gplx/mediawiki/includes/libs/rdbms/loadbalancer/ILoadBalancer.php

501 lines
22 KiB
PHP

<?php
/**
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
* http://www.gnu.org/copyleft/gpl.html
*
* @file
*/
namespace Wikimedia\Rdbms;
/**
* This class is a delegate to ILBFactory for a given database cluster
*
* ILoadBalancer tracks the database connections and transactions for a given database cluster.
* A "cluster" is considered to be the set of database servers that manage a given dataset.
* Within a given cluster, each database server can have one of the following roles:
* - sole-primary: the server used by this datacenter for writes from the application
* - co-primary: one of several servers used by this datacenter for writes from the application,
* relying on asynchronous replication to synchronize their copies of the dataset
* - replica: a server used only for reads from the application, relying on asynchronous
* replication to apply writes from the primary server or co-primary servers
* - static clone: a server that only accepts reads from the application, does not replicate,
* and has a copy of the final dataset, which must be static (all the servers reject writes)
*
* Single-datacenter database clusters consist of either:
* - A sole-primary server and zero or more replica servers
* - A set of static clone servers
*
* Multi-datacenter database clusters either consist of either:
* - A sole-primary server and zero or more replica servers in the "primary datacenter"
* (the one datacenter meant to handle requests/jobs that mutate the database), and zero or
* more replica servers in each "secondary datacenter" (all the other datacenters)
* - A co-primary server and zero or more replica servers within each datacenter
* - A set of static clone servers in each datacenter
*
* The term "primary" refers to the server used by this datacenter for handling writes,
* whether it is a sole-primary or co-primary.
*
* The "servers" configuration array contains the list of database servers to use for operations
* originating from the the local datacenter. The first entry must refer to the server to use for
* write and read-for-write operations (e.g. the "writer server"):
* - If there is a primary server, then the first entry must refer to it, even if the primary
* server resides in a remote datacenter
* - If there are co-primary servers, then the first entry must refer to the one in the local
* datacenter
* - If the servers are static clones, then the first entry can refer to any of them, since the
* concept of a "writer server" is merely nominal
*
* On an infrastructure level, circular replication setups can have more than one database server
* act as a replication "source" within the same datacenter, provided that no more than one of the
* servers are writable at any time, namely the "writer server". The other source servers will be
* treated as replicas by the load balancer, but can be quickly promoted to the "writer server" by
* the site admin as needed.
*
* Likewise, Galera Cluster setups still require the choice of a single "writer server" for each
* datacenter. Limiting the number of servers that initiate transactions helps reduce the rate of
* aborted transactions due to wsrep conflicts.
*
* By default, each DB server uses DBO_DEFAULT for its 'flags' setting, unless explicitly set
* otherwise in configuration. DBO_DEFAULT behavior depends on whether 'cliMode' is set:
* - In CLI mode, the flag has no effect with regards to LoadBalancer.
* - In non-CLI mode, the flag causes implicit transactions to be used; the first query on
* a database starts a transaction on that database. The transactions are meant to remain
* pending until either commitPrimaryChanges() or rollbackPrimaryChanges() is called. The
* application must have some point where it calls commitPrimaryChanges() near the end of
* the PHP request.
* Every iteration of beginPrimaryChanges()/commitPrimaryChanges() is called a "transaction round".
* Rounds are useful on the primary DB connections because they make single-DB (and by and large
* multi-DB) updates in web requests all-or-nothing. Also, transactions on replica DBs are useful
* when REPEATABLE-READ or SERIALIZABLE isolation is used because all foreign keys and constraints
* hold across separate queries in the DB transaction since the data appears within a consistent
* point-in-time snapshot.
*
* The typical caller will use LoadBalancer::getConnection( DB_* ) to yield a database
* connection handle. The choice of which DB server to use is based on pre-defined loads for
* weighted random selection, adjustments thereof by LoadMonitor, and the amount of replication
* lag on each DB server. Lag checks might cause problems in certain setups, so they should be
* tuned in the server configuration maps as follows:
* - Sole-primary + N Replica(s): set 'max lag' to an appropriate threshold for avoiding any
* replica database lagged by this much or more. If all replicas are this lagged, then the
* load balancer considers the cluster to be read-only.
* - Per-datacenter co-primary + N Replica(s): set 'max lag' to an appropriate threshold for
* avoiding any replica database lagged by this much or more. If all replicas are this
* lagged, then the load balancer considers the cluster to be read-only.
* - Read-only archive clones: set 'is static' in the server configuration maps. This will
* treat all such DBs as having 0 lag.
* - Externally updated dataset clones: set 'is static' in the server configuration maps.
