1094 $procCache->set( $key, $value, $pcTTL );
1095 }
1096 }
1097
1098 return $value;
1099 }
1100
1101 /**
1102 * Do the actual I/O for getWithSetCallback() when needed
1103 *
1104 * @see WANObjectCache::getWithSetCallback()
1105 *
1106 * @param string $key
1107 * @param int $ttl
1108 * @param callback $callback
1109 * @param array $opts Options map for getWithSetCallback()
1110 * @param float &$asOf Cache generation timestamp of returned value [returned]
1111 * @return mixed
1112 * @note Callable type hints are not used to avoid class-autoloading
1113 */
1114 protected function doGetWithSetCallback( $key, $ttl, $callback, array $opts, &$asOf = null ) {
1115 $lowTTL = isset( $opts['lowTTL'] ) ? $opts['lowTTL'] : min( self::LOW_TTL, $ttl );
1116 $lockTSE = isset( $opts['lockTSE'] ) ? $opts['lockTSE'] : self::TSE_NONE;
1117 $staleTTL = isset( $opts['staleTTL'] ) ? $opts['staleTTL'] : self::STALE_TTL_NONE;
1118 $graceTTL = isset( $opts['graceTTL'] ) ? $opts['graceTTL'] : self::GRACE_TTL_NONE;
1119 $checkKeys = isset( $opts['checkKeys'] ) ? $opts['checkKeys'] : [];
1120 $busyValue = isset( $opts['busyValue'] ) ? $opts['busyValue'] : null;
1121 $popWindow = isset( $opts['hotTTR'] ) ? $opts['hotTTR'] : self::HOT_TTR;
1122 $ageNew = isset( $opts['ageNew'] ) ? $opts['ageNew'] : self::AGE_NEW;
1123 $minTime = isset( $opts['minAsOf'] ) ? $opts['minAsOf'] : self::MIN_TIMESTAMP_NONE;
1124 $versioned = isset( $opts['version'] );
1125
1126 // Get a collection name to describe this class of key
1127 $kClass = $this->determineKeyClass( $key );
1128
1129 // Get the current key value
1130 $curTTL = null;
1131 $cValue = $this->get( $key, $curTTL, $checkKeys, $asOf ); // current value
1132 $value = $cValue; // return value
1133
1134 $preCallbackTime = $this->getCurrentTime();
1135 // Determine if a cached value regeneration is needed or desired
1136 if ( $value !== false
1137 && $this->isAliveOrInGracePeriod( $curTTL, $graceTTL )
1138 && $this->isValid( $value, $versioned, $asOf, $minTime )
1139 ) {
1140 $preemptiveRefresh = (
1141 $this->worthRefreshExpiring( $curTTL, $lowTTL ) ||
1142 $this->worthRefreshPopular( $asOf, $ageNew, $popWindow, $preCallbackTime )
1143 );
1144
1145 if ( !$preemptiveRefresh ) {
1146 $this->stats->increment( "wanobjectcache.$kClass.hit.good" );
1147
1148 return $value;
1149 } elseif ( $this->asyncHandler ) {
1150 // Update the cache value later, such during post-send of an HTTP request
1151 $func = $this->asyncHandler;
1152 $func( function () use ( $key, $ttl, $callback, $opts, $asOf ) {
1153 $opts['minAsOf'] = INF; // force a refresh
1154 $this->doGetWithSetCallback( $key, $ttl, $callback, $opts, $asOf );
1155 } );
1156 $this->stats->increment( "wanobjectcache.$kClass.hit.refresh" );
1157
1158 return $value;
1159 }
1160 }
1161
1162 // A deleted key with a negative TTL left must be tombstoned
1163 $isTombstone = ( $curTTL !== null && $value === false );
1164 if ( $isTombstone && $lockTSE <= 0 ) {
1165 // Use the INTERIM value for tombstoned keys to reduce regeneration load
1166 $lockTSE = self::INTERIM_KEY_TTL;
1167 }
1168 // Assume a key is hot if requested soon after invalidation
1169 $isHot = ( $curTTL !== null && $curTTL <= 0 && abs( $curTTL ) <= $lockTSE );
1170 // Use the mutex if there is no value and a busy fallback is given
1171 $checkBusy = ( $busyValue !== null && $value === false );
1172 // Decide whether a single thread should handle regenerations.
1173 // This avoids stampedes when $checkKeys are bumped and when preemptive
1174 // renegerations take too long. It also reduces regenerations while $key
1175 // is tombstoned. This balances cache freshness with avoiding DB load.
1176 $useMutex = ( $isHot || ( $isTombstone && $lockTSE > 0 ) || $checkBusy );
1177
1178 $lockAcquired = false;
1179 if ( $useMutex ) {
1180 // Acquire a datacenter-local non-blocking lock
1181 if ( $this->cache->add( self::MUTEX_KEY_PREFIX . $key, 1, self::LOCK_TTL ) ) {
1182 // Lock acquired; this thread should update the key
1183 $lockAcquired = true;
1184 } elseif ( $value !== false && $this->isValid( $value, $versioned, $asOf, $minTime ) ) {
1185 $this->stats->increment( "wanobjectcache.$kClass.hit.stale" );
1186 // If it cannot be acquired; then the stale value can be used
1187 return $value;
1188 } else {
1189 // Use the INTERIM value for tombstoned keys to reduce regeneration load.
1190 // For hot keys, either another thread has the lock or the lock failed;
1191 // use the INTERIM value from the last thread that regenerated it.
