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.
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.
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
22 use Psr\Log\LoggerAwareInterface
;
23 use Psr\Log\LoggerInterface
;
24 use Psr\Log\NullLogger
;
27 * Multi-datacenter aware caching interface
29 * All operations go to the local datacenter cache, except for delete(),
30 * touchCheckKey(), and resetCheckKey(), which broadcast to all datacenters.
32 * This class is intended for caching data from primary stores.
33 * If the get() method does not return a value, then the caller
34 * should query the new value and backfill the cache using set().
35 * The preferred way to do this logic is through getWithSetCallback().
36 * When querying the store on cache miss, the closest DB replica
37 * should be used. Try to avoid heavyweight DB master or quorum reads.
38 * When the source data changes, a purge method should be called.
39 * Since purges are expensive, they should be avoided. One can do so if:
40 * - a) The object cached is immutable; or
41 * - b) Validity is checked against the source after get(); or
42 * - c) Using a modest TTL is reasonably correct and performant
44 * The simplest purge method is delete().
46 * There are three supported ways to handle broadcasted operations:
47 * - a) Configure the 'purge' EventRelayer to point to a valid PubSub endpoint
48 * that has subscribed listeners on the cache servers applying the cache updates.
49 * - b) Ignore the 'purge' EventRelayer configuration (default is NullEventRelayer)
50 * and set up mcrouter as the underlying cache backend, using one of the memcached
51 * BagOStuff classes as 'cache'. Use OperationSelectorRoute in the mcrouter settings
52 * to configure 'set' and 'delete' operations to go to all DCs via AllAsyncRoute and
53 * configure other operations to go to the local DC via PoolRoute (for reference,
54 * see https://github.com/facebook/mcrouter/wiki/List-of-Route-Handles).
55 * - c) Ignore the 'purge' EventRelayer configuration (default is NullEventRelayer)
56 * and set up dynomite as cache middleware between the web servers and either
57 * memcached or redis. This will also broadcast all key setting operations, not just purges,
58 * which can be useful for cache warming. Writes are eventually consistent via the
59 * Dynamo replication model (see https://github.com/Netflix/dynomite).
61 * Broadcasted operations like delete() and touchCheckKey() are done asynchronously
62 * in all datacenters this way, though the local one should likely be near immediate.
64 * This means that callers in all datacenters may see older values for however many
65 * milliseconds that the purge took to reach that datacenter. As with any cache, this
66 * should not be relied on for cases where reads are used to determine writes to source
67 * (e.g. non-cache) data stores, except when reading immutable data.
69 * All values are wrapped in metadata arrays. Keys use a "WANCache:" prefix
70 * to avoid collisions with keys that are not wrapped as metadata arrays. The
71 * prefixes are as follows:
72 * - a) "WANCache:v" : used for regular value keys
73 * - b) "WANCache:i" : used for temporarily storing values of tombstoned keys
74 * - c) "WANCache:t" : used for storing timestamp "check" keys
75 * - d) "WANCache:m" : used for temporary mutex keys to avoid cache stampedes
80 class WANObjectCache
implements IExpiringStore
, LoggerAwareInterface
{
81 /** @var BagOStuff The local datacenter cache */
83 /** @var HashBagOStuff[] Map of group PHP instance caches */
84 protected $processCaches = [];
85 /** @var string Purge channel name */
86 protected $purgeChannel;
87 /** @var EventRelayer Bus that handles purge broadcasts */
88 protected $purgeRelayer;
89 /** @var LoggerInterface */
92 /** @var int ERR_* constant for the "last error" registry */
93 protected $lastRelayError = self
::ERR_NONE
;
95 /** @var int Callback stack depth for getWithSetCallback() */
96 private $callbackDepth = 0;
97 /** @var mixed[] Temporary warm-up cache */
98 private $warmupCache = [];
99 /** @var int Key fetched */
100 private $warmupKeyMisses = 0;
102 /** Max time expected to pass between delete() and DB commit finishing */
103 const MAX_COMMIT_DELAY
= 3;
104 /** Max replication+snapshot lag before applying TTL_LAGGED or disallowing set() */
105 const MAX_READ_LAG
= 7;
106 /** Seconds to tombstone keys on delete() */
107 const HOLDOFF_TTL
= 11; // MAX_COMMIT_DELAY + MAX_READ_LAG + 1
109 /** Seconds to keep dependency purge keys around */
110 const CHECK_KEY_TTL
= self
::TTL_YEAR
;
111 /** Seconds to keep lock keys around */
113 /** Default remaining TTL at which to consider pre-emptive regeneration */
115 /** Default time-since-expiry on a miss that makes a key "hot" */
118 /** Never consider performing "popularity" refreshes until a key reaches this age */
120 /** The time length of the "popularity" refresh window for hot keys */
122 /** Hits/second for a refresh to be expected within the "popularity" window */
123 const HIT_RATE_HIGH
= 1;
124 /** Seconds to ramp up to the "popularity" refresh chance after a key is no longer new */
125 const RAMPUP_TTL
= 30;
127 /** Idiom for getWithSetCallback() callbacks to avoid calling set() */
128 const TTL_UNCACHEABLE
= -1;
129 /** Idiom for getWithSetCallback() callbacks to 'lockTSE' logic */
131 /** Max TTL to store keys when a data sourced is lagged */
132 const TTL_LAGGED
= 30;
133 /** Idiom for delete() for "no hold-off" */
134 const HOLDOFF_NONE
= 0;
135 /** Idiom for getWithSetCallback() for "no minimum required as-of timestamp" */
136 const MIN_TIMESTAMP_NONE
= 0.0;
138 /** Tiny negative float to use when CTL comes up >= 0 due to clock skew */
139 const TINY_NEGATIVE
= -0.000001;
141 /** Cache format version number */
144 const FLD_VERSION
= 0; // key to cache version number
145 const FLD_VALUE
= 1; // key to the cached value
146 const FLD_TTL
= 2; // key to the original TTL
147 const FLD_TIME
= 3; // key to the cache time
148 const FLD_FLAGS
= 4; // key to the flags bitfield
149 const FLD_HOLDOFF
= 5; // key to any hold-off TTL
151 /** @var int Treat this value as expired-on-arrival */
154 const ERR_NONE
= 0; // no error
155 const ERR_NO_RESPONSE
= 1; // no response
156 const ERR_UNREACHABLE
= 2; // can't connect
157 const ERR_UNEXPECTED
= 3; // response gave some error
158 const ERR_RELAY
= 4; // relay broadcast failed
160 const VALUE_KEY_PREFIX
= 'WANCache:v:';
161 const INTERIM_KEY_PREFIX
= 'WANCache:i:';
162 const TIME_KEY_PREFIX
= 'WANCache:t:';
163 const MUTEX_KEY_PREFIX
= 'WANCache:m:';
165 const PURGE_VAL_PREFIX
= 'PURGED:';
167 const VFLD_DATA
= 'WOC:d'; // key to the value of versioned data
168 const VFLD_VERSION
= 'WOC:v'; // key to the version of the value present
170 const PC_PRIMARY
= 'primary:1000'; // process cache name and max key count
172 const DEFAULT_PURGE_CHANNEL
= 'wancache-purge';
175 * @param array $params
176 * - cache : BagOStuff object for a persistent cache
177 * - channels : Map of (action => channel string). Actions include "purge".
178 * - relayers : Map of (action => EventRelayer object). Actions include "purge".
179 * - logger : LoggerInterface object
181 public function __construct( array $params ) {
182 $this->cache
= $params['cache'];
183 $this->purgeChannel
= isset( $params['channels']['purge'] )
184 ?
$params['channels']['purge']
185 : self
::DEFAULT_PURGE_CHANNEL
;
186 $this->purgeRelayer
= isset( $params['relayers']['purge'] )
187 ?
$params['relayers']['purge']
188 : new EventRelayerNull( [] );
189 $this->setLogger( isset( $params['logger'] ) ?
$params['logger'] : new NullLogger() );
192 public function setLogger( LoggerInterface
$logger ) {
193 $this->logger
= $logger;
197 * Get an instance that wraps EmptyBagOStuff
199 * @return WANObjectCache
201 public static function newEmpty() {
203 'cache' => new EmptyBagOStuff(),
209 * Fetch the value of a key from cache
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
217 * If a key is tombstoned, $curTTL will reflect the time since delete().
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
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
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.
239 * Consider using getWithSetCallback() instead of get() and set() cycles.
