MediaWiki master
Wikimedia\Rdbms\IDatabase Interface Reference

Basic database interface for live and lazy-loaded relation database handles. More...

Inherits Wikimedia\Rdbms\IReadableDatabase.

Inherited by Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\IMaintainableDatabase.

Collaboration diagram for Wikimedia\Rdbms\IDatabase:

Public Member Functions

 affectedRows ()
 Get the number of rows affected by the last query method call.
 
 begin ( $fname=__METHOD__, $mode=self::TRANSACTION_EXPLICIT)
 Begin a transaction.
 
 cancelAtomic ( $fname=__METHOD__, AtomicSectionIdentifier $sectionId=null)
 Cancel an atomic section of SQL statements.
 
 commit ( $fname=__METHOD__, $flush=self::FLUSHING_ONE)
 Commits a transaction previously started using begin()
 
 delete ( $table, $conds, $fname=__METHOD__)
 Delete all rows in a table that match a condition.
 
 deleteJoin ( $delTable, $joinTable, $delVar, $joinVar, $conds, $fname=__METHOD__)
 Delete all rows in a table that match a condition which includes a join.
 
 doAtomicSection ( $fname, callable $callback, $cancelable=self::ATOMIC_NOT_CANCELABLE)
 Perform an atomic section of reversible SQL statements from a callback.
 
 endAtomic ( $fname=__METHOD__)
 Ends an atomic section of SQL statements.
 
 explicitTrxActive ()
 Check whether there is a transaction open at the specific request of a caller.
 
 flushSession ( $fname=__METHOD__, $flush=self::FLUSHING_ONE)
 Release important session-level state (named lock, table locks) as post-rollback cleanup.
 
 flushSnapshot ( $fname=__METHOD__, $flush=self::FLUSHING_ONE)
 Commit any transaction but error out if writes or callbacks are pending.
 
 getLBInfo ( $name=null)
 Get properties passed down from the server info array of the load balancer.
 
 getPrimaryPos ()
 Get the replication position of this primary DB server.
 
 getScopedLockAndFlush ( $lockKey, $fname, $timeout)
 Acquire a named lock, flush any transaction, and return an RAII style unlocker object.
 
 getTopologyRole ()
 Get the replication topology role of this server.
 
 insert ( $table, $rows, $fname=__METHOD__, $options=[])
 Insert row(s) into a table, in the provided order.
 
 insertId ()
 Get the sequence-based ID assigned by the last query method call.
 
 insertSelect ( $destTable, $srcTable, $varMap, $conds, $fname=__METHOD__, $insertOptions=[], $selectOptions=[], $selectJoinConds=[])
 INSERT SELECT wrapper.
 
 isReadOnly ()
 Check if this DB server is marked as read-only according to load balancer info.
 
 lastDoneWrites ()
 Get the last time the connection may have been used for a write query.
 
 lock ( $lockName, $method, $timeout=5, $flags=0)
 Acquire a named lock.
 
 lockForUpdate ( $table, $conds='', $fname=__METHOD__, $options=[], $join_conds=[])
 Lock all rows meeting the given conditions/options FOR UPDATE.
 
 lockIsFree ( $lockName, $method)
 Check to see if a named lock is not locked by any thread (non-blocking)
 
 newDeleteQueryBuilder ()
 Get an DeleteQueryBuilder bound to this connection.
 
 newInsertQueryBuilder ()
 Get an InsertQueryBuilder bound to this connection.
 
 newReplaceQueryBuilder ()
 Get an ReplaceQueryBuilder bound to this connection.
 
 newUpdateQueryBuilder ()
 Get an UpdateQueryBuilder bound to this connection.
 
 nextSequenceValue ( $seqName)
 Deprecated method, calls should be removed.
 
 onAtomicSectionCancel (callable $callback, $fname=__METHOD__)
 Run a callback when the atomic section is cancelled.
 
 onTransactionCommitOrIdle (callable $callback, $fname=__METHOD__)
 Run a callback when the current transaction commits or now if there is none.
 
 onTransactionPreCommitOrIdle (callable $callback, $fname=__METHOD__)
 Run a callback before the current transaction commits or now if there is none.
 