* This will treat all such DBs as having 0 lag.
* - SQL load balancing proxy: any proxy should handle lag checks on its own, so the 'max lag'
* parameter should probably be set to INF in the server configuration maps. This will make
* the load balancer ignore whatever it detects as the lag of the logical replica is (which
* would probably just randomly bounce around).
*
* If using a SQL proxy service, it would probably be best to have two proxy hosts for the load
* balancer to talk to. One would be the 'host' of the "writer server" entry and another for the
* (logical) replica server entry. The proxy could map the load balancer's "replica" DB to any
* number of physical replica DBs.
*
* @since 1.28
* @ingroup Database
*/
interface ILoadBalancer {
/**
* Request a replica DB connection. Can't be used as a binary flag with bitwise operators!
*/
public const DB_REPLICA = -1;
/**
* Request a primary, write-enabled DB connection. Can't be used as a binary flag with bitwise
* operators!
* @since 1.36
*/
public const DB_PRIMARY = -2;
/** Domain specifier when no specific database needs to be selected */
public const DOMAIN_ANY = '';
/** The generic query group */
public const GROUP_GENERIC = '';
/** Yield a tracked autocommit-mode handle (reuse existing ones) */
public const CONN_TRX_AUTOCOMMIT = 1;
/**
* Yield an untracked, low-timeout, autocommit-mode handle (to gauge server health)
* @internal
*/
public const CONN_UNTRACKED_GAUGE = 2;
/**
* Yield null on connection failure instead of throwing an exception
* @internal
*/
public const CONN_SILENCE_ERRORS = 4;
/**
* Get the name of the overall cluster of database servers managing the dataset
*
* Note that the cluster might contain servers in multiple datacenters.
* The load balancer instance only needs to be aware of the local replica servers,
* along with either the sole-primary server or the local co-primary server.
*
* This is useful for identifying a cluster or replicated dataset, even when:
* - The primary server is sometimes swapped with another one
* - The cluster/dataset is replicated among multiple datacenters, with one "primary"
* datacenter having the writable primary server and the other datacenters having a
* read-only replica in the "primary" server slot
* - The dataset is replicated among multiple datacenters, via circular replication,
* with each datacenter having its own "co-primary" server
*
* @return string
* @since 1.36
*/
public function getClusterName(): string;
/**
* Get the local (and default) database domain ID of connection handles
*
* @see DatabaseDomain
* @return string Database domain ID; this specifies DB name, schema, and table prefix
* @since 1.31
*/
public function getLocalDomainID(): string;
/**
* @param DatabaseDomain|string|false $domain Database domain
* @return string Value of $domain if it is foreign or the local domain otherwise
* @since 1.32
*/
public function resolveDomainID( $domain ): string;
/**
* Indicate whether the tables on this domain are only temporary tables for testing
*
* In "temporary tables mode", the CONN_TRX_AUTOCOMMIT flag is ignored
*
* @param bool $value
* @param string $domain
* @return bool Whether "temporary tables mode" was active
* @since 1.34
*/
public function setTempTablesOnlyMode( $value, $domain );
/**
* Get the specific server index of the reader connection for a given group
*
* This takes into account load ratios and lag times. It should return a consistent
* index during the life time of the load balancer. This initially checks replica DBs
* for connectivity to avoid returning an unusable server. This means that connections
* might be attempted by calling this method (usually one at the most but possibly more).
* Subsequent calls with the same $group will not need to make new connection attempts
* since the acquired connection for each group is preserved.
*
* @param string|false $group Query group or false for the generic group
* @return int|false Specific server index, or false if no DB handle can be obtained
*/
public function getReaderIndex( $group = false );
/**
* Get a lazy-connecting database handle for a specific or virtual (DB_PRIMARY/DB_REPLICA) server index
*
* The server index, $i, can be one of the following:
* - DB_REPLICA: a server index will be selected by the load balancer based on read
* weight, connectivity, and replication lag. Note that the primary server might be
* configured with read weight. If $groups is empty then it means "the generic group",
* in which case all servers defined with read weight will be considered. Additional
* query groups can be configured, having their own list of server indexes and read
* weights. If a query group list is provided in $groups, then each recognized group
* will be tried, left-to-right, until server index selection succeeds or all groups
* have been tried, in which case the generic group will be tried.
* - DB_PRIMARY: the primary server index will be used; the same as ServerInfo::WRITER_INDEX.
* The value of $groups should be [] when using this server index.
* - Specific server index: a positive integer can be provided to use the server with
* that index. An error will be thrown in no such server index is recognized. This
* server selection method is usually only useful for internal load balancing logic.
* The value of $groups should be [] when using a specific server index.