240 * That method has cache slam avoiding features for hot/expensive keys.
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
248 final public function get( $key, &$curTTL = null, array $checkKeys = [], &$asOf = null ) {
251 $values = $this->getMulti( [ $key ], $curTTLs, $checkKeys, $asOfs );
252 $curTTL = isset( $curTTLs[$key] ) ?
$curTTLs[$key] : null;
253 $asOf = isset( $asOfs[$key] ) ?
$asOfs[$key] : null;
255 return isset( $values[$key] ) ?
$values[$key] : false;
259 * Fetch the value of several keys from cache
261 * @see WANObjectCache::get()
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 and are not tombstoned
270 final public function getMulti(
271 array $keys, &$curTTLs = [], array $checkKeys = [], array &$asOfs = []
277 $vPrefixLen = strlen( self
::VALUE_KEY_PREFIX
);
278 $valueKeys = self
::prefixCacheKeys( $keys, self
::VALUE_KEY_PREFIX
);
280 $checkKeysForAll = [];
281 $checkKeysByKey = [];
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 );
290 $checkKeysByKey[$i] = isset( $checkKeysByKey[$i] )
291 ?
array_merge( $checkKeysByKey[$i], $prefixed )
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 $this->warmupKeyMisses +
= count( $keysGet );
306 $wrappedValues +
= $this->cache
->getMulti( $keysGet );
308 // Time used to compare/init "check" keys (derived after getMulti() to be pessimistic)
309 $now = microtime( true );
311 // Collect timestamps from all "check" keys
312 $purgeValuesForAll = $this->processCheckKeys( $checkKeysForAll, $wrappedValues, $now );
313 $purgeValuesByKey = [];
314 foreach ( $checkKeysByKey as $cacheKey => $checks ) {
315 $purgeValuesByKey[$cacheKey] =
316 $this->processCheckKeys( $checks, $wrappedValues, $now );
319 // Get the main cache value for each key and validate them
320 foreach ( $valueKeys as $vKey ) {
321 if ( !isset( $wrappedValues[$vKey] ) ) {
322 continue; // not found
325 $key = substr( $vKey, $vPrefixLen ); // unprefix
327 list( $value, $curTTL ) = $this->unwrap( $wrappedValues[$vKey], $now );
328 if ( $value !== false ) {
329 $result[$key] = $value;
331 // Force dependant keys to be invalid for a while after purging
332 // to reduce race conditions involving stale data getting cached
333 $purgeValues = $purgeValuesForAll;
334 if ( isset( $purgeValuesByKey[$key] ) ) {
335 $purgeValues = array_merge( $purgeValues, $purgeValuesByKey[$key] );
337 foreach ( $purgeValues as $purge ) {
338 $safeTimestamp = $purge[self
::FLD_TIME
] +
$purge[self
::FLD_HOLDOFF
];
339 if ( $safeTimestamp >= $wrappedValues[$vKey][self
::FLD_TIME
] ) {
340 // How long ago this value was expired by *this* check key
341 $ago = min( $purge[self
::FLD_TIME
] - $now, self
::TINY_NEGATIVE
);
342 // How long ago this value was expired by *any* known check key
343 $curTTL = min( $curTTL, $ago );
347 $curTTLs[$key] = $curTTL;
348 $asOfs[$key] = ( $value !== false ) ?
$wrappedValues[$vKey][self
::FLD_TIME
] : null;
356 * @param array $timeKeys List of prefixed time check keys
357 * @param array $wrappedValues
359 * @return array List of purge value arrays
361 private function processCheckKeys( array $timeKeys, array $wrappedValues, $now ) {
363 foreach ( $timeKeys as $timeKey ) {
364 $purge = isset( $wrappedValues[$timeKey] )
365 ? self
::parsePurgeValue( $wrappedValues[$timeKey] )
367 if ( $purge === false ) {
368 // Key is not set or invalid; regenerate
369 $newVal = $this->makePurgeValue( $now, self
::HOLDOFF_TTL
);
370 $this->cache
->add( $timeKey, $newVal, self
::CHECK_KEY_TTL
);
371 $purge = self
::parsePurgeValue( $newVal );
373 $purgeValues[] = $purge;
379 * Set the value of a key in cache
381 * Simply calling this method when source data changes is not valid because
382 * the changes do not replicate to the other WAN sites. In that case, delete()
383 * should be used instead. This method is intended for use on cache misses.
385 * If the data was read from a snapshot-isolated transactions (e.g. the default
386 * REPEATABLE-READ in innoDB), use 'since' to avoid the following race condition:
388 * - b) T2 updates a row, calls delete(), and commits
389 * - c) The HOLDOFF_TTL passes, expiring the delete() tombstone
390 * - d) T1 reads the row and calls set() due to a cache miss
391 * - e) Stale value is stuck in cache
393 * Setting 'lag' and 'since' help avoids keys getting stuck in stale states.
397 * $dbr = wfGetDB( DB_REPLICA );
398 * $setOpts = Database::getCacheSetOptions( $dbr );
399 * // Fetch the row from the DB
400 * $row = $dbr->selectRow( ... );
401 * $key = $cache->makeKey( 'building', $buildingId );
402 * $cache->set( $key, $row, $cache::TTL_DAY, $setOpts );
405 * @param string $key Cache key
406 * @param mixed $value
407 * @param int $ttl Seconds to live. Special values are:
408 * - WANObjectCache::TTL_INDEFINITE: Cache forever
409 * @param array $opts Options map:
410 * - lag : Seconds of replica DB lag. Typically, this is either the replica DB lag
411 * before the data was read or, if applicable, the replica DB lag before
412 * the snapshot-isolated transaction the data was read from started.
413 * Use false to indicate that replication is not running.
415 * - since : UNIX timestamp of the data in $value. Typically, this is either
416 * the current time the data was read or (if applicable) the time when
417 * the snapshot-isolated transaction the data was read from started.
419 * - pending : Whether this data is possibly from an uncommitted write transaction.
420 * Generally, other threads should not see values from the future and
421 * they certainly should not see ones that ended up getting rolled back.
423 * - lockTSE : if excessive replication/snapshot lag is detected, then store the value
424 * with this TTL and flag it as stale. This is only useful if the reads for
425 * this key use getWithSetCallback() with "lockTSE" set.
426 * Default: WANObjectCache::TSE_NONE
427 * - staleTTL : Seconds to keep the key around if it is stale. The get()/getMulti()
428 * methods return such stale values with a $curTTL of 0, and getWithSetCallback()
429 * will call the regeneration callback in such cases, passing in the old value
430 * and its as-of time to the callback. This is useful if adaptiveTTL() is used
431 * on the old value's as-of time when it is verified as still being correct.
433 * @note Options added in 1.28: staleTTL
434 * @return bool Success
436 final public function set( $key, $value, $ttl = 0, array $opts = [] ) {
437 $now = microtime( true );
438 $lockTSE = isset( $opts['lockTSE'] ) ?
$opts['lockTSE'] : self
::TSE_NONE
;
439 $age = isset( $opts['since'] ) ?
max( 0, $now - $opts['since'] ) : 0;
440 $lag = isset( $opts['lag'] ) ?
$opts['lag'] : 0;
441 $staleTTL = isset( $opts['staleTTL'] ) ?