 onTransactionResolution (callable $callback, $fname=__METHOD__)
 Run a callback when the current transaction commits or rolls back.
 
 pendingWriteCallers ()
 Get the list of method names that did write queries for this transaction.
 
 pendingWriteQueryDuration ( $type=self::ESTIMATE_TOTAL)
 Get the time spend running write queries for this transaction.
 
 query ( $sql, $fname=__METHOD__, $flags=0)
 Run an SQL query statement and return the result.
 
 replace ( $table, $uniqueKeys, $rows, $fname=__METHOD__)
 Insert row(s) into a table, in the provided order, while deleting conflicting rows.
 
 rollback ( $fname=__METHOD__, $flush=self::FLUSHING_ONE)
 Rollback a transaction previously started using begin()
 
 serverIsReadOnly ()
 
 setLBInfo ( $nameOrArray, $value=null)
 Set the entire array or a particular key of the managing load balancer info array.
 
 setSessionOptions (array $options)
 Override database's default behavior.
 
 setTransactionListener ( $name, callable $callback=null)
 Run a callback after each time any transaction commits or rolls back.
 
 startAtomic ( $fname=__METHOD__, $cancelable=self::ATOMIC_NOT_CANCELABLE)
 Begin an atomic section of SQL statements.
 
 trxLevel ()
 Gets the current transaction level.
 
 trxTimestamp ()
 Get the UNIX timestamp of the time that the transaction was established.
 
 unlock ( $lockName, $method)
 Release a lock.
 
 update ( $table, $set, $conds, $fname=__METHOD__, $options=[])
 Update all rows in a table that match a given condition.
 
 upsert ( $table, array $rows, $uniqueKeys, array $set, $fname=__METHOD__)
 Upsert row(s) into a table, in the provided order, while updating conflicting rows.
 
 writesOrCallbacksPending ()
 Whether there is a transaction open with either possible write queries or unresolved pre-commit/commit/resolution callbacks pending.
 
 writesPending ()
 
- Public Member Functions inherited from Wikimedia\Rdbms\IReadableDatabase
 __toString ()
 Get a debugging string that mentions the database type, the ID of this instance, and the ID of any underlying connection resource or driver object if one is present.
 
 close ( $fname=__METHOD__)
 Close the database connection.
 
 databasesAreIndependent ()
 Returns true if DBs are assumed to be on potentially different servers.
 
 dbSchema ( $schema=null)
 Get/set the db schema.
 
 decodeBlob ( $b)
 Some DBMSs return a special placeholder object representing blob fields in result objects.
 
 encodeBlob ( $b)
 Some DBMSs have a special format for inserting into blob fields, they don't allow simple quoted strings to be inserted.
 
 estimateRowCount ( $tables, $var=' *', $conds='', $fname=__METHOD__, $options=[], $join_conds=[])
 Estimate the number of rows in dataset.
 
 expr (string $field, string $op, $value)
 See Expression::__construct()
 
 getDBname ()
 Get the current database name; null if there isn't one.
 
 getDomainID ()
 Return the currently selected domain ID.
 
 getLag ()
 Get the seconds of replication lag on this database server.
 
 getServer ()
 Get the hostname or IP address of the server.
 
 getServerInfo ()
 Get a human-readable string describing the current software version.
 
 getServerName ()
 Get the readable name for the server.
 
 getServerVersion ()
 A string describing the current software version.
 
 getSessionLagStatus ()
 Get a cached estimate of the seconds of replication lag on this database server, using the estimate obtained at the start of the current transaction if one is active.
 
 getSoftwareLink ()
 Returns a wikitext style link to the DB's website (e.g.
 
 getType ()
 Get the RDBMS type of the server (e.g.
 
 isOpen ()
 
 lastErrno ()
 Get the RDBMS-specific error code from the last attempted query statement.
 
 lastError ()
 Get the RDBMS-specific error description from the last attempted query statement.
 
 newSelectQueryBuilder ()
 Create an empty SelectQueryBuilder which can be used to run queries against this connection.
 
 newUnionQueryBuilder ()
 Create an empty UnionQueryBuilder which can be used to run queries against this connection.
 
 ping ()
 Ping the server and try to reconnect if it there is no connection.
 
