MediaWiki  1.28.0
WANObjectCache Class Reference

Multi-datacenter aware caching interface. More...

Inheritance diagram for WANObjectCache:
Collaboration diagram for WANObjectCache:

Public Member Functions

 __construct (array $params)
 
 adaptiveTTL ($mtime, $maxTTL, $minTTL=30, $factor=.2)
 Get a TTL that is higher for objects that have not changed recently. More...
 
 clearLastError ()
 Clear the "last error" registry. More...
 
 clearProcessCache ()
 Clear the in-process caches; useful for testing. More...
 
 delete ($key, $ttl=self::HOLDOFF_TTL)
 Purge a key from all datacenters. More...
 
 get ($key, &$curTTL=null, array $checkKeys=[], &$asOf=null)
 Fetch the value of a key from cache. More...
 
 getCheckKeyTime ($key)
 Fetch the value of a timestamp "check" key. More...
 
 getLastError ()
 Get the "last error" registered; clearLastError() should be called manually. More...
 
 getMulti (array $keys, &$curTTLs=[], array $checkKeys=[], array &$asOfs=[])
 Fetch the value of several keys from cache. More...
 
 getMultiWithSetCallback (ArrayIterator $keyedIds, $ttl, callable $callback, array $opts=[])
 Method to fetch/regenerate multiple cache keys at once. More...
 
 getQoS ($flag)
 
 getWithSetCallback ($key, $ttl, $callback, array $opts=[])
 Method to fetch/regenerate cache keys. More...
 
 makeGlobalKey ()
 
 makeKey ()
 
 makeMultiKeys (array $entities, callable $keyFunc)
 
 resetCheckKey ($key)
 Delete a "check" key from all datacenters, invalidating keys that use it. More...
 
 set ($key, $value, $ttl=0, array $opts=[])
 Set the value of a key in cache. More...
 
 setLogger (LoggerInterface $logger)
 
 touchCheckKey ($key, $holdoff=self::HOLDOFF_TTL)
 Purge a "check" key from all datacenters, invalidating keys that use it. More...
 

Static Public Member Functions

static newEmpty ()
 Get an instance that wraps EmptyBagOStuff. More...
 

Public Attributes

const AGE_NEW = 60
 Never consider performing "popularity" refreshes until a key reaches this age. More...
 
const CHECK_KEY_TTL = self::TTL_YEAR
 Seconds to keep dependency purge keys around. More...
 
const DEFAULT_PURGE_CHANNEL = 'wancache-purge'
 
const ERR_NO_RESPONSE = 1
 
const ERR_NONE = 0
 
const ERR_RELAY = 4
 
const ERR_UNEXPECTED = 3
 
const ERR_UNREACHABLE = 2
 
const FLD_FLAGS = 4
 
const FLD_HOLDOFF = 5
 
const FLD_TIME = 3
 
const FLD_TTL = 2
 
const FLD_VALUE = 1
 
const FLD_VERSION = 0
 
const HIT_RATE_HIGH = 1
 Hits/second for a refresh to be expected within the "popularity" window. More...
 
const HOLDOFF_NONE = 0
 Idiom for delete() for "no hold-off". More...
 
const HOLDOFF_TTL = 11
 Seconds to tombstone keys on delete() More...
 
const HOT_TTR = 900
 The time length of the "popularity" refresh window for hot keys. More...
 
const INTERIM_KEY_PREFIX = 'WANCache:i:'
 
const LOCK_TSE = 1
 Default time-since-expiry on a miss that makes a key "hot". More...
 
const LOCK_TTL = 10
 Seconds to keep lock keys around. More...
 
const LOW_TTL = 30
 Default remaining TTL at which to consider pre-emptive regeneration. More...
 
const MAX_COMMIT_DELAY = 3
 Max time expected to pass between delete() and DB commit finishing. More...
 
const MAX_READ_LAG = 7
 Max replication+snapshot lag before applying TTL_LAGGED or disallowing set() More...
 
const MIN_TIMESTAMP_NONE = 0.0
 Idiom for getWithSetCallback() for "no minimum required as-of timestamp". More...
 
const MUTEX_KEY_PREFIX = 'WANCache:m:'
 
const PC_PRIMARY = 'primary:1000'
 
const PURGE_VAL_PREFIX = 'PURGED:'
 
const RAMPUP_TTL = 30
 Seconds to ramp up to the "popularity" refresh chance after a key is no longer new. More...
 
const TIME_KEY_PREFIX = 'WANCache:t:'
 
const TINY_NEGATIVE = -0.000001
 Tiny negative float to use when CTL comes up >= 0 due to clock skew. More...
 
const TSE_NONE = -1
 Idiom for getWithSetCallback() callbacks to 'lockTSE' logic. More...
 
const TTL_LAGGED = 30
 Max TTL to store keys when a data sourced is lagged. More...
 
const TTL_UNCACHEABLE = -1
 Idiom for getWithSetCallback() callbacks to avoid calling set() More...
 
const VALUE_KEY_PREFIX = 'WANCache:v:'
 
const VERSION = 1
 Cache format version number. More...
 
const VFLD_DATA = 'WOC:d'
 
const VFLD_VERSION = 'WOC:v'
 
- Public Attributes inherited from IExpiringStore
const ATTR_EMULATION = 1
 
const ATTR_SYNCWRITES = 2
 
const QOS_EMULATION_SQL = 1
 
const QOS_SYNCWRITES_BE = 2
 
const QOS_SYNCWRITES_NONE = 1
 
const QOS_SYNCWRITES_QC = 3
 
const QOS_SYNCWRITES_SS = 4
 
const QOS_UNKNOWN = INF
 
const TTL_DAY = 86400
 
const TTL_HOUR = 3600
 
const TTL_INDEFINITE = 0
 
const TTL_MINUTE = 60
 
const TTL_MONTH = 2592000
 
const TTL_PROC_LONG = 30
 
const TTL_PROC_SHORT = 3
 
const TTL_WEEK = 604800
 
const TTL_YEAR = 31536000
 

Protected Member Functions

 doGetWithSetCallback ($key, $ttl, $callback, array $opts, &$asOf=null)
 Do the actual I/O for getWithSetCallback() when needed. More...
 
 getProcessCache ($group)
 
 isValid ($value, $versioned, $asOf, $minTime)
 Check whether $value is appropriately versioned and not older than $minTime (if set) More...
 
 makePurgeValue ($timestamp, $holdoff)
 
 relayDelete ($key)
 Do the actual async bus delete of a key. More...
 
 relayPurge ($key, $ttl, $holdoff)
 Do the actual async bus purge of a key. More...
 
 unwrap ($wrapped, $now)
 Do not use this method outside WANObjectCache. More...
 
 worthRefreshExpiring ($curTTL, $lowTTL)
 Check if a key should be regenerated (using random probability) More...
 
 worthRefreshPopular ($asOf, $ageNew, $timeTillRefresh, $now)
 Check if a key is due for randomized regeneration due to its popularity. More...
 
 wrap ($value, $ttl, $now)
 Do not use this method outside WANObjectCache. More...
 