$opts['staleTTL'] : 0;
443 // Do not cache potentially uncommitted data as it might get rolled back
444 if ( !empty( $opts['pending'] ) ) {
445 $this->logger
->info( "Rejected set() for $key due to pending writes." );
447 return true; // no-op the write for being unsafe
450 $wrapExtra = []; // additional wrapped value fields
451 // Check if there's a risk of writing stale data after the purge tombstone expired
452 if ( $lag === false ||
( $lag +
$age ) > self
::MAX_READ_LAG
) {
453 // Case A: read lag with "lockTSE"; save but record value as stale
454 if ( $lockTSE >= 0 ) {
455 $ttl = max( 1, (int)$lockTSE ); // set() expects seconds
456 $wrapExtra[self
::FLD_FLAGS
] = self
::FLG_STALE
; // mark as stale
457 // Case B: any long-running transaction; ignore this set()
458 } elseif ( $age > self
::MAX_READ_LAG
) {
459 $this->logger
->info( "Rejected set() for $key due to snapshot lag." );
461 return true; // no-op the write for being unsafe
462 // Case C: high replication lag; lower TTL instead of ignoring all set()s
463 } elseif ( $lag === false ||
$lag > self
::MAX_READ_LAG
) {
464 $ttl = $ttl ?
min( $ttl, self
::TTL_LAGGED
) : self
::TTL_LAGGED
;
465 $this->logger
->warning( "Lowered set() TTL for $key due to replication lag." );
466 // Case D: medium length request with medium replication lag; ignore this set()
468 $this->logger
->info( "Rejected set() for $key due to high read lag." );
470 return true; // no-op the write for being unsafe
474 // Wrap that value with time/TTL/version metadata
475 $wrapped = $this->wrap( $value, $ttl, $now ) +
$wrapExtra;
477 $func = function ( $cache, $key, $cWrapped ) use ( $wrapped ) {
478 return ( is_string( $cWrapped ) )
479 ?
false // key is tombstoned; do nothing
483 return $this->cache
->merge( self
::VALUE_KEY_PREFIX
. $key, $func, $ttl +
$staleTTL, 1 );
487 * Purge a key from all datacenters
489 * This should only be called when the underlying data (being cached)
490 * changes in a significant way. This deletes the key and starts a hold-off
491 * period where the key cannot be written to for a few seconds (HOLDOFF_TTL).
492 * This is done to avoid the following race condition:
493 * - a) Some DB data changes and delete() is called on a corresponding key
494 * - b) A request refills the key with a stale value from a lagged DB
495 * - c) The stale value is stuck there until the key is expired/evicted
497 * This is implemented by storing a special "tombstone" value at the cache
498 * key that this class recognizes; get() calls will return false for the key
499 * and any set() calls will refuse to replace tombstone values at the key.
500 * For this to always avoid stale value writes, the following must hold:
501 * - a) Replication lag is bounded to being less than HOLDOFF_TTL; or
502 * - b) If lag is higher, the DB will have gone into read-only mode already
504 * Note that set() can also be lag-aware and lower the TTL if it's high.
506 * When using potentially long-running ACID transactions, a good pattern is
507 * to use a pre-commit hook to issue the delete. This means that immediately
508 * after commit, callers will see the tombstone in cache upon purge relay.
509 * It also avoids the following race condition:
510 * - a) T1 begins, changes a row, and calls delete()
511 * - b) The HOLDOFF_TTL passes, expiring the delete() tombstone
512 * - c) T2 starts, reads the row and calls set() due to a cache miss
513 * - d) T1 finally commits
514 * - e) Stale value is stuck in cache
518 * $dbw->startAtomic( __METHOD__ ); // start of request
519 * ... <execute some stuff> ...
520 * // Update the row in the DB
521 * $dbw->update( ... );
522 * $key = $cache->makeKey( 'homes', $homeId );
523 * // Purge the corresponding cache entry just before committing
524 * $dbw->onTransactionPreCommitOrIdle( function() use ( $cache, $key ) {
525 * $cache->delete( $key );
527 * ... <execute some stuff> ...
528 * $dbw->endAtomic( __METHOD__ ); // end of request
531 * The $ttl parameter can be used when purging values that have not actually changed
532 * recently. For example, a cleanup script to purge cache entries does not really need
533 * a hold-off period, so it can use HOLDOFF_NONE. Likewise for user-requested purge.
534 * Note that $ttl limits the effective range of 'lockTSE' for getWithSetCallback().
536 * If called twice on the same key, then the last hold-off TTL takes precedence. For
537 * idempotence, the $ttl should not vary for different delete() calls on the same key.
539 * @param string $key Cache key
540 * @param int $ttl Tombstone TTL; Default: WANObjectCache::HOLDOFF_TTL
541 * @return bool True if the item was purged or not found, false on failure
543 final public function delete( $key, $ttl = self
::HOLDOFF_TTL
) {
544 $key = self
::VALUE_KEY_PREFIX
. $key;
547 // Publish the purge to all datacenters
548 $ok = $this->relayDelete( $key );
550 // Publish the purge to all datacenters
551 $ok = $this->relayPurge( $key, $ttl, self
::HOLDOFF_NONE
);
558 * Fetch the value of a timestamp "check" key
560 * The key will be *initialized* to the current time if not set,
561 * so only call this method if this behavior is actually desired
563 * The timestamp can be used to check whether a cached value is valid.
564 * Callers should not assume that this returns the same timestamp in
565 * all datacenters due to relay delays.
567 * The level of staleness can roughly be estimated from this key, but
568 * if the key was evicted from cache, such calculations may show the
569 * time since expiry as ~0 seconds.
571 * Note that "check" keys won't collide with other regular keys.
574 * @return float UNIX timestamp of the check key
576 final public function getCheckKeyTime( $key ) {
577 $key = self
::TIME_KEY_PREFIX
. $key;
579 $purge = self
::parsePurgeValue( $this->cache
->get( $key ) );
580 if ( $purge !== false ) {
581 $time = $purge[self
::FLD_TIME
];
583 // Casting assures identical floats for the next getCheckKeyTime() calls
584 $now = (string)microtime( true );
585 $this->cache
->add( $key,
586 $this->makePurgeValue( $now, self
::HOLDOFF_TTL
),
596 * Purge a "check" key from all datacenters, invalidating keys that use it
598 * This should only be called when the underlying data (being cached)
599 * changes in a significant way, and it is impractical to call delete()
600 * on all keys that should be changed. When get() is called on those
601 * keys, the relevant "check" keys must be supplied for this to work.
603 * The "check" key essentially represents a last-modified field.
604 * When touched, the field will be updated on all cache servers.
605 * Keys using it via get(), getMulti(), or getWithSetCallback() will
606 * be invalidated. It is treated as being HOLDOFF_TTL seconds in the future
607 * by those methods to avoid race conditions where dependent keys get updated
608 * with stale values (e.g. from a DB replica DB).
610 * This is typically useful for keys with hardcoded names or in some cases
611 * dynamically generated names where a low number of combinations exist.
612 * When a few important keys get a large number of hits, a high cache
613 * time is usually desired as well as "lockTSE" logic. The resetCheckKey()
614 * method is less appropriate in such cases since the "time since expiry"
615 * cannot be inferred, causing any get() after the reset to treat the key
616 * as being "hot", resulting in more stale value usage.
618 * Note that "check" keys won't collide with other regular keys.
620 * @see WANObjectCache::get()
621 * @see WANObjectCache::getWithSetCallback()
622 * @see WANObjectCache::resetCheckKey()
624 * @param string $key Cache key
625 * @param int $holdoff HOLDOFF_TTL or HOLDOFF_NONE constant
626 * @return bool True if the item was purged or not found, false on failure
628 final public function touchCheckKey( $key, $holdoff = self
::HOLDOFF_TTL
) {
629 // Publish the purge to all datacenters
630 return $this->relayPurge( self
::TIME_KEY_PREFIX
. $key, self
::CHECK_KEY_TTL
, $holdoff );
634 * Delete a "check" key from all datacenters, invalidating keys that use it
636 * This is similar to touchCheckKey() in that keys using it via get(), getMulti(),
637 * or getWithSetCallback() will be invalidated. The differences are:
638 * - a) The "check" key will be deleted from all caches and lazily
639 * re-initialized when accessed (rather than set everywhere)
640 * - b) Thus, dependent keys will be known to be invalid, but not
641 * for how long (they are treated as "just" purged), which
642 * effects any lockTSE logic in getWithSetCallback()
643 * - c) Since "check" keys are initialized only on the server the key hashes
644 * to, any temporary ejection of that server will cause the value to be
645 * seen as purged as a new server will initialize the "check" key.
647 * The advantage is that this does not place high TTL keys on every cache
648 * server, making it better for code that will cache many different keys
649 * and either does not use lockTSE or uses a low enough TTL anyway.
651 * This is typically useful for keys with dynamically generated names
652 * where a high number of combinations exist.
654 * Note that "check" keys won't collide with other regular keys.
656 * @see WANObjectCache::get()
657 * @see WANObjectCache::getWithSetCallback()
658 * @see WANObjectCache::touchCheckKey()
660 * @param string $key Cache key
661 * @return bool True if the item was purged or not found, false on failure
663 final public function resetCheckKey( $key ) {
664 // Publish the purge to all datacenters
665 return $this->relayDelete( self
::TIME_KEY_PREFIX
. $key );
669 * Method to fetch/regenerate cache keys
671 * On cache miss, the key will be set to the callback result via set()
672 * (unless the callback returns false) and that result will be returned.
