MediaWiki master
Wikimedia\Rdbms\LBFactoryMulti Class Reference

LoadBalancer manager for sites with several "main" database clusters. More...

Inherits Wikimedia\Rdbms\LBFactory.

Collaboration diagram for Wikimedia\Rdbms\LBFactoryMulti:

Public Member Functions

 __construct (array $conf)
 Template override precedence (highest => lowest):
 
 getAllExternalLBs ()
 Get the tracked load balancer instances for all external clusters.
 
 getAllMainLBs ()
 Get the tracked load balancer instances for all main clusters.
 
 getExternalLB ( $cluster)
 Get the tracked load balancer instance for an external cluster.If no tracked instances exists, then one will be instantiated
Parameters
string$clusterExternal cluster name
Exceptions
InvalidArgumentExceptionIf $cluster is not recognized
Returns
ILoadBalancer

 
 getMainLB ( $domain=false)
 Get the tracked load balancer instance for the main cluster that handles the given domain.If no tracked instances exists, then one will be instantiated
Note
The local/default database domain used by the load balancer instance will still inherit from this ILBFactory instance, regardless of the $domain parameter.
Parameters
string | false$domainDomain ID, or false for the current domain
Returns
ILoadBalancer

 
 newExternalLB ( $cluster)
 Create a new load balancer instance for an external cluster.The resulting object will be untracked and the caller is responsible for cleaning it up. Database replication positions will not be saved by ChronologyProtector.This method is for only advanced usage and callers should almost always use getExternalLB() instead. This method can be useful when a table is used as a key/value store. In that cases, one might want to query it in autocommit mode (DBO_TRX off) but still use DBO_TRX transaction rounds on other tables.
Parameters
string$clusterExternal cluster name
Exceptions
InvalidArgumentExceptionIf $cluster is not recognized
Returns
ILoadBalancerForOwner

 
 newMainLB ( $domain=false)
 Create a new load balancer instance for the main cluster that handles the given domain.The resulting object is considered to be owned by the caller. Namely, it will be untracked, the caller is responsible for cleaning it up, and replication positions from it will not be saved by ChronologyProtector.This method is for only advanced usage and callers should almost always use getMainLB() instead. This method can be useful when a table is used as a key/value store. In that cases, one might want to query it in autocommit mode (DBO_TRX off) but still use DBO_TRX transaction rounds on other tables.
Note
The local/default database domain used by the load balancer instance will still inherit from this ILBFactory instance, regardless of the $domain parameter.
Parameters
string | false$domainDomain ID, or false for the current domain
Returns
ILoadBalancerForOwner

 
 reconfigure (array $conf)
 Reconfigure using the given config array.
 
- Public Member Functions inherited from Wikimedia\Rdbms\LBFactory
 autoReconfigure ()
 Reload config using the callback passed defined $config['configCallback'].
 
 beginPrimaryChanges ( $fname=__METHOD__)
 Wrap subsequent queries for all transaction round aware primary connections in a transaction.Each of these transactions will be owned by this ILBFactory instance such that direct calls to IDatabase::commit() or IDatabase::rollback() will be disabled. These transactions get resolved by a single call to either commitPrimaryChanges() or rollbackPrimaryChanges(), after which, the transaction wrapping and ownership behavior revert back to the default. When there are multiple connections involved, these methods perform best-effort distributed transactions. When using distributed transactions, the RDBMS should be configured to used pessimistic concurrency control such that the commit step of each transaction is unlikely to fail.Transactions on replication connections are flushed so that future reads will not keep using the same point-in-time view snapshots (e.g. from MySQL REPEATABLE-READ). However, this does not wait for replication to catch up, so subsequent reads from replicas might not reflect recently committed changes.This only applies to the tracked load balancer instances.This allows for custom transaction rounds from any outer transaction scope.
Parameters
string$fname
Exceptions
DBTransactionError
Since
1.37