*
* Handle sharing is very useful when callers get DB_PRIMARY handles that are transaction
* round aware (the default). All such callers will operate within a single transaction as
* a consequence. The same applies to DB_REPLICA that are samely query grouped (the default)
* and transaction round aware (the default).
*
* Use CONN_TRX_AUTOCOMMIT to use a separate pool of only autocommit handles. This flag is
* ignored for databases with ATTR_DB_LEVEL_LOCKING (e.g. sqlite) in order to avoid deadlocks.
* getServerAttributes() can be used to check this attribute beforehand. Avoid using begin()
* and commit() on such handles. If handle methods like startAtomic() and endAtomic() must be
* used on the handles, callers should at least make sure that the atomic sections are closed
* on failure via try/catch and cancelAtomic().
*
* Use CONN_UNTRACKED_GAUGE to get a new, untracked, handle, that uses a low connection timeout, a low
* read timeout, and autocommit mode. This flag is intended for use only be internal callers.
*
* CONN_UNTRACKED_GAUGE and CONN_TRX_AUTOCOMMIT are incompatible.
*
* @see ILoadBalancer::getServerAttributes()
*
* @param int $i Specific (overrides $groups) or virtual (DB_PRIMARY/DB_REPLICA) server index
* @param string[]|string $groups Query group(s) in preference order; [] for the default group
* @param string|false $domain DB domain ID or false for the local domain
* @param int $flags Bitfield of CONN_* class constants
* @return IDatabase|false This returns false on failure if CONN_SILENCE_ERRORS is set
*/
public function getConnection( $i, $groups = [], $domain = false, $flags = 0 );
/**
* Get a DB handle for a specific server index
*
* This is an internal utility method for methods like LoadBalancer::getConnectionInternal()
* and DBConnRef to create the underlying connection to a concrete server.
*
* The following is the responsibility of the caller:
*
* - translate any virtual server indexes (DB_PRIMARY/DB_REPLICA) to a real server index.
* - enforce read-only mode on primary DB handle if there is high replication lag.
*
* @see ILoadBalancer::getConnection()
*
* @internal Only for use within ILoadBalancer/ILoadMonitor
* @param int $i Specific server index
* @param string $domain Resolved DB domain
* @param int $flags Bitfield of class CONN_* constants
* @return IDatabaseForOwner|false This returns false on failure if CONN_SILENCE_ERRORS is set
* @throws DBError If no DB handle could be obtained and CONN_SILENCE_ERRORS is not set
*/
public function getServerConnection( $i, $domain, $flags = 0 );
/**
* @deprecated since 1.39, use ILoadBalancer::getConnection() instead.
* @param int $i Specific or virtual (DB_PRIMARY/DB_REPLICA) server index
* @param string[]|string $groups Query group(s) in preference order; [] for the default group
* @param string|false $domain DB domain ID or false for the local domain
* @param int $flags Bitfield of CONN_* class constants (e.g. CONN_TRX_AUTOCOMMIT)
* @return DBConnRef
*/
public function getConnectionRef( $i, $groups = [], $domain = false, $flags = 0 ): DBConnRef;
/**
* @internal Only to be used by DBConnRef
* @param int $i Specific (overrides $groups) or virtual (DB_PRIMARY/DB_REPLICA) server index
* @param string[]|string $groups Query group(s) in preference order; [] for the default group
* @param string|false $domain DB domain ID or false for the local domain
* @param int $flags Bitfield of CONN_* class constants (e.g. CONN_TRX_AUTOCOMMIT)
* @return IDatabase
*/
public function getConnectionInternal( $i, $groups = [], $domain = false, $flags = 0 ): IDatabase;
/**
* Get a DB handle, suitable for migrations and schema changes, for a server index
*
* The DBConnRef methods simply proxy an underlying IDatabase object which
* takes care of the actual connection and query logic.
*
* The CONN_TRX_AUTOCOMMIT flag is ignored for databases with ATTR_DB_LEVEL_LOCKING
* (e.g. sqlite) in order to avoid deadlocks. getServerAttributes()
* can be used to check such flags beforehand. Avoid the use of begin() or startAtomic()
* on any CONN_TRX_AUTOCOMMIT connections.