673 * The arguments supplied to the callback are:
674 * - $oldValue : current cache value or false if not present
675 * - &$ttl : a reference to the TTL which can be altered
676 * - &$setOpts : a reference to options for set() which can be altered
677 * - $oldAsOf : generation UNIX timestamp of $oldValue or null if not present (since 1.28)
679 * It is strongly recommended to set the 'lag' and 'since' fields to avoid race conditions
680 * that can cause stale values to get stuck at keys. Usually, callbacks ignore the current
681 * value, but it can be used to maintain "most recent X" values that come from time or
682 * sequence based source data, provided that the "as of" id/time is tracked. Note that
683 * preemptive regeneration and $checkKeys can result in a non-false current value.
685 * Usage of $checkKeys is similar to get() and getMulti(). However, rather than the caller
686 * having to inspect a "current time left" variable (e.g. $curTTL, $curTTLs), a cache
687 * regeneration will automatically be triggered using the callback.
689 * The simplest way to avoid stampedes for hot keys is to use
690 * the 'lockTSE' option in $opts. If cache purges are needed, also:
691 * - a) Pass $key into $checkKeys
692 * - b) Use touchCheckKey( $key ) instead of delete( $key )
694 * Example usage (typical key):
696 * $catInfo = $cache->getWithSetCallback(
697 * // Key to store the cached value under
698 * $cache->makeKey( 'cat-attributes', $catId ),
699 * // Time-to-live (in seconds)
700 * $cache::TTL_MINUTE,
701 * // Function that derives the new key value
702 * function ( $oldValue, &$ttl, array &$setOpts ) {
703 * $dbr = wfGetDB( DB_REPLICA );
704 * // Account for any snapshot/replica DB lag
705 * $setOpts += Database::getCacheSetOptions( $dbr );
707 * return $dbr->selectRow( ... );
712 * Example usage (key that is expensive and hot):
714 * $catConfig = $cache->getWithSetCallback(
715 * // Key to store the cached value under
716 * $cache->makeKey( 'site-cat-config' ),
717 * // Time-to-live (in seconds)
719 * // Function that derives the new key value
720 * function ( $oldValue, &$ttl, array &$setOpts ) {
721 * $dbr = wfGetDB( DB_REPLICA );
722 * // Account for any snapshot/replica DB lag
723 * $setOpts += Database::getCacheSetOptions( $dbr );
725 * return CatConfig::newFromRow( $dbr->selectRow( ... ) );
728 * // Calling touchCheckKey() on this key invalidates the cache
729 * 'checkKeys' => [ $cache->makeKey( 'site-cat-config' ) ],
730 * // Try to only let one datacenter thread manage cache updates at a time
732 * // Avoid querying cache servers multiple times in a web request
733 * 'pcTTL' => $cache::TTL_PROC_LONG
738 * Example usage (key with dynamic dependencies):
740 * $catState = $cache->getWithSetCallback(
741 * // Key to store the cached value under
742 * $cache->makeKey( 'cat-state', $cat->getId() ),
743 * // Time-to-live (seconds)
745 * // Function that derives the new key value
746 * function ( $oldValue, &$ttl, array &$setOpts ) {
747 * // Determine new value from the DB
748 * $dbr = wfGetDB( DB_REPLICA );
749 * // Account for any snapshot/replica DB lag
750 * $setOpts += Database::getCacheSetOptions( $dbr );
752 * return CatState::newFromResults( $dbr->select( ... ) );
755 * // The "check" keys that represent things the value depends on;
756 * // Calling touchCheckKey() on any of them invalidates the cache
758 * $cache->makeKey( 'sustenance-bowls', $cat->getRoomId() ),
759 * $cache->makeKey( 'people-present', $cat->getHouseId() ),
760 * $cache->makeKey( 'cat-laws', $cat->getCityId() ),
766 * Example usage (hot key holding most recent 100 events):
768 * $lastCatActions = $cache->getWithSetCallback(
769 * // Key to store the cached value under
770 * $cache->makeKey( 'cat-last-actions', 100 ),
771 * // Time-to-live (in seconds)
773 * // Function that derives the new key value
774 * function ( $oldValue, &$ttl, array &$setOpts ) {
775 * $dbr = wfGetDB( DB_REPLICA );
776 * // Account for any snapshot/replica DB lag
777 * $setOpts += Database::getCacheSetOptions( $dbr );
779 * // Start off with the last cached list
780 * $list = $oldValue ?: [];
781 * // Fetch the last 100 relevant rows in descending order;
782 * // only fetch rows newer than $list[0] to reduce scanning
783 * $rows = iterator_to_array( $dbr->select( ... ) );
784 * // Merge them and get the new "last 100" rows
785 * return array_slice( array_merge( $new, $list ), 0, 100 );
788 * // Try to only let one datacenter thread manage cache updates at a time
790 * // Use a magic value when no cache value is ready rather than stampeding
791 * 'busyValue' => 'computing'
796 * @see WANObjectCache::get()
797 * @see WANObjectCache::set()
799 * @param string $key Cache key
800 * @param int $ttl Seconds to live for key updates. Special values are:
801 * - WANObjectCache::TTL_INDEFINITE: Cache forever
802 * - WANObjectCache::TTL_UNCACHEABLE: Do not cache at all
803 * @param callable $callback Value generation function
804 * @param array $opts Options map:
805 * - checkKeys: List of "check" keys. The key at $key will be seen as invalid when either
806 * touchCheckKey() or resetCheckKey() is called on any of these keys.
808 * - lockTSE: If the key is tombstoned or expired (by checkKeys) less than this many seconds
809 * ago, then try to have a single thread handle cache regeneration at any given time.
810 * Other threads will try to use stale values if possible. If, on miss, the time since
811 * expiration is low, the assumption is that the key is hot and that a stampede is worth
812 * avoiding. Setting this above WANObjectCache::HOLDOFF_TTL makes no difference. The
813 * higher this is set, the higher the worst-case staleness can be.
814 * Use WANObjectCache::TSE_NONE to disable this logic.
815 * Default: WANObjectCache::TSE_NONE.
816 * - busyValue: If no value exists and another thread is currently regenerating it, use this
817 * as a fallback value (or a callback to generate such a value). This assures that cache
818 * stampedes cannot happen if the value falls out of cache. This can be used as insurance
819 * against cache regeneration becoming very slow for some reason (greater than the TTL).
821 * - pcTTL: Process cache the value in this PHP instance for this many seconds. This avoids
822 * network I/O when a key is read several times. This will not cache when the callback
823 * returns false, however. Note that any purges will not be seen while process cached;
824 * since the callback should use replica DBs and they may be lagged or have snapshot
825 * isolation anyway, this should not typically matter.
826 * Default: WANObjectCache::TTL_UNCACHEABLE.
827 * - pcGroup: Process cache group to use instead of the primary one. If set, this must be
828 * of the format ALPHANUMERIC_NAME:MAX_KEY_SIZE, e.g. "mydata:10". Use this for storing
829 * large values, small yet numerous values, or some values with a high cost of eviction.
830 * It is generally preferable to use a class constant when setting this value.
831 * This has no effect unless pcTTL is used.
832 * Default: WANObjectCache::PC_PRIMARY.
833 * - version: Integer version number. This allows for callers to make breaking changes to
834 * how values are stored while maintaining compatability and correct cache purges. New
835 * versions are stored alongside older versions concurrently. Avoid storing class objects
836 * however, as this reduces compatibility (due to serialization).
838 * - minAsOf: Reject values if they were generated before this UNIX timestamp.
839 * This is useful if the source of a key is suspected of having possibly changed
840 * recently, and the caller wants any such changes to be reflected.
841 * Default: WANObjectCache::MIN_TIMESTAMP_NONE.
842 * - hotTTR: Expected time-till-refresh for keys that average ~1 hit/second.
843 * This should be greater than "ageNew". Keys with higher hit rates will regenerate
844 * more often. This is useful when a popular key is changed but the cache purge was
845 * delayed or lost. Seldom used keys are rarely affected by this setting, unless an
846 * extremely low "hotTTR" value is passed in.
847 * Default: WANObjectCache::HOT_TTR.
848 * - lowTTL: Consider pre-emptive updates when the current TTL (seconds) of the key is less
849 * than this. It becomes more likely over time, becoming certain once the key is expired.
850 * Default: WANObjectCache::LOW_TTL.
851 * - ageNew: Consider popularity refreshes only once a key reaches this age in seconds.