1192 $value = $this->getInterimValue( $key, $versioned, $minTime, $asOf );
1193 if ( $value !== false ) {
1194 $this->stats->increment( "wanobjectcache.$kClass.hit.volatile" );
1195
1196 return $value;
1197 }
1198 // Use the busy fallback value if nothing else
1199 if ( $busyValue !== null ) {
1200 $this->stats->increment( "wanobjectcache.$kClass.miss.busy" );
1201
1202 return is_callable( $busyValue ) ? $busyValue() : $busyValue;
1203 }
1204 }
1205 }
1206
1207 if ( !is_callable( $callback ) ) {
1208 throw new InvalidArgumentException( "Invalid cache miss callback provided." );
1209 }
1210
1211 // Generate the new value from the callback...
1212 $setOpts = [];
1213 ++$this->callbackDepth;
1214 try {
1215 $value = call_user_func_array( $callback, [ $cValue, &$ttl, &$setOpts, $asOf ] );
1216 } finally {
1217 --$this->callbackDepth;
1218 }
1219 $valueIsCacheable = ( $value !== false && $ttl >= 0 );
1220
1221 // When delete() is called, writes are write-holed by the tombstone,
1222 // so use a special INTERIM key to pass the new value around threads.
1223 if ( ( $isTombstone && $lockTSE > 0 ) && $valueIsCacheable ) {
1224 $tempTTL = max( 1, (int)$lockTSE ); // set() expects seconds
1225 $newAsOf = $this->getCurrentTime();
1226 $wrapped = $this->wrap( $value, $tempTTL, $newAsOf );
1227 // Avoid using set() to avoid pointless mcrouter broadcasting
1228 $this->setInterimValue( $key, $wrapped, $tempTTL );
1229 }
1230
1231 if ( $valueIsCacheable ) {
1232 $setOpts['lockTSE'] = $lockTSE;
1233 $setOpts['staleTTL'] = $staleTTL;
1234 // Use best known "since" timestamp if not provided
1235 $setOpts += [ 'since' => $preCallbackTime ];
1236 // Update the cache; this will fail if the key is tombstoned
1237 $this->set( $key, $value, $ttl, $setOpts );
1238 }
1239
1240 if ( $lockAcquired ) {
1241 // Avoid using delete() to avoid pointless mcrouter broadcasting
1242 $this->cache->changeTTL( self::MUTEX_KEY_PREFIX . $key, (int)$preCallbackTime - 60 );
1243 }
1244
1245 $this->stats->increment( "wanobjectcache.$kClass.miss.compute" );
1246
1247 return $value;
1248 }
1249
1250 /**
1251 * @param string $key
1252 * @param bool $versioned
1253 * @param float $minTime
1254 * @param mixed &$asOf
1255 * @return mixed
1256 */
1257 protected function getInterimValue( $key, $versioned, $minTime, &$asOf ) {
1258 if ( !$this->useInterimHoldOffCaching ) {
1259 return false; // disabled
1260 }
1261
1262 $wrapped = $this->cache->get( self::INTERIM_KEY_PREFIX . $key );
1263 list( $value ) = $this->unwrap( $wrapped, $this->getCurrentTime() );
1264 if ( $value !== false && $this->isValid( $value, $versioned, $asOf, $minTime ) ) {
1265 $asOf = $wrapped[self::FLD_TIME];
1266
1267 return $value;
1268 }
1269
1270 return false;
1271 }
1272
1273 /**
1274 * @param string $key
1275 * @param array $wrapped
1276 * @param int $tempTTL
1277 */
1278 protected function setInterimValue( $key, $wrapped, $tempTTL ) {
1279 $this->cache->merge(
1280 self::INTERIM_KEY_PREFIX . $key,
1281 function () use ( $wrapped ) {
1282 return $wrapped;
1283 },
1284 $tempTTL,
1285 1
1286 );
1287 }
1288
1289 /**
1290 * Method to fetch multiple cache keys at once with regeneration
1291 *
1292 * This works the same as getWithSetCallback() except:
1293 * - a) The $keys argument expects the result of WANObjectCache::makeMultiKeys()
1294 * - b) The $callback argument expects a callback taking the following arguments:
1295 * - $id: ID of an entity to query
1296 * - $oldValue : the prior cache value or false if none was present
1297 * - &$ttl : a reference to the new value TTL in seconds
1298 * - &$setOpts : a reference to options for set() which can be altered
1299 * - $oldAsOf : generation UNIX timestamp of $oldValue or null if not present
1300 * Aside from the additional $id argument, the other arguments function the same
1301 * way they do in getWithSetCallback().
