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