Static Protected Member Functions

static parsePurgeValue ($value)
 
static prefixCacheKeys (array $keys, $prefix)
 

Protected Attributes

BagOStuff $cache
 The local datacenter cache. More...
 
int $lastRelayError = self::ERR_NONE
 ERR_* constant for the "last error" registry. More...
 
LoggerInterface $logger
 
HashBagOStuff[] $processCaches = []
 Map of group PHP instance caches. More...
 
string $purgeChannel
 Purge channel name. More...
 
EventRelayer $purgeRelayer
 Bus that handles purge broadcasts. More...
 

Private Member Functions

 processCheckKeys (array $timeKeys, array $wrappedValues, $now)
 

Private Attributes

integer $callbackDepth = 0
 Callback stack depth for getWithSetCallback() More...
 
mixed[] $warmupCache = []
 Temporary warm-up cache. More...
 

Detailed Description

Multi-datacenter aware caching interface.

All operations go to the local datacenter cache, except for delete(), touchCheckKey(), and resetCheckKey(), which broadcast to all datacenters.

This class is intended for caching data from primary stores. If the get() method does not return a value, then the caller should query the new value and backfill the cache using set(). The preferred way to do this logic is through getWithSetCallback(). When querying the store on cache miss, the closest DB replica should be used. Try to avoid heavyweight DB master or quorum reads. When the source data changes, a purge method should be called. Since purges are expensive, they should be avoided. One can do so if:

  • a) The object cached is immutable; or
  • b) Validity is checked against the source after get(); or
  • c) Using a modest TTL is reasonably correct and performant

The simplest purge method is delete().

There are two supported ways to handle broadcasted operations:

  • a) Configure the 'purge' EventRelayer to point to a valid PubSub endpoint that has subscribed listeners on the cache servers applying the cache updates.
  • b) Ignore the 'purge' EventRelayer configuration (default is NullEventRelayer) and set up mcrouter as the underlying cache backend, using one of the memcached BagOStuff classes as 'cache'. Use OperationSelectorRoute in the mcrouter settings to configure 'set' and 'delete' operations to go to all DCs via AllAsyncRoute and configure other operations to go to the local DC via PoolRoute (for reference, see https://github.com/facebook/mcrouter/wiki/List-of-Route-Handles).

Broadcasted operations like delete() and touchCheckKey() are done asynchronously in all datacenters this way, though the local one should likely be near immediate.

This means that callers in all datacenters may see older values for however many milliseconds that the purge took to reach that datacenter. As with any cache, this should not be relied on for cases where reads are used to determine writes to source (e.g. non-cache) data stores, except when reading immutable data.

All values are wrapped in metadata arrays. Keys use a "WANCache:" prefix to avoid collisions with keys that are not wrapped as metadata arrays. The prefixes are as follows:

  • a) "WANCache:v" : used for regular value keys
  • b) "WANCache:i" : used for temporarily storing values of tombstoned keys
  • c) "WANCache:t" : used for storing timestamp "check" keys
  • d) "WANCache:m" : used for temporary mutex keys to avoid cache stampedes
Since
1.26

Definition at line 76 of file WANObjectCache.php.

Constructor & Destructor Documentation

WANObjectCache::__construct ( array  $params)
Parameters
array$params
  • cache : BagOStuff object for a persistent cache
  • channels : Map of (action => channel string). Actions include "purge".
  • relayers : Map of (action => EventRelayer object). Actions include "purge".
  • logger : LoggerInterface object

Definition at line 175 of file WANObjectCache.php.

References cache, and setLogger().

Member Function Documentation

WANObjectCache::adaptiveTTL (   $mtime,
  $maxTTL,
  $minTTL = 30,
  $factor = .2 
)

Get a TTL that is higher for objects that have not changed recently.

This is useful for keys that get explicit purges and DB or purge relay lag is a potential concern (especially how it interacts with CDN cache)

Example usage:

// Last-modified time of page
$mtime = wfTimestamp( TS_UNIX, $page->getTimestamp() );
// Get adjusted TTL. If $mtime is 3600 seconds ago and $minTTL/$factor left at
// defaults, then $ttl is 3600 * .2 = 720. If $minTTL was greater than 720, then
// $ttl would be $minTTL. If $maxTTL was smaller than 720, $ttl would be $maxTTL.
$ttl = $cache->adaptiveTTL( $mtime, $cache::TTL_DAY );
Parameters
integer | float$mtimeUNIX timestamp
integer$maxTTLMaximum TTL (seconds)
integer$minTTLMinimum TTL (seconds); Default: 30
float$factorValue in the range (0,1); Default: .2
Returns
integer Adaptive TTL
Since
1.28

Definition at line 1241 of file WANObjectCache.php.

WANObjectCache::clearLastError ( )
final

Clear the "last error" registry.

Definition at line 1195 of file WANObjectCache.php.

References cache.

WANObjectCache::clearProcessCache ( )

Clear the in-process caches; useful for testing.

Since
1.27

Definition at line 1205 of file WANObjectCache.php.

WANObjectCache::delete (   $key,
  $ttl = self::HOLDOFF_TTL 
)
final

Purge a key from all datacenters.