1302 * - c) The return value is a map of (cache key => value) in the order of $keyedIds
1303 *
1304 * @see WANObjectCache::getWithSetCallback()
1305 * @see WANObjectCache::getMultiWithUnionSetCallback()
1306 *
1307 * Example usage:
1308 * @code
1309 * $rows = $cache->getMultiWithSetCallback(
1310 * // Map of cache keys to entity IDs
1311 * $cache->makeMultiKeys(
1312 * $this->fileVersionIds(),
1313 * function ( $id, WANObjectCache $cache ) {
1314 * return $cache->makeKey( 'file-version', $id );
1315 * }
1316 * ),
1317 * // Time-to-live (in seconds)
1318 * $cache::TTL_DAY,
1319 * // Function that derives the new key value
1320 * function ( $id, $oldValue, &$ttl, array &$setOpts ) {
1321 * $dbr = wfGetDB( DB_REPLICA );
1322 * // Account for any snapshot/replica DB lag
1323 * $setOpts += Database::getCacheSetOptions( $dbr );
1324 *
1325 * // Load the row for this file
1326 * $queryInfo = File::getQueryInfo();
1327 * $row = $dbr->selectRow(
1328 * $queryInfo['tables'],
1329 * $queryInfo['fields'],
1330 * [ 'id' => $id ],
1331 * __METHOD__,
1332 * [],
1333 * $queryInfo['joins']
1334 * );
1335 *
1336 * return $row ? (array)$row : false;
1337 * },
1338 * [
1339 * // Process cache for 30 seconds
1340 * 'pcTTL' => 30,
1341 * // Use a dedicated 500 item cache (initialized on-the-fly)
1342 * 'pcGroup' => 'file-versions:500'
1343 * ]
1344 * );
1345 * $files = array_map( [ __CLASS__, 'newFromRow' ], $rows );
1346 * @endcode
1347 *
1348 * @param ArrayIterator $keyedIds Result of WANObjectCache::makeMultiKeys()
1349 * @param int $ttl Seconds to live for key updates
1350 * @param callable $callback Callback the yields entity regeneration callbacks
1351 * @param array $opts Options map
1352 * @return array Map of (cache key => value) in the same order as $keyedIds
1353 * @since 1.28
1354 */
1355 final public function getMultiWithSetCallback(
1356 ArrayIterator $keyedIds, $ttl, callable $callback, array $opts = []
1357 ) {
1358 $valueKeys = array_keys( $keyedIds->getArrayCopy() );
1359 $checkKeys = isset( $opts['checkKeys'] ) ? $opts['checkKeys'] : [];
1360
1361 // Load required keys into process cache in one go
1362 $this->warmupCache = $this->getRawKeysForWarmup(
1363 $this->getNonProcessCachedKeys( $valueKeys, $opts ),
1364 $checkKeys
1365 );
1366 $this->warmupKeyMisses = 0;
1367
1368 // Wrap $callback to match the getWithSetCallback() format while passing $id to $callback
1369 $id = null; // current entity ID
1370 $func = function ( $oldValue, &$ttl, &$setOpts, $oldAsOf ) use ( $callback, &$id ) {
1371 return $callback( $id, $oldValue, $ttl, $setOpts, $oldAsOf );
1372 };
1373
1374 $values = [];
1375 foreach ( $keyedIds as $key => $id ) { // preserve order
1376 $values[$key] = $this->getWithSetCallback( $key, $ttl, $func, $opts );
1377 }
1378
1379 $this->warmupCache = [];
1380
1381 return $values;
1382 }
1383
1384 /**
1385 * Method to fetch/regenerate multiple cache keys at once
1386 *
1387 * This works the same as getWithSetCallback() except:
1388 * - a) The $keys argument expects the result of WANObjectCache::makeMultiKeys()
1389 * - b) The $callback argument expects a callback returning a map of (ID => new value)
1390 * for all entity IDs in $ids and it takes the following arguments:
1391 * - $ids: a list of entity IDs that require cache regeneration
1392 * - &$ttls: a reference to the (entity ID => new TTL) map
1393 * - &$setOpts: a reference to options for set() which can be altered
1394 * - c) The return value is a map of (cache key => value) in the order of $keyedIds
1395 * - d) The "lockTSE" and "busyValue" options are ignored
1396 *
1397 * @see WANObjectCache::getWithSetCallback()
1398 * @see WANObjectCache::getMultiWithSetCallback()
1399 *
1400 * Example usage:
1401 * @code
1402 * $rows = $cache->getMultiWithUnionSetCallback(
1403 * // Map of cache keys to entity IDs
1404 * $cache->makeMultiKeys(
1405 * $this->fileVersionIds(),
1406 * function ( $id, WANObjectCache $cache ) {
1407 * return $cache->makeKey( 'file-version', $id );
1408 * }
1409 * ),
1410 * // Time-to-live (in seconds)
1411 * $cache::TTL_DAY,
1412 * // Function that derives the new key value
1413 * function ( array $ids, array &$ttls, array &$setOpts ) {
1414 * $dbr = wfGetDB( DB_REPLICA );
1415 * // Account for any snapshot/replica DB lag
1416 * $setOpts += Database::getCacheSetOptions( $dbr );
1417 *
1418 * // Load the rows for these files
1419 * $rows = [];
1420 * $queryInfo = File::getQueryInfo();
1421 * $res = $dbr->select(
1422 * $queryInfo['tables'],
1423 * $queryInfo['fields'],
1424 * [ 'id' => $ids ],
1425 * __METHOD__,
1426 * [],
1427 * $queryInfo['joins']
1428 * );
1429 * foreach ( $res as $row ) {
1430 * $rows[$row->id] = $row;
1431 * $mtime = wfTimestamp( TS_UNIX, $row->timestamp );
1432 * $ttls[$row->id] = $this->adaptiveTTL( $mtime, $ttls[$row->id] );
1433 * }
1434 *
1435 * return $rows;
1436 * },
1437 * ]
1438 * );
1439 * $files = array_map( [ __CLASS__, 'newFromRow' ], $rows );
1440 * @endcode
1441 *
1442 * @param ArrayIterator $keyedIds Result of WANObjectCache::makeMultiKeys()
1443 * @param int $ttl Seconds to live for key updates
1444 * @param callable $callback Callback the yields entity regeneration callbacks
1445 * @param array $opts Options map
1446 * @return array Map of (cache key => value) in the same order as $keyedIds
1447 * @since 1.30
1448 */
1449 final public function getMultiWithUnionSetCallback(
1450 ArrayIterator $keyedIds, $ttl, callable $callback, array $opts = []
1451 ) {
1452 $idsByValueKey = $keyedIds->getArrayCopy();
1453 $valueKeys = array_keys( $idsByValueKey );
1454 $checkKeys = isset( $opts['checkKeys'] ) ? $opts['checkKeys'] : [];
1455 unset( $opts['lockTSE'] ); // incompatible
1456 unset( $opts['busyValue'] ); // incompatible
1457
1458 // Load required keys into process cache in one go
1459 $keysGet = $this->getNonProcessCachedKeys( $valueKeys, $opts );
