Merge "Handle missing namespace prefix in XML dumps more gracefully"
[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 * @author Aaron Schulz
21 */
22
23 use Psr\Log\LoggerAwareInterface;
24 use Psr\Log\LoggerInterface;
25 use Psr\Log\NullLogger;
26
27 /**
28 * Multi-datacenter aware caching interface
29 *
30 * All operations go to the local datacenter cache, except for delete(),
31 * touchCheckKey(), and resetCheckKey(), which broadcast to all datacenters.
32 *
33 * This class is intended for caching data from primary stores.
34 * If the get() method does not return a value, then the caller
35 * should query the new value and backfill the cache using set().
36 * The preferred way to do this logic is through getWithSetCallback().
37 * When querying the store on cache miss, the closest DB replica
38 * should be used. Try to avoid heavyweight DB master or quorum reads.
39 * When the source data changes, a purge method should be called.
40 * Since purges are expensive, they should be avoided. One can do so if:
41 * - a) The object cached is immutable; or
42 * - b) Validity is checked against the source after get(); or
43 * - c) Using a modest TTL is reasonably correct and performant
44 *
45 * The simplest purge method is delete().
46 *
47 * There are 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
93 /** @var int ERR_* constant for the "last error" registry */
94 protected $lastRelayError = self::ERR_NONE;
95
96 /** @var integer Callback stack depth for getWithSetCallback() */
97 private $callbackDepth = 0;
98 /** @var mixed[] Temporary warm-up cache */
99 private $warmupCache = [];
100
101 /** Max time expected to pass between delete() and DB commit finishing */
102 const MAX_COMMIT_DELAY = 3;
103 /** Max replication+snapshot lag before applying TTL_LAGGED or disallowing set() */
104 const MAX_READ_LAG = 7;
105 /** Seconds to tombstone keys on delete() */
106 const HOLDOFF_TTL = 11; // MAX_COMMIT_DELAY + MAX_READ_LAG + 1
107
108 /** Seconds to keep dependency purge keys around */
109 const CHECK_KEY_TTL = self::TTL_YEAR;
110 /** Seconds to keep lock keys around */
111 const LOCK_TTL = 10;
112 /** Default remaining TTL at which to consider pre-emptive regeneration */
113 const LOW_TTL = 30;
114 /** Default time-since-expiry on a miss that makes a key "hot" */
115 const LOCK_TSE = 1;
116
117 /** Never consider performing "popularity" refreshes until a key reaches this age */
118 const AGE_NEW = 60;
119 /** The time length of the "popularity" refresh window for hot keys */
120 const HOT_TTR = 900;
121 /** Hits/second for a refresh to be expected within the "popularity" window */
122 const HIT_RATE_HIGH = 1;
123 /** Seconds to ramp up to the "popularity" refresh chance after a key is no longer new */
124 const RAMPUP_TTL = 30;
125
126 /** Idiom for getWithSetCallback() callbacks to avoid calling set() */
127 const TTL_UNCACHEABLE = -1;
128 /** Idiom for getWithSetCallback() callbacks to 'lockTSE' logic */
129 const TSE_NONE = -1;
130 /** Max TTL to store keys when a data sourced is lagged */
131 const TTL_LAGGED = 30;
132 /** Idiom for delete() for "no hold-off" */
133 const HOLDOFF_NONE = 0;
134 /** Idiom for getWithSetCallback() for "no minimum required as-of timestamp" */
135 const MIN_TIMESTAMP_NONE = 0.0;
136
137 /** Tiny negative float to use when CTL comes up >= 0 due to clock skew */
138 const TINY_NEGATIVE = -0.000001;
139
140 /** Cache format version number */
141 const VERSION = 1;
142
143 const FLD_VERSION = 0; // key to cache version number
144 const FLD_VALUE = 1; // key to the cached value
145 const FLD_TTL = 2; // key to the original TTL
146 const FLD_TIME = 3; // key to the cache time
147 const FLD_FLAGS = 4; // key to the flags bitfield
148 const FLD_HOLDOFF = 5; // key to any hold-off TTL
149
150 /** @var integer Treat this value as expired-on-arrival */
151 const FLG_STALE = 1;
152
153 const ERR_NONE = 0; // no error
154 const ERR_NO_RESPONSE = 1; // no response
155 const ERR_UNREACHABLE = 2; // can't connect
156 const ERR_UNEXPECTED = 3; // response gave some error
157 const ERR_RELAY = 4; // relay broadcast failed
158
159 const VALUE_KEY_PREFIX = 'WANCache:v:';
160 const INTERIM_KEY_PREFIX = 'WANCache:i:';
161 const TIME_KEY_PREFIX = 'WANCache:t:';
162 const MUTEX_KEY_PREFIX = 'WANCache:m:';
163
164 const PURGE_VAL_PREFIX = 'PURGED:';
165
166 const VFLD_DATA = 'WOC:d'; // key to the value of versioned data
167 const VFLD_VERSION = 'WOC:v'; // key to the version of the value present
168
169 const PC_PRIMARY = 'primary:1000'; // process cache name and max key count
170
171 const DEFAULT_PURGE_CHANNEL = 'wancache-purge';
172
173 /**
174 * @param array $params
175 * - cache : BagOStuff object for a persistent cache
176 * - channels : Map of (action => channel string). Actions include "purge".
177 * - relayers : Map of (action => EventRelayer object). Actions include "purge".
178 * - logger : LoggerInterface object
179 */
180 public function __construct( array $params ) {
181 $this->cache = $params['cache'];
182 $this->purgeChannel = isset( $params['channels']['purge'] )
183 ? $params['channels']['purge']
184 : self::DEFAULT_PURGE_CHANNEL;
185 $this->purgeRelayer = isset( $params['relayers']['purge'] )
186 ? $params['relayers']['purge']
187 : new EventRelayerNull( [] );
188 $this->setLogger( isset( $params['logger'] ) ? $params['logger'] : new NullLogger() );
189 }
190
191 public function setLogger( LoggerInterface $logger ) {
192 $this->logger = $logger;
193 }
194
195 /**
196 * Get an instance that wraps EmptyBagOStuff
197 *
198 * @return WANObjectCache
199 */
200 public static function newEmpty() {
201 return new self( [
202 'cache' => new EmptyBagOStuff(),
203 'pool' => 'empty',
204 'relayer' => new EventRelayerNull( [] )
205 ] );
206 }
207
208 /**
209 * Fetch the value of a key from cache
210 *
211 * If supplied, $curTTL is set to the remaining TTL (current time left):
212 * - a) INF; if $key exists, has no TTL, and is not expired by $checkKeys
213 * - b) float (>=0); if $key exists, has a TTL, and is not expired by $checkKeys
214 * - c) float (<0); if $key is tombstoned, stale, or existing but expired by $checkKeys
215 * - d) null; if $key does not exist and is not tombstoned
216 *
217 * If a key is tombstoned, $curTTL will reflect the time since delete().
218 *
219 * The timestamp of $key will be checked against the last-purge timestamp
220 * of each of $checkKeys. Those $checkKeys not in cache will have the last-purge
221 * initialized to the current timestamp. If any of $checkKeys have a timestamp
222 * greater than that of $key, then $curTTL will reflect how long ago $key
223 * became invalid. Callers can use $curTTL to know when the value is stale.
224 * The $checkKeys parameter allow mass invalidations by updating a single key:
225 * - a) Each "check" key represents "last purged" of some source data
226 * - b) Callers pass in relevant "check" keys as $checkKeys in get()
227 * - c) When the source data that "check" keys represent changes,
228 * the touchCheckKey() method is called on them
229 *
230 * Source data entities might exists in a DB that uses snapshot isolation
231 * (e.g. the default REPEATABLE-READ in innoDB). Even for mutable data, that
232 * isolation can largely be maintained by doing the following:
233 * - a) Calling delete() on entity change *and* creation, before DB commit
234 * - b) Keeping transaction duration shorter than delete() hold-off TTL
235 *
236 * However, pre-snapshot values might still be seen if an update was made
237 * in a remote datacenter but the purge from delete() didn't relay yet.
238 *
239 * Consider using getWithSetCallback() instead of get() and set() cycles.
240 * That method has cache slam avoiding features for hot/expensive keys.