This should only be called when the underlying data (being cached) changes in a significant way. This deletes the key and starts a hold-off period where the key cannot be written to for a few seconds (HOLDOFF_TTL). This is done to avoid the following race condition:

  • a) Some DB data changes and delete() is called on a corresponding key
  • b) A request refills the key with a stale value from a lagged DB
  • c) The stale value is stuck there until the key is expired/evicted

This is implemented by storing a special "tombstone" value at the cache key that this class recognizes; get() calls will return false for the key and any set() calls will refuse to replace tombstone values at the key. For this to always avoid stale value writes, the following must hold:

  • a) Replication lag is bounded to being less than HOLDOFF_TTL; or
  • b) If lag is higher, the DB will have gone into read-only mode already

Note that set() can also be lag-aware and lower the TTL if it's high.

When using potentially long-running ACID transactions, a good pattern is to use a pre-commit hook to issue the delete. This means that immediately after commit, callers will see the tombstone in cache upon purge relay. It also avoids the following race condition:

  • a) T1 begins, changes a row, and calls delete()
  • b) The HOLDOFF_TTL passes, expiring the delete() tombstone
  • c) T2 starts, reads the row and calls set() due to a cache miss
  • d) T1 finally commits
  • e) Stale value is stuck in cache

Example usage:

$dbw->startAtomic( __METHOD__ ); // start of request
... <execute some stuff> ...
// Update the row in the DB
$dbw->update( ... );
$key = $cache->makeKey( 'homes', $homeId );
// Purge the corresponding cache entry just before committing
$dbw->onTransactionPreCommitOrIdle( function() use ( $cache, $key ) {
$cache->delete( $key );
} );
... <execute some stuff> ...
$dbw->endAtomic( __METHOD__ ); // end of request

The $ttl parameter can be used when purging values that have not actually changed recently. For example, a cleanup script to purge cache entries does not really need a hold-off period, so it can use HOLDOFF_NONE. Likewise for user-requested purge. Note that $ttl limits the effective range of 'lockTSE' for getWithSetCallback().

If called twice on the same key, then the last hold-off TTL takes precedence. For idempotence, the $ttl should not vary for different delete() calls on the same key.

Parameters
string$keyCache key
integer$ttlTombstone TTL; Default: WANObjectCache::HOLDOFF_TTL
Returns
bool True if the item was purged or not found, false on failure

Definition at line 535 of file WANObjectCache.php.

References relayDelete(), and relayPurge().

Referenced by WANObjectCacheTest\testBusyValue(), WANObjectCacheTest\testGetMultiWithSetCallback(), WANObjectCacheTest\testGetWithSetCallback(), WANObjectCacheTest\testGetWithSetCallback_versions(), WANObjectCacheTest\testLockTSE(), and WANObjectCacheTest\testLockTSESlow().

WANObjectCache::doGetWithSetCallback (   $key,
  $ttl,
  $callback,
array  $opts,
$asOf = null 
)
protected

Do the actual I/O for getWithSetCallback() when needed.

See also
WANObjectCache::getWithSetCallback()
Parameters
string$key
integer$ttl
callback$callback
array$optsOptions map for getWithSetCallback()
float&$asOfCache generation timestamp of returned value [returned]
Returns
mixed
Note
Callable type hints are not used to avoid class-autoloading

Definition at line 931 of file WANObjectCache.php.

References $callbackDepth, $value, cache, false, isValid(), list, set(), unwrap(), use, worthRefreshExpiring(), worthRefreshPopular(), and wrap().

Referenced by getWithSetCallback().

WANObjectCache::get (   $key,
$curTTL = null,
array  $checkKeys = [],
$asOf = null 
)
final

Fetch the value of a key from cache.

If supplied, $curTTL is set to the remaining TTL (current time left):

  • a) INF; if $key exists, has no TTL, and is not expired by $checkKeys
  • b) float (>=0); if $key exists, has a TTL, and is not expired by $checkKeys
  • c) float (<0); if $key is tombstoned, stale, or existing but expired by $checkKeys
  • d) null; if $key does not exist and is not tombstoned

If a key is tombstoned, $curTTL will reflect the time since delete().

The timestamp of $key will be checked against the last-purge timestamp of each of $checkKeys. Those $checkKeys not in cache will have the last-purge initialized to the current timestamp. If any of $checkKeys have a timestamp greater than that of $key, then $curTTL will reflect how long ago $key became invalid. Callers can use $curTTL to know when the value is stale. The $checkKeys parameter allow mass invalidations by updating a single key:

  • a) Each "check" key represents "last purged" of some source data
  • b) Callers pass in relevant "check" keys as $checkKeys in get()
  • c) When the source data that "check" keys represent changes, the touchCheckKey() method is called on them

Source data entities might exists in a DB that uses snapshot isolation (e.g. the default REPEATABLE-READ in innoDB). Even for mutable data, that isolation can largely be maintained by doing the following:

  • a) Calling delete() on entity change and creation, before DB commit
  • b) Keeping transaction duration shorter than delete() hold-off TTL

However, pre-snapshot values might still be seen if an update was made in a remote datacenter but the purge from delete() didn't relay yet.

Consider using getWithSetCallback() instead of get() and set() cycles. That method has cache slam avoiding features for hot/expensive keys.

Parameters
string$keyCache key
mixed$curTTLApproximate TTL left on the key if present/tombstoned [returned]
array$checkKeysList of "check" keys
float&$asOfUNIX timestamp of cached value; null on failure [returned]
Returns
mixed Value of cache key or false on failure

Definition at line 243 of file WANObjectCache.php.

References getMulti().

Referenced by JobQueueDB\doDeduplicateRootJob(), WANObjectCacheTest\testCheckKeyInitHoldoff(), WANObjectCacheTest\testGetMultiWithSetCallback(), WANObjectCacheTest\testGetWithSetCallback(), and WANObjectCacheTest\testLockTSESlow().

WANObjectCache::getCheckKeyTime (   $key)
final

Fetch the value of a timestamp "check" key.

The key will be initialized to the current time if not set, so only call this method if this behavior is actually desired

The timestamp can be used to check whether a cached value is valid. Callers should not assume that this returns the same timestamp in all datacenters due to relay delays.

The level of staleness can roughly be estimated from this key, but if the key was evicted from cache, such calculations may show the time since expiry as ~0 seconds.