1460 $this->warmupCache = $this->getRawKeysForWarmup( $keysGet, $checkKeys );
1461 $this->warmupKeyMisses = 0;
1462
1463 // IDs of entities known to be in need of regeneration
1464 $idsRegen = [];
1465
1466 // Find out which keys are missing/deleted/stale
1467 $curTTLs = [];
1468 $asOfs = [];
1469 $curByKey = $this->getMulti( $keysGet, $curTTLs, $checkKeys, $asOfs );
1470 foreach ( $keysGet as $key ) {
1471 if ( !array_key_exists( $key, $curByKey ) || $curTTLs[$key] < 0 ) {
1472 $idsRegen[] = $idsByValueKey[$key];
1473 }
1474 }
1475
1476 // Run the callback to populate the regeneration value map for all required IDs
1477 $newSetOpts = [];
1478 $newTTLsById = array_fill_keys( $idsRegen, $ttl );
1479 $newValsById = $idsRegen ? $callback( $idsRegen, $newTTLsById, $newSetOpts ) : [];
1480
1481 // Wrap $callback to match the getWithSetCallback() format while passing $id to $callback
1482 $id = null; // current entity ID
1483 $func = function ( $oldValue, &$ttl, &$setOpts, $oldAsOf )
1484 use ( $callback, &$id, $newValsById, $newTTLsById, $newSetOpts )
1485 {
1486 if ( array_key_exists( $id, $newValsById ) ) {
1487 // Value was already regerated as expected, so use the value in $newValsById
1488 $newValue = $newValsById[$id];
1489 $ttl = $newTTLsById[$id];
1490 $setOpts = $newSetOpts;
1491 } else {
1492 // Pre-emptive/popularity refresh and version mismatch cases are not detected
1493 // above and thus $newValsById has no entry. Run $callback on this single entity.
1494 $ttls = [ $id => $ttl ];
1495 $newValue = $callback( [ $id ], $ttls, $setOpts )[$id];
1496 $ttl = $ttls[$id];
1497 }
1498
1499 return $newValue;
1500 };
1501
1502 // Run the cache-aside logic using warmupCache instead of persistent cache queries
1503 $values = [];
1504 foreach ( $idsByValueKey as $key => $id ) { // preserve order
1505 $values[$key] = $this->getWithSetCallback( $key, $ttl, $func, $opts );
1506 }
1507
1508 $this->warmupCache = [];
1509
1510 return $values;
1511 }
1512
1513 /**
1514 * Locally set a key to expire soon if it is stale based on $purgeTimestamp
1515 *
1516 * This sets stale keys' time-to-live at HOLDOFF_TTL seconds, which both avoids
1517 * broadcasting in mcrouter setups and also avoids races with new tombstones.
1518 *
1519 * @param string $key Cache key
1520 * @param int $purgeTimestamp UNIX timestamp of purge
1521 * @param bool &$isStale Whether the key is stale
1522 * @return bool Success
1523 * @since 1.28
1524 */
1525 final public function reap( $key, $purgeTimestamp, &$isStale = false ) {
1526 $minAsOf = $purgeTimestamp + self::HOLDOFF_TTL;
1527 $wrapped = $this->cache->get( self::VALUE_KEY_PREFIX . $key );
1528 if ( is_array( $wrapped ) && $wrapped[self::FLD_TIME] < $minAsOf ) {
1529 $isStale = true;
1530 $this->logger->warning( "Reaping stale value key '$key'." );
1531 $ttlReap = self::HOLDOFF_TTL; // avoids races with tombstone creation
1532 $ok = $this->cache->changeTTL( self::VALUE_KEY_PREFIX . $key, $ttlReap );
1533 if ( !$ok ) {
1534 $this->logger->error( "Could not complete reap of key '$key'." );
1535 }
1536
1537 return $ok;
1538 }
1539
1540 $isStale = false;
1541
1542 return true;
1543 }
1544
1545 /**
1546 * Locally set a "check" key to expire soon if it is stale based on $purgeTimestamp
1547 *
1548 * @param string $key Cache key
1549 * @param int $purgeTimestamp UNIX timestamp of purge
1550 * @param bool &$isStale Whether the key is stale
1551 * @return bool Success
1552 * @since 1.28
1553 */
1554 final public function reapCheckKey( $key, $purgeTimestamp, &$isStale = false ) {
1555 $purge = $this->parsePurgeValue( $this->cache->get( self::TIME_KEY_PREFIX . $key ) );
1556 if ( $purge && $purge[self::FLD_TIME] < $purgeTimestamp ) {
1557 $isStale = true;
1558 $this->logger->warning( "Reaping stale check key '$key'." );
1559 $ok = $this->cache->changeTTL( self::TIME_KEY_PREFIX . $key, 1 );
1560 if ( !$ok ) {
1561 $this->logger->error( "Could not complete reap of check key '$key'." );
1562 }
1563
1564 return $ok;
1565 }
1566
1567 $isStale = false;
1568
1569 return false;
1570 }
1571
1572 /**
1573 * @see BagOStuff::makeKey()
1574 * @param string $class Key class
1575 * @param string $component [optional] Key component (starting with a key collection name)
1576 * @return string Colon-delimited list of $keyspace followed by escaped components of $args
1577 * @since 1.27
1578 */
1579 public function makeKey( $class, $component = null ) {
1580 return call_user_func_array( [ $this->cache, __FUNCTION__ ], func_get_args() );
1581 }
1582
1583 /**
1584 * @see BagOStuff::makeGlobalKey()
1585 * @param string $class Key class
1586 * @param string $component [optional] Key component (starting with a key collection name)
1587 * @return string Colon-delimited list of $keyspace followed by escaped components of $args
1588 * @since 1.27
1589 */
1590 public function makeGlobalKey( $class, $component = null ) {
1591 return call_user_func_array( [ $this->cache, __FUNCTION__ ], func_get_args() );
1592 }
1593
1594 /**
1595 * @param array $entities List of entity IDs
1596 * @param callable $keyFunc Callback yielding a key from (entity ID, this WANObjectCache)
1597 * @return ArrayIterator Iterator yielding (cache key => entity ID) in $entities order
1598 * @since 1.28
1599 */
1600 final public function makeMultiKeys( array $entities, callable $keyFunc ) {
1601 $map = [];
1602 foreach ( $entities as $entity ) {
1603 $map[$keyFunc( $entity, $this )] = $entity;
1604 }
1605
1606 return new ArrayIterator( $map );
1607 }
1608
1609 /**
1610 * Get the "last error" registered; clearLastError() should be called manually
1611 * @return int ERR_* class constant for the "last error" registry
1612 */
1613 final public function getLastError() {
1614 if ( $this->lastRelayError ) {
1615 // If the cache and the relayer failed, focus on the latter.