 primaryPosWait (DBPrimaryPos $pos, $timeout)
 Wait for the replica server to catch up to a given primary server position.
 
 select ( $table, $vars, $conds='', $fname=__METHOD__, $options=[], $join_conds=[])
 Execute a SELECT query constructed using the various parameters provided.
 
 selectDomain ( $domain)
 Set the current domain (database, schema, and table prefix)
 
 selectField ( $table, $var, $cond='', $fname=__METHOD__, $options=[], $join_conds=[])
 A SELECT wrapper which returns a single field from a single result row.
 
 selectFieldValues ( $table, $var, $cond='', $fname=__METHOD__, $options=[], $join_conds=[])
 A SELECT wrapper which returns a list of single field values from result rows.
 
 selectRow ( $table, $vars, $conds, $fname=__METHOD__, $options=[], $join_conds=[])
 Wrapper to IDatabase::select() that only fetches one row (via LIMIT)
 
 selectRowCount ( $tables, $var=' *', $conds='', $fname=__METHOD__, $options=[], $join_conds=[])
 Get the number of rows in dataset.
 
 tablePrefix ( $prefix=null)
 Get/set the table prefix.
 
 wasDeadlock ()
 Determines if the last failure was due to a deadlock.
 
 wasReadOnlyError ()
 Determines if the last failure was due to the database being read-only.
 
- Public Member Functions inherited from Wikimedia\Rdbms\Platform\ISQLPlatform
 addIdentifierQuotes ( $s)
 Escape a SQL identifier (e.g.
 
 anyChar ()
 Returns a token for buildLike() that denotes a '_' to be used in a LIKE query.
 
 anyString ()
 Returns a token for buildLike() that denotes a '' to be used in a LIKE query.
 
 bitAnd ( $fieldLeft, $fieldRight)
 
 bitNot ( $field)
 
 bitOr ( $fieldLeft, $fieldRight)
 
 buildComparison (string $op, array $conds)
 Build a condition comparing multiple values, for use with indexes that cover multiple fields, common when e.g.
 
 buildConcat ( $stringList)
 Build a concatenation list to feed into a SQL query.
 
 buildExcludedValue ( $column)
 Build a reference to a column value from the conflicting proposed upsert() row.
 
 buildGreatest ( $fields, $values)
 Build a GREATEST function statement comparing columns/values.
 
 buildGroupConcatField ( $delim, $table, $field, $conds='', $join_conds=[])
 Build a GROUP_CONCAT or equivalent statement for a query.
 
 buildIntegerCast ( $field)
 
 buildLeast ( $fields, $values)
 Build a LEAST function statement comparing columns/values.
 
 buildLike ( $param,... $params)
 LIKE statement wrapper.
 
 buildSelectSubquery ( $table, $vars, $conds='', $fname=__METHOD__, $options=[], $join_conds=[])
 Equivalent to IDatabase::selectSQLText() except wraps the result in Subquery.
 
 buildStringCast ( $field)
 
 buildSubString ( $input, $startPosition, $length=null)
 Build a SUBSTRING function.
 
 conditional ( $cond, $caseTrueExpression, $caseFalseExpression)
 Returns an SQL expression for a simple conditional.
 
 decodeExpiry ( $expiry, $format=TS_MW)
 Decode an expiry time into a DBMS independent format.
 
 encodeExpiry ( $expiry)
 Encode an expiry time into the DBMS dependent format.
 
 factorConds ( $condsArray)
 Given an array of condition arrays representing an OR list of AND lists, for example:
 
 getInfinity ()
 Find out when 'infinity' is.
 
 getTableAliases ()
 Return current table aliases.
 
 implicitOrderby ()
 Returns true if this database does an implicit order by when the column has an index For example: SELECT page_title FROM page LIMIT 1.
 
 limitResult ( $sql, $limit, $offset=false)
 Construct a LIMIT query with optional offset.
 
 makeList (array $a, $mode=self::LIST_COMMA)
 Makes an encoded list of strings from an array.
 
 makeWhereFrom2d ( $data, $baseKey, $subKey)
 Build a "OR" condition with pairs from a two-dimensional array.
 
 selectSQLText ( $table, $vars, $conds='', $fname=__METHOD__, $options=[], $join_conds=[])
 Take the same arguments as IDatabase::select() and return the SQL it would use.
 
 setIndexAliases (array $aliases)
 Convert certain index names to alternative names before querying the DB.
 
 setSchemaVars ( $vars)
 Set schema variables to be used when streaming commands from SQL files or stdin.
 