852 * Default: WANObjectCache::AGE_NEW.
853 * @return mixed Value found or written to the key
854 * @note Options added in 1.28: version, busyValue, hotTTR, ageNew, pcGroup, minAsOf
855 * @note Callable type hints are not used to avoid class-autoloading
857 final public function getWithSetCallback( $key, $ttl, $callback, array $opts = [] ) {
858 $pcTTL = isset( $opts['pcTTL'] ) ?
$opts['pcTTL'] : self
::TTL_UNCACHEABLE
;
860 // Try the process cache if enabled and the cache callback is not within a cache callback.
861 // Process cache use in nested callbacks is not lag-safe with regard to HOLDOFF_TTL since
862 // the in-memory value is further lagged than the shared one since it uses a blind TTL.
863 if ( $pcTTL >= 0 && $this->callbackDepth
== 0 ) {
864 $group = isset( $opts['pcGroup'] ) ?
$opts['pcGroup'] : self
::PC_PRIMARY
;
865 $procCache = $this->getProcessCache( $group );
866 $value = $procCache->get( $key );
872 if ( $value === false ) {
873 // Fetch the value over the network
874 if ( isset( $opts['version'] ) ) {
875 $version = $opts['version'];
877 $cur = $this->doGetWithSetCallback(
880 function ( $oldValue, &$ttl, &$setOpts, $oldAsOf )
881 use ( $callback, $version ) {
882 if ( is_array( $oldValue )
883 && array_key_exists( self
::VFLD_DATA
, $oldValue )
885 $oldData = $oldValue[self
::VFLD_DATA
];
887 // VFLD_DATA is not set if an old, unversioned, key is present
892 self
::VFLD_DATA
=> $callback( $oldData, $ttl, $setOpts, $oldAsOf ),
893 self
::VFLD_VERSION
=> $version
899 if ( $cur[self
::VFLD_VERSION
] === $version ) {
900 // Value created or existed before with version; use it
901 $value = $cur[self
::VFLD_DATA
];
903 // Value existed before with a different version; use variant key.
904 // Reflect purges to $key by requiring that this key value be newer.
905 $value = $this->doGetWithSetCallback(
906 'cache-variant:' . md5( $key ) . ":$version",
909 // Regenerate value if not newer than $key
910 [ 'version' => null, 'minAsOf' => $asOf ] +
$opts
914 $value = $this->doGetWithSetCallback( $key, $ttl, $callback, $opts );
917 // Update the process cache if enabled
918 if ( $procCache && $value !== false ) {
919 $procCache->set( $key, $value, $pcTTL );
927 * Do the actual I/O for getWithSetCallback() when needed
929 * @see WANObjectCache::getWithSetCallback()
933 * @param callback $callback
934 * @param array $opts Options map for getWithSetCallback()
935 * @param float &$asOf Cache generation timestamp of returned value [returned]
937 * @note Callable type hints are not used to avoid class-autoloading
939 protected function doGetWithSetCallback( $key, $ttl, $callback, array $opts, &$asOf = null ) {
940 $lowTTL = isset( $opts['lowTTL'] ) ?
$opts['lowTTL'] : min( self
::LOW_TTL
, $ttl );
941 $lockTSE = isset( $opts['lockTSE'] ) ?
$opts['lockTSE'] : self
::TSE_NONE
;
942 $checkKeys = isset( $opts['checkKeys'] ) ?
$opts['checkKeys'] : [];
943 $busyValue = isset( $opts['busyValue'] ) ?
$opts['busyValue'] : null;
944 $popWindow = isset( $opts['hotTTR'] ) ?
$opts['hotTTR'] : self
::HOT_TTR
;
945 $ageNew = isset( $opts['ageNew'] ) ?
$opts['ageNew'] : self
::AGE_NEW
;
946 $minTime = isset( $opts['minAsOf'] ) ?
$opts['minAsOf'] : self
::MIN_TIMESTAMP_NONE
;
947 $versioned = isset( $opts['version'] );
949 // Get the current key value
951 $cValue = $this->get( $key, $curTTL, $checkKeys, $asOf ); // current value
952 $value = $cValue; // return value
954 $preCallbackTime = microtime( true );
955 // Determine if a cached value regeneration is needed or desired
956 if ( $value !== false
958 && $this->isValid( $value, $versioned, $asOf, $minTime )
959 && !$this->worthRefreshExpiring( $curTTL, $lowTTL )
960 && !$this->worthRefreshPopular( $asOf, $ageNew, $popWindow, $preCallbackTime )
965 // A deleted key with a negative TTL left must be tombstoned
966 $isTombstone = ( $curTTL !== null && $value === false );
967 // Assume a key is hot if requested soon after invalidation
968 $isHot = ( $curTTL !== null && $curTTL <= 0 && abs( $curTTL ) <= $lockTSE );
969 // Use the mutex if there is no value and a busy fallback is given
970 $checkBusy = ( $busyValue !== null && $value === false );
971 // Decide whether a single thread should handle regenerations.
972 // This avoids stampedes when $checkKeys are bumped and when preemptive
973 // renegerations take too long. It also reduces regenerations while $key
974 // is tombstoned. This balances cache freshness with avoiding DB load.
975 $useMutex = ( $isHot ||
( $isTombstone && $lockTSE > 0 ) ||
$checkBusy );
977 $lockAcquired = false;
979 // Acquire a datacenter-local non-blocking lock
980 if ( $this->cache
->add( self
::MUTEX_KEY_PREFIX
. $key, 1, self
::LOCK_TTL
) ) {
981 // Lock acquired; this thread should update the key
982 $lockAcquired = true;
983 } elseif ( $value !== false && $this->isValid( $value, $versioned, $asOf, $minTime ) ) {
984 // If it cannot be acquired; then the stale value can be used
987 // Use the INTERIM value for tombstoned keys to reduce regeneration load.
988 // For hot keys, either another thread has the lock or the lock failed;
989 // use the INTERIM value from the last thread that regenerated it.
990 $wrapped = $this->cache
->get( self
::INTERIM_KEY_PREFIX
. $key );
991 list( $value ) = $this->unwrap( $wrapped, microtime( true ) );
992 if ( $value !== false && $this->isValid( $value, $versioned, $asOf, $minTime ) ) {
993 $asOf = $wrapped[self
::FLD_TIME
];
997 // Use the busy fallback value if nothing else
998 if ( $busyValue !== null ) {
999 return is_callable( $busyValue ) ?
$busyValue() : $busyValue;
1004 if ( !is_callable( $callback ) ) {
1005 throw new InvalidArgumentException( "Invalid cache miss callback provided." );
1008 // Generate the new value from the callback...
1010 ++
$this->callbackDepth
;
1012 $value = call_user_func_array( $callback, [ $cValue, &$ttl, &$setOpts, $asOf ] );
1014 --$this->callbackDepth
;
1016 // When delete() is called, writes are write-holed by the tombstone,
1017 // so use a special INTERIM key to pass the new value around threads.
1018 if ( ( $isTombstone && $lockTSE > 0 ) && $value !== false && $ttl >= 0 ) {
1019 $tempTTL = max( 1, (int)$lockTSE ); // set() expects seconds
1020 $newAsOf = microtime( true );
1021 $wrapped = $this->wrap( $value, $tempTTL, $newAsOf );
1022 // Avoid using set() to avoid pointless mcrouter broadcasting
1023 $this->cache
->merge(
1024 self
::INTERIM_KEY_PREFIX
. $key,
1025 function () use ( $wrapped ) {
1033 if ( $value !== false && $ttl >= 0 ) {
1034 $setOpts['lockTSE'] = $lockTSE;
1035 // Use best known "since" timestamp if not provided
1036 $setOpts +
= [ 'since' => $preCallbackTime ];
1037 // Update the cache; this will fail if the key is tombstoned
1038 $this->set( $key, $value, $ttl, $setOpts );
1041 if ( $lockAcquired ) {
1042 // Avoid using delete() to avoid pointless mcrouter broadcasting
1043 $this->cache
->changeTTL( self
::MUTEX_KEY_PREFIX
. $key, 1 );
1050 * Method to fetch multiple cache keys at once with regeneration
1052 * This works the same as getWithSetCallback() except:
1053 * - a) The $keys argument expects the result of WANObjectCache::makeMultiKeys()
1054 * - b) The $callback argument expects a callback taking the following arguments:
1055 * - $id: ID of an entity to query
1056 * - $oldValue : the prior cache value or false if none was present
1057 * - &$ttl : a reference to the new value TTL in seconds
1058 * - &$setOpts : a reference to options for set() which can be altered
1059 * - $oldAsOf : generation UNIX timestamp of $oldValue or null if not present
1060 * Aside from the additional $id argument, the other arguments function the same
1061 * way they do in getWithSetCallback().