1616 // An update not making it to the relayer means it won't show up
1617 // in other DCs (nor will consistent re-hashing see up-to-date values).
1618 // On the other hand, if just the cache update failed, then it should
1619 // eventually be applied by the relayer.
1620 return $this->lastRelayError;
1621 }
1622
1623 $code = $this->cache->getLastError();
1624 switch ( $code ) {
1625 case BagOStuff::ERR_NONE:
1626 return self::ERR_NONE;
1627 case BagOStuff::ERR_NO_RESPONSE:
1628 return self::ERR_NO_RESPONSE;
1629 case BagOStuff::ERR_UNREACHABLE:
1630 return self::ERR_UNREACHABLE;
1631 default:
1632 return self::ERR_UNEXPECTED;
1633 }
1634 }
1635
1636 /**
1637 * Clear the "last error" registry
1638 */
1639 final public function clearLastError() {
1640 $this->cache->clearLastError();
1641 $this->lastRelayError = self::ERR_NONE;
1642 }
1643
1644 /**
1645 * Clear the in-process caches; useful for testing
1646 *
1647 * @since 1.27
1648 */
1649 public function clearProcessCache() {
1650 $this->processCaches = [];
1651 }
1652
1653 /**
1654 * Enable or disable the use of brief caching for tombstoned keys
1655 *
1656 * When a key is purged via delete(), there normally is a period where caching
1657 * is hold-off limited to an extremely short time. This method will disable that
1658 * caching, forcing the callback to run for any of:
1659 * - WANObjectCache::getWithSetCallback()
1660 * - WANObjectCache::getMultiWithSetCallback()
1661 * - WANObjectCache::getMultiWithUnionSetCallback()
1662 *
1663 * This is useful when both:
1664 * - a) the database used by the callback is known to be up-to-date enough
1665 * for some particular purpose (e.g. replica DB has applied transaction X)
1666 * - b) the caller needs to exploit that fact, and therefore needs to avoid the
1667 * use of inherently volatile and possibly stale interim keys
1668 *
1669 * @see WANObjectCache::delete()
1670 * @param bool $enabled Whether to enable interim caching
1671 * @since 1.31
1672 */
1673 final public function useInterimHoldOffCaching( $enabled ) {
1674 $this->useInterimHoldOffCaching = $enabled;
1675 }
1676
1677 /**
1678 * @param int $flag ATTR_* class constant
1679 * @return int QOS_* class constant
1680 * @since 1.28
1681 */
1682 public function getQoS( $flag ) {
1683 return $this->cache->getQoS( $flag );
1684 }
1685
1686 /**
1687 * Get a TTL that is higher for objects that have not changed recently
1688 *
1689 * This is useful for keys that get explicit purges and DB or purge relay
1690 * lag is a potential concern (especially how it interacts with CDN cache)
1691 *
1692 * Example usage:
1693 * @code
1694 * // Last-modified time of page
1695 * $mtime = wfTimestamp( TS_UNIX, $page->getTimestamp() );
1696 * // Get adjusted TTL. If $mtime is 3600 seconds ago and $minTTL/$factor left at
1697 * // defaults, then $ttl is 3600 * .2 = 720. If $minTTL was greater than 720, then
1698 * // $ttl would be $minTTL. If $maxTTL was smaller than 720, $ttl would be $maxTTL.
1699 * $ttl = $cache->adaptiveTTL( $mtime, $cache::TTL_DAY );
1700 * @endcode
1701 *
1702 * Another use case is when there are no applicable "last modified" fields in the DB,
1703 * and there are too many dependencies for explicit purges to be viable, and the rate of
1704 * change to relevant content is unstable, and it is highly valued to have the cached value
1705 * be as up-to-date as possible.
1706 *
1707 * Example usage:
1708 * @code
1709 * $query = "<some complex query>";
1710 * $idListFromComplexQuery = $cache->getWithSetCallback(
1711 * $cache->makeKey( 'complex-graph-query', $hashOfQuery ),
1712 * GraphQueryClass::STARTING_TTL,
1713 * function ( $oldValue, &$ttl, array &$setOpts, $oldAsOf ) use ( $query, $cache ) {
1714 * $gdb = $this->getReplicaGraphDbConnection();
1715 * // Account for any snapshot/replica DB lag
1716 * $setOpts += GraphDatabase::getCacheSetOptions( $gdb );
1717 *
1718 * $newList = iterator_to_array( $gdb->query( $query ) );
1719 * sort( $newList, SORT_NUMERIC ); // normalize
1720 *
1721 * $minTTL = GraphQueryClass::MIN_TTL;
1722 * $maxTTL = GraphQueryClass::MAX_TTL;
1723 * if ( $oldValue !== false ) {
1724 * // Note that $oldAsOf is the last time this callback ran
1725 * $ttl = ( $newList === $oldValue )
1726 * // No change: cache for 150% of the age of $oldValue
1727 * ? $cache->adaptiveTTL( $oldAsOf, $maxTTL, $minTTL, 1.5 )
1728 * // Changed: cache for 50% of the age of $oldValue
1729 * : $cache->adaptiveTTL( $oldAsOf, $maxTTL, $minTTL, .5 );
1730 * }
1731 *
1732 * return $newList;
1733 * },
1734 * [
1735 * // Keep stale values around for doing comparisons for TTL calculations.