241 *
242 * @param string $key Cache key
243 * @param mixed $curTTL Approximate TTL left on the key if present/tombstoned [returned]
244 * @param array $checkKeys List of "check" keys
245 * @param float &$asOf UNIX timestamp of cached value; null on failure [returned]
246 * @return mixed Value of cache key or false on failure
247 */
248 final public function get( $key, &$curTTL = null, array $checkKeys = [], &$asOf = null ) {
249 $curTTLs = [];
250 $asOfs = [];
251 $values = $this->getMulti( [ $key ], $curTTLs, $checkKeys, $asOfs );
252 $curTTL = isset( $curTTLs[$key] ) ? $curTTLs[$key] : null;
253 $asOf = isset( $asOfs[$key] ) ? $asOfs[$key] : null;
254
255 return isset( $values[$key] ) ? $values[$key] : false;
256 }
257
258 /**
259 * Fetch the value of several keys from cache
260 *
261 * @see WANObjectCache::get()
262 *
263 * @param array $keys List of cache keys
264 * @param array $curTTLs Map of (key => approximate TTL left) for existing keys [returned]
265 * @param array $checkKeys List of check keys to apply to all $keys. May also apply "check"
266 * keys to specific cache keys only by using cache keys as keys in the $checkKeys array.
267 * @param float[] &$asOfs Map of (key => UNIX timestamp of cached value; null on failure)
268 * @return array Map of (key => value) for keys that exist
269 */
270 final public function getMulti(
271 array $keys, &$curTTLs = [], array $checkKeys = [], array &$asOfs = []
272 ) {
273 $result = [];
274 $curTTLs = [];
275 $asOfs = [];
276
277 $vPrefixLen = strlen( self::VALUE_KEY_PREFIX );
278 $valueKeys = self::prefixCacheKeys( $keys, self::VALUE_KEY_PREFIX );
279
280 $checkKeysForAll = [];
281 $checkKeysByKey = [];
282 $checkKeysFlat = [];
283 foreach ( $checkKeys as $i => $checkKeyGroup ) {
284 $prefixed = self::prefixCacheKeys( (array)$checkKeyGroup, self::TIME_KEY_PREFIX );
285 $checkKeysFlat = array_merge( $checkKeysFlat, $prefixed );
286 // Is this check keys for a specific cache key, or for all keys being fetched?
287 if ( is_int( $i ) ) {
288 $checkKeysForAll = array_merge( $checkKeysForAll, $prefixed );
289 } else {
290 $checkKeysByKey[$i] = isset( $checkKeysByKey[$i] )
291 ? array_merge( $checkKeysByKey[$i], $prefixed )
292 : $prefixed;
293 }
294 }
295
296 // Fetch all of the raw values
297 $keysGet = array_merge( $valueKeys, $checkKeysFlat );
298 if ( $this->warmupCache ) {
299 $wrappedValues = array_intersect_key( $this->warmupCache, array_flip( $keysGet ) );
300 $keysGet = array_diff( $keysGet, array_keys( $wrappedValues ) ); // keys left to fetch
301 } else {
302 $wrappedValues = [];
303 }
304 $wrappedValues += $this->cache->getMulti( $keysGet );
305 // Time used to compare/init "check" keys (derived after getMulti() to be pessimistic)
306 $now = microtime( true );
307
308 // Collect timestamps from all "check" keys
309 $purgeValuesForAll = $this->processCheckKeys( $checkKeysForAll, $wrappedValues, $now );
310 $purgeValuesByKey = [];
311 foreach ( $checkKeysByKey as $cacheKey => $checks ) {
312 $purgeValuesByKey[$cacheKey] =
313 $this->processCheckKeys( $checks, $wrappedValues, $now );
314 }
315
316 // Get the main cache value for each key and validate them
317 foreach ( $valueKeys as $vKey ) {
318 if ( !isset( $wrappedValues[$vKey] ) ) {
319 continue; // not found
320 }
321
322 $key = substr( $vKey, $vPrefixLen ); // unprefix
323
324 list( $value, $curTTL ) = $this->unwrap( $wrappedValues[$vKey], $now );
325 if ( $value !== false ) {
326 $result[$key] = $value;
327
328 // Force dependant keys to be invalid for a while after purging
329 // to reduce race conditions involving stale data getting cached
330 $purgeValues = $purgeValuesForAll;
331 if ( isset( $purgeValuesByKey[$key] ) ) {
332 $purgeValues = array_merge( $purgeValues, $purgeValuesByKey[$key] );
333 }
334 foreach ( $purgeValues as $purge ) {
335 $safeTimestamp = $purge[self::FLD_TIME] + $purge[self::FLD_HOLDOFF];
336 if ( $safeTimestamp >= $wrappedValues[$vKey][self::FLD_TIME] ) {
337 // How long ago this value was expired by *this* check key
338 $ago = min( $purge[self::FLD_TIME] - $now, self::TINY_NEGATIVE );
339 // How long ago this value was expired by *any* known check key
340 $curTTL = min( $curTTL, $ago );
341 }
342 }
343 }
344 $curTTLs[$key] = $curTTL;
345 $asOfs[$key] = ( $value !== false ) ? $wrappedValues[$vKey][self::FLD_TIME] : null;
346 }
347
348 return $result;
349 }
350
351 /**
352 * @since 1.27
353 * @param array $timeKeys List of prefixed time check keys
354 * @param array $wrappedValues
355 * @param float $now
356 * @return array List of purge value arrays
357 */
358 private function processCheckKeys( array $timeKeys, array $wrappedValues, $now ) {
359 $purgeValues = [];
360 foreach ( $timeKeys as $timeKey ) {
361 $purge = isset( $wrappedValues[$timeKey] )
362 ? self::parsePurgeValue( $wrappedValues[$timeKey] )
363 : false;
364 if ( $purge === false ) {
365 // Key is not set or invalid; regenerate
366 $newVal = $this->makePurgeValue( $now, self::HOLDOFF_TTL );
367 $this->cache->add( $timeKey, $newVal, self::CHECK_KEY_TTL );
368 $purge = self::parsePurgeValue( $newVal );
369 }
370 $purgeValues[] = $purge;
371 }
372 return $purgeValues;
373 }
374
375 /**
376 * Set the value of a key in cache
377 *
378 * Simply calling this method when source data changes is not valid because
379 * the changes do not replicate to the other WAN sites. In that case, delete()
380 * should be used instead. This method is intended for use on cache misses.
381 *
382 * If the data was read from a snapshot-isolated transactions (e.g. the default
383 * REPEATABLE-READ in innoDB), use 'since' to avoid the following race condition:
384 * - a) T1 starts
385 * - b) T2 updates a row, calls delete(), and commits
386 * - c) The HOLDOFF_TTL passes, expiring the delete() tombstone
387 * - d) T1 reads the row and calls set() due to a cache miss
388 * - e) Stale value is stuck in cache
389 *
390 * Setting 'lag' and 'since' help avoids keys getting stuck in stale states.
391 *
392 * Example usage:
393 * @code
394 * $dbr = wfGetDB( DB_REPLICA );
395 * $setOpts = Database::getCacheSetOptions( $dbr );
396 * // Fetch the row from the DB
397 * $row = $dbr->selectRow( ... );
398 * $key = $cache->makeKey( 'building', $buildingId );
399 * $cache->set( $key, $row, $cache::TTL_DAY, $setOpts );
400 * @endcode
401 *
402 * @param string $key Cache key
403 * @param mixed $value
404 * @param integer $ttl Seconds to live. Special values are:
405 * - WANObjectCache::TTL_INDEFINITE: Cache forever
406 * @param array $opts Options map:
407 * - lag : Seconds of replica DB lag. Typically, this is either the replica DB lag
408 * before the data was read or, if applicable, the replica DB lag before
409 * the snapshot-isolated transaction the data was read from started.
410 * Use false to indicate that replication is not running.
411 * Default: 0 seconds
412 * - since : UNIX timestamp of the data in $value. Typically, this is either
413 * the current time the data was read or (if applicable) the time when
414 * the snapshot-isolated transaction the data was read from started.
415 * Default: 0 seconds
416 * - pending : Whether this data is possibly from an uncommitted write transaction.
417 * Generally, other threads should not see values from the future and
418 * they certainly should not see ones that ended up getting rolled back.
419 * Default: false
420 * - lockTSE : if excessive replication/snapshot lag is detected, then store the value
421 * with this TTL and flag it as stale. This is only useful if the reads for
422 * this key use getWithSetCallback() with "lockTSE" set.
423 * Default: WANObjectCache::TSE_NONE
424 * - staleTTL : Seconds to keep the key around if it is stale. The get()/getMulti()
425 * methods return such stale values with a $curTTL of 0, and getWithSetCallback()
426 * will call the regeneration callback in such cases, passing in the old value
427 * and its as-of time to the callback. This is useful if adaptiveTTL() is used
428 * on the old value's as-of time when it is verified as still being correct.