 
 closeAll ( $fname=__METHOD__)
 Close all connections on instantiated tracked load balancer instances.
Parameters
string$fnameCaller name (e.g. METHOD)

 
 commitAndWaitForReplication ( $fname, $ticket, array $opts=[])
 Commit primary DB transactions and wait for replication (if $ticket indicates it is safe).This is mostly used in jobs or deferred updates dealing with batching.The ticket is used to check that the caller owns the transaction round or can act on behalf of the caller that owns the transaction round.
See also
ILBFactory::commitPrimaryChanges()
ILBFactory::waitForReplication()
Since
1.28
Parameters
string$fnameCaller name (e.g. METHOD)
mixed$ticketResult of getEmptyTransactionTicket()
array$optsOptions to waitForReplication()
Returns
bool True if the wait was successful, false on timeout

 
 commitPrimaryChanges ( $fname=__METHOD__, int $maxWriteDuration=0)
 Commit all primary connection transactions and flush all replica connection transactions.Transactions on replication connections are flushed so that future reads will not keep using the same point-in-time view snapshots (e.g. from MySQL REPEATABLE-READ). However, this does not wait for replication to catch up, so subsequent reads from replicas might not reflect the committed changes.This only applies to the instantiated tracked load balancer instances.
Parameters
string$fnameCaller name
int$maxWriteDurationabort if more than this much time was spent in write queries
Exceptions
DBTransactionError
Since
1.37

 
 destroy ()
 Close all connections and make further attempts to open connections result in DBAccessError.
 
 disableChronologyProtection ()
 Disable the ChronologyProtector on all instantiated tracked load balancer instances.
 
 flushPrimarySessions ( $fname=__METHOD__)
 Release important session-level state (named lock, table locks) as post-rollback cleanup.This only applies to the instantiated tracked load balancer instances.This should only be called by application entry point functions, since there must be no chance that a future caller will still be expecting some of the lost session state.
Parameters
string$fnameCaller name
Since
1.38

 
 flushReplicaSnapshots ( $fname=__METHOD__)
 Commit all replica database server transactions, clearing any point-in-time view snapshots.This only applies to the instantiated tracked load balancer instances.This is useful for getting rid of stale data from an implicit transaction round
Parameters
string$fnameCaller name
Deprecated
Since 1.43

 
 getAllLBs ()
 Get all tracked load balancer instances (generator)
Returns
Generator|ILoadBalancer[]
Since
1.39

 
 getAutoCommitPrimaryConnection ( $domain=false)
 Like IConnectionProvider::getPrimaryDatabase() but with AUTOCOMMIT mode.This is useful for whether the caller needs to use AUTOCOMMIT (no transaction wrapping) or it needs a new connection outside of the current transaction to bypass REPEATABLE READ isolation.This method accepts virtual domains ({

See also
\MediaWiki\MainConfigSchema::VirtualDomainsMapping}).
Since
1.44
Parameters
string | false$domainDomain ID, or false for the current domain
Returns
IDatabase

 
 getEmptyTransactionTicket ( $fname)
 Get a token asserting that no write transactions are active on tracked connections.This is mostly used in jobs or deferred updates dealing with batching.
Since
1.28
Parameters
string$fnameCaller name (e.g. METHOD)
Returns
mixed A value to pass to commitAndWaitForReplication()

 
 getLoadBalancer ( $domain=false)
 Get the tracked load balancer instance for a given domain.If no tracked instances exists, then one will be instantiated.This method accepts virtual domains ({

See also
\MediaWiki\MainConfigSchema::VirtualDomainsMapping}).
Since
1.43
Parameters
string | false$domainDomain ID, or false for the current domain
Returns
ILoadBalancer

 
 getLocalDomainID ()
 Get the local (and default) database domain ID of connection handles.
 
 getMappedDomain (string|false $domain)
 
 getPrimaryDatabase ( $domain=false)
 Get connection to the primary database.This should be used when there the code needs to write to the database.This method accepts virtual domains ({