 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.
 
 strreplace ( $orig, $old, $new)
 Returns a SQL expression for simple string replacement (e.g.
 
 tableName (string $name, $format='quoted')
 Format a table name ready for use in constructing an SQL query.
 
 tableNames (... $tables)
 Fetch a number of table names into an associative array.
 
 tableNamesN (... $tables)
 Fetch a number of table names into a zero-indexed numerical array.
 
 timestamp ( $ts=0)
 Convert a timestamp in one of the formats accepted by ConvertibleTimestamp to the format used for inserting into timestamp fields in this DBMS.
 
 timestampOrNull ( $ts=null)
 Convert a timestamp in one of the formats accepted by ConvertibleTimestamp to the format used for inserting into timestamp fields in this DBMS.
 
 unionQueries ( $sqls, $all, $options=[])
 Construct a UNION query.
 
 unionSupportsOrderAndLimit ()
 Determine if the RDBMS supports ORDER BY and LIMIT for separate subqueries within UNION.
 
- Public Member Functions inherited from Wikimedia\Rdbms\Database\DbQuoter
 addQuotes ( $s)
 Escape and quote a raw value string for use in a SQL query.
 
- Public Member Functions inherited from Wikimedia\Rdbms\Database\IDatabaseFlags
 clearFlag ( $flag, $remember=self::REMEMBER_NOTHING)
 Clear a flag for this connection.
 
 getFlag ( $flag)
 Returns a boolean whether the flag $flag is set for this connection.
 
 restoreFlags ( $state=self::RESTORE_PRIOR)
 Restore the flags to their prior state before the last setFlag/clearFlag call.
 
 setFlag ( $flag, $remember=self::REMEMBER_NOTHING)
 Set a flag for this connection.
 

Public Attributes

const LOCK_TIMESTAMP = 1
 Flag to return the lock acquisition timestamp (null if not acquired)
 
- Public Attributes inherited from Wikimedia\Rdbms\Platform\ISQLPlatform
const QUERY_PSEUDO_PERMANENT = 2
 Track a TEMPORARY table CREATE as if it was for a permanent table (for testing)
 
- Public Attributes inherited from Wikimedia\Rdbms\Database\IDatabaseFlags
const DBO_COMPRESS = 512
 Enable compression in connection protocol.
 
const DBO_DDLMODE = 128
 Schema file mode; was used by Oracle.
 
const DBO_DEBUG = 1
 Enable debug logging of all SQL queries.
 
const DBO_DEFAULT = 16
 Join load balancer transaction rounds (which control DBO_TRX) in non-CLI mode.
 
const DBO_GAUGE = 1024
 Optimize connection for guaging server state (e.g.
 
const DBO_IGNORE = 4
 Unused since 1.31.
 
const DBO_NOBUFFER = 2
 Unused since 1.34.
 
const DBO_PERSISTENT = 32
 Use DB persistent connections if possible.
 
const DBO_SSL = 256
 Enable SSL/TLS in connection protocol.
 
const DBO_SYSDBA = 64
 DBA session mode; was used by Oracle.
 
const DBO_TRX = 8
 Automatically start a transaction before running a query if none is active.
 
const REMEMBER_NOTHING = ''
 Do not remember the prior flags.
 
const REMEMBER_PRIOR = 'remember'
 Remember the prior flags.
 
const RESTORE_INITIAL = 'initial'
 Restore to the initial flag state.
 
const RESTORE_PRIOR = 'prior'
 Restore to the prior flag state.
 

Detailed Description

Basic database interface for live and lazy-loaded relation database handles.

Definition at line 36 of file IDatabase.php.

Member Function Documentation

◆ affectedRows()

Wikimedia\Rdbms\IDatabase::affectedRows ( )

Get the number of rows affected by the last query method call.

