3 * Database load balancing interface
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
18 * http://www.gnu.org/copyleft/gpl.html
23 namespace Wikimedia\Rdbms
;
26 use InvalidArgumentException
;
29 * Database cluster connection, tracking, load balancing, and transaction manager interface
31 * A "cluster" is considered to be one master database and zero or more replica databases.
32 * Typically, the replica DBs replicate from the master asynchronously. The first node in the
33 * "servers" configuration array is always considered the "master". However, this class can still
34 * be used when all or some of the "replica" DBs are multi-master peers of the master or even
35 * when all the DBs are non-replicating clones of each other holding read-only data. Thus, the
36 * role of "master" is in some cases merely nominal.
38 * By default, each DB server uses DBO_DEFAULT for its 'flags' setting, unless explicitly set
39 * otherwise in configuration. DBO_DEFAULT behavior depends on whether 'cliMode' is set:
40 * - In CLI mode, the flag has no effect with regards to LoadBalancer.
41 * - In non-CLI mode, the flag causes implicit transactions to be used; the first query on
42 * a database starts a transaction on that database. The transactions are meant to remain
43 * pending until either commitMasterChanges() or rollbackMasterChanges() is called. The
44 * application must have some point where it calls commitMasterChanges() near the end of
46 * Every iteration of beginMasterChanges()/commitMasterChanges() is called a "transaction round".
47 * Rounds are useful on the master DB connections because they make single-DB (and by and large
48 * multi-DB) updates in web requests all-or-nothing. Also, transactions on replica DBs are useful
49 * when REPEATABLE-READ or SERIALIZABLE isolation is used because all foriegn keys and constraints
50 * hold across separate queries in the DB transaction since the data appears within a consistent
51 * point-in-time snapshot.
53 * The typical caller will use LoadBalancer::getConnection( DB_* ) to yield a live database
54 * connection handle. The choice of which DB server to use is based on pre-defined loads for
55 * weighted random selection, adjustments thereof by LoadMonitor, and the amount of replication
56 * lag on each DB server. Lag checks might cause problems in certain setups, so they should be
57 * tuned in the server configuration maps as follows:
58 * - Master + N Replica(s): set 'max lag' to an appropriate threshold for avoiding any database
59 * lagged by this much or more. If all DBs are this lagged, then the load balancer considers
60 * the cluster to be read-only.
61 * - Galera Cluster: Seconds_Behind_Master will be 0, so there probably is nothing to tune.
62 * Note that lag is still possible depending on how wsrep-sync-wait is set server-side.
63 * - Read-only archive clones: set 'is static' in the server configuration maps. This will
64 * treat all such DBs as having 0 lag.
65 * - SQL load balancing proxy: any proxy should handle lag checks on its own, so the 'max lag'
66 * parameter should probably be set to INF in the server configuration maps. This will make
67 * the load balancer ignore whatever it detects as the lag of the logical replica is (which
68 * would probably just randomly bounce around).
70 * If using a SQL proxy service, it would probably be best to have two proxy hosts for the
71 * load balancer to talk to. One would be the 'host' of the master server entry and another for
72 * the (logical) replica server entry. The proxy could map the load balancer's "replica" DB to
73 * any number of physical replica DBs.
78 interface ILoadBalancer
{
79 /** @var int Request a replica DB connection */
80 const DB_REPLICA
= -1;
81 /** @var int Request a master DB connection */
84 /** @var string Domain specifier when no specific database needs to be selected */
85 const DOMAIN_ANY
= '';
87 /** @var int DB handle should have DBO_TRX disabled and the caller will leave it as such */
88 const CONN_TRX_AUTOCOMMIT
= 1;
89 /** @var int Alias for CONN_TRX_AUTOCOMMIT for b/c; deprecated since 1.31 */
90 const CONN_TRX_AUTO
= 1;
93 * Construct a manager of IDatabase connection objects
95 * @param array $params Parameter map with keys:
96 * - servers : Required. Array of server info structures.
97 * - localDomain: A DatabaseDomain or domain ID string.
98 * - loadMonitor : Name of a class used to fetch server lag and load.
99 * - readOnlyReason : Reason the master DB is read-only if so [optional]
100 * - waitTimeout : Maximum time to wait for replicas for consistency [optional]
101 * - maxLag: Avoid replica DB servers with more lag than this [optional]
102 * - srvCache : BagOStuff object for server cache [optional]
103 * - wanCache : WANObjectCache object [optional]
104 * - chronologyCallback: Callback to run before the first connection attempt [optional]
105 * - hostname : The name of the current server [optional]
106 * - cliMode: Whether the execution context is a CLI script. [optional]
107 * - profiler : Class name or instance with profileIn()/profileOut() methods. [optional]
108 * - trxProfiler: TransactionProfiler instance. [optional]
109 * - replLogger: PSR-3 logger instance. [optional]
110 * - connLogger: PSR-3 logger instance. [optional]
111 * - queryLogger: PSR-3 logger instance. [optional]
112 * - perfLogger: PSR-3 logger instance. [optional]
113 * - errorLogger : Callback that takes an Exception and logs it. [optional]
114 * - deprecationLogger: Callback to log a deprecation warning. [optional]
115 * @throws InvalidArgumentException
117 public function __construct( array $params );
120 * Get the index of the reader connection, which may be a replica DB
122 * This takes into account load ratios and lag times. It should
123 * always return a consistent index during a given invocation.
125 * Side effect: opens connections to databases
126 * @param string|bool $group Query group, or false for the generic reader
127 * @param string|bool $domain Domain ID, or false for the current domain
129 * @return bool|int|string
131 public function getReaderIndex( $group = false, $domain = false );
134 * Set the master wait position
136 * If a DB_REPLICA connection has been opened already, then wait immediately.
137 * Otherwise sets a variable telling it to wait if such a connection is opened.
139 * This only applies to connections to the generic replica DB for this request.
140 * If a timeout happens when waiting, then getLaggedReplicaMode()/laggedReplicaUsed()
143 * @param DBMasterPos|bool $pos Master position or false
145 public function waitFor( $pos );
148 * Set the master wait position and wait for a "generic" replica DB to catch up to it
150 * This can be used a faster proxy for waitForAll()
152 * @param DBMasterPos|bool $pos Master position or false
153 * @param int $timeout Max seconds to wait; default is mWaitTimeout
154 * @return bool Success (able to connect and no timeouts reached)
156 public function waitForOne( $pos, $timeout = null );
159 * Set the master wait position and wait for ALL replica DBs to catch up to it
161 * @param DBMasterPos|bool $pos Master position or false
162 * @param int $timeout Max seconds to wait; default is mWaitTimeout
163 * @return bool Success (able to connect and no timeouts reached)
165 public function waitForAll( $pos, $timeout = null );
168 * Get any open connection to a given server index, local or foreign
170 * Use CONN_TRX_AUTOCOMMIT to only look for connections opened with that flag
172 * @param int $i Server index or DB_MASTER/DB_REPLICA
173 * @param int $flags Bitfield of CONN_* class constants
174 * @return Database|bool False if no such connection is open
176 public function getAnyOpenConnection( $i, $flags = 0 );
179 * Get a connection handle by server index
181 * The CONN_TRX_AUTOCOMMIT flag is ignored for databases with ATTR_DB_LEVEL_LOCKING
182 * (e.g. sqlite) in order to avoid deadlocks. ILoadBalancer::getServerAttributes()
183 * can be used to check such flags beforehand.
185 * If the caller uses $domain or sets CONN_TRX_AUTOCOMMIT in $flags, then it must also
186 * call ILoadBalancer::reuseConnection() on the handle when finished using it.
187 * In all other cases, this is not necessary, though not harmful either.
189 * @param int $i Server index (overrides $groups) or DB_MASTER/DB_REPLICA
190 * @param array|string|bool $groups Query group(s), or false for the generic reader
191 * @param string|bool $domain Domain ID, or false for the current domain
192 * @param int $flags Bitfield of CONN_* class constants
194 * @note This method throws DBAccessError if ILoadBalancer::disable() was called
199 public function getConnection( $i, $groups = [], $domain = false, $flags = 0 );
202 * Mark a foreign connection as being available for reuse under a different DB domain
204 * This mechanism is reference-counted, and must be called the same number of times
205 * as getConnection() to work.
207 * @param IDatabase $conn
208 * @throws InvalidArgumentException
210 public function reuseConnection( IDatabase
$conn );
213 * Get a database connection handle reference
215 * The handle's methods simply wrap those of a Database handle
217 * The CONN_TRX_AUTOCOMMIT flag is ignored for databases with ATTR_DB_LEVEL_LOCKING
218 * (e.g. sqlite) in order to avoid deadlocks. ILoadBalancer::getServerAttributes()
219 * can be used to check such flags beforehand.
221 * @see ILoadBalancer::getConnection() for parameter information
223 * @param int $i Server index or DB_MASTER/DB_REPLICA
224 * @param array|string|bool $groups Query group(s), or false for the generic reader
225 * @param string|bool $domain Domain ID, or false for the current domain
226 * @param int $flags Bitfield of CONN_* class constants (e.g. CONN_TRX_AUTOCOMMIT)
229 public function getConnectionRef( $i, $groups = [], $domain = false, $flags = 0 );
232 * Get a database connection handle reference without connecting yet
234 * The handle's methods simply wrap those of a Database handle
236 * The CONN_TRX_AUTOCOMMIT flag is ignored for databases with ATTR_DB_LEVEL_LOCKING
237 * (e.g. sqlite) in order to avoid deadlocks. ILoadBalancer::getServerAttributes()
238 * can be used to check such flags beforehand.
240 * @see ILoadBalancer::getConnection() for parameter information
242 * @param int $i Server index or DB_MASTER/DB_REPLICA
243 * @param array|string|bool $groups Query group(s), or false for the generic reader
244 * @param string|bool $domain Domain ID, or false for the current domain
245 * @param int $flags Bitfield of CONN_* class constants (e.g. CONN_TRX_AUTOCOMMIT)
248 public function getLazyConnectionRef( $i, $groups = [], $domain = false, $flags = 0 );
251 * Get a maintenance database connection handle reference for migrations and schema changes
253 * The handle's methods simply wrap those of a Database handle
255 * The CONN_TRX_AUTOCOMMIT flag is ignored for databases with ATTR_DB_LEVEL_LOCKING
256 * (e.g. sqlite) in order to avoid deadlocks. ILoadBalancer::getServerAttributes()
257 * can be used to check such flags beforehand.
259 * @see ILoadBalancer::getConnection() for parameter information
261 * @param int $db Server index or DB_MASTER/DB_REPLICA
262 * @param array|string|bool $groups Query group(s), or false for the generic reader
263 * @param string|bool $domain Domain ID, or false for the current domain
264 * @param int $flags Bitfield of CONN_* class constants (e.g. CONN_TRX_AUTOCOMMIT)
265 * @return MaintainableDBConnRef
267 public function getMaintenanceConnectionRef( $db, $groups = [], $domain = false, $flags = 0 );
270 * Open a connection to the server given by the specified index
272 * The index must be an actual index into the array. If a connection to the server is
273 * already open and not considered an "in use" foreign connection, this simply returns it.
275 * Avoid using CONN_TRX_AUTOCOMMIT for databases with ATTR_DB_LEVEL_LOCKING (e.g. sqlite) in
276 * order to avoid deadlocks. ILoadBalancer::getServerAttributes() can be used to check
277 * such flags beforehand.
279 * If the caller uses $domain or sets CONN_TRX_AUTOCOMMIT in $flags, then it must also
280 * call ILoadBalancer::reuseConnection() on the handle when finished using it.
281 * In all other cases, this is not necessary, though not harmful either.
283 * @note This method throws DBAccessError if ILoadBalancer::disable() was called
285 * @param int $i Server index (does not support DB_MASTER/DB_REPLICA)
286 * @param string|bool $domain Domain ID, or false for the current domain
287 * @param int $flags Bitfield of CONN_* class constants (e.g. CONN_TRX_AUTOCOMMIT)
288 * @return Database|bool Returns false on errors
289 * @throws DBAccessError
291 public function openConnection( $i, $domain = false, $flags = 0 );
296 public function getWriterIndex();
299 * Returns true if the specified index is a valid server index
304 public function haveIndex( $i );
307 * Returns true if the specified index is valid and has non-zero load
312 public function isNonZeroLoad( $i );
315 * Get the number of defined servers (not the number of open connections)
319 public function getServerCount();
322 * Get the host name or IP address of the server with the specified index
325 * @return string Readable name if available or IP/host otherwise
327 public function getServerName( $i );
330 * Get DB type of the server with the specified index
333 * @return string One of (mysql,postgres,sqlite,...) or "unknown" for bad indexes
336 public function getServerType( $i );
339 * @param int $i Server index
340 * @return array (Database::ATTRIBUTE_* constant => value) for all such constants
343 public function getServerAttributes( $i );
346 * Get the current master position for chronology control purposes
347 * @return DBMasterPos|bool Returns false if not applicable
349 public function getMasterPos();
352 * Disable this load balancer. All connections are closed, and any attempt to
353 * open a new connection will result in a DBAccessError.
355 public function disable();
358 * Close all open connections
360 public function closeAll();
365 * Using this function makes sure the LoadBalancer knows the connection is closed.
366 * If you use $conn->close() directly, the load balancer won't update its state.
368 * @param IDatabase $conn
370 public function closeConnection( IDatabase
$conn );
373 * Commit transactions on all open connections
374 * @param string $fname Caller name
375 * @throws DBExpectedError
377 public function commitAll( $fname = __METHOD__
);
380 * Perform all pre-commit callbacks that remain part of the atomic transactions
381 * and disable any post-commit callbacks until runMasterPostTrxCallbacks()
383 * Use this only for mutli-database commits
385 public function finalizeMasterChanges();
388 * Perform all pre-commit checks for things like replication safety
390 * Use this only for mutli-database commits
392 * @param array $options Includes:
393 * - maxWriteDuration : max write query duration time in seconds
394 * @throws DBTransactionError
396 public function approveMasterChanges( array $options );
399 * Flush any master transaction snapshots and set DBO_TRX (if DBO_DEFAULT is set)
401 * The DBO_TRX setting will be reverted to the default in each of these methods:
402 * - commitMasterChanges()
403 * - rollbackMasterChanges()
405 * This allows for custom transaction rounds from any outer transaction scope.
407 * @param string $fname
408 * @throws DBExpectedError
410 public function beginMasterChanges( $fname = __METHOD__
);
413 * Issue COMMIT on all master connections where writes where done
414 * @param string $fname Caller name
415 * @throws DBExpectedError
417 public function commitMasterChanges( $fname = __METHOD__
);
420 * Issue all pending post-COMMIT/ROLLBACK callbacks
422 * Use this only for mutli-database commits
424 * @param int $type IDatabase::TRIGGER_* constant
425 * @return Exception|null The first exception or null if there were none
427 public function runMasterPostTrxCallbacks( $type );
430 * Issue ROLLBACK only on master, only if queries were done on connection
431 * @param string $fname Caller name
432 * @throws DBExpectedError
434 public function rollbackMasterChanges( $fname = __METHOD__
);
437 * Suppress all pending post-COMMIT/ROLLBACK callbacks
439 * Use this only for mutli-database commits
441 * @return Exception|null The first exception or null if there were none
443 public function suppressTransactionEndCallbacks();
446 * Commit all replica DB transactions so as to flush any REPEATABLE-READ or SSI snapshot
448 * @param string $fname Caller name
450 public function flushReplicaSnapshots( $fname = __METHOD__
);
453 * @return bool Whether a master connection is already open
455 public function hasMasterConnection();
458 * Determine if there are pending changes in a transaction by this thread
461 public function hasMasterChanges();
464 * Get the timestamp of the latest write query done by this thread
465 * @return float|bool UNIX timestamp or false
467 public function lastMasterChangeTimestamp();
470 * Check if this load balancer object had any recent or still
471 * pending writes issued against it by this PHP thread
473 * @param float $age How many seconds ago is "recent" [defaults to mWaitTimeout]
476 public function hasOrMadeRecentMasterChanges( $age = null );
479 * Get the list of callers that have pending master changes
481 * @return string[] List of method names
483 public function pendingMasterChangeCallers();
486 * @note This method will trigger a DB connection if not yet done
487 * @param string|bool $domain Domain ID, or false for the current domain
488 * @return bool Whether the database for generic connections this request is highly "lagged"
490 public function getLaggedReplicaMode( $domain = false );
493 * Checks whether the database for generic connections this request was both:
494 * - a) Already choosen due to a prior connection attempt
495 * - b) Considered highly "lagged"
497 * @note This method will never cause a new DB connection
500 public function laggedReplicaUsed();
503 * @note This method may trigger a DB connection if not yet done
504 * @param string|bool $domain Domain ID, or false for the current domain
505 * @param IDatabase|null $conn DB master connection; used to avoid loops [optional]
506 * @return string|bool Reason the master is read-only or false if it is not
508 public function getReadOnlyReason( $domain = false, IDatabase
$conn = null );
511 * Disables/enables lag checks
512 * @param null|bool $mode
515 public function allowLagged( $mode = null );
520 public function pingAll();
523 * Call a function with each open connection object
524 * @param callable $callback
525 * @param array $params
527 public function forEachOpenConnection( $callback, array $params = [] );
530 * Call a function with each open connection object to a master
531 * @param callable $callback
532 * @param array $params
534 public function forEachOpenMasterConnection( $callback, array $params = [] );
537 * Call a function with each open replica DB connection object
538 * @param callable $callback
539 * @param array $params
541 public function forEachOpenReplicaConnection( $callback, array $params = [] );
544 * Get the hostname and lag time of the most-lagged replica DB
546 * This is useful for maintenance scripts that need to throttle their updates.
547 * May attempt to open connections to replica DBs on the default DB. If there is
548 * no lag, the maximum lag will be reported as -1.
550 * @param bool|string $domain Domain ID, or false for the default database
551 * @return array ( host, max lag, index of max lagged host )
553 public function getMaxLag( $domain = false );
556 * Get an estimate of replication lag (in seconds) for each server
558 * Results are cached for a short time in memcached/process cache
560 * Values may be "false" if replication is too broken to estimate
562 * @param string|bool $domain
563 * @return int[] Map of (server index => float|int|bool)
565 public function getLagTimes( $domain = false );
568 * Get the lag in seconds for a given connection, or zero if this load
569 * balancer does not have replication enabled.
571 * This should be used in preference to Database::getLag() in cases where
572 * replication may not be in use, since there is no way to determine if
573 * replication is in use at the connection level without running
574 * potentially restricted queries such as SHOW SLAVE STATUS. Using this
575 * function instead of Database::getLag() avoids a fatal error in this
576 * case on many installations.
578 * @param IDatabase $conn
579 * @return int|bool Returns false on error
581 public function safeGetLag( IDatabase
$conn );
584 * Wait for a replica DB to reach a specified master position
586 * This will connect to the master to get an accurate position if $pos is not given
588 * @param IDatabase $conn Replica DB
589 * @param DBMasterPos|bool $pos Master position; default: current position
590 * @param int $timeout Timeout in seconds [optional]
591 * @return bool Success
593 public function safeWaitForMasterPos( IDatabase
$conn, $pos = false, $timeout = 10 );
596 * Set a callback via IDatabase::setTransactionListener() on
597 * all current and future master connections of this load balancer
599 * @param string $name Callback name
600 * @param callable|null $callback
602 public function setTransactionListener( $name, callable
$callback = null );
605 * Set a new table prefix for the existing local domain ID for testing
607 * @param string $prefix
609 public function setDomainPrefix( $prefix );
612 * Make certain table names use their own database, schema, and table prefix
613 * when passed into SQL queries pre-escaped and without a qualified database name
615 * For example, "user" can be converted to "myschema.mydbname.user" for convenience.
616 * Appearances like `user`, somedb.user, somedb.someschema.user will used literally.
618 * Calling this twice will completely clear any old table aliases. Also, note that
619 * callers are responsible for making sure the schemas and databases actually exist.
621 * @param array[] $aliases Map of (table => (dbname, schema, prefix) map)
623 public function setTableAliases( array $aliases );
626 * Convert certain index names to alternative names before querying the DB
628 * Note that this applies to indexes regardless of the table they belong to.
630 * This can be employed when an index was renamed X => Y in code, but the new Y-named
631 * indexes were not yet built on all DBs. After all the Y-named ones are added by the DBA,
632 * the aliases can be removed, and then the old X-named indexes dropped.
634 * @param string[] $aliases
638 public function setIndexAliases( array $aliases );