See also
\MediaWiki\MainConfigSchema::VirtualDomainsMapping}).
Since
1.40
Parameters
string | false$domainDomain ID, or false for the current domain
Returns
IDatabase

 
 getReplicaDatabase (string|false $domain=false, $group=null)
 Get connection to a replica database.Note that a read can have replication lag.This method accepts virtual domains ({

See also
\MediaWiki\MainConfigSchema::VirtualDomainsMapping}).
Since
1.40
Parameters
string | false$domainDomain ID, or false for the current domain
string | null$groupQuery group; null for the default group
Returns
IReadableDatabase

 
 getTransactionProfiler ()
 Get the TransactionProfiler used by this instance.
Returns
TransactionProfiler
Since
1.35

 
 hasOrMadeRecentPrimaryChanges ( $age=null)
 Determine if any primary connection has pending/written changes from this request.This only applies to the instantiated tracked load balancer instances.
Parameters
float | null$ageHow many seconds ago is "recent" [defaults to LB lag wait timeout]
Returns
bool

 
 hasPrimaryChanges ()
 Determine if any primary connection has pending changes.This only applies to the instantiated tracked load balancer instances.
Returns
bool
Since
1.37

 
 hasStreamingReplicaServers ()
 Whether it has streaming replica servers.
Since
1.41
Returns
bool

 
 hasTransactionRound ()
 Check if an explicit transaction round is active.
Returns
bool
Since
1.29

 
 isLocalDomain ( $domain)
 Determine whether, after mapping, the domain refers to the main domain of the local wiki.
 
 isReadyForRoundOperations ()
 Check if transaction rounds can be started, committed, or rolled back right now.This can be used as a recursion guard to avoid exceptions in transaction callbacks.
Returns
bool
Since
1.32

 
 isSharedVirtualDomain ( $domain)
 Is the domain a virtual domain with a statically configured database name?
 
 laggedReplicaUsed ()
 Determine if any lagged replica database server connection was used.This only applies to the instantiated tracked load balancer instances.
Returns
bool

 
 redefineLocalDomain ( $domain)
 Close all connections and redefine the local database domain.This only applies to the tracked load balancer instances.This method is only intended for use with schema creation or integration testing
Parameters
DatabaseDomain | string$domain
Since
1.33

 
 rollbackPrimaryChanges ( $fname=__METHOD__)
 Rollback all primary connection transactions and flush all replica connection transactions.This only applies to the instantiated tracked load balancer instances.
Parameters
string$fnameCaller name
Since
1.37

 
 setAgentName ( $agent)
 
Parameters
string$agentAgent name for query profiling

 
 setDefaultGroupName (string $defaultGroup)
 Set the default group name for all load balancers.
 
 setDefaultReplicationWaitTimeout ( $seconds)
 Set the default timeout for replication wait checks.
Parameters
int$secondsTimeout, in seconds
Returns
int The previous default timeout
Since
1.35

 
 setDomainAliases (array $aliases)
 Convert certain database domains to alternative ones.This can be used for backwards compatibility logic.
Parameters
DatabaseDomain[] | string[]$aliasesMap of (domain alias => domain)
Since
1.35

 
 setLocalDomainPrefix ( $prefix)
 Set a new table prefix for the existing local domain ID for testing.
Parameters
string$prefix
Since
1.33

 
 setTableAliases (array $aliases)
 Make certain table names use their own database, schema, and table prefix when passed into SQL queries pre-escaped and without a qualified database name.For example, "user" can be converted to "myschema.mydbname.user" for convenience. Appearances like user, somedb.user, somedb.someschema.user will used literally.Calling this twice will completely clear any old table aliases. Also, note that callers are responsible for making sure the schemas and databases actually exist.
Parameters
array[]$aliasesMap of (table => (dbname, schema, prefix) map)
Since
1.31