1736 * // High values improve long-tail keys hit-rates, though might waste space.
1737 * 'staleTTL' => GraphQueryClass::GRACE_TTL
1738 * ]
1739 * );
1740 * @endcode
1741 *
1742 * @param int|float $mtime UNIX timestamp
1743 * @param int $maxTTL Maximum TTL (seconds)
1744 * @param int $minTTL Minimum TTL (seconds); Default: 30
1745 * @param float $factor Value in the range (0,1); Default: .2
1746 * @return int Adaptive TTL
1747 * @since 1.28
1748 */
1749 public function adaptiveTTL( $mtime, $maxTTL, $minTTL = 30, $factor = 0.2 ) {
1750 if ( is_float( $mtime ) || ctype_digit( $mtime ) ) {
1751 $mtime = (int)$mtime; // handle fractional seconds and string integers
1752 }
1753
1754 if ( !is_int( $mtime ) || $mtime <= 0 ) {
1755 return $minTTL; // no last-modified time provided
1756 }
1757
1758 $age = $this->getCurrentTime() - $mtime;
1759
1760 return (int)min( $maxTTL, max( $minTTL, $factor * $age ) );
1761 }
1762
1763 /**
1764 * @return int Number of warmup key cache misses last round
1765 * @since 1.30
1766 */
1767 final public function getWarmupKeyMisses() {
1768 return $this->warmupKeyMisses;
1769 }
1770
1771 /**
1772 * Do the actual async bus purge of a key
1773 *
1774 * This must set the key to "PURGED:<UNIX timestamp>:<holdoff>"
1775 *
1776 * @param string $key Cache key
1777 * @param int $ttl How long to keep the tombstone [seconds]
1778 * @param int $holdoff HOLDOFF_* constant controlling how long to ignore sets for this key
1779 * @return bool Success
1780 */
1781 protected function relayPurge( $key, $ttl, $holdoff ) {
1782 if ( $this->purgeRelayer instanceof EventRelayerNull ) {
1783 // This handles the mcrouter and the single-DC case
1784 $ok = $this->cache->set( $key,
1785 $this->makePurgeValue( $this->getCurrentTime(), self::HOLDOFF_NONE ),
1786 $ttl
1787 );
1788 } else {
1789 $event = $this->cache->modifySimpleRelayEvent( [
1790 'cmd' => 'set',
1791 'key' => $key,
1792 'val' => 'PURGED:$UNIXTIME$:' . (int)$holdoff,
1793 'ttl' => max( $ttl, 1 ),
1794 'sbt' => true, // substitute $UNIXTIME$ with actual microtime
1795 ] );
1796
1797 $ok = $this->purgeRelayer->notify( $this->purgeChannel, $event );
1798 if ( !$ok ) {
1799 $this->lastRelayError = self::ERR_RELAY;
1800 }
1801 }
1802
1803 return $ok;
1804 }
1805
1806 /**
1807 * Do the actual async bus delete of a key
1808 *
1809 * @param string $key Cache key
1810 * @return bool Success
1811 */
1812 protected function relayDelete( $key ) {
1813 if ( $this->purgeRelayer instanceof EventRelayerNull ) {
1814 // This handles the mcrouter and the single-DC case
1815 $ok = $this->cache->delete( $key );
1816 } else {
1817 $event = $this->cache->modifySimpleRelayEvent( [
1818 'cmd' => 'delete',
1819 'key' => $key,
1820 ] );
1821
1822 $ok = $this->purgeRelayer->notify( $this->purgeChannel, $event );
1823 if ( !$ok ) {
1824 $this->lastRelayError = self::ERR_RELAY;
1825 }
1826 }
1827
1828 return $ok;
1829 }
1830
1831 /**
1832 * Check if a key is fresh or in the grace window and thus due for randomized reuse
1833 *
1834 * If $curTTL > 0 (e.g. not expired) this returns true. Otherwise, the chance of returning
1835 * true decrease steadily from 100% to 0% as the |$curTTL| moves from 0 to $graceTTL seconds.
1836 * This handles widely varying levels of cache access traffic.
1837 *
1838 * If $curTTL <= -$graceTTL (e.g. already expired), then this returns false.
1839 *
1840 * @param float $curTTL Approximate TTL left on the key if present
1841 * @param int $graceTTL Consider using stale values if $curTTL is greater than this
1842 * @return bool
1843 */
1844 protected function isAliveOrInGracePeriod( $curTTL, $graceTTL ) {
1845 if ( $curTTL > 0 ) {
1846 return true;
1847 } elseif ( $graceTTL <= 0 ) {
1848 return false;
1849 }
1850
1851 $ageStale = abs( $curTTL ); // seconds of staleness
1852 $curGTTL = ( $graceTTL - $ageStale ); // current grace-time-to-live
1853 if ( $curGTTL <= 0 ) {
1854 return false; // already out of grace period
1855 }
1856
1857 // Chance of using a stale value is the complement of the chance of refreshing it
1858 return !$this->worthRefreshExpiring( $curGTTL, $graceTTL );
1859 }
1860
1861 /**
1862 * Check if a key is nearing expiration and thus due for randomized regeneration
1863 *
1864 * This returns false if $curTTL >= $lowTTL. Otherwise, the chance of returning true
1865 * increases steadily from 0% to 100% as the $curTTL moves from $lowTTL to 0 seconds.