This method should only be called when all the following hold true:

  • (a) A method call was made to insert(), upsert(), replace(), update(), delete(), insertSelect(), query() with a non-SELECT statement, or queryMulti() with a non-SELECT terminal statement
  • (b) The method call succeeded
  • (c) No subsequent method calls were made, with the exception of affectedRows(), lastErrno(), lastError(), and getType()

In all other cases, the return value is unspecified.

UPDATE queries consider rows affected even when all their new column values match the previous values. Such rows can be excluded from the count by changing the WHERE clause to filter them out.

If the last query method call was to query() or queryMulti(), then the results are based on the (last) statement provided to that call and are driver-specific.

Returns
int

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by MediaWiki\User\User\addToDatabase(), DeleteLocalPasswords\processUsers(), and MediaWiki\User\User\saveSettings().

◆ begin()

Wikimedia\Rdbms\IDatabase::begin ( $fname = __METHOD__,
$mode = self::TRANSACTION_EXPLICIT )

Begin a transaction.

Only call this from code with outer transaction scope. See https://www.mediawiki.org/wiki/Database_transactions for details. Nesting of transactions is not supported.

Note that when the DBO_TRX flag is set (which is usually the case for web requests, but not for maintenance scripts), any previous database query will have started a transaction automatically.

Nesting of transactions is not supported. Attempts to nest transactions will cause a warning, unless the current transaction was started automatically because of the DBO_TRX flag.

Parameters
string$fnameCalling function name
string$modeA situationally valid IDatabase::TRANSACTION_* constant [optional]
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\DBConnRef, and Wikimedia\Rdbms\Database.

Referenced by Maintenance\beginTransaction().

◆ cancelAtomic()

Wikimedia\Rdbms\IDatabase::cancelAtomic ( $fname = __METHOD__,
AtomicSectionIdentifier $sectionId = null )

Cancel an atomic section of SQL statements.

This will roll back only the statements executed since the start of the most recent atomic section, and close that section. If a transaction was open before the corresponding startAtomic() call, any statements before that call are not rolled back and the transaction remains open. If the corresponding startAtomic() implicitly started a transaction, that transaction is rolled back.

Note
callers must use additional measures for situations involving two or more (peer) transactions (e.g. updating two database servers at once). The transaction and savepoint logic of startAtomic() are bound to specific IDatabase instances.

Note that a call to IDatabase::rollback() will also roll back any open atomic sections.

Note
As an optimization to save rountrips, this method may only be called when startAtomic() was called with the ATOMIC_CANCELABLE flag.
Since
1.31
See also
IDatabase::startAtomic
Parameters
string$fname
AtomicSectionIdentifier | null$sectionIdSection ID from startAtomic(); passing this enables cancellation of unclosed nested sections [optional]
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ commit()

Wikimedia\Rdbms\IDatabase::commit ( $fname = __METHOD__,
$flush = self::FLUSHING_ONE )

Commits a transaction previously started using begin()

If no transaction is in progress, a warning is issued.

Only call this from code with outer transaction scope. See https://www.mediawiki.org/wiki/Database_transactions for details. Nesting of transactions is not supported.

Parameters
string$fname
string$flushFlush flag, set to situationally valid IDatabase::FLUSHING_* constant to disable warnings about explicitly committing implicit transactions, or calling commit when no transaction is in progress. This will trigger an exception if there is an ongoing explicit transaction. Only set the flush flag if you are sure that these warnings are not applicable, and no explicit transactions are open.
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by Wikimedia\Rdbms\LoadBalancer\commitPrimaryChanges(), and Maintenance\commitTransaction().

◆ delete()

Wikimedia\Rdbms\IDatabase::delete ( $table,
$conds,
$fname = __METHOD__ )

Delete all rows in a table that match a condition.

This operation will be seen by affectedRows()/insertId() as one query statement, regardless of how many statements are actually sent by the class implementation.