 
 setWaitForReplicationListener ( $name, ?callable $callback=null)
 Add a callback to be run in every call to waitForReplication() prior to any waiting.Callbacks must clear any transactions that they start.
Parameters
string$nameCallback name
callable | null$callbackUse null to unset a callback
Deprecated
Since 1.44

 
 shutdown ( $flags=self::SHUTDOWN_NORMAL, ?callable $workCallback=null, &$cpIndex=null, &$cpClientId=null)
 Prepare all instantiated tracked load balancer instances for shutdown.
Parameters
int$flagsBit field of ILBFactory::SHUTDOWN_* constants
callable | null$workCallbackWork to mask ChronologyProtector writes
int | null&$cpIndexPosition key write counter for ChronologyProtector [returned]
string | null&$cpClientIdClient ID hash for ChronologyProtector [returned]

 
 waitForReplication (array $opts=[])
 Waits for the replica database server to catch up to the current primary position.Use this when updating very large numbers of rows, as in maintenance scripts, to avoid causing too much lag. This is a no-op if there are no replica database servers.By default this waits on all DB clusters actually used in this request. This makes sense when lag being waiting on is caused by the code that does this check. In that case, setting "ifWritesSince" can avoid the overhead of waiting for clusters that were not changed since the last wait check.Never call this function after a large DB write that is still in a transaction. It only makes sense to call this after the possible lag inducing changes were committed.This only applies to the instantiated tracked load balancer instances.
Parameters
array$optsOptional fields that include:
  • timeout: Max wait time. Default: 60 seconds for CLI, 1 second for web.
  • ifWritesSince: Only wait if writes were done since this UNIX timestamp.
Returns
bool True on success, false if a timeout or error occurred while waiting

 

Protected Member Functions

 getLBsForOwner ()
 Get all tracked load balancers with the internal "for owner" interface.
Returns
Generator|ILoadBalancerForOwner[]

 
- Protected Member Functions inherited from Wikimedia\Rdbms\LBFactory
 baseLoadBalancerParams ()
 Get parameters to ILoadBalancer::__construct()
 
 configure (array $conf)
 
 initLoadBalancer (ILoadBalancerForOwner $lb)
 

Additional Inherited Members

- Public Attributes inherited from Wikimedia\Rdbms\ILBFactory
const CLUSTER_MAIN_DEFAULT = 'DEFAULT'
 Default main cluster name (do not change this)
 
const SHUTDOWN_NO_CHRONPROT = 1
 Do not save "session consistency" DB replication positions.
 
const SHUTDOWN_NORMAL = 0
 Idiom for "no special shutdown flags".
 
- Protected Attributes inherited from Wikimedia\Rdbms\LBFactory
ChronologyProtector $chronologyProtector
 
DatabaseDomain[] string[] $domainAliases = []
 Map of (domain alias => DB domain)
 
DatabaseDomain $localDomain
 Local domain.
 
string false $readOnlyReason = false
 Reason all LBs are read-only or false if not.
 
BagOStuff $srvCache
 
array string[] $virtualDomains = []
 List of registered virtual domains.
 
array array[] $virtualDomainsMapping = []
 Map of virtual domain to array of cluster and domain.
 
WANObjectCache $wanCache
 

Detailed Description

LoadBalancer manager for sites with several "main" database clusters.

Each database cluster consists of a "primary" server and any number of replica servers, all of which converge, as soon as possible, to contain the same schemas and records. If a replication topology has multiple primaries, then the "primary" is merely the preferred co-primary for the current context (e.g. datacenter).

For single-primary topologies, the schemas and records of the primary define the "dataset". For multiple-primary topologies, the "dataset" is the convergent result of applying/merging all committed events (regardless of the co-primary they originated on); it possible that no co-primary has yet converged upon this state at any given time (especially when there are frequent writes and co-primaries are geographically distant).

A "main" cluster contain a "main" dataset, which consists of data that is compact, highly relational (e.g. read by JOIN queries), and essential to one or more sites. The "external" clusters each store an "external" dataset, which consists of data that is non-relational (e.g. key/value pairs), self-contained (e.g. JOIN queries and transactions thereof never involve a main dataset), or too bulky to reside in a main dataset (e.g. text blobs).