Note that "check" keys won't collide with other regular keys.

Parameters
string$key
Returns
float UNIX timestamp of the check key

Definition at line 568 of file WANObjectCache.php.

References $time, cache, makePurgeValue(), and string.

Referenced by WANObjectCacheTest\testGetMulti(), WANObjectCacheTest\testGetMultiWithSetCallback(), and WANObjectCacheTest\testGetWithSetCallback().

WANObjectCache::getLastError ( )
final

Get the "last error" registered; clearLastError() should be called manually.

Returns
int ERR_* class constant for the "last error" registry

Definition at line 1169 of file WANObjectCache.php.

References $code, $lastRelayError, cache, BagOStuff\ERR_NO_RESPONSE, BagOStuff\ERR_NONE, and BagOStuff\ERR_UNREACHABLE.

WANObjectCache::getMulti ( array  $keys,
$curTTLs = [],
array  $checkKeys = [],
array $asOfs = [] 
)
final

Fetch the value of several keys from cache.

See also
WANObjectCache::get()
Parameters
array$keysList of cache keys
array$curTTLsMap of (key => approximate TTL left) for existing keys [returned]
array$checkKeysList of check keys to apply to all $keys. May also apply "check" keys to specific cache keys only by using cache keys as keys in the $checkKeys array.
float[]&$asOfs Map of (key => UNIX timestamp of cached value; null on failure)
Returns
array Map of (key => value) for keys that exist

Definition at line 265 of file WANObjectCache.php.

References $value, array(), as, cache, false, list, processCheckKeys(), and unwrap().

Referenced by get(), WANObjectCacheTest\testGetMulti(), and WANObjectCacheTest\testGetMultiCheckKeys().

WANObjectCache::getMultiWithSetCallback ( ArrayIterator  $keyedIds,
  $ttl,
callable  $callback,
array  $opts = [] 
)
final

Method to fetch/regenerate multiple cache keys at once.

This works the same as getWithSetCallback() except:

  • a) The $keys argument expects the result of WANObjectCache::makeMultiKeys()
  • b) The $callback argument expects a callback taking the following arguments:
    • $id: ID of an entity to query
    • $oldValue : the prior cache value or false if none was present
    • &$ttl : a reference to the new value TTL in seconds
    • &$setOpts : a reference to options for set() which can be altered
    • $oldAsOf : generation UNIX timestamp of $oldValue or null if not present Aside from the additional $id argument, the other arguments function the same way they do in getWithSetCallback().
  • c) The return value is a map of (cache key => value) in the order of $keyedIds
See also
WANObjectCache::getWithSetCallback()

Example usage:

$rows = $cache->getMultiWithSetCallback(
// Map of cache keys to entity IDs
$cache->makeMultiKeys(
$this->fileVersionIds(),
function ( $id, WANObjectCache $cache ) {
return $cache->makeKey( 'file-version', $id );
}
),
// Time-to-live (in seconds)
$cache::TTL_DAY,
// Function that derives the new key value
return function ( $id, $oldValue, &$ttl, array &$setOpts ) {
// Account for any snapshot/replica DB lag
// Load the row for this file
$row = $dbr->selectRow( 'file', '*', [ 'id' => $id ], __METHOD__ );
return $row ? (array)$row : false;
},
[
// Process cache for 30 seconds
'pcTTL' => 30,
// Use a dedicated 500 item cache (initialized on-the-fly)
'pcGroup' => 'file-versions:500'
]
);
$files = array_map( [ __CLASS__, 'newFromRow' ], $rows );
Parameters
ArrayIterator$keyedIdsResult of WANObjectCache::makeMultiKeys()
integer$ttlSeconds to live for key updates
callable$callbackCallback the yields entity regeneration callbacks
array$optsOptions map
Returns
array Map of (cache key => value) in the same order as $keyedIds
Since
1.28

Definition at line 1098 of file WANObjectCache.php.

References array(), as, cache, getWithSetCallback(), and use.

Referenced by WANObjectCacheTest\testGetMultiWithSetCallback().

WANObjectCache::getProcessCache (   $group)
protected
Parameters
string$group
Returns
HashBagOStuff

Definition at line 1502 of file WANObjectCache.php.

References list.

Referenced by getWithSetCallback().

WANObjectCache::getQoS (   $flag)
Parameters
integer$flagATTR_* class constant
Returns
integer QOS_* class constant
Since
1.28

Definition at line 1214 of file WANObjectCache.php.

References cache.

WANObjectCache::getWithSetCallback (   $key,
  $ttl,
  $callback,
array  $opts = [] 
)
final

Method to fetch/regenerate cache keys.

On cache miss, the key will be set to the callback result via set() (unless the callback returns false) and that result will be returned. The arguments supplied to the callback are:

  • $oldValue : current cache value or false if not present
  • &$ttl : a reference to the TTL which can be altered
  • &$setOpts : a reference to options for set() which can be altered
  • $oldAsOf : generation UNIX timestamp of $oldValue or null if not present (since 1.28)

It is strongly recommended to set the 'lag' and 'since' fields to avoid race conditions that can cause stale values to get stuck at keys. Usually, callbacks ignore the current value, but it can be used to maintain "most recent X" values that come from time or sequence based source data, provided that the "as of" id/time is tracked. Note that preemptive regeneration and $checkKeys can result in a non-false current value.

Usage of $checkKeys is similar to get() and getMulti(). However, rather than the caller having to inspect a "current time left" variable (e.g. $curTTL, $curTTLs), a cache regeneration will automatically be triggered using the callback.