1062 * - c) The return value is a map of (cache key => value) in the order of $keyedIds
1064 * @see WANObjectCache::getWithSetCallback()
1065 * @see WANObjectCache::getMultiWithUnionSetCallback()
1069 * $rows = $cache->getMultiWithSetCallback(
1070 * // Map of cache keys to entity IDs
1071 * $cache->makeMultiKeys(
1072 * $this->fileVersionIds(),
1073 * function ( $id, WANObjectCache $cache ) {
1074 * return $cache->makeKey( 'file-version', $id );
1077 * // Time-to-live (in seconds)
1079 * // Function that derives the new key value
1080 * function ( $id, $oldValue, &$ttl, array &$setOpts ) {
1081 * $dbr = wfGetDB( DB_REPLICA );
1082 * // Account for any snapshot/replica DB lag
1083 * $setOpts += Database::getCacheSetOptions( $dbr );
1085 * // Load the row for this file
1086 * $row = $dbr->selectRow( 'file', '*', [ 'id' => $id ], __METHOD__ );
1088 * return $row ? (array)$row : false;
1091 * // Process cache for 30 seconds
1093 * // Use a dedicated 500 item cache (initialized on-the-fly)
1094 * 'pcGroup' => 'file-versions:500'
1097 * $files = array_map( [ __CLASS__, 'newFromRow' ], $rows );
1100 * @param ArrayIterator $keyedIds Result of WANObjectCache::makeMultiKeys()
1101 * @param int $ttl Seconds to live for key updates
1102 * @param callable $callback Callback the yields entity regeneration callbacks
1103 * @param array $opts Options map
1104 * @return array Map of (cache key => value) in the same order as $keyedIds
1107 final public function getMultiWithSetCallback(
1108 ArrayIterator
$keyedIds, $ttl, callable
$callback, array $opts = []
1110 $valueKeys = array_keys( $keyedIds->getArrayCopy() );
1111 $checkKeys = isset( $opts['checkKeys'] ) ?
$opts['checkKeys'] : [];
1113 // Load required keys into process cache in one go
1114 $this->warmupCache
= $this->getRawKeysForWarmup(
1115 $this->getNonProcessCachedKeys( $valueKeys, $opts ),
1118 $this->warmupKeyMisses
= 0;
1120 // Wrap $callback to match the getWithSetCallback() format while passing $id to $callback
1121 $id = null; // current entity ID
1122 $func = function ( $oldValue, &$ttl, &$setOpts, $oldAsOf ) use ( $callback, &$id ) {
1123 return $callback( $id, $oldValue, $ttl, $setOpts, $oldAsOf );
1127 foreach ( $keyedIds as $key => $id ) { // preserve order
1128 $values[$key] = $this->getWithSetCallback( $key, $ttl, $func, $opts );
1131 $this->warmupCache
= [];
1137 * Method to fetch/regenerate multiple cache keys at once
1139 * This works the same as getWithSetCallback() except:
1140 * - a) The $keys argument expects the result of WANObjectCache::makeMultiKeys()
1141 * - b) The $callback argument expects a callback returning a map of (ID => new value)
1142 * for all entity IDs in $regenById and it takes the following arguments:
1143 * - $ids: a list of entity IDs to regenerate
1144 * - &$ttls: a reference to the (entity ID => new TTL) map
1145 * - &$setOpts: a reference to options for set() which can be altered
1146 * - c) The return value is a map of (cache key => value) in the order of $keyedIds
1147 * - d) The "lockTSE" and "busyValue" options are ignored
1149 * @see WANObjectCache::getWithSetCallback()
1150 * @see WANObjectCache::getMultiWithSetCallback()
1154 * $rows = $cache->getMultiWithUnionSetCallback(
1155 * // Map of cache keys to entity IDs
1156 * $cache->makeMultiKeys(
1157 * $this->fileVersionIds(),
1158 * function ( $id, WANObjectCache $cache ) {
1159 * return $cache->makeKey( 'file-version', $id );
1162 * // Time-to-live (in seconds)
1164 * // Function that derives the new key value
1165 * function ( array $ids, array &$ttls, array &$setOpts ) {
1166 * $dbr = wfGetDB( DB_REPLICA );
1167 * // Account for any snapshot/replica DB lag
1168 * $setOpts += Database::getCacheSetOptions( $dbr );
1170 * // Load the rows for these files
1172 * $res = $dbr->select( 'file', '*', [ 'id' => $ids ], __METHOD__ );
1173 * foreach ( $res as $row ) {
1174 * $rows[$row->id] = $row;
1175 * $mtime = wfTimestamp( TS_UNIX, $row->timestamp );
1176 * $ttls[$row->id] = $this->adaptiveTTL( $mtime, $ttls[$row->id] );
1183 * $files = array_map( [ __CLASS__, 'newFromRow' ], $rows );
1186 * @param ArrayIterator $keyedIds Result of WANObjectCache::makeMultiKeys()
1187 * @param int $ttl Seconds to live for key updates
1188 * @param callable $callback Callback the yields entity regeneration callbacks
1189 * @param array $opts Options map
1190 * @return array Map of (cache key => value) in the same order as $keyedIds
1193 final public function getMultiWithUnionSetCallback(
1194 ArrayIterator
$keyedIds, $ttl, callable
$callback, array $opts = []
1196 $idsByValueKey = $keyedIds->getArrayCopy();
1197 $valueKeys = array_keys( $idsByValueKey );
1198 $checkKeys = isset( $opts['checkKeys'] ) ?
$opts['checkKeys'] : [];
1199 unset( $opts['lockTSE'] ); // incompatible
1200 unset( $opts['busyValue'] ); // incompatible
1202 // Load required keys into process cache in one go
1203 $keysGet = $this->getNonProcessCachedKeys( $valueKeys, $opts );
1204 $this->warmupCache
= $this->getRawKeysForWarmup( $keysGet, $checkKeys );
1205 $this->warmupKeyMisses
= 0;
1207 // IDs of entities known to be in need of regeneration
1210 // Find out which keys are missing/deleted/stale
1213 $curByKey = $this->getMulti( $keysGet, $curTTLs, $checkKeys, $asOfs );
1214 foreach ( $keysGet as $key ) {
1215 if ( !array_key_exists( $key, $curByKey ) ||
$curTTLs[$key] < 0 ) {
1216 $idsRegen[] = $idsByValueKey[$key];
1220 // Run the callback to populate the regeneration value map for all required IDs
1222 $newTTLsById = array_fill_keys( $idsRegen, $ttl );
1223 $newValsById = $idsRegen ?
$callback( $idsRegen, $newTTLsById, $newSetOpts ) : [];
1225 // Wrap $callback to match the getWithSetCallback() format while passing $id to $callback
1226 $id = null; // current entity ID
1227 $func = function ( $oldValue, &$ttl, &$setOpts, $oldAsOf )
1228 use ( $callback, &$id, $newValsById, $newTTLsById, $newSetOpts )
1230 if ( array_key_exists( $id, $newValsById ) ) {
1231 // Value was already regerated as expected, so use the value in $newValsById
1232 $newValue = $newValsById[$id];
1233 $ttl = $newTTLsById[$id];
1234 $setOpts = $newSetOpts;
1236 // Pre-emptive/popularity refresh and version mismatch cases are not detected
1237 // above and thus $newValsById has no entry. Run $callback on this single entity.
1238 $ttls = [ $id => $ttl ];
1239 $newValue = $callback( [ $id ], $ttls, $setOpts )[$id];
1246 // Run the cache-aside logic using warmupCache instead of persistent cache queries
1248 foreach ( $idsByValueKey as $key => $id ) { // preserve order
1249 $values[$key] = $this->getWithSetCallback( $key, $ttl, $func, $opts );
1252 $this->warmupCache
= [];
1258 * Locally set a key to expire soon if it is stale based on $purgeTimestamp
1260 * This sets stale keys' time-to-live at HOLDOFF_TTL seconds, which both avoids
1261 * broadcasting in mcrouter setups and also avoids races with new tombstones.