The class allows for large site farms to split up their data in the following ways:

  • Vertically shard compact site-specific data by site (e.g. page/comment metadata)
  • Vertically shard compact global data by module (e.g. account/notification data)
  • Horizontally shard any bulk data by blob key (e.g. page/comment content blobs)

Definition at line 39 of file LBFactoryMulti.php.

Constructor & Destructor Documentation

◆ __construct()

Wikimedia\Rdbms\LBFactoryMulti::__construct ( array $conf)

Template override precedence (highest => lowest):

  • templateOverridesByServer
  • masterTemplateOverrides
  • templateOverridesBySection/templateOverridesByCluster
  • externalTemplateOverrides
  • serverTemplate Overrides only work on top level keys (so nested values will not be merged).

Server config maps should be of the format Database::factory() requires. Additionally, a 'max lag' key should also be set on server maps, indicating how stale the data can be before the load balancer tries to avoid using it. The map can have 'is static' set to disable blocking replication sync checks (intended for archive servers with unchanging data).

See also
LBFactory::__construct()
Parameters
array$confAdditional parameters include:
  • hostsByName: map of (server name => IP address). [optional]
  • sectionsByDB: map of (database => main section). The database name "DEFAULT" is interpreted as a catch-all for all databases not otherwise mentioned. If no section name is specified for "DEFAULT", then the catch-all section is assumed to be named "DEFAULT". [optional]
  • sectionLoads: map of (main section => server name => load ratio); the first host listed in each section is the primary DB server for that section. [optional]
  • externalLoads: map of (cluster => server name => load ratio) map. [optional]
  • serverTemplate: server config map for Database::factory(). Note that "host", "serverName" and "load" entries will be overridden by "hostsByName". [optional]
  • externalTemplateOverrides: server config map overrides for external stores; respects the override precedence described above. [optional]
  • templateOverridesBySection: map of (main section => server config map overrides); respects the override precedence described above. [optional]
  • templateOverridesByCluster: map of (external cluster => server config map overrides); respects the override precedence described above. [optional]
  • masterTemplateOverrides: server config map overrides for masters; respects the override precedence described above. [optional]
  • templateOverridesByServer: map of (server name => server config map overrides); respects the override precedence described above and applies to both core and external storage. [optional]
  • loadMonitor: LoadMonitor::__construct() parameters with "class" field. [optional]
  • readOnlyBySection: map of (main section => message text or false). String values make sections read only, whereas anything else does not restrict read/write mode. [optional]
  • configCallback: A callback that returns a conf array that can be passed to the reconfigure() method. This will be used to autoReconfigure() to load any updated configuration.

Reimplemented from Wikimedia\Rdbms\LBFactory.

Definition at line 118 of file LBFactoryMulti.php.

References Wikimedia\Rdbms\ILBFactory\CLUSTER_MAIN_DEFAULT.

Member Function Documentation

◆ getAllExternalLBs()

Wikimedia\Rdbms\LBFactoryMulti::getAllExternalLBs ( )

Get the tracked load balancer instances for all external clusters.

If no tracked instance exists for a cluster, then one will be instantiated

Returns
ILoadBalancer[] Map of (cluster name => ILoadBalancer)
Since
1.29

Implements Wikimedia\Rdbms\ILBFactory.

Definition at line 253 of file LBFactoryMulti.php.

◆ getAllMainLBs()

Wikimedia\Rdbms\LBFactoryMulti::getAllMainLBs ( )

Get the tracked load balancer instances for all main clusters.

If no tracked instance exists for a cluster, then one will be instantiated

Note that default main cluster name is ILoadBalancer::CLUSTER_MAIN_DEFAULT

Returns
ILoadBalancer[] Map of (cluster name => ILoadBalancer)
Since
1.29

Implements Wikimedia\Rdbms\ILBFactory.