The simplest way to avoid stampedes for hot keys is to use the 'lockTSE' option in $opts. If cache purges are needed, also:

  • a) Pass $key into $checkKeys
  • b) Use touchCheckKey( $key ) instead of delete( $key )

Example usage (typical key):

// Key to store the cached value under
$cache->makeKey( 'cat-attributes', $catId ),
// Time-to-live (in seconds)
$cache::TTL_MINUTE,
// Function that derives the new key value
function ( $oldValue, &$ttl, array &$setOpts ) {
// Account for any snapshot/replica DB lag
return $dbr->selectRow( ... );
}
);

Example usage (key that is expensive and hot):

$catConfig = $cache->getWithSetCallback(
// Key to store the cached value under
$cache->makeKey( 'site-cat-config' ),
// Time-to-live (in seconds)
$cache::TTL_DAY,
// Function that derives the new key value
function ( $oldValue, &$ttl, array &$setOpts ) {
// Account for any snapshot/replica DB lag
return CatConfig::newFromRow( $dbr->selectRow( ... ) );
},
[
// Calling touchCheckKey() on this key invalidates the cache
'checkKeys' => [ $cache->makeKey( 'site-cat-config' ) ],
// Try to only let one datacenter thread manage cache updates at a time
'lockTSE' => 30,
// Avoid querying cache servers multiple times in a web request
'pcTTL' => $cache::TTL_PROC_LONG
]
);

Example usage (key with dynamic dependencies):

// Key to store the cached value under
$cache->makeKey( 'cat-state', $cat->getId() ),
// Time-to-live (seconds)
$cache::TTL_HOUR,
// Function that derives the new key value
function ( $oldValue, &$ttl, array &$setOpts ) {
// Determine new value from the DB
// Account for any snapshot/replica DB lag
return CatState::newFromResults( $dbr->select( ... ) );
},
[
// The "check" keys that represent things the value depends on;
// Calling touchCheckKey() on any of them invalidates the cache
'checkKeys' => [
$cache->makeKey( 'sustenance-bowls', $cat->getRoomId() ),
$cache->makeKey( 'people-present', $cat->getHouseId() ),
$cache->makeKey( 'cat-laws', $cat->getCityId() ),
]
]
);

Example usage (hot key holding most recent 100 events):

$lastCatActions = $cache->getWithSetCallback(
// Key to store the cached value under
$cache->makeKey( 'cat-last-actions', 100 ),
// Time-to-live (in seconds)
10,
// Function that derives the new key value
function ( $oldValue, &$ttl, array &$setOpts ) {
// Account for any snapshot/replica DB lag
// Start off with the last cached list
$list = $oldValue ?: [];
// Fetch the last 100 relevant rows in descending order;
// only fetch rows newer than $list[0] to reduce scanning
$rows = iterator_to_array( $dbr->select( ... ) );
// Merge them and get the new "last 100" rows
return array_slice( array_merge( $new, $list ), 0, 100 );
},
[
// Try to only let one datacenter thread manage cache updates at a time
'lockTSE' => 30,
// Use a magic value when no cache value is ready rather than stampeding
'busyValue' => 'computing'
]
);
See also
WANObjectCache::get()
WANObjectCache::set()
Parameters
string$keyCache key
integer$ttlSeconds to live for key updates. Special values are:
callable$callbackValue generation function
array$optsOptions map:
  • checkKeys: List of "check" keys. The key at $key will be seen as invalid when either touchCheckKey() or resetCheckKey() is called on any of these keys. Default: [].
  • lockTSE: If the key is tombstoned or expired (by checkKeys) less than this many seconds ago, then try to have a single thread handle cache regeneration at any given time. Other threads will try to use stale values if possible. If, on miss, the time since expiration is low, the assumption is that the key is hot and that a stampede is worth avoiding. Setting this above WANObjectCache::HOLDOFF_TTL makes no difference. The higher this is set, the higher the worst-case staleness can be. Use WANObjectCache::TSE_NONE to disable this logic. Default: WANObjectCache::TSE_NONE.
  • busyValue: If no value exists and another thread is currently regenerating it, use this as a fallback value (or a callback to generate such a value). This assures that cache stampedes cannot happen if the value falls out of cache. This can be used as insurance against cache regeneration becoming very slow for some reason (greater than the TTL). Default: null.
  • pcTTL: Process cache the value in this PHP instance for this many seconds. This avoids network I/O when a key is read several times. This will not cache when the callback returns false, however. Note that any purges will not be seen while process cached; since the callback should use replica DBs and they may be lagged or have snapshot isolation anyway, this should not typically matter. Default: WANObjectCache::TTL_UNCACHEABLE.
  • pcGroup: Process cache group to use instead of the primary one. If set, this must be of the format ALPHANUMERIC_NAME:MAX_KEY_SIZE, e.g. "mydata:10". Use this for storing large values, small yet numerous values, or some values with a high cost of eviction. It is generally preferable to use a class constant when setting this value. This has no effect unless pcTTL is used. Default: WANObjectCache::PC_PRIMARY.
  • version: Integer version number. This allows for callers to make breaking changes to how values are stored while maintaining compatability and correct cache purges. New versions are stored alongside older versions concurrently. Avoid storing class objects however, as this reduces compatibility (due to serialization). Default: null.
  • minAsOf: Reject values if they were generated before this UNIX timestamp. This is useful if the source of a key is suspected of having possibly changed recently, and the caller wants any such changes to be reflected. Default: WANObjectCache::MIN_TIMESTAMP_NONE.
  • hotTTR: Expected time-till-refresh for keys that average ~1 hit/second. This should be greater than "ageNew". Keys with higher hit rates will regenerate more often. This is useful when a popular key is changed but the cache purge was delayed or lost. Seldom used keys are rarely affected by this setting, unless an extremely low "hotTTR" value is passed in. Default: WANObjectCache::HOT_TTR.
  • lowTTL: Consider pre-emptive updates when the current TTL (seconds) of the key is less than this. It becomes more likely over time, becoming certain once the key is expired. Default: WANObjectCache::LOW_TTL.
  • ageNew: Consider popularity refreshes only once a key reaches this age in seconds. Default: WANObjectCache::AGE_NEW.
Returns
mixed Value found or written to the key
Note
Options added in 1.28: version, busyValue, hotTTR, ageNew, pcGroup, minAsOf
Callable type hints are not used to avoid class-autoloading

Definition at line 849 of file WANObjectCache.php.

References $value, doGetWithSetCallback(), getProcessCache(), and use.

Referenced by getMultiWithSetCallback(), WANObjectCacheTest\testBusyValue(), WANObjectCacheTest\testGetWithSetCallback(), WANObjectCacheTest\testGetWithSetCallback_versions(), WANObjectCacheTest\testLockTSE(), and WANObjectCacheTest\testLockTSESlow().