1263 * @param string $key Cache key
1264 * @param int $purgeTimestamp UNIX timestamp of purge
1265 * @param bool &$isStale Whether the key is stale
1266 * @return bool Success
1269 public function reap( $key, $purgeTimestamp, &$isStale = false ) {
1270 $minAsOf = $purgeTimestamp + self
::HOLDOFF_TTL
;
1271 $wrapped = $this->cache
->get( self
::VALUE_KEY_PREFIX
. $key );
1272 if ( is_array( $wrapped ) && $wrapped[self
::FLD_TIME
] < $minAsOf ) {
1274 $this->logger
->warning( "Reaping stale value key '$key'." );
1275 $ttlReap = self
::HOLDOFF_TTL
; // avoids races with tombstone creation
1276 $ok = $this->cache
->changeTTL( self
::VALUE_KEY_PREFIX
. $key, $ttlReap );
1278 $this->logger
->error( "Could not complete reap of key '$key'." );
1290 * Locally set a "check" key to expire soon if it is stale based on $purgeTimestamp
1292 * @param string $key Cache key
1293 * @param int $purgeTimestamp UNIX timestamp of purge
1294 * @param bool &$isStale Whether the key is stale
1295 * @return bool Success
1298 public function reapCheckKey( $key, $purgeTimestamp, &$isStale = false ) {
1299 $purge = $this->parsePurgeValue( $this->cache
->get( self
::TIME_KEY_PREFIX
. $key ) );
1300 if ( $purge && $purge[self
::FLD_TIME
] < $purgeTimestamp ) {
1302 $this->logger
->warning( "Reaping stale check key '$key'." );
1303 $ok = $this->cache
->changeTTL( self
::TIME_KEY_PREFIX
. $key, 1 );
1305 $this->logger
->error( "Could not complete reap of check key '$key'." );
1317 * @see BagOStuff::makeKey()
1318 * @param string $keys,... Key component
1322 public function makeKey() {
1323 return call_user_func_array( [ $this->cache
, __FUNCTION__
], func_get_args() );
1327 * @see BagOStuff::makeGlobalKey()
1328 * @param string $keys,... Key component
1332 public function makeGlobalKey() {
1333 return call_user_func_array( [ $this->cache
, __FUNCTION__
], func_get_args() );
1337 * @param array $entities List of entity IDs
1338 * @param callable $keyFunc Callback yielding a key from (entity ID, this WANObjectCache)
1339 * @return ArrayIterator Iterator yielding (cache key => entity ID) in $entities order
1342 public function makeMultiKeys( array $entities, callable
$keyFunc ) {
1344 foreach ( $entities as $entity ) {
1345 $map[$keyFunc( $entity, $this )] = $entity;
1348 return new ArrayIterator( $map );
1352 * Get the "last error" registered; clearLastError() should be called manually
1353 * @return int ERR_* class constant for the "last error" registry
1355 final public function getLastError() {
1356 if ( $this->lastRelayError
) {
1357 // If the cache and the relayer failed, focus on the latter.
1358 // An update not making it to the relayer means it won't show up
1359 // in other DCs (nor will consistent re-hashing see up-to-date values).
1360 // On the other hand, if just the cache update failed, then it should
1361 // eventually be applied by the relayer.
1362 return $this->lastRelayError
;
1365 $code = $this->cache
->getLastError();
1367 case BagOStuff
::ERR_NONE
:
1368 return self
::ERR_NONE
;
1369 case BagOStuff
::ERR_NO_RESPONSE
:
1370 return self
::ERR_NO_RESPONSE
;
1371 case BagOStuff
::ERR_UNREACHABLE
:
1372 return self
::ERR_UNREACHABLE
;
1374 return self
::ERR_UNEXPECTED
;
1379 * Clear the "last error" registry
1381 final public function clearLastError() {
1382 $this->cache
->clearLastError();
1383 $this->lastRelayError
= self
::ERR_NONE
;
1387 * Clear the in-process caches; useful for testing
1391 public function clearProcessCache() {
1392 $this->processCaches
= [];
1396 * @param int $flag ATTR_* class constant
1397 * @return int QOS_* class constant
1400 public function getQoS( $flag ) {
1401 return $this->cache
->getQoS( $flag );
1405 * Get a TTL that is higher for objects that have not changed recently
1407 * This is useful for keys that get explicit purges and DB or purge relay
1408 * lag is a potential concern (especially how it interacts with CDN cache)
1412 * // Last-modified time of page
1413 * $mtime = wfTimestamp( TS_UNIX, $page->getTimestamp() );
1414 * // Get adjusted TTL. If $mtime is 3600 seconds ago and $minTTL/$factor left at
1415 * // defaults, then $ttl is 3600 * .2 = 720. If $minTTL was greater than 720, then
1416 * // $ttl would be $minTTL. If $maxTTL was smaller than 720, $ttl would be $maxTTL.
1417 * $ttl = $cache->adaptiveTTL( $mtime, $cache::TTL_DAY );
1420 * @param int|float $mtime UNIX timestamp
1421 * @param int $maxTTL Maximum TTL (seconds)
1422 * @param int $minTTL Minimum TTL (seconds); Default: 30
1423 * @param float $factor Value in the range (0,1); Default: .2
1424 * @return int Adaptive TTL
1427 public function adaptiveTTL( $mtime, $maxTTL, $minTTL = 30, $factor = .2 ) {
1428 if ( is_float( $mtime ) ||
ctype_digit( $mtime ) ) {
1429 $mtime = (int)$mtime; // handle fractional seconds and string integers
1432 if ( !is_int( $mtime ) ||
$mtime <= 0 ) {
1433 return $minTTL; // no last-modified time provided
1436 $age = time() - $mtime;
1438 return (int)min( $maxTTL, max( $minTTL, $factor * $age ) );
1442 * @return int Number of warmup key cache misses last round
1445 public function getWarmupKeyMisses() {
1446 return $this->warmupKeyMisses
;
1450 * Do the actual async bus purge of a key
1452 * This must set the key to "PURGED:<UNIX timestamp>:<holdoff>"
1454 * @param string $key Cache key
1455 * @param int $ttl How long to keep the tombstone [seconds]
1456 * @param int $holdoff HOLDOFF_* constant controlling how long to ignore sets for this key
1457 * @return bool Success
1459 protected function relayPurge( $key, $ttl, $holdoff ) {
1460 if ( $this->purgeRelayer
instanceof EventRelayerNull
) {
1461 // This handles the mcrouter and the single-DC case
1462 $ok = $this->cache
->set( $key,
1463 $this->makePurgeValue( microtime( true ), self
::HOLDOFF_NONE
),
1467 $event = $this->cache
->modifySimpleRelayEvent( [
1470 'val' => 'PURGED:$UNIXTIME$:' . (int)$holdoff,
1471 'ttl' => max( $ttl, 1 ),
1472 'sbt' => true, // substitute $UNIXTIME$ with actual microtime
1475 $ok = $this->purgeRelayer
->notify( $this->purgeChannel
, $event );
1477 $this->lastRelayError
= self
::ERR_RELAY
;
1485 * Do the actual async bus delete of a key
1487 * @param string $key Cache key
1488 * @return bool Success
1490 protected function relayDelete( $key ) {
1491 if ( $this->purgeRelayer
instanceof EventRelayerNull
) {
1492 // This handles the mcrouter and the single-DC case
1493 $ok = $this->cache
->delete( $key );
1495 $event = $this->cache
->modifySimpleRelayEvent( [
1500 $ok = $this->purgeRelayer
->notify( $this->purgeChannel
, $event );
1502 $this->lastRelayError
= self
::ERR_RELAY
;
1510 * Check if a key should be regenerated (using random probability)
1512 * This returns false if $curTTL >= $lowTTL. Otherwise, the chance
1513 * of returning true increases steadily from 0% to 100% as the $curTTL
1514 * moves from $lowTTL to 0 seconds. This handles widely varying
1515 * levels of cache access traffic.
1517 * @param float $curTTL Approximate TTL left on the key if present
1518 * @param float $lowTTL Consider a refresh when $curTTL is less than this
1521 protected function worthRefreshExpiring( $curTTL, $lowTTL ) {
1522 if ( $curTTL >= $lowTTL ) {
1524 } elseif ( $curTTL <= 0 ) {
1528 $chance = ( 1 - $curTTL / $lowTTL );
1530 return mt_rand( 1, 1e9
) <= 1e9
* $chance;
1534 * Check if a key is due for randomized regeneration due to its popularity
1536 * This is used so that popular keys can preemptively refresh themselves for higher
1537 * consistency (especially in the case of purge loss/delay). Unpopular keys can remain
1538 * in cache with their high nominal TTL. This means popular keys keep good consistency,
1539 * whether the data changes frequently or not, and long-tail keys get to stay in cache
1540 * and get hits too. Similar to worthRefreshExpiring(), randomization is used.