1866 * This handles widely varying levels of cache access traffic.
1867 *
1868 * If $curTTL <= 0 (e.g. already expired), then this returns false.
1869 *
1870 * @param float $curTTL Approximate TTL left on the key if present
1871 * @param float $lowTTL Consider a refresh when $curTTL is less than this
1872 * @return bool
1873 */
1874 protected function worthRefreshExpiring( $curTTL, $lowTTL ) {
1875 if ( $lowTTL <= 0 ) {
1876 return false;
1877 } elseif ( $curTTL >= $lowTTL ) {
1878 return false;
1879 } elseif ( $curTTL <= 0 ) {
1880 return false;
1881 }
1882
1883 $chance = ( 1 - $curTTL / $lowTTL );
1884
1885 return mt_rand( 1, 1e9 ) <= 1e9 * $chance;
1886 }
1887
1888 /**
1889 * Check if a key is due for randomized regeneration due to its popularity
1890 *
1891 * This is used so that popular keys can preemptively refresh themselves for higher
1892 * consistency (especially in the case of purge loss/delay). Unpopular keys can remain
1893 * in cache with their high nominal TTL. This means popular keys keep good consistency,
1894 * whether the data changes frequently or not, and long-tail keys get to stay in cache
1895 * and get hits too. Similar to worthRefreshExpiring(), randomization is used.
1896 *
1897 * @param float $asOf UNIX timestamp of the value
1898 * @param int $ageNew Age of key when this might recommend refreshing (seconds)
1899 * @param int $timeTillRefresh Age of key when it should be refreshed if popular (seconds)
1900 * @param float $now The current UNIX timestamp
1901 * @return bool
1902 */
1903 protected function worthRefreshPopular( $asOf, $ageNew, $timeTillRefresh, $now ) {
1904 if ( $ageNew < 0 || $timeTillRefresh <= 0 ) {
1905 return false;
1906 }
1907
1908 $age = $now - $asOf;
1909 $timeOld = $age - $ageNew;
1910 if ( $timeOld <= 0 ) {
1911 return false;
1912 }
1913
1914 // Lifecycle is: new, ramp-up refresh chance, full refresh chance.
1915 // Note that the "expected # of refreshes" for the ramp-up time range is half of what it
1916 // would be if P(refresh) was at its full value during that time range.
1917 $refreshWindowSec = max( $timeTillRefresh - $ageNew - self::RAMPUP_TTL / 2, 1 );
1918 // P(refresh) * (# hits in $refreshWindowSec) = (expected # of refreshes)
1919 // P(refresh) * ($refreshWindowSec * $popularHitsPerSec) = 1
1920 // P(refresh) = 1/($refreshWindowSec * $popularHitsPerSec)
1921 $chance = 1 / ( self::HIT_RATE_HIGH * $refreshWindowSec );
1922
1923 // Ramp up $chance from 0 to its nominal value over RAMPUP_TTL seconds to avoid stampedes
1924 $chance *= ( $timeOld <= self::RAMPUP_TTL ) ? $timeOld / self::RAMPUP_TTL : 1;
1925
1926 return mt_rand( 1, 1e9 ) <= 1e9 * $chance;
1927 }
1928
1929 /**
1930 * Check whether $value is appropriately versioned and not older than $minTime (if set)
1931 *
1932 * @param array $value
1933 * @param bool $versioned
1934 * @param float $asOf The time $value was generated
1935 * @param float $minTime The last time the main value was generated (0.0 if unknown)
1936 * @return bool
1937 */
1938 protected function isValid( $value, $versioned, $asOf, $minTime ) {
1939 if ( $versioned && !isset( $value[self::VFLD_VERSION] ) ) {
1940 return false;
1941 } elseif ( $minTime > 0 && $asOf < $minTime ) {
1942 return false;
1943 }
1944
1945 return true;
1946 }
1947
1948 /**
1949 * Do not use this method outside WANObjectCache
1950 *
1951 * @param mixed $value
1952 * @param int $ttl [0=forever]
1953 * @param float $now Unix Current timestamp just before calling set()
1954 * @return array
1955 */
1956 protected function wrap( $value, $ttl, $now ) {
1957 return [
1958 self::FLD_VERSION => self::VERSION,
1959 self::FLD_VALUE => $value,
1960 self::FLD_TTL => $ttl,
1961 self::FLD_TIME => $now
1962 ];
1963 }
1964
1965 /**
1966 * Do not use this method outside WANObjectCache
1967 *
1968 * @param array|string|bool $wrapped
1969 * @param float $now Unix Current timestamp (preferrably pre-query)
1970 * @return array (mixed; false if absent/tombstoned/invalid, current time left)
1971 */
1972 protected function unwrap( $wrapped, $now ) {
1973 // Check if the value is a tombstone
1974 $purge = $this->parsePurgeValue( $wrapped );
1975 if ( $purge !== false ) {
1976 // Purged values should always have a negative current $ttl
1977 $curTTL = min( $purge[self::FLD_TIME] - $now, self::TINY_NEGATIVE );
1978 return [ false, $curTTL ];
1979 }
1980
1981 if ( !is_array( $wrapped ) // not found
1982 || !isset( $wrapped[self::FLD_VERSION] ) // wrong format
1983 || $wrapped[self::FLD_VERSION] !== self::VERSION // wrong version
1984 ) {
1985 return [ false, null ];
1986 }
1987
1988 $flags = isset( $wrapped[self::FLD_FLAGS] ) ? $wrapped[self::FLD_FLAGS] : 0;
1989 if ( ( $flags & self::FLG_STALE ) == self::FLG_STALE ) {
1990 // Treat as expired, with the cache time as the expiration
1991 $age = $now - $wrapped[self::FLD_TIME];
1992 $curTTL = min( -$age, self::TINY_NEGATIVE );
1993 } elseif ( $wrapped[self::FLD_TTL] > 0 ) {
1994 // Get the approximate time left on the key
1995 $age = $now - $wrapped[self::FLD_TIME];
1996 $curTTL = max( $wrapped[self::FLD_TTL] - $age, 0.0 );
1997 } else {
1998 // Key had no TTL, so the time left is unbounded
1999 $curTTL = INF;
2000 }
2001
2002 return [ $wrapped[self::FLD_VALUE], $curTTL ];
2003 }
2004
2005 /**
2006 * @param array $keys
2007 * @param string $prefix
2008 * @return string[]
2009 */
2010 protected static function prefixCacheKeys( array $keys, $prefix ) {
2011 $res = [];
2012 foreach ( $keys as $key ) {
2013 $res[] = $prefix . $key;
2014 }
2015
2016 return $res;
2017 }
2018
2019 /**
2020 * @param string $key String of the format <scope>:<class>[:<class or variable>]...