Access: internal
callers outside of rdbms library should use DeleteQueryBuilder instead.
Parameters
string$tableUnqualified name of table
string|IExpression|array<string,?scalar|non-empty-array<?scalar>>|array<int,string|IExpression>$conds Array of conditions. See $conds in IDatabase::select() In order to prevent possible performance or replication issues or damaging a data accidentally, an empty condition for 'delete' queries isn't allowed. IDatabase::ALL_ROWS should be passed explicitly in order to delete all rows.
string$fnameName of the calling function
Returns
bool Return true if no exception was thrown (deprecated since 1.33)
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ deleteJoin()

Wikimedia\Rdbms\IDatabase::deleteJoin ( $delTable,
$joinTable,
$delVar,
$joinVar,
$conds,
$fname = __METHOD__ )

Delete all rows in a table that match a condition which includes a join.

For safety, an empty $conds will not delete everything. If you want to delete all rows where the join condition matches, set $conds=IDatabase::ALL_ROWS.

DO NOT put the join condition in $conds.

This operation will be seen by affectedRows()/insertId() as one query statement, regardless of how many statements are actually sent by the class implementation.

Parameters
string$delTableUnqualified name of table to delete rows from.
string$joinTableUnqualified name of reference table to join on.
string$delVarThe variable to join on, in the first table.
string$joinVarThe variable to join on, in the second table.
string|IExpression|array<string,?scalar|non-empty-array<?scalar>>|array<int,string|IExpression>$conds Condition array of field names mapped to variables, ANDed together in the WHERE clause
string$fnameCalling function name (use METHOD) for logs/profiling
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ doAtomicSection()

Wikimedia\Rdbms\IDatabase::doAtomicSection ( $fname,
callable $callback,
$cancelable = self::ATOMIC_NOT_CANCELABLE )

Perform an atomic section of reversible SQL statements from a callback.

The $callback takes the following arguments:

  • This database object
  • The value of $fname

This will execute the callback inside a pair of startAtomic()/endAtomic() calls. If any exception occurs during execution of the callback, it will be handled as follows:

  • If $cancelable is ATOMIC_CANCELABLE, cancelAtomic() will be called to back out any (and only) statements executed during the atomic section. If that succeeds, then the exception will be re-thrown; if it fails, then a different exception will be thrown and any further query attempts will fail until rollback() is called.
  • If $cancelable is ATOMIC_NOT_CANCELABLE, cancelAtomic() will be called to mark the end of the section and the error will be re-thrown. Any further query attempts will fail until rollback() is called.

This method is convenient for letting calls to the caller of this method be wrapped in a try/catch blocks for exception types that imply that the caller failed but was able to properly discard the changes it made in the transaction. This method can be an alternative to explicit calls to startAtomic()/endAtomic()/cancelAtomic().

Example usage, "RecordStore::save" method:

$dbw->doAtomicSection( __METHOD__, function ( $dbw ) use ( $record ) {
// Create new record metadata row
$dbw->insert( 'records', $record->toArray(), __METHOD__ );
// Figure out where to store the data based on the new row's ID
$path = $this->recordDirectory . '/' . $dbw->insertId();
// Write the record data to the storage system;
// blob store throws StoreFailureException on failure
$this->blobStore->create( $path, $record->getJSON() );
// Try to cleanup files orphaned by transaction rollback
$dbw->onTransactionResolution(
function ( $type ) use ( $path ) {
if ( $type === IDatabase::TRIGGER_ROLLBACK ) {
$this->blobStore->delete( $path );
}
},
__METHOD__
);
}, $dbw::ATOMIC_CANCELABLE );

Example usage, caller of the "RecordStore::save" method:

$dbw->startAtomic( __METHOD__ );
// ...various SQL writes happen...
try {
$recordStore->save( $record );
} catch ( StoreFailureException $e ) {
// ...various SQL writes happen...
}
// ...various SQL writes happen...
$dbw->endAtomic( __METHOD__ );
See also
Database::startAtomic
Database::endAtomic
Database::cancelAtomic
Parameters
string$fnameCaller name (usually METHOD)
callable$callbackCallback that issues write queries
string$cancelablePass self::ATOMIC_CANCELABLE to use a savepoint and enable self::cancelAtomic() for this section.
Returns
mixed Result of the callback (since 1.28)
Exceptions
DBErrorIf an error occurs, {
See also
query}
Exceptions
ExceptionIf an error occurs in the callback
Since
1.27; prior to 1.31 this did a rollback() instead of cancelAtomic(), and assumed no callers up the stack would ever try to catch the exception.