429 * Default: 0.
430 * @note Options added in 1.28: staleTTL
431 * @return bool Success
432 */
433 final public function set( $key, $value, $ttl = 0, array $opts = [] ) {
434 $now = microtime( true );
435 $lockTSE = isset( $opts['lockTSE'] ) ? $opts['lockTSE'] : self::TSE_NONE;
436 $age = isset( $opts['since'] ) ? max( 0, $now - $opts['since'] ) : 0;
437 $lag = isset( $opts['lag'] ) ? $opts['lag'] : 0;
438 $staleTTL = isset( $opts['staleTTL'] ) ? $opts['staleTTL'] : 0;
439
440 // Do not cache potentially uncommitted data as it might get rolled back
441 if ( !empty( $opts['pending'] ) ) {
442 $this->logger->info( "Rejected set() for $key due to pending writes." );
443
444 return true; // no-op the write for being unsafe
445 }
446
447 $wrapExtra = []; // additional wrapped value fields
448 // Check if there's a risk of writing stale data after the purge tombstone expired
449 if ( $lag === false || ( $lag + $age ) > self::MAX_READ_LAG ) {
450 // Case A: read lag with "lockTSE"; save but record value as stale
451 if ( $lockTSE >= 0 ) {
452 $ttl = max( 1, (int)$lockTSE ); // set() expects seconds
453 $wrapExtra[self::FLD_FLAGS] = self::FLG_STALE; // mark as stale
454 // Case B: any long-running transaction; ignore this set()
455 } elseif ( $age > self::MAX_READ_LAG ) {
456 $this->logger->info( "Rejected set() for $key due to snapshot lag." );
457
458 return true; // no-op the write for being unsafe
459 // Case C: high replication lag; lower TTL instead of ignoring all set()s
460 } elseif ( $lag === false || $lag > self::MAX_READ_LAG ) {
461 $ttl = $ttl ? min( $ttl, self::TTL_LAGGED ) : self::TTL_LAGGED;
462 $this->logger->warning( "Lowered set() TTL for $key due to replication lag." );
463 // Case D: medium length request with medium replication lag; ignore this set()
464 } else {
465 $this->logger->info( "Rejected set() for $key due to high read lag." );
466
467 return true; // no-op the write for being unsafe
468 }
469 }
470
471 // Wrap that value with time/TTL/version metadata
472 $wrapped = $this->wrap( $value, $ttl, $now ) + $wrapExtra;
473
474 $func = function ( $cache, $key, $cWrapped ) use ( $wrapped ) {
475 return ( is_string( $cWrapped ) )
476 ? false // key is tombstoned; do nothing
477 : $wrapped;
478 };
479
480 return $this->cache->merge( self::VALUE_KEY_PREFIX . $key, $func, $ttl + $staleTTL, 1 );
481 }
482
483 /**
484 * Purge a key from all datacenters
485 *
486 * This should only be called when the underlying data (being cached)
487 * changes in a significant way. This deletes the key and starts a hold-off
488 * period where the key cannot be written to for a few seconds (HOLDOFF_TTL).
489 * This is done to avoid the following race condition:
490 * - a) Some DB data changes and delete() is called on a corresponding key
491 * - b) A request refills the key with a stale value from a lagged DB
492 * - c) The stale value is stuck there until the key is expired/evicted
493 *
494 * This is implemented by storing a special "tombstone" value at the cache
495 * key that this class recognizes; get() calls will return false for the key
496 * and any set() calls will refuse to replace tombstone values at the key.
497 * For this to always avoid stale value writes, the following must hold:
498 * - a) Replication lag is bounded to being less than HOLDOFF_TTL; or
499 * - b) If lag is higher, the DB will have gone into read-only mode already
500 *
501 * Note that set() can also be lag-aware and lower the TTL if it's high.
502 *
503 * When using potentially long-running ACID transactions, a good pattern is
504 * to use a pre-commit hook to issue the delete. This means that immediately
505 * after commit, callers will see the tombstone in cache upon purge relay.
506 * It also avoids the following race condition:
507 * - a) T1 begins, changes a row, and calls delete()
508 * - b) The HOLDOFF_TTL passes, expiring the delete() tombstone
509 * - c) T2 starts, reads the row and calls set() due to a cache miss
510 * - d) T1 finally commits
511 * - e) Stale value is stuck in cache
512 *
513 * Example usage:
514 * @code
515 * $dbw->startAtomic( __METHOD__ ); // start of request
516 * ... <execute some stuff> ...
517 * // Update the row in the DB
518 * $dbw->update( ... );
519 * $key = $cache->makeKey( 'homes', $homeId );
520 * // Purge the corresponding cache entry just before committing
521 * $dbw->onTransactionPreCommitOrIdle( function() use ( $cache, $key ) {
522 * $cache->delete( $key );
523 * } );
524 * ... <execute some stuff> ...
525 * $dbw->endAtomic( __METHOD__ ); // end of request
526 * @endcode
527 *
528 * The $ttl parameter can be used when purging values that have not actually changed
529 * recently. For example, a cleanup script to purge cache entries does not really need
530 * a hold-off period, so it can use HOLDOFF_NONE. Likewise for user-requested purge.
531 * Note that $ttl limits the effective range of 'lockTSE' for getWithSetCallback().
532 *
533 * If called twice on the same key, then the last hold-off TTL takes precedence. For
534 * idempotence, the $ttl should not vary for different delete() calls on the same key.
535 *
536 * @param string $key Cache key
537 * @param integer $ttl Tombstone TTL; Default: WANObjectCache::HOLDOFF_TTL
538 * @return bool True if the item was purged or not found, false on failure
539 */
540 final public function delete( $key, $ttl = self::HOLDOFF_TTL ) {
541 $key = self::VALUE_KEY_PREFIX . $key;
542
543 if ( $ttl <= 0 ) {
544 // Publish the purge to all datacenters
545 $ok = $this->relayDelete( $key );
546 } else {
547 // Publish the purge to all datacenters
548 $ok = $this->relayPurge( $key, $ttl, self::HOLDOFF_NONE );
549 }
550
551 return $ok;
552 }
553
554 /**
555 * Fetch the value of a timestamp "check" key
556 *
557 * The key will be *initialized* to the current time if not set,
558 * so only call this method if this behavior is actually desired
559 *
560 * The timestamp can be used to check whether a cached value is valid.
561 * Callers should not assume that this returns the same timestamp in
562 * all datacenters due to relay delays.
563 *
564 * The level of staleness can roughly be estimated from this key, but
565 * if the key was evicted from cache, such calculations may show the
566 * time since expiry as ~0 seconds.
567 *
568 * Note that "check" keys won't collide with other regular keys.
569 *
570 * @param string $key
571 * @return float UNIX timestamp of the check key
572 */
573 final public function getCheckKeyTime( $key ) {
574 $key = self::TIME_KEY_PREFIX . $key;
575
576 $purge = self::parsePurgeValue( $this->cache->get( $key ) );
577 if ( $purge !== false ) {
578 $time = $purge[self::FLD_TIME];
579 } else {
580 // Casting assures identical floats for the next getCheckKeyTime() calls
581 $now = (string)microtime( true );
582 $this->cache->add( $key,
583 $this->makePurgeValue( $now, self::HOLDOFF_TTL ),
584 self::CHECK_KEY_TTL
585 );
586 $time = (float)$now;
587 }
588
589 return $time;
590 }
591
592 /**
593 * Purge a "check" key from all datacenters, invalidating keys that use it
594 *
595 * This should only be called when the underlying data (being cached)
596 * changes in a significant way, and it is impractical to call delete()
597 * on all keys that should be changed. When get() is called on those
598 * keys, the relevant "check" keys must be supplied for this to work.
599 *
600 * The "check" key essentially represents a last-modified field.
601 * When touched, the field will be updated on all cache servers.
602 * Keys using it via get(), getMulti(), or getWithSetCallback() will
603 * be invalidated. It is treated as being HOLDOFF_TTL seconds in the future
604 * by those methods to avoid race conditions where dependent keys get updated
605 * with stale values (e.g. from a DB replica DB).
606 *
607 * This is typically useful for keys with hardcoded names or in some cases
608 * dynamically generated names where a low number of combinations exist.
609 * When a few important keys get a large number of hits, a high cache
610 * time is usually desired as well as "lockTSE" logic. The resetCheckKey()
611 * method is less appropriate in such cases since the "time since expiry"
612 * cannot be inferred, causing any get() after the reset to treat the key
613 * as being "hot", resulting in more stale value usage.