WANObjectCache::isValid (   $value,
  $versioned,
  $asOf,
  $minTime 
)
protected

Check whether $value is appropriately versioned and not older than $minTime (if set)

Parameters
array$value
bool$versioned
float$asOfThe time $value was generated
float$minTimeThe last time the main value was generated (0.0 if unknown)
Returns
bool

Definition at line 1383 of file WANObjectCache.php.

References $value.

Referenced by doGetWithSetCallback().

WANObjectCache::makeGlobalKey ( )
See also
BagOStuff::makeGlobalKey()
Parameters
string... Key component
Returns
string
Since
1.27

Definition at line 1146 of file WANObjectCache.php.

References cache.

Referenced by User\getCacheKey().

WANObjectCache::makeKey ( )
See also
BagOStuff::makeKey()
Parameters
string... Key component
Returns
string
Since
1.27

Definition at line 1136 of file WANObjectCache.php.

References cache.

WANObjectCache::makeMultiKeys ( array  $entities,
callable  $keyFunc 
)
Parameters
array$entitiesList of entity IDs
callable$keyFuncCallback yielding a key from (entity ID, this WANObjectCache)
Returns
ArrayIterator Iterator yielding (cache key => entity ID) in $entities order
Since
1.28

Definition at line 1156 of file WANObjectCache.php.

References as.

Referenced by WANObjectCacheTest\testGetMultiWithSetCallback().

WANObjectCache::makePurgeValue (   $timestamp,
  $holdoff 
)
protected
Parameters
float$timestamp
int$holdoffIn seconds
Returns
string Wrapped purge value

Definition at line 1494 of file WANObjectCache.php.

References $timestamp.

Referenced by getCheckKeyTime(), processCheckKeys(), and relayPurge().

static WANObjectCache::parsePurgeValue (   $value)
staticprotected
Parameters
string$valueWrapped value like "PURGED:<timestamp>:<holdoff>"
Returns
array|bool Array containing a UNIX timestamp (float) and holdoff period (integer), or false if value isn't a valid purge value

Definition at line 1469 of file WANObjectCache.php.

References $value.

static WANObjectCache::prefixCacheKeys ( array  $keys,
  $prefix 
)
staticprotected
Parameters
array$keys
string$prefix
Returns
string[]

Definition at line 1455 of file WANObjectCache.php.

References $res, and as.

WANObjectCache::processCheckKeys ( array  $timeKeys,
array  $wrappedValues,
  $now 
)
private
Since
1.27
Parameters
array$timeKeysList of prefixed time check keys
array$wrappedValues
float$now
Returns
array List of purge value arrays

Definition at line 353 of file WANObjectCache.php.

References as, cache, and makePurgeValue().

Referenced by getMulti().

WANObjectCache::relayDelete (   $key)
protected

Do the actual async bus delete of a key.

Parameters
string$keyCache key
Returns
bool Success

Definition at line 1296 of file WANObjectCache.php.

References cache.

Referenced by delete(), and resetCheckKey().

WANObjectCache::relayPurge (   $key,
  $ttl,
  $holdoff 
)
protected

Do the actual async bus purge of a key.

This must set the key to "PURGED:<UNIX timestamp>:<holdoff>"

Parameters
string$keyCache key
integer$ttlHow long to keep the tombstone [seconds]
integer$holdoffHOLDOFF_* constant controlling how long to ignore sets for this key
Returns
bool Success

Definition at line 1265 of file WANObjectCache.php.

References cache, and makePurgeValue().

Referenced by delete(), and touchCheckKey().

WANObjectCache::resetCheckKey (   $key)
final

Delete a "check" key from all datacenters, invalidating keys that use it.

This is similar to touchCheckKey() in that keys using it via get(), getMulti(), or getWithSetCallback() will be invalidated. The differences are:

  • a) The "check" key will be deleted from all caches and lazily re-initialized when accessed (rather than set everywhere)
  • b) Thus, dependent keys will be known to be invalid, but not for how long (they are treated as "just" purged), which effects any lockTSE logic in getWithSetCallback()
  • c) Since "check" keys are initialized only on the server the key hashes to, any temporary ejection of that server will cause the value to be seen as purged as a new server will initialize the "check" key.

The advantage is that this does not place high TTL keys on every cache server, making it better for code that will cache many different keys and either does not use lockTSE or uses a low enough TTL anyway.

This is typically useful for keys with dynamically generated names where a high number of combinations exist.

Note that "check" keys won't collide with other regular keys.

See also
WANObjectCache::get()
WANObjectCache::getWithSetCallback()
WANObjectCache::touchCheckKey()
Parameters
string$keyCache key
Returns
bool True if the item was purged or not found, false on failure

Definition at line 655 of file WANObjectCache.php.

References relayDelete().

WANObjectCache::set (   $key,
  $value,
  $ttl = 0,
array  $opts = [] 
)
final

Set the value of a key in cache.

Simply calling this method when source data changes is not valid because the changes do not replicate to the other WAN sites. In that case, delete() should be used instead. This method is intended for use on cache misses.

If the data was read from a snapshot-isolated transactions (e.g. the default REPEATABLE-READ in innoDB), use 'since' to avoid the following race condition:

  • a) T1 starts
  • b) T2 updates a row, calls delete(), and commits
  • c) The HOLDOFF_TTL passes, expiring the delete() tombstone
  • d) T1 reads the row and calls set() due to a cache miss
  • e) Stale value is stuck in cache

Setting 'lag' and 'since' help avoids keys getting stuck in stale states.