1542 * @param float $asOf UNIX timestamp of the value
1543 * @param int $ageNew Age of key when this might recommend refreshing (seconds)
1544 * @param int $timeTillRefresh Age of key when it should be refreshed if popular (seconds)
1545 * @param float $now The current UNIX timestamp
1548 protected function worthRefreshPopular( $asOf, $ageNew, $timeTillRefresh, $now ) {
1549 $age = $now - $asOf;
1550 $timeOld = $age - $ageNew;
1551 if ( $timeOld <= 0 ) {
1555 // Lifecycle is: new, ramp-up refresh chance, full refresh chance.
1556 // Note that the "expected # of refreshes" for the ramp-up time range is half of what it
1557 // would be if P(refresh) was at its full value during that time range.
1558 $refreshWindowSec = max( $timeTillRefresh - $ageNew - self
::RAMPUP_TTL
/ 2, 1 );
1559 // P(refresh) * (# hits in $refreshWindowSec) = (expected # of refreshes)
1560 // P(refresh) * ($refreshWindowSec * $popularHitsPerSec) = 1
1561 // P(refresh) = 1/($refreshWindowSec * $popularHitsPerSec)
1562 $chance = 1 / ( self
::HIT_RATE_HIGH
* $refreshWindowSec );
1564 // Ramp up $chance from 0 to its nominal value over RAMPUP_TTL seconds to avoid stampedes
1565 $chance *= ( $timeOld <= self
::RAMPUP_TTL
) ?
$timeOld / self
::RAMPUP_TTL
: 1;
1567 return mt_rand( 1, 1e9
) <= 1e9
* $chance;
1571 * Check whether $value is appropriately versioned and not older than $minTime (if set)
1573 * @param array $value
1574 * @param bool $versioned
1575 * @param float $asOf The time $value was generated
1576 * @param float $minTime The last time the main value was generated (0.0 if unknown)
1579 protected function isValid( $value, $versioned, $asOf, $minTime ) {
1580 if ( $versioned && !isset( $value[self
::VFLD_VERSION
] ) ) {
1582 } elseif ( $minTime > 0 && $asOf < $minTime ) {
1590 * Do not use this method outside WANObjectCache
1592 * @param mixed $value
1593 * @param int $ttl [0=forever]
1594 * @param float $now Unix Current timestamp just before calling set()
1597 protected function wrap( $value, $ttl, $now ) {
1599 self
::FLD_VERSION
=> self
::VERSION
,
1600 self
::FLD_VALUE
=> $value,
1601 self
::FLD_TTL
=> $ttl,
1602 self
::FLD_TIME
=> $now
1607 * Do not use this method outside WANObjectCache
1609 * @param array|string|bool $wrapped
1610 * @param float $now Unix Current timestamp (preferrably pre-query)
1611 * @return array (mixed; false if absent/tombstoned/invalid, current time left)
1613 protected function unwrap( $wrapped, $now ) {
1614 // Check if the value is a tombstone
1615 $purge = self
::parsePurgeValue( $wrapped );
1616 if ( $purge !== false ) {
1617 // Purged values should always have a negative current $ttl
1618 $curTTL = min( $purge[self
::FLD_TIME
] - $now, self
::TINY_NEGATIVE
);
1619 return [ false, $curTTL ];
1622 if ( !is_array( $wrapped ) // not found
1623 ||
!isset( $wrapped[self
::FLD_VERSION
] ) // wrong format
1624 ||
$wrapped[self
::FLD_VERSION
] !== self
::VERSION
// wrong version
1626 return [ false, null ];
1629 $flags = isset( $wrapped[self
::FLD_FLAGS
] ) ?
$wrapped[self
::FLD_FLAGS
] : 0;
1630 if ( ( $flags & self
::FLG_STALE
) == self
::FLG_STALE
) {
1631 // Treat as expired, with the cache time as the expiration
1632 $age = $now - $wrapped[self
::FLD_TIME
];
1633 $curTTL = min( -$age, self
::TINY_NEGATIVE
);
1634 } elseif ( $wrapped[self
::FLD_TTL
] > 0 ) {
1635 // Get the approximate time left on the key
1636 $age = $now - $wrapped[self
::FLD_TIME
];
1637 $curTTL = max( $wrapped[self
::FLD_TTL
] - $age, 0.0 );
1639 // Key had no TTL, so the time left is unbounded
1643 return [ $wrapped[self
::FLD_VALUE
], $curTTL ];
1647 * @param array $keys
1648 * @param string $prefix
1651 protected static function prefixCacheKeys( array $keys, $prefix ) {
1653 foreach ( $keys as $key ) {
1654 $res[] = $prefix . $key;
1661 * @param string $value Wrapped value like "PURGED:<timestamp>:<holdoff>"
1662 * @return array|bool Array containing a UNIX timestamp (float) and holdoff period (integer),
1663 * or false if value isn't a valid purge value
1665 protected static function parsePurgeValue( $value ) {
1666 if ( !is_string( $value ) ) {
1669 $segments = explode( ':', $value, 3 );
1670 if ( !isset( $segments[0] ) ||
!isset( $segments[1] )
1671 ||
"{$segments[0]}:" !== self
::PURGE_VAL_PREFIX
1675 if ( !isset( $segments[2] ) ) {
1676 // Back-compat with old purge values without holdoff
1677 $segments[2] = self
::HOLDOFF_TTL
;
1680 self
::FLD_TIME
=> (float)$segments[1],
1681 self
::FLD_HOLDOFF
=> (int)$segments[2],
1686 * @param float $timestamp
1687 * @param int $holdoff In seconds
1688 * @return string Wrapped purge value
1690 protected function makePurgeValue( $timestamp, $holdoff ) {
1691 return self
::PURGE_VAL_PREFIX
. (float)$timestamp . ':' . (int)$holdoff;
1695 * @param string $group
1696 * @return HashBagOStuff
1698 protected function getProcessCache( $group ) {
1699 if ( !isset( $this->processCaches
[$group] ) ) {
1700 list( , $n ) = explode( ':', $group );
1701 $this->processCaches
[$group] = new HashBagOStuff( [ 'maxKeys' => (int)$n ] );
1704 return $this->processCaches
[$group];
1708 * @param array $keys
1709 * @param array $opts
1710 * @return array List of keys
1712 private function getNonProcessCachedKeys( array $keys, array $opts ) {
1714 if ( isset( $opts['pcTTL'] ) && $opts['pcTTL'] > 0 && $this->callbackDepth
== 0 ) {
1715 $pcGroup = isset( $opts['pcGroup'] ) ?
$opts['pcGroup'] : self
::PC_PRIMARY
;
1716 $procCache = $this->getProcessCache( $pcGroup );
1717 foreach ( $keys as $key ) {
1718 if ( $procCache->get( $key ) !== false ) {
1719 $keysFound[] = $key;
1724 return array_diff( $keys, $keysFound );
1728 * @param array $keys
1729 * @param array $checkKeys
1730 * @return array Map of (cache key => mixed)
1732 private function getRawKeysForWarmup( array $keys, array $checkKeys ) {
1738 // Get all the value keys to fetch...
1739 foreach ( $keys as $key ) {
1740 $keysWarmUp[] = self
::VALUE_KEY_PREFIX
. $key;
1742 // Get all the check keys to fetch...
1743 foreach ( $checkKeys as $i => $checkKeyOrKeys ) {
1744 if ( is_int( $i ) ) {
1745 // Single check key that applies to all value keys
1746 $keysWarmUp[] = self
::TIME_KEY_PREFIX
. $checkKeyOrKeys;
1748 // List of check keys that apply to value key $i
1749 $keysWarmUp = array_merge(
1751 self
::prefixCacheKeys( $checkKeyOrKeys, self
::TIME_KEY_PREFIX
)
1756 $warmupCache = $this->cache
->getMulti( $keysWarmUp );
1757 $warmupCache +
= array_fill_keys( $keysWarmUp, false );
1759 return $warmupCache;