614 *
615 * Note that "check" keys won't collide with other regular keys.
616 *
617 * @see WANObjectCache::get()
618 * @see WANObjectCache::getWithSetCallback()
619 * @see WANObjectCache::resetCheckKey()
620 *
621 * @param string $key Cache key
622 * @param int $holdoff HOLDOFF_TTL or HOLDOFF_NONE constant
623 * @return bool True if the item was purged or not found, false on failure
624 */
625 final public function touchCheckKey( $key, $holdoff = self::HOLDOFF_TTL ) {
626 // Publish the purge to all datacenters
627 return $this->relayPurge( self::TIME_KEY_PREFIX . $key, self::CHECK_KEY_TTL, $holdoff );
628 }
629
630 /**
631 * Delete a "check" key from all datacenters, invalidating keys that use it
632 *
633 * This is similar to touchCheckKey() in that keys using it via get(), getMulti(),
634 * or getWithSetCallback() will be invalidated. The differences are:
635 * - a) The "check" key will be deleted from all caches and lazily
636 * re-initialized when accessed (rather than set everywhere)
637 * - b) Thus, dependent keys will be known to be invalid, but not
638 * for how long (they are treated as "just" purged), which
639 * effects any lockTSE logic in getWithSetCallback()
640 * - c) Since "check" keys are initialized only on the server the key hashes
641 * to, any temporary ejection of that server will cause the value to be
642 * seen as purged as a new server will initialize the "check" key.
643 *
644 * The advantage is that this does not place high TTL keys on every cache
645 * server, making it better for code that will cache many different keys
646 * and either does not use lockTSE or uses a low enough TTL anyway.
647 *
648 * This is typically useful for keys with dynamically generated names
649 * where a high number of combinations exist.
650 *
651 * Note that "check" keys won't collide with other regular keys.
652 *
653 * @see WANObjectCache::get()
654 * @see WANObjectCache::getWithSetCallback()
655 * @see WANObjectCache::touchCheckKey()
656 *
657 * @param string $key Cache key
658 * @return bool True if the item was purged or not found, false on failure
659 */
660 final public function resetCheckKey( $key ) {
661 // Publish the purge to all datacenters
662 return $this->relayDelete( self::TIME_KEY_PREFIX . $key );
663 }
664
665 /**
666 * Method to fetch/regenerate cache keys
667 *
668 * On cache miss, the key will be set to the callback result via set()
669 * (unless the callback returns false) and that result will be returned.
670 * The arguments supplied to the callback are:
671 * - $oldValue : current cache value or false if not present
672 * - &$ttl : a reference to the TTL which can be altered
673 * - &$setOpts : a reference to options for set() which can be altered
674 * - $oldAsOf : generation UNIX timestamp of $oldValue or null if not present (since 1.28)
675 *
676 * It is strongly recommended to set the 'lag' and 'since' fields to avoid race conditions
677 * that can cause stale values to get stuck at keys. Usually, callbacks ignore the current
678 * value, but it can be used to maintain "most recent X" values that come from time or
679 * sequence based source data, provided that the "as of" id/time is tracked. Note that
680 * preemptive regeneration and $checkKeys can result in a non-false current value.
681 *
682 * Usage of $checkKeys is similar to get() and getMulti(). However, rather than the caller
683 * having to inspect a "current time left" variable (e.g. $curTTL, $curTTLs), a cache
684 * regeneration will automatically be triggered using the callback.
685 *
686 * The simplest way to avoid stampedes for hot keys is to use
687 * the 'lockTSE' option in $opts. If cache purges are needed, also:
688 * - a) Pass $key into $checkKeys
689 * - b) Use touchCheckKey( $key ) instead of delete( $key )
690 *
691 * Example usage (typical key):
692 * @code
693 * $catInfo = $cache->getWithSetCallback(
694 * // Key to store the cached value under
695 * $cache->makeKey( 'cat-attributes', $catId ),
696 * // Time-to-live (in seconds)
697 * $cache::TTL_MINUTE,
698 * // Function that derives the new key value
699 * function ( $oldValue, &$ttl, array &$setOpts ) {
700 * $dbr = wfGetDB( DB_REPLICA );
701 * // Account for any snapshot/replica DB lag
702 * $setOpts += Database::getCacheSetOptions( $dbr );
703 *
704 * return $dbr->selectRow( ... );
705 * }
706 * );
707 * @endcode
708 *
709 * Example usage (key that is expensive and hot):
710 * @code
711 * $catConfig = $cache->getWithSetCallback(
712 * // Key to store the cached value under
713 * $cache->makeKey( 'site-cat-config' ),
714 * // Time-to-live (in seconds)
715 * $cache::TTL_DAY,
716 * // Function that derives the new key value
717 * function ( $oldValue, &$ttl, array &$setOpts ) {
718 * $dbr = wfGetDB( DB_REPLICA );
719 * // Account for any snapshot/replica DB lag
720 * $setOpts += Database::getCacheSetOptions( $dbr );
721 *
722 * return CatConfig::newFromRow( $dbr->selectRow( ... ) );
723 * },
724 * [
725 * // Calling touchCheckKey() on this key invalidates the cache
726 * 'checkKeys' => [ $cache->makeKey( 'site-cat-config' ) ],
727 * // Try to only let one datacenter thread manage cache updates at a time
728 * 'lockTSE' => 30,
729 * // Avoid querying cache servers multiple times in a web request
730 * 'pcTTL' => $cache::TTL_PROC_LONG
731 * ]
732 * );
733 * @endcode
734 *
735 * Example usage (key with dynamic dependencies):
736 * @code
737 * $catState = $cache->getWithSetCallback(
738 * // Key to store the cached value under
739 * $cache->makeKey( 'cat-state', $cat->getId() ),
740 * // Time-to-live (seconds)
741 * $cache::TTL_HOUR,
742 * // Function that derives the new key value
743 * function ( $oldValue, &$ttl, array &$setOpts ) {
744 * // Determine new value from the DB
745 * $dbr = wfGetDB( DB_REPLICA );
746 * // Account for any snapshot/replica DB lag
747 * $setOpts += Database::getCacheSetOptions( $dbr );
748 *
749 * return CatState::newFromResults( $dbr->select( ... ) );
750 * },
751 * [
752 * // The "check" keys that represent things the value depends on;
753 * // Calling touchCheckKey() on any of them invalidates the cache
754 * 'checkKeys' => [
755 * $cache->makeKey( 'sustenance-bowls', $cat->getRoomId() ),
756 * $cache->makeKey( 'people-present', $cat->getHouseId() ),
757 * $cache->makeKey( 'cat-laws', $cat->getCityId() ),
758 * ]
759 * ]
760 * );
761 * @endcode
762 *
763 * Example usage (hot key holding most recent 100 events):
764 * @code
765 * $lastCatActions = $cache->getWithSetCallback(
766 * // Key to store the cached value under
767 * $cache->makeKey( 'cat-last-actions', 100 ),
768 * // Time-to-live (in seconds)
769 * 10,
770 * // Function that derives the new key value
771 * function ( $oldValue, &$ttl, array &$setOpts ) {
772 * $dbr = wfGetDB( DB_REPLICA );
773 * // Account for any snapshot/replica DB lag
774 * $setOpts += Database::getCacheSetOptions( $dbr );
775 *
776 * // Start off with the last cached list
777 * $list = $oldValue ?: [];
778 * // Fetch the last 100 relevant rows in descending order;
779 * // only fetch rows newer than $list[0] to reduce scanning
780 * $rows = iterator_to_array( $dbr->select( ... ) );
781 * // Merge them and get the new "last 100" rows
782 * return array_slice( array_merge( $new, $list ), 0, 100 );
783 * },
784 * [
785 * // Try to only let one datacenter thread manage cache updates at a time
786 * 'lockTSE' => 30,
787 * // Use a magic value when no cache value is ready rather than stampeding
788 * 'busyValue' => 'computing'
789 * ]
790 * );
791 * @endcode
792 *
793 * @see WANObjectCache::get()
794 * @see WANObjectCache::set()
795 *
796 * @param string $key Cache key
797 * @param integer $ttl Seconds to live for key updates. Special values are:
798 * - WANObjectCache::TTL_INDEFINITE: Cache forever
799 * - WANObjectCache::TTL_UNCACHEABLE: Do not cache at all
800 * @param callable $callback Value generation function
801 * @param array $opts Options map:
802 * - checkKeys: List of "check" keys. The key at $key will be seen as invalid when either
803 * touchCheckKey() or resetCheckKey() is called on any of these keys.
804 * Default: [].
805 * - lockTSE: If the key is tombstoned or expired (by checkKeys) less than this many seconds
806 * ago, then try to have a single thread handle cache regeneration at any given time.
807 * Other threads will try to use stale values if possible. If, on miss, the time since
808 * expiration is low, the assumption is that the key is hot and that a stampede is worth
809 * avoiding. Setting this above WANObjectCache::HOLDOFF_TTL makes no difference. The
810 * higher this is set, the higher the worst-case staleness can be.
811 * Use WANObjectCache::TSE_NONE to disable this logic.
812 * Default: WANObjectCache::TSE_NONE.
813 * - busyValue: If no value exists and another thread is currently regenerating it, use this
814 * as a fallback value (or a callback to generate such a value). This assures that cache
815 * stampedes cannot happen if the value falls out of cache. This can be used as insurance
816 * against cache regeneration becoming very slow for some reason (greater than the TTL).
817 * Default: null.
818 * - pcTTL: Process cache the value in this PHP instance for this many seconds. This avoids
819 * network I/O when a key is read several times. This will not cache when the callback
820 * returns false, however. Note that any purges will not be seen while process cached;
821 * since the callback should use replica DBs and they may be lagged or have snapshot
822 * isolation anyway, this should not typically matter.
823 * Default: WANObjectCache::TTL_UNCACHEABLE.
824 * - pcGroup: Process cache group to use instead of the primary one. If set, this must be
825 * of the format ALPHANUMERIC_NAME:MAX_KEY_SIZE, e.g. "mydata:10". Use this for storing
826 * large values, small yet numerous values, or some values with a high cost of eviction.
827 * It is generally preferable to use a class constant when setting this value.
828 * This has no effect unless pcTTL is used.
829 * Default: WANObjectCache::PC_PRIMARY.
830 * - version: Integer version number. This allows for callers to make breaking changes to
831 * how values are stored while maintaining compatability and correct cache purges. New
832 * versions are stored alongside older versions concurrently. Avoid storing class objects
833 * however, as this reduces compatibility (due to serialization).
834 * Default: null.
835 * - minAsOf: Reject values if they were generated before this UNIX timestamp.
836 * This is useful if the source of a key is suspected of having possibly changed
837 * recently, and the caller wants any such changes to be reflected.
838 * Default: WANObjectCache::MIN_TIMESTAMP_NONE.
839 * - hotTTR: Expected time-till-refresh for keys that average ~1 hit/second.
840 * This should be greater than "ageNew". Keys with higher hit rates will regenerate
841 * more often. This is useful when a popular key is changed but the cache purge was
842 * delayed or lost. Seldom used keys are rarely affected by this setting, unless an
843 * extremely low "hotTTR" value is passed in.
844 * Default: WANObjectCache::HOT_TTR.
845 * - lowTTL: Consider pre-emptive updates when the current TTL (seconds) of the key is less
846 * than this. It becomes more likely over time, becoming certain once the key is expired.
847 * Default: WANObjectCache::LOW_TTL.
848 * - ageNew: Consider popularity refreshes only once a key reaches this age in seconds.
849 * Default: WANObjectCache::AGE_NEW.
850 * @return mixed Value found or written to the key
851 * @note Options added in 1.28: version, busyValue, hotTTR, ageNew, pcGroup, minAsOf
852 * @note Callable type hints are not used to avoid class-autoloading
853 */
854 final public function getWithSetCallback( $key, $ttl, $callback, array $opts = [] ) {
855 $pcTTL = isset( $opts['pcTTL'] ) ? $opts['pcTTL'] : self::TTL_UNCACHEABLE;
856
857 // Try the process cache if enabled and the cache callback is not within a cache callback.
858 // Process cache use in nested callbacks is not lag-safe with regard to HOLDOFF_TTL since
859 // the in-memory value is further lagged than the shared one since it uses a blind TTL.
860 if ( $pcTTL >= 0 && $this->callbackDepth == 0 ) {
861 $group = isset( $opts['pcGroup'] ) ? $opts['pcGroup'] : self::PC_PRIMARY;
862 $procCache = $this->getProcessCache( $group );
863 $value = $procCache->get( $key );
864 } else {
865 $procCache = false;
866 $value = false;
867 }
868
869 if ( $value === false ) {
870 // Fetch the value over the network
871 if ( isset( $opts['version'] ) ) {
872 $version = $opts['version'];
873 $asOf = null;
874 $cur = $this->doGetWithSetCallback(
875 $key,
876 $ttl,
877 function ( $oldValue, &$ttl, &$setOpts, $oldAsOf )
878 use ( $callback, $version ) {
879 if ( is_array( $oldValue )
880 && array_key_exists( self::VFLD_DATA, $oldValue )
881 ) {
882 $oldData = $oldValue[self::VFLD_DATA];
883 } else {
884 // VFLD_DATA is not set if an old, unversioned, key is present
885 $oldData = false;
886 }
887
888 return [
889 self::VFLD_DATA => $callback( $oldData, $ttl, $setOpts, $oldAsOf ),
890 self::VFLD_VERSION => $version
891 ];
892 },
893 $opts,
894 $asOf
895 );
896 if ( $cur[self::VFLD_VERSION] === $version ) {
897 // Value created or existed before with version; use it
898 $value = $cur[self::VFLD_DATA];
899 } else {
900 // Value existed before with a different version; use variant key.
901 // Reflect purges to $key by requiring that this key value be newer.
902 $value = $this->doGetWithSetCallback(
903 'cache-variant:' . md5( $key ) . ":$version",
904 $ttl,
905 $callback,
906 // Regenerate value if not newer than $key
907 [ 'version' => null, 'minAsOf' => $asOf ] + $opts
908 );
909 }
910 } else {
911 $value = $this->doGetWithSetCallback( $key, $ttl, $callback, $opts );
912 }
913
914 // Update the process cache if enabled
915 if ( $procCache && $value !== false ) {
916 $procCache->set( $key, $value, $pcTTL );
917 }
918 }
919
920 return $value;
921 }
922
923 /**
924 * Do the actual I/O for getWithSetCallback() when needed
925 *
926 * @see WANObjectCache::getWithSetCallback()
927 *
928 * @param string $key
929 * @param integer $ttl
930 * @param callback $callback
931 * @param array $opts Options map for getWithSetCallback()
932 * @param float &$asOf Cache generation timestamp of returned value [returned]
933 * @return mixed
934 * @note Callable type hints are not used to avoid class-autoloading
935 */
936 protected function doGetWithSetCallback( $key, $ttl, $callback, array $opts, &$asOf = null ) {
937 $lowTTL = isset( $opts['lowTTL'] ) ? $opts['lowTTL'] : min( self::LOW_TTL, $ttl );
938 $lockTSE = isset( $opts['lockTSE'] ) ? $opts['lockTSE'] : self::TSE_NONE;
939 $checkKeys = isset( $opts['checkKeys'] ) ? $opts['checkKeys'] : [];
940 $busyValue = isset( $opts['busyValue'] ) ? $opts['busyValue'] : null;
941 $popWindow = isset( $opts['hotTTR'] ) ? $opts['hotTTR'] : self::HOT_TTR;
942 $ageNew = isset( $opts['ageNew'] ) ? $opts['ageNew'] : self::AGE_NEW;
943 $minTime = isset( $opts['minAsOf'] ) ? $opts['minAsOf'] : self::MIN_TIMESTAMP_NONE;
944 $versioned = isset( $opts['version'] );
945
946 // Get the current key value
947 $curTTL = null;
948 $cValue = $this->get( $key, $curTTL, $checkKeys, $asOf ); // current value
949 $value = $cValue; // return value
950
951 $preCallbackTime = microtime( true );
952 // Determine if a cached value regeneration is needed or desired
953 if ( $value !== false
954 && $curTTL > 0
955 && $this->isValid( $value, $versioned, $asOf, $minTime )
956 && !$this->worthRefreshExpiring( $curTTL, $lowTTL )
957 && !$this->worthRefreshPopular( $asOf, $ageNew, $popWindow, $preCallbackTime )
958 ) {
959 return $value;
960 }
961
962 // A deleted key with a negative TTL left must be tombstoned
963 $isTombstone = ( $curTTL !== null && $value === false );
964 // Assume a key is hot if requested soon after invalidation
965 $isHot = ( $curTTL !== null && $curTTL <= 0 && abs( $curTTL ) <= $lockTSE );
966 // Use the mutex if there is no value and a busy fallback is given
967 $checkBusy = ( $busyValue !== null && $value === false );
968 // Decide whether a single thread should handle regenerations.
969 // This avoids stampedes when $checkKeys are bumped and when preemptive
970 // renegerations take too long. It also reduces regenerations while $key
971 // is tombstoned. This balances cache freshness with avoiding DB load.
972 $useMutex = ( $isHot || ( $isTombstone && $lockTSE > 0 ) || $checkBusy );
973
974 $lockAcquired = false;
975 if ( $useMutex ) {
976 // Acquire a datacenter-local non-blocking lock
977 if ( $this->cache->add( self::MUTEX_KEY_PREFIX . $key, 1, self::LOCK_TTL ) ) {
978 // Lock acquired; this thread should update the key
979 $lockAcquired = true;
980 } elseif ( $value !== false && $this->isValid( $value, $versioned, $asOf, $minTime ) ) {
981 // If it cannot be acquired; then the stale value can be used
982 return $value;
983 } else {
984 // Use the INTERIM value for tombstoned keys to reduce regeneration load.
985 // For hot keys, either another thread has the lock or the lock failed;
986 // use the INTERIM value from the last thread that regenerated it.
987 $wrapped = $this->cache->get( self::INTERIM_KEY_PREFIX . $key );
988 list( $value ) = $this->unwrap( $wrapped, microtime( true ) );
989 if ( $value !== false && $this->isValid( $value, $versioned, $asOf, $minTime ) ) {
990 $asOf = $wrapped[self::FLD_TIME];
991
992 return $value;
993 }
994 // Use the busy fallback value if nothing else
995 if ( $busyValue !== null ) {
996 return is_callable( $busyValue ) ? $busyValue() : $busyValue;
997 }
998 }
999 }
1000
1001 if ( !is_callable( $callback ) ) {
1002 throw new InvalidArgumentException( "Invalid cache miss callback provided." );
1003 }
1004
1005 // Generate the new value from the callback...
1006 $setOpts = [];
1007 ++$this->callbackDepth;
1008 try {
1009 $value = call_user_func_array( $callback, [ $cValue, &$ttl, &$setOpts, $asOf ] );
1010 } finally {
1011 --$this->callbackDepth;
1012 }
1013 // When delete() is called, writes are write-holed by the tombstone,
1014 // so use a special INTERIM key to pass the new value around threads.
1015 if ( ( $isTombstone && $lockTSE > 0 ) && $value !== false && $ttl >= 0 ) {
1016 $tempTTL = max( 1, (int)$lockTSE ); // set() expects seconds
1017 $newAsOf = microtime( true );
1018 $wrapped = $this->wrap( $value, $tempTTL, $newAsOf );
1019 // Avoid using set() to avoid pointless mcrouter broadcasting
1020 $this->cache->merge(
1021 self::INTERIM_KEY_PREFIX . $key,
1022 function () use ( $wrapped ) {
1023 return $wrapped;
1024 },
1025 $tempTTL,
1026 1
1027 );
1028 }
1029
1030 if ( $value !== false && $ttl >= 0 ) {
1031 $setOpts['lockTSE'] = $lockTSE;
1032 // Use best known "since" timestamp if not provided
1033 $setOpts += [ 'since' => $preCallbackTime ];
1034 // Update the cache; this will fail if the key is tombstoned
1035 $this->set( $key, $value, $ttl, $setOpts );
1036 }
1037
1038 if ( $lockAcquired ) {
1039 // Avoid using delete() to avoid pointless mcrouter broadcasting
1040 $this->cache->changeTTL( self::MUTEX_KEY_PREFIX . $key, 1 );
1041 }
1042
1043 return $value;
1044 }
1045
1046 /**
1047 * Method to fetch/regenerate multiple cache keys at once
1048 *
1049 * This works the same as getWithSetCallback() except:
1050 * - a) The $keys argument expects the result of WANObjectCache::makeMultiKeys()
1051 * - b) The $callback argument expects a callback taking the following arguments:
1052 * - $id: ID of an entity to query
1053 * - $oldValue : the prior cache value or false if none was present
1054 * - &$ttl : a reference to the new value TTL in seconds
1055 * - &$setOpts : a reference to options for set() which can be altered
1056 * - $oldAsOf : generation UNIX timestamp of $oldValue or null if not present
1057 * Aside from the additional $id argument, the other arguments function the same
1058 * way they do in getWithSetCallback().
1059 * - c) The return value is a map of (cache key => value) in the order of $keyedIds
1060 *
1061 * @see WANObjectCache::getWithSetCallback()
1062 *
1063 * Example usage:
1064 * @code
1065 * $rows = $cache->getMultiWithSetCallback(
1066 * // Map of cache keys to entity IDs
1067 * $cache->makeMultiKeys(
1068 * $this->fileVersionIds(),
1069 * function ( $id, WANObjectCache $cache ) {
1070 * return $cache->makeKey( 'file-version', $id );
1071 * }
1072 * ),
1073 * // Time-to-live (in seconds)
1074 * $cache::TTL_DAY,
1075 * // Function that derives the new key value
1076 * return function ( $id, $oldValue, &$ttl, array &$setOpts ) {
1077 * $dbr = wfGetDB( DB_REPLICA );
1078 * // Account for any snapshot/replica DB lag
1079 * $setOpts += Database::getCacheSetOptions( $dbr );
1080 *
1081 * // Load the row for this file
1082 * $row = $dbr->selectRow( 'file', '*', [ 'id' => $id ], __METHOD__ );
1083 *
1084 * return $row ? (array)$row : false;
1085 * },
1086 * [
1087 * // Process cache for 30 seconds
1088 * 'pcTTL' => 30,
1089 * // Use a dedicated 500 item cache (initialized on-the-fly)
1090 * 'pcGroup' => 'file-versions:500'
1091 * ]
1092 * );
1093 * $files = array_map( [ __CLASS__, 'newFromRow' ], $rows );
1094 * @endcode
1095 *
1096 * @param ArrayIterator $keyedIds Result of WANObjectCache::makeMultiKeys()
1097 * @param integer $ttl Seconds to live for key updates
1098 * @param callable $callback Callback the yields entity regeneration callbacks
1099 * @param array $opts Options map
1100 * @return array Map of (cache key => value) in the same order as $keyedIds
1101 * @since 1.28
1102 */
1103 final public function getMultiWithSetCallback(
1104 ArrayIterator $keyedIds, $ttl, callable $callback, array $opts = []
1105 ) {
1106 $keysWarmUp = iterator_to_array( $keyedIds, true );
1107 $checkKeys = isset( $opts['checkKeys'] ) ? $opts['checkKeys'] : [];
1108 foreach ( $checkKeys as $i => $checkKeyOrKeys ) {
1109 if ( is_int( $i ) ) {
1110 $keysWarmUp[] = $checkKeyOrKeys;
1111 } else {
1112 $keysWarmUp = array_merge( $keysWarmUp, $checkKeyOrKeys );
1113 }
1114 }
1115
1116 $this->warmupCache = $this->cache->getMulti( $keysWarmUp );
1117 $this->warmupCache += array_fill_keys( $keysWarmUp, false );
1118
1119 // Wrap $callback to match the getWithSetCallback() format while passing $id to $callback
1120 $id = null;
1121 $func = function ( $oldValue, &$ttl, array $setOpts, $oldAsOf ) use ( $callback, &$id ) {
1122 return $callback( $id, $oldValue, $ttl, $setOpts, $oldAsOf );
1123 };
1124
1125 $values = [];
1126 foreach ( $keyedIds as $key => $id ) {
1127 $values[$key] = $this->getWithSetCallback( $key, $ttl, $func, $opts );
1128 }
1129
1130 $this->warmupCache = [];
1131
1132 return $values;
1133 }
1134
1135 /**
1136 * Locally set a key to expire soon if it is stale based on $purgeTimestamp
1137 *
1138 * This sets stale keys' time-to-live at HOLDOFF_TTL seconds, which both avoids
1139 * broadcasting in mcrouter setups and also avoids races with new tombstones.
1140 *
1141 * @param string $key Cache key
1142 * @param int $purgeTimestamp UNIX timestamp of purge
1143 * @param bool &$isStale Whether the key is stale
1144 * @return bool Success
1145 * @since 1.28
1146 */
1147 public function reap( $key, $purgeTimestamp, &$isStale = false ) {
1148 $minAsOf = $purgeTimestamp + self::HOLDOFF_TTL;
1149 $wrapped = $this->cache->get( self::VALUE_KEY_PREFIX . $key );
1150 if ( is_array( $wrapped ) && $wrapped[self::FLD_TIME] < $minAsOf ) {
1151 $isStale = true;
1152 $this->logger->warning( "Reaping stale value key '$key'." );
1153 $ttlReap = self::HOLDOFF_TTL; // avoids races with tombstone creation
1154 $ok = $this->cache->changeTTL( self::VALUE_KEY_PREFIX . $key, $ttlReap );
1155 if ( !$ok ) {
1156 $this->logger->error( "Could not complete reap of key '$key'." );
1157 }
1158
1159 return $ok;
1160 }
1161
1162 $isStale = false;
1163
1164 return true;
1165 }
1166
1167 /**
1168 * Locally set a "check" key to expire soon if it is stale based on $purgeTimestamp
1169 *
1170 * @param string $key Cache key
1171 * @param int $purgeTimestamp UNIX timestamp of purge
1172 * @param bool &$isStale Whether the key is stale
1173 * @return bool Success
1174 * @since 1.28
1175 */
1176 public function reapCheckKey( $key, $purgeTimestamp, &$isStale = false ) {
1177 $purge = $this->parsePurgeValue( $this->cache->get( self::TIME_KEY_PREFIX . $key ) );
1178 if ( $purge && $purge[self::FLD_TIME] < $purgeTimestamp ) {
1179 $isStale = true;
1180 $this->logger->warning( "Reaping stale check key '$key'." );
1181 $ok = $this->cache->changeTTL( self::TIME_KEY_PREFIX . $key, 1 );
1182 if ( !$ok ) {
1183 $this->logger->error( "Could not complete reap of check key '$key'." );
1184 }
1185
1186 return $ok;
1187 }
1188
1189 $isStale = false;
1190
1191 return false;
1192 }
1193
1194 /**
1195 * @see BagOStuff::makeKey()
1196 * @param string ... Key component
1197 * @return string
1198 * @since 1.27
1199 */
1200 public function makeKey() {
1201 return call_user_func_array( [ $this->cache, __FUNCTION__ ], func_get_args() );
1202 }
1203
1204 /**
1205 * @see BagOStuff::makeGlobalKey()
1206 * @param string ... Key component
1207 * @return string
1208 * @since 1.27
1209 */
1210 public function makeGlobalKey() {
1211 return call_user_func_array( [ $this->cache, __FUNCTION__ ], func_get_args() );
1212 }
1213
1214 /**
1215 * @param array $entities List of entity IDs
1216 * @param callable $keyFunc Callback yielding a key from (entity ID, this WANObjectCache)
1217 * @return ArrayIterator Iterator yielding (cache key => entity ID) in $entities order
1218 * @since 1.28
1219 */
1220 public function makeMultiKeys( array $entities, callable $keyFunc ) {
1221 $map = [];
1222 foreach ( $entities as $entity ) {
1223 $map[$keyFunc( $entity, $this )] = $entity;
1224 }
1225
1226 return new ArrayIterator( $map );
1227 }
1228
1229 /**
1230 * Get the "last error" registered; clearLastError() should be called manually
1231 * @return int ERR_* class constant for the "last error" registry
1232 */
1233 final public function getLastError() {
1234 if ( $this->lastRelayError ) {
1235 // If the cache and the relayer failed, focus on the latter.
1236 // An update not making it to the relayer means it won't show up
1237 // in other DCs (nor will consistent re-hashing see up-to-date values).
1238 // On the other hand, if just the cache update failed, then it should
1239 // eventually be applied by the relayer.
1240 return $this->lastRelayError;
1241 }
1242
1243 $code = $this->cache->getLastError();
1244 switch ( $code ) {
1245 case BagOStuff::ERR_NONE:
1246 return self::ERR_NONE;
1247 case BagOStuff::ERR_NO_RESPONSE:
1248 return self::ERR_NO_RESPONSE;
1249 case BagOStuff::ERR_UNREACHABLE:
1250 return self::ERR_UNREACHABLE;
1251 default:
1252 return self::ERR_UNEXPECTED;
1253 }
1254 }
1255
1256 /**
1257 * Clear the "last error" registry
1258 */
1259 final public function clearLastError() {
1260 $this->cache->clearLastError();
1261 $this->lastRelayError = self::ERR_NONE;
1262 }
1263
1264 /**
1265 * Clear the in-process caches; useful for testing
1266 *
1267 * @since 1.27
1268 */
1269 public function clearProcessCache() {
1270 $this->processCaches = [];
1271 }
1272
1273 /**
1274 * @param integer $flag ATTR_* class constant
1275 * @return integer QOS_* class constant
1276 * @since 1.28
1277 */
1278 public function getQoS( $flag ) {
1279 return $this->cache->getQoS( $flag );
1280 }
1281
1282 /**
1283 * Get a TTL that is higher for objects that have not changed recently
1284 *
1285 * This is useful for keys that get explicit purges and DB or purge relay
1286 * lag is a potential concern (especially how it interacts with CDN cache)
1287 *
1288 * Example usage:
1289 * @code
1290 * // Last-modified time of page
1291 * $mtime = wfTimestamp( TS_UNIX, $page->getTimestamp() );
1292 * // Get adjusted TTL. If $mtime is 3600 seconds ago and $minTTL/$factor left at
1293 * // defaults, then $ttl is 3600 * .2 = 720. If $minTTL was greater than 720, then
1294 * // $ttl would be $minTTL. If $maxTTL was smaller than 720, $ttl would be $maxTTL.
1295 * $ttl = $cache->adaptiveTTL( $mtime, $cache::TTL_DAY );
1296 * @endcode
1297 *
1298 * @param integer|float $mtime UNIX timestamp
1299 * @param integer $maxTTL Maximum TTL (seconds)
1300 * @param integer $minTTL Minimum TTL (seconds); Default: 30
1301 * @param float $factor Value in the range (0,1); Default: .2
1302 * @return integer Adaptive TTL
1303 * @since 1.28
1304 */
1305 public function adaptiveTTL( $mtime, $maxTTL, $minTTL = 30, $factor = .2 ) {
1306 if ( is_float( $mtime ) || ctype_digit( $mtime ) ) {
1307 $mtime = (int)$mtime; // handle fractional seconds and string integers
1308 }
1309
1310 if ( !is_int( $mtime ) || $mtime <= 0 ) {
1311 return $minTTL; // no last-modified time provided
1312 }
1313
1314 $age = time() - $mtime;
1315
1316 return (int)min( $maxTTL, max( $minTTL, $factor * $age ) );
1317 }
1318
1319 /**
1320 * Do the actual async bus purge of a key
1321 *
1322 * This must set the key to "PURGED:<UNIX timestamp>:<holdoff>"
1323 *
1324 * @param string $key Cache key
1325 * @param integer $ttl How long to keep the tombstone [seconds]
1326 * @param integer $holdoff HOLDOFF_* constant controlling how long to ignore sets for this key
1327 * @return bool Success
1328 */
1329 protected function relayPurge( $key, $ttl, $holdoff ) {
1330 if ( $this->purgeRelayer instanceof EventRelayerNull ) {
1331 // This handles the mcrouter and the single-DC case
1332 $ok = $this->cache->set( $key,
1333 $this->makePurgeValue( microtime( true ), self::HOLDOFF_NONE ),
1334 $ttl
1335 );
1336 } else {
1337 $event = $this->cache->modifySimpleRelayEvent( [
1338 'cmd' => 'set',
1339 'key' => $key,
1340 'val' => 'PURGED:$UNIXTIME$:' . (int)$holdoff,
1341 'ttl' => max( $ttl, 1 ),
1342 'sbt' => true, // substitute $UNIXTIME$ with actual microtime
1343 ] );
1344
1345 $ok = $this->purgeRelayer->notify( $this->purgeChannel, $event );
1346 if ( !$ok ) {
1347 $this->lastRelayError = self::ERR_RELAY;
1348 }
1349 }
1350
1351 return $ok;
1352 }
1353
1354 /**
1355 * Do the actual async bus delete of a key
1356 *
1357 * @param string $key Cache key
1358 * @return bool Success
1359 */
1360 protected function relayDelete( $key ) {
1361 if ( $this->purgeRelayer instanceof EventRelayerNull ) {
1362 // This handles the mcrouter and the single-DC case
1363 $ok = $this->cache->delete( $key );
1364 } else {
1365 $event = $this->cache->modifySimpleRelayEvent( [
1366 'cmd' => 'delete',
1367 'key' => $key,
1368 ] );
1369
1370 $ok = $this->purgeRelayer->notify( $this->purgeChannel, $event );
1371 if ( !$ok ) {
1372 $this->lastRelayError = self::ERR_RELAY;
1373 }
1374 }
1375
1376 return $ok;
1377 }
1378
1379 /**
1380 * Check if a key should be regenerated (using random probability)
1381 *
1382 * This returns false if $curTTL >= $lowTTL. Otherwise, the chance
1383 * of returning true increases steadily from 0% to 100% as the $curTTL
1384 * moves from $lowTTL to 0 seconds. This handles widely varying
1385 * levels of cache access traffic.
1386 *
1387 * @param float $curTTL Approximate TTL left on the key if present
1388 * @param float $lowTTL Consider a refresh when $curTTL is less than this
1389 * @return bool
1390 */
1391 protected function worthRefreshExpiring( $curTTL, $lowTTL ) {
1392 if ( $curTTL >= $lowTTL ) {
1393 return false;
1394 } elseif ( $curTTL <= 0 ) {
1395 return true;
1396 }
1397
1398 $chance = ( 1 - $curTTL / $lowTTL );
1399
1400 return mt_rand( 1, 1e9 ) <= 1e9 * $chance;
1401 }
1402
1403 /**
1404 * Check if a key is due for randomized regeneration due to its popularity
1405 *
1406 * This is used so that popular keys can preemptively refresh themselves for higher
1407 * consistency (especially in the case of purge loss/delay). Unpopular keys can remain
1408 * in cache with their high nominal TTL. This means popular keys keep good consistency,
1409 * whether the data changes frequently or not, and long-tail keys get to stay in cache
1410 * and get hits too. Similar to worthRefreshExpiring(), randomization is used.
1411 *
1412 * @param float $asOf UNIX timestamp of the value
1413 * @param integer $ageNew Age of key when this might recommend refreshing (seconds)
1414 * @param integer $timeTillRefresh Age of key when it should be refreshed if popular (seconds)
1415 * @param float $now The current UNIX timestamp
1416 * @return bool
1417 */
1418 protected function worthRefreshPopular( $asOf, $ageNew, $timeTillRefresh, $now ) {
1419 $age = $now - $asOf;
1420 $timeOld = $age - $ageNew;
1421 if ( $timeOld <= 0 ) {
1422 return false;
1423 }
1424
1425 // Lifecycle is: new, ramp-up refresh chance, full refresh chance.
1426 // Note that the "expected # of refreshes" for the ramp-up time range is half of what it
1427 // would be if P(refresh) was at its full value during that time range.
1428 $refreshWindowSec = max( $timeTillRefresh - $ageNew - self::RAMPUP_TTL / 2, 1 );
1429 // P(refresh) * (# hits in $refreshWindowSec) = (expected # of refreshes)
1430 // P(refresh) * ($refreshWindowSec * $popularHitsPerSec) = 1
1431 // P(refresh) = 1/($refreshWindowSec * $popularHitsPerSec)
1432 $chance = 1 / ( self::HIT_RATE_HIGH * $refreshWindowSec );
1433
1434 // Ramp up $chance from 0 to its nominal value over RAMPUP_TTL seconds to avoid stampedes
1435 $chance *= ( $timeOld <= self::RAMPUP_TTL ) ? $timeOld / self::RAMPUP_TTL : 1;
1436
1437 return mt_rand( 1, 1e9 ) <= 1e9 * $chance;
1438 }
1439
1440 /**
1441 * Check whether $value is appropriately versioned and not older than $minTime (if set)
1442 *
1443 * @param array $value
1444 * @param bool $versioned
1445 * @param float $asOf The time $value was generated
1446 * @param float $minTime The last time the main value was generated (0.0 if unknown)
1447 * @return bool
1448 */
1449 protected function isValid( $value, $versioned, $asOf, $minTime ) {
1450 if ( $versioned && !isset( $value[self::VFLD_VERSION] ) ) {
1451 return false;
1452 } elseif ( $minTime > 0 && $asOf < $minTime ) {
1453 return false;
1454 }
1455
1456 return true;
1457 }
1458
1459 /**
1460 * Do not use this method outside WANObjectCache
1461 *
1462 * @param mixed $value
1463 * @param integer $ttl [0=forever]
1464 * @param float $now Unix Current timestamp just before calling set()
1465 * @return array
1466 */
1467 protected function wrap( $value, $ttl, $now ) {
1468 return [
1469 self::FLD_VERSION => self::VERSION,
1470 self::FLD_VALUE => $value,
1471 self::FLD_TTL => $ttl,
1472 self::FLD_TIME => $now
1473 ];
1474 }
1475
1476 /**
1477 * Do not use this method outside WANObjectCache
1478 *
1479 * @param array|string|bool $wrapped
1480 * @param float $now Unix Current timestamp (preferrably pre-query)
1481 * @return array (mixed; false if absent/invalid, current time left)
1482 */
1483 protected function unwrap( $wrapped, $now ) {
1484 // Check if the value is a tombstone
1485 $purge = self::parsePurgeValue( $wrapped );
1486 if ( $purge !== false ) {
1487 // Purged values should always have a negative current $ttl
1488 $curTTL = min( $purge[self::FLD_TIME] - $now, self::TINY_NEGATIVE );
1489 return [ false, $curTTL ];
1490 }
1491
1492 if ( !is_array( $wrapped ) // not found
1493 || !isset( $wrapped[self::FLD_VERSION] ) // wrong format
1494 || $wrapped[self::FLD_VERSION] !== self::VERSION // wrong version
1495 ) {
1496 return [ false, null ];
1497 }
1498
1499 $flags = isset( $wrapped[self::FLD_FLAGS] ) ? $wrapped[self::FLD_FLAGS] : 0;
1500 if ( ( $flags & self::FLG_STALE ) == self::FLG_STALE ) {
1501 // Treat as expired, with the cache time as the expiration
1502 $age = $now - $wrapped[self::FLD_TIME];
1503 $curTTL = min( -$age, self::TINY_NEGATIVE );
1504 } elseif ( $wrapped[self::FLD_TTL] > 0 ) {
1505 // Get the approximate time left on the key
1506 $age = $now - $wrapped[self::FLD_TIME];
1507 $curTTL = max( $wrapped[self::FLD_TTL] - $age, 0.0 );
1508 } else {
1509 // Key had no TTL, so the time left is unbounded
1510 $curTTL = INF;
1511 }
1512
1513 return [ $wrapped[self::FLD_VALUE], $curTTL ];
1514 }
1515
1516 /**
1517 * @param array $keys
1518 * @param string $prefix
1519 * @return string[]
1520 */
1521 protected static function prefixCacheKeys( array $keys, $prefix ) {
1522 $res = [];
1523 foreach ( $keys as $key ) {
1524 $res[] = $prefix . $key;
1525 }
1526
1527 return $res;
1528 }
1529
1530 /**
1531 * @param string $value Wrapped value like "PURGED:<timestamp>:<holdoff>"
1532 * @return array|bool Array containing a UNIX timestamp (float) and holdoff period (integer),
1533 * or false if value isn't a valid purge value
1534 */
1535 protected static function parsePurgeValue( $value ) {
1536 if ( !is_string( $value ) ) {
1537 return false;
1538 }
1539 $segments = explode( ':', $value, 3 );
1540 if ( !isset( $segments[0] ) || !isset( $segments[1] )
1541 || "{$segments[0]}:" !== self::PURGE_VAL_PREFIX
1542 ) {
1543 return false;
1544 }
1545 if ( !isset( $segments[2] ) ) {
1546 // Back-compat with old purge values without holdoff
1547 $segments[2] = self::HOLDOFF_TTL;
1548 }
1549 return [
1550 self::FLD_TIME => (float)$segments[1],
1551 self::FLD_HOLDOFF => (int)$segments[2],
1552 ];
1553 }
1554
1555 /**
1556 * @param float $timestamp
1557 * @param int $holdoff In seconds
1558 * @return string Wrapped purge value
1559 */
1560 protected function makePurgeValue( $timestamp, $holdoff ) {
1561 return self::PURGE_VAL_PREFIX . (float)$timestamp . ':' . (int)$holdoff;
1562 }
1563
1564 /**
1565 * @param string $group
1566 * @return HashBagOStuff
1567 */
1568 protected function getProcessCache( $group ) {
1569 if ( !isset( $this->processCaches[$group] ) ) {
1570 list( , $n ) = explode( ':', $group );
1571 $this->processCaches[$group] = new HashBagOStuff( [ 'maxKeys' => (int)$n ] );
1572 }
1573
1574 return $this->processCaches[$group];
1575 }
1576 }