Example usage:

// Fetch the row from the DB
$row = $dbr->selectRow( ... );
$key = $cache->makeKey( 'building', $buildingId );
$cache->set( $key, $row, $cache::TTL_DAY, $setOpts );
Parameters
string$keyCache key
mixed$value
integer$ttlSeconds to live. Special values are:
array$optsOptions map:
  • lag : Seconds of replica DB lag. Typically, this is either the replica DB lag before the data was read or, if applicable, the replica DB lag before the snapshot-isolated transaction the data was read from started. Use false to indicate that replication is not running. Default: 0 seconds
  • since : UNIX timestamp of the data in $value. Typically, this is either the current time the data was read or (if applicable) the time when the snapshot-isolated transaction the data was read from started. Default: 0 seconds
  • pending : Whether this data is possibly from an uncommitted write transaction. Generally, other threads should not see values from the future and they certainly should not see ones that ended up getting rolled back. Default: false
  • lockTSE : if excessive replication/snapshot lag is detected, then store the value with this TTL and flag it as stale. This is only useful if the reads for this key use getWithSetCallback() with "lockTSE" set. Default: WANObjectCache::TSE_NONE
  • staleTTL : Seconds to keep the key around if it is stale. The get()/getMulti() methods return such stale values with a $curTTL of 0, and getWithSetCallback() will call the regeneration callback in such cases, passing in the old value and its as-of time to the callback. This is useful if adaptiveTTL() is used on the old value's as-of time when it is verified as still being correct. Default: 0.
Note
Options added in 1.28: staleTTL
Returns
bool Success

Definition at line 428 of file WANObjectCache.php.

References $cache, $value, cache, use, and wrap().

Referenced by JobQueueDB\doDeduplicateRootJob(), doGetWithSetCallback(), WANObjectCacheTest\testCheckKeyInitHoldoff(), WANObjectCacheTest\testGetMulti(), and WANObjectCacheTest\testGetMultiCheckKeys().

WANObjectCache::setLogger ( LoggerInterface  $logger)

Definition at line 186 of file WANObjectCache.php.

References $logger.

Referenced by __construct().

WANObjectCache::touchCheckKey (   $key,
  $holdoff = self::HOLDOFF_TTL 
)
final

Purge a "check" key from all datacenters, invalidating keys that use it.

This should only be called when the underlying data (being cached) changes in a significant way, and it is impractical to call delete() on all keys that should be changed. When get() is called on those keys, the relevant "check" keys must be supplied for this to work.

The "check" key essentially represents a last-modified field. When touched, the field will be updated on all cache servers. Keys using it via get(), getMulti(), or getWithSetCallback() will be invalidated. It is treated as being HOLDOFF_TTL seconds in the future by those methods to avoid race conditions where dependent keys get updated with stale values (e.g. from a DB replica DB).

This is typically useful for keys with hardcoded names or in some cases dynamically generated names where a low number of combinations exist. When a few important keys get a large number of hits, a high cache time is usually desired as well as "lockTSE" logic. The resetCheckKey() method is less appropriate in such cases since the "time since expiry" cannot be inferred, causing any get() after the reset to treat the key as being "hot", resulting in more stale value usage.

Note that "check" keys won't collide with other regular keys.

See also
WANObjectCache::get()
WANObjectCache::getWithSetCallback()
WANObjectCache::resetCheckKey()
Parameters
string$keyCache key
int$holdoffHOLDOFF_TTL or HOLDOFF_NONE constant
Returns
bool True if the item was purged or not found, false on failure

Definition at line 620 of file WANObjectCache.php.

References relayPurge().

Referenced by WANObjectCacheTest\testGetMultiCheckKeys().

WANObjectCache::unwrap (   $wrapped,
  $now 
)
protected

Do not use this method outside WANObjectCache.

Parameters
array | string | bool$wrapped
float$nowUnix Current timestamp (preferrably pre-query)
Returns
array (mixed; false if absent/invalid, current time left)

Definition at line 1417 of file WANObjectCache.php.

References $flags.

Referenced by doGetWithSetCallback(), and getMulti().

WANObjectCache::worthRefreshExpiring (   $curTTL,
  $lowTTL 
)
protected

Check if a key should be regenerated (using random probability)

This returns false if $curTTL >= $lowTTL. Otherwise, the chance of returning true increases steadily from 0% to 100% as the $curTTL moves from $lowTTL to 0 seconds. This handles widely varying levels of cache access traffic.

Parameters
float$curTTLApproximate TTL left on the key if present
float$lowTTLConsider a refresh when $curTTL is less than this
Returns
bool

Definition at line 1327 of file WANObjectCache.php.

Referenced by doGetWithSetCallback().

WANObjectCache::worthRefreshPopular (   $asOf,
  $ageNew,
  $timeTillRefresh,
  $now 
)
protected

Check if a key is due for randomized regeneration due to its popularity.

This is used so that popular keys can preemptively refresh themselves for higher consistency (especially in the case of purge loss/delay). Unpopular keys can remain in cache with their high nominal TTL. This means popular keys keep good consistency, whether the data changes frequently or not, and long-tail keys get to stay in cache and get hits too. Similar to worthRefreshExpiring(), randomization is used.

Parameters
float$asOfUNIX timestamp of the value
integer$ageNewAge of key when this might recommend refreshing (seconds)
integer$timeTillRefreshAge of key when it should be refreshed if popular (seconds)
float$nowThe current UNIX timestamp
Returns
bool

Definition at line 1354 of file WANObjectCache.php.

Referenced by doGetWithSetCallback().

WANObjectCache::wrap (   $value,
  $ttl,
  $now 
)
protected

Do not use this method outside WANObjectCache.

Parameters
mixed$value
integer$ttl[0=forever]
float$nowUnix Current timestamp just before calling set()
Returns
array

Definition at line 1401 of file WANObjectCache.php.

References $value.

Referenced by doGetWithSetCallback(), and set().

Member Data Documentation

BagOStuff WANObjectCache::$cache
protected

The local datacenter cache.

Definition at line 78 of file WANObjectCache.php.

Referenced by set().

integer WANObjectCache::$callbackDepth = 0
private

Callback stack depth for getWithSetCallback()

Definition at line 92 of file WANObjectCache.php.

Referenced by doGetWithSetCallback().

int WANObjectCache::$lastRelayError = self::ERR_NONE
protected

ERR_* constant for the "last error" registry.

Definition at line 89 of file WANObjectCache.php.

Referenced by getLastError().

LoggerInterface WANObjectCache::$logger
protected

Definition at line 86 of file WANObjectCache.php.

Referenced by setLogger().

HashBagOStuff [] WANObjectCache::$processCaches = []
protected

Map of group PHP instance caches.