Definition at line 242 of file LBFactoryMulti.php.

◆ getExternalLB()

Wikimedia\Rdbms\LBFactoryMulti::getExternalLB ( $cluster)

Get the tracked load balancer instance for an external cluster.If no tracked instances exists, then one will be instantiated

Parameters
string$clusterExternal cluster name
Exceptions
InvalidArgumentExceptionIf $cluster is not recognized
Returns
ILoadBalancer

Implements Wikimedia\Rdbms\ILBFactory.

Definition at line 232 of file LBFactoryMulti.php.

◆ getLBsForOwner()

Wikimedia\Rdbms\LBFactoryMulti::getLBsForOwner ( )
protected

Get all tracked load balancers with the internal "for owner" interface.

Returns
Generator|ILoadBalancerForOwner[]

Reimplemented from Wikimedia\Rdbms\LBFactory.

Definition at line 263 of file LBFactoryMulti.php.

◆ getMainLB()

Wikimedia\Rdbms\LBFactoryMulti::getMainLB ( $domain = false)

Get the tracked load balancer instance for the main cluster that handles the given domain.If no tracked instances exists, then one will be instantiated

Note
The local/default database domain used by the load balancer instance will still inherit from this ILBFactory instance, regardless of the $domain parameter.
Parameters
string | false$domainDomain ID, or false for the current domain
Returns
ILoadBalancer

Implements Wikimedia\Rdbms\ILBFactory.

Definition at line 203 of file LBFactoryMulti.php.

◆ newExternalLB()

Wikimedia\Rdbms\LBFactoryMulti::newExternalLB ( $cluster)

Create a new load balancer instance for an external cluster.The resulting object will be untracked and the caller is responsible for cleaning it up. Database replication positions will not be saved by ChronologyProtector.This method is for only advanced usage and callers should almost always use getExternalLB() instead. This method can be useful when a table is used as a key/value store. In that cases, one might want to query it in autocommit mode (DBO_TRX off) but still use DBO_TRX transaction rounds on other tables.

Parameters
string$clusterExternal cluster name
Exceptions
InvalidArgumentExceptionIf $cluster is not recognized
Returns
ILoadBalancerForOwner

Implements Wikimedia\Rdbms\ILBFactory.

Definition at line 215 of file LBFactoryMulti.php.

◆ newMainLB()

Wikimedia\Rdbms\LBFactoryMulti::newMainLB ( $domain = false)

Create a new load balancer instance for the main cluster that handles the given domain.The resulting object is considered to be owned by the caller. Namely, it will be untracked, the caller is responsible for cleaning it up, and replication positions from it will not be saved by ChronologyProtector.This method is for only advanced usage and callers should almost always use getMainLB() instead. This method can be useful when a table is used as a key/value store. In that cases, one might want to query it in autocommit mode (DBO_TRX off) but still use DBO_TRX transaction rounds on other tables.

Note
The local/default database domain used by the load balancer instance will still inherit from this ILBFactory instance, regardless of the $domain parameter.
Parameters
string | false$domainDomain ID, or false for the current domain
Returns
ILoadBalancerForOwner

Implements Wikimedia\Rdbms\ILBFactory.

Definition at line 152 of file LBFactoryMulti.php.

◆ reconfigure()

Wikimedia\Rdbms\LBFactoryMulti::reconfigure ( array $conf)

Reconfigure using the given config array.

Any fields omitted from $conf will be taken from the current config.

If the config changed, this calls reconfigure() on all load balancers, which causes them to close all existing connections.

Note
This invalidates the current transaction ticket.
Warning
This must only be called in top level code such as the execute() method of a maintenance script. Any database connection in use when this method is called will become defunct.
Since
1.39
Parameters
array$confA configuration array, using the same structure as the one passed to the constructor (see also $wgLBFactoryConf).

Reimplemented from Wikimedia\Rdbms\LBFactory.

Definition at line 337 of file LBFactoryMulti.php.


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