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by MediaWiki\Revision\RevisionStore\insertRevisionOn().

◆ endAtomic()

Wikimedia\Rdbms\IDatabase::endAtomic ( $fname = __METHOD__)

Ends an atomic section of SQL statements.

Ends the next section of atomic SQL statements and commits the transaction if necessary.

Since
1.23
See also
IDatabase::startAtomic
Parameters
string$fname
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by JobQueueDB\doBatchPushInternal(), and LCStoreDB\finishWrite().

◆ explicitTrxActive()

Wikimedia\Rdbms\IDatabase::explicitTrxActive ( )

Check whether there is a transaction open at the specific request of a caller.

Explicit transactions are spawned by begin(), startAtomic(), and doAtomicSection(). Note that explicit transactions should not be confused with explicit transaction rounds.

Returns
bool
Since
1.28

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ flushSession()

Wikimedia\Rdbms\IDatabase::flushSession ( $fname = __METHOD__,
$flush = self::FLUSHING_ONE )

Release important session-level state (named lock, table locks) as post-rollback cleanup.

This should only be called by a load balancer or if the handle is not attached to one. Also, there must be no chance that a future caller will still be expecting some of the lost session state.

Connection and query errors will be suppressed and logged

Parameters
string$fnameCalling function name
string$flushFlush flag, set to a situationally valid IDatabase::FLUSHING_* constant to disable warnings about explicitly rolling back implicit transactions. This will silently break any ongoing explicit transaction. Only set the flush flag if you are sure that it is safe to ignore these warnings in your context.
Exceptions
DBErrorIf an error occurs, {
See also
query}
Since
1.38

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ flushSnapshot()

Wikimedia\Rdbms\IDatabase::flushSnapshot ( $fname = __METHOD__,
$flush = self::FLUSHING_ONE )

Commit any transaction but error out if writes or callbacks are pending.

This is intended for clearing out REPEATABLE-READ snapshots so that callers can see a new point-in-time of the database. This is useful when one of many transaction rounds finished and significant time will pass in the script's lifetime. It is also useful to call on a replica server after waiting on replication to catch up to the primary server.

Parameters
string$fnameCalling function name
string$flushFlush flag, set to situationally valid IDatabase::FLUSHING_* constant to disable warnings about explicitly committing implicit transactions, or calling commit when no transaction is in progress. This will trigger an exception if there is an ongoing explicit transaction. Only set the flush flag if you are sure that these warnings are not applicable, and no explicit transactions are open.
Exceptions
DBErrorIf an error occurs, {
See also
query}
Since
1.28
1.34 Added $flush parameter

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ getLBInfo()

Wikimedia\Rdbms\IDatabase::getLBInfo ( $name = null)

Get properties passed down from the server info array of the load balancer.

Access: internal
should not be called outside of rdbms library.
Parameters
string | null$nameThe entry of the info array to get, or null to get the whole array
Returns
array|mixed|null

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by Wikimedia\Rdbms\LoadBalancer\waitForPrimaryPos().

◆ getPrimaryPos()

Wikimedia\Rdbms\IDatabase::getPrimaryPos ( )

Get the replication position of this primary DB server.

Returns
DBPrimaryPos|false Position; false if this is not a primary DB
Exceptions
DBErrorIf an error occurs, {
See also
query}
Since
1.37

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ getScopedLockAndFlush()

Wikimedia\Rdbms\IDatabase::getScopedLockAndFlush ( $lockKey,
$fname,
$timeout )

Acquire a named lock, flush any transaction, and return an RAII style unlocker object.

Only call this from outer transaction scope and when only one DB server will be affected. See https://www.mediawiki.org/wiki/Database_transactions for details.

This is suitable for transactions that need to be serialized using cooperative locks, where each transaction can see each others' changes. Any transaction is flushed to clear out stale REPEATABLE-READ snapshot data. Once the returned object falls out of PHP scope, the lock will be released unless a transaction is active. If one is active, then the lock will be released when it either commits or rolls back.

If the lock acquisition failed, then no transaction flush happens, and null is returned.