Definition at line 80 of file WANObjectCache.php.

string WANObjectCache::$purgeChannel
protected

Purge channel name.

Definition at line 82 of file WANObjectCache.php.

EventRelayer WANObjectCache::$purgeRelayer
protected

Bus that handles purge broadcasts.

Definition at line 84 of file WANObjectCache.php.

mixed [] WANObjectCache::$warmupCache = []
private

Temporary warm-up cache.

Definition at line 94 of file WANObjectCache.php.

const WANObjectCache::AGE_NEW = 60

Never consider performing "popularity" refreshes until a key reaches this age.

Definition at line 113 of file WANObjectCache.php.

const WANObjectCache::CHECK_KEY_TTL = self::TTL_YEAR

Seconds to keep dependency purge keys around.

Definition at line 104 of file WANObjectCache.php.

const WANObjectCache::DEFAULT_PURGE_CHANNEL = 'wancache-purge'

Definition at line 166 of file WANObjectCache.php.

const WANObjectCache::ERR_NO_RESPONSE = 1

Definition at line 149 of file WANObjectCache.php.

const WANObjectCache::ERR_NONE = 0

Definition at line 148 of file WANObjectCache.php.

const WANObjectCache::ERR_RELAY = 4

Definition at line 152 of file WANObjectCache.php.

const WANObjectCache::ERR_UNEXPECTED = 3

Definition at line 151 of file WANObjectCache.php.

const WANObjectCache::ERR_UNREACHABLE = 2

Definition at line 150 of file WANObjectCache.php.

const WANObjectCache::FLD_FLAGS = 4

Definition at line 142 of file WANObjectCache.php.

const WANObjectCache::FLD_HOLDOFF = 5

Definition at line 143 of file WANObjectCache.php.

const WANObjectCache::FLD_TIME = 3

Definition at line 141 of file WANObjectCache.php.

const WANObjectCache::FLD_TTL = 2

Definition at line 140 of file WANObjectCache.php.

const WANObjectCache::FLD_VALUE = 1

Definition at line 139 of file WANObjectCache.php.

const WANObjectCache::FLD_VERSION = 0

Definition at line 138 of file WANObjectCache.php.

const WANObjectCache::HIT_RATE_HIGH = 1

Hits/second for a refresh to be expected within the "popularity" window.

Definition at line 117 of file WANObjectCache.php.

const WANObjectCache::HOLDOFF_NONE = 0

Idiom for delete() for "no hold-off".

Definition at line 128 of file WANObjectCache.php.

Referenced by WANObjectCacheTest\testDelete(), and WANObjectCacheTest\testGetMultiCheckKeys().

const WANObjectCache::HOLDOFF_TTL = 11

Seconds to tombstone keys on delete()

Definition at line 101 of file WANObjectCache.php.

Referenced by MessageCache\getValidationHash().

const WANObjectCache::HOT_TTR = 900

The time length of the "popularity" refresh window for hot keys.

Definition at line 115 of file WANObjectCache.php.

const WANObjectCache::INTERIM_KEY_PREFIX = 'WANCache:i:'

Definition at line 155 of file WANObjectCache.php.

const WANObjectCache::LOCK_TSE = 1

Default time-since-expiry on a miss that makes a key "hot".

Definition at line 110 of file WANObjectCache.php.

const WANObjectCache::LOCK_TTL = 10

Seconds to keep lock keys around.

Definition at line 106 of file WANObjectCache.php.

const WANObjectCache::LOW_TTL = 30

Default remaining TTL at which to consider pre-emptive regeneration.

Definition at line 108 of file WANObjectCache.php.

const WANObjectCache::MAX_COMMIT_DELAY = 3

Max time expected to pass between delete() and DB commit finishing.

Definition at line 97 of file WANObjectCache.php.

const WANObjectCache::MAX_READ_LAG = 7

Max replication+snapshot lag before applying TTL_LAGGED or disallowing set()

Definition at line 99 of file WANObjectCache.php.

const WANObjectCache::MIN_TIMESTAMP_NONE = 0.0

Idiom for getWithSetCallback() for "no minimum required as-of timestamp".

Definition at line 130 of file WANObjectCache.php.

const WANObjectCache::MUTEX_KEY_PREFIX = 'WANCache:m:'

Definition at line 157 of file WANObjectCache.php.

const WANObjectCache::PC_PRIMARY = 'primary:1000'

Definition at line 164 of file WANObjectCache.php.

const WANObjectCache::PURGE_VAL_PREFIX = 'PURGED:'
const WANObjectCache::RAMPUP_TTL = 30

Seconds to ramp up to the "popularity" refresh chance after a key is no longer new.

Definition at line 119 of file WANObjectCache.php.

const WANObjectCache::TIME_KEY_PREFIX = 'WANCache:t:'

Definition at line 156 of file WANObjectCache.php.

Referenced by WANObjectCacheTest\testGetWithSeveralCheckKeys().

const WANObjectCache::TINY_NEGATIVE = -0.000001

Tiny negative float to use when CTL comes up >= 0 due to clock skew.

Definition at line 133 of file WANObjectCache.php.

const WANObjectCache::TSE_NONE = -1

Idiom for getWithSetCallback() callbacks to 'lockTSE' logic.

Definition at line 124 of file WANObjectCache.php.

const WANObjectCache::TTL_LAGGED = 30

Max TTL to store keys when a data sourced is lagged.

Definition at line 126 of file WANObjectCache.php.

const WANObjectCache::TTL_UNCACHEABLE = -1

Idiom for getWithSetCallback() callbacks to avoid calling set()

Definition at line 122 of file WANObjectCache.php.

Referenced by ForeignDBFile\getDescriptionText(), and File\getDescriptionText().

const WANObjectCache::VALUE_KEY_PREFIX = 'WANCache:v:'

Definition at line 154 of file WANObjectCache.php.

const WANObjectCache::VERSION = 1

Cache format version number.

Definition at line 136 of file WANObjectCache.php.

const WANObjectCache::VFLD_DATA = 'WOC:d'

Definition at line 161 of file WANObjectCache.php.

const WANObjectCache::VFLD_VERSION = 'WOC:v'

Definition at line 162 of file WANObjectCache.php.


The documentation for this class was generated from the following file: