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