2021 * @return string
2022 */
2023 protected function determineKeyClass( $key ) {
2024 $parts = explode( ':', $key );
2025
2026 return isset( $parts[1] ) ? $parts[1] : $parts[0]; // sanity
2027 }
2028
2029 /**
2030 * @return float UNIX timestamp
2031 * @codeCoverageIgnore
2032 */
2033 protected function getCurrentTime() {
2034 return microtime( true );
2035 }
2036
2037 /**
2038 * @param string $value Wrapped value like "PURGED:<timestamp>:<holdoff>"
2039 * @return array|bool Array containing a UNIX timestamp (float) and holdoff period (integer),
2040 * or false if value isn't a valid purge value
2041 */
2042 protected function parsePurgeValue( $value ) {
2043 if ( !is_string( $value ) ) {
2044 return false;
2045 }
2046 $segments = explode( ':', $value, 3 );
2047 if ( !isset( $segments[0] ) || !isset( $segments[1] )
2048 || "{$segments[0]}:" !== self::PURGE_VAL_PREFIX
2049 ) {
2050 return false;
2051 }
2052 if ( !isset( $segments[2] ) ) {
2053 // Back-compat with old purge values without holdoff
2054 $segments[2] = self::HOLDOFF_TTL;
2055 }
2056 return [
2057 self::FLD_TIME => (float)$segments[1],
2058 self::FLD_HOLDOFF => (int)$segments[2],
2059 ];
2060 }
2061
2062 /**
2063 * @param float $timestamp
2064 * @param int $holdoff In seconds
2065 * @return string Wrapped purge value
2066 */
2067 protected function makePurgeValue( $timestamp, $holdoff ) {
2068 return self::PURGE_VAL_PREFIX . (float)$timestamp . ':' . (int)$holdoff;
2069 }
2070
2071 /**
2072 * @param string $group
2073 * @return HashBagOStuff
2074 */
2075 protected function getProcessCache( $group ) {
2076 if ( !isset( $this->processCaches[$group] ) ) {
2077 list( , $n ) = explode( ':', $group );
2078 $this->processCaches[$group] = new HashBagOStuff( [ 'maxKeys' => (int)$n ] );
2079 }
2080
2081 return $this->processCaches[$group];
2082 }
2083
2084 /**
2085 * @param array $keys
2086 * @param array $opts
2087 * @return array List of keys
2088 */
2089 private function getNonProcessCachedKeys( array $keys, array $opts ) {
2090 $keysFound = [];
2091 if ( isset( $opts['pcTTL'] ) && $opts['pcTTL'] > 0 && $this->callbackDepth == 0 ) {
2092 $pcGroup = isset( $opts['pcGroup'] ) ? $opts['pcGroup'] : self::PC_PRIMARY;
2093 $procCache = $this->getProcessCache( $pcGroup );
2094 foreach ( $keys as $key ) {
2095 if ( $procCache->get( $key ) !== false ) {
2096 $keysFound[] = $key;
2097 }
2098 }
2099 }
2100
2101 return array_diff( $keys, $keysFound );
2102 }
2103
2104 /**
2105 * @param array $keys
2106 * @param array $checkKeys
2107 * @return array Map of (cache key => mixed)
2108 */
2109 private function getRawKeysForWarmup( array $keys, array $checkKeys ) {
2110 if ( !$keys ) {
2111 return [];
2112 }
2113
2114 $keysWarmUp = [];
2115 // Get all the value keys to fetch...
2116 foreach ( $keys as $key ) {
2117 $keysWarmUp[] = self::VALUE_KEY_PREFIX . $key;
2118 }
2119 // Get all the check keys to fetch...
2120 foreach ( $checkKeys as $i => $checkKeyOrKeys ) {
2121 if ( is_int( $i ) ) {
2122 // Single check key that applies to all value keys
2123 $keysWarmUp[] = self::TIME_KEY_PREFIX . $checkKeyOrKeys;
2124 } else {
2125 // List of check keys that apply to value key $i
2126 $keysWarmUp = array_merge(
2127 $keysWarmUp,
2128 self::prefixCacheKeys( $checkKeyOrKeys, self::TIME_KEY_PREFIX )
2129 );
2130 }
2131 }
2132
2133 $warmupCache = $this->cache->getMulti( $keysWarmUp );
2134 $warmupCache += array_fill_keys( $keysWarmUp, false );
2135
2136 return $warmupCache;
2137 }
2138 }