*
* @see ILoadBalancer::getConnection() for parameter information
* @param int $i Specific or virtual (DB_PRIMARY/DB_REPLICA) server index
* @param string[]|string $groups Query group(s) in preference order; [] for the default group
* @param string|false $domain DB domain ID or false for the local domain
* @param int $flags Bitfield of CONN_* class constants (e.g. CONN_TRX_AUTOCOMMIT)
* @return DBConnRef
*/
public function getMaintenanceConnectionRef( $i, $groups = [], $domain = false, $flags = 0 ): DBConnRef;
/**
* Get the number of servers defined in configuration
*
* @return int
*/
public function getServerCount();
/**
* Whether there are any replica servers configured
*
* This scans the list of servers defined in configuration, checking for:
* - Servers that are listed after the primary and not flagged with "is static";
* such servers are assumed to be typical streaming replicas
* - Servers that are listed after the primary and flagged with "is static";
* such servers are assumed to have a clone of the static dataset (matching the primary)
*
* @return bool
* @since 1.34
*/
public function hasReplicaServers();
/**
* Whether any replica servers use streaming replication from the primary server
*
* This scans the list of servers defined in configuration, checking for:
* - Servers that are listed after the primary and not flagged with "is static";
* such servers are assumed to be typical streaming replicas
*
* It is possible for some replicas to be configured with "is static" but not
* others, though it generally should either be set for all or none of the replicas.
*
* If this returns false, this means that there is generally no reason to execute
* replication wait logic for session consistency and lag reduction.
*
* @return bool
* @since 1.34
*/
public function hasStreamingReplicaServers();
/**
* Get the readable name of the server with the specified index
*
* @param int $i Specific server index
* @return string Readable server name, falling back to the hostname or IP address
*/
public function getServerName( $i ): string;
/**
* Return the server configuration map for the server with the specified index
*
* @param int $i Specific server index
* @return array|false Server configuration map; false if the index is invalid
* @since 1.31
*/
public function getServerInfo( $i );
/**
* Get the RDBMS type of the server with the specified index (e.g. "mysql", "sqlite")
*
* @param int $i Specific server index
* @return string One of (mysql,postgres,sqlite,...) or "unknown" for bad indexes
* @since 1.30
*/
public function getServerType( $i );
/**
* Get basic attributes of the server with the specified index without connecting
*
* @param int $i Specific server index
* @return array (Database::ATTRIBUTE_* constant => value) for all such constants
* @since 1.31
*/
public function getServerAttributes( $i );
/**
* Get the current primary replication position
*
* @return DBPrimaryPos|false Returns false if not applicable
* @throws DBError
* @since 1.37
*/
public function getPrimaryPos();
/**
* Whether there are pending changes or callbacks in a transaction by this thread
* @return bool
* @since 1.37
*/
public function hasPrimaryChanges();
/**
* Determine whether an explicit transaction is active on any open primary
* connection.
* @return bool
* @since 1.39
*/
public function explicitTrxActive();
/**
* Check if this load balancer object had any recent or still
* pending writes issued against it by this PHP thread
*
* @param float|null $age How many seconds ago is "recent" [defaults to mWaitTimeout]
* @return bool
* @since 1.37
*/
public function hasOrMadeRecentPrimaryChanges( $age = null );
/**
* @note This method may trigger a DB connection if not yet done
* @return string|false Reason the primary is read-only or false if it is not
*/
public function getReadOnlyReason();
/**
* @return bool
*/
public function pingAll();
/**
* Get the name and lag time of the most-lagged replica server
*
* This is useful for maintenance scripts that need to throttle their updates.
* May attempt to open connections to replica DBs on the default DB. If there is
* no lag, the maximum lag will be reported as -1.
*
* @return array{0:string,1:float|int|false,2:int} (host, max lag, index of max lagged host)
*/
public function getMaxLag();
/**
* Get an estimate of replication lag (in seconds) for each server
*
* Results are cached for a short time in memcached/process cache
*
* Values may be "false" if replication is too broken to estimate
*
* @return float[]|int[]|false[] Map of (server index => lag) in order of server index
*/
public function getLagTimes();
/**
* Wait for a replica DB to reach a specified primary position
*
* If $conn is not a replica server connection, then this will return true.
* Otherwise, if $pos is not provided, this will connect to the primary server
* to get an accurate position.
*
* @param IDatabase $conn Replica DB
* @return bool Success
* @since 1.37
*/
public function waitForPrimaryPos( IDatabase $conn );
/**
* Set a callback via IDatabase::setTransactionListener() on
* all current and future primary connections of this load balancer
*
* @param string $name Callback name
* @param callable|null $callback
*/
public function setTransactionListener( $name, ?callable $callback = null );
/**
* 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.
*
* @param array[] $aliases Map of (table => (dbname, schema, prefix) map)
*/
public function setTableAliases( array $aliases );
/**
* Convert certain database domains to alternative ones.
*
* This can be used for backwards compatibility logic.
*
* @param DatabaseDomain[]|string[] $aliases Map of (domain alias => domain)
* @since 1.35
*/
public function setDomainAliases( array $aliases );
}