Parameters
string$lockKeyName of lock to release
string$fnameName of the calling method
int$timeoutAcquisition timeout in seconds
Returns
ScopedCallback|null
Exceptions
DBErrorIf an error occurs, {
See also
query}
Since
1.27

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by MediaWiki\Deferred\LinksUpdate\LinksUpdate\acquirePageLock().

◆ getTopologyRole()

Wikimedia\Rdbms\IDatabase::getTopologyRole ( )

Get the replication topology role of this server.

A replication topology defines which servers can originate changes to a given dataset and how those changes propagate among database servers. It is assumed that the server only participates in the replication of a single relevant dataset.

Returns
string One of the class ROLE_* constants
Exceptions
DBQueryError
Since
1.34

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ insert()

Wikimedia\Rdbms\IDatabase::insert ( $table,
$rows,
$fname = __METHOD__,
$options = [] )

Insert row(s) into a table, in the provided order.

This operation will be seen by affectedRows()/insertId() as one query statement, regardless of how many statements are actually sent by the class implementation.

Access: internal
callers outside of rdbms library should use InsertQueryBuilder instead.
Parameters
string$tableUnqualified name of table
array | array[]$rowsRow(s) to insert, as either:
  • A string-keyed map of (column name => value) defining a new row. Values are treated as literals and quoted appropriately; null is interpreted as NULL.
  • An integer-keyed list of such string-keyed maps, defining a list of new rows. The keys in each map must be identical to each other and in the same order. The rows must not collide with each other.
string$fnameCalling function name (use METHOD) for logs/profiling
string | array$optionsCombination map/list where each string-keyed entry maps a non-boolean option to the option parameters and each integer-keyed value is the name of a boolean option. Supported options are:
  • IGNORE: Boolean: skip insertion of rows that would cause unique key conflicts. IDatabase::affectedRows() can be used to determine how many rows were inserted.
Returns
bool Return true if no exception was thrown (deprecated since 1.33)
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ insertId()

Wikimedia\Rdbms\IDatabase::insertId ( )

Get the sequence-based ID assigned by the last query method call.

This method should only be called when all the following hold true:

  • (a) A method call was made to insert(), upsert(), replace(), or insertSelect()
  • (b) The method call attempts to insert exactly one row
  • (c) The method call omits the value of exactly one auto-increment column
  • (d) The method call succeeded
  • (e) No subsequent method calls were made, with the exception of affectedRows(), lastErrno(), lastError(), and getType()

In all other cases, the return value is unspecified.

When the query method is either insert() with "IGNORE", upsert(), or insertSelect(), callers should first check affectedRows() before calling this method, making sure that the query method actually created a row. Otherwise, an ID from a previous insert might be incorrectly assumed to belong to last insert.

Returns
int

Implemented in Wikimedia\Rdbms\Database, Wikimedia\Rdbms\DatabaseMySQL, and Wikimedia\Rdbms\DBConnRef.

Referenced by MediaWiki\User\User\addToDatabase(), and MediaWiki\CommentStore\CommentStore\createComment().

◆ insertSelect()

Wikimedia\Rdbms\IDatabase::insertSelect ( $destTable,
$srcTable,
$varMap,
$conds,
$fname = __METHOD__,
$insertOptions = [],
$selectOptions = [],
$selectJoinConds = [] )

INSERT SELECT wrapper.

Warning
If the insert will use an auto-increment or sequence to determine the value of a column, this may break replication on databases using statement-based replication if the SELECT is not deterministically ordered.

This operation will be seen by affectedRows()/insertId() as one query statement, regardless of how many statements are actually sent by the class implementation.

Parameters
string$destTableUnqualified name of destination table
string | array$srcTableUnqualified name of source table(s) (use an array for a join)
array$varMapMust be an associative array of the form [ 'dest1' => 'source1', ... ]. Source items may be literals rather than field names, but strings should be quoted with IDatabase::addQuotes()
string|IExpression|array<string,?scalar|non-empty-array<?scalar>>|array<int,string|IExpression>$conds Condition array. See $conds in IDatabase::select() for the details of the format of condition arrays. May be "*" to copy the whole table.
string$fnameThe function name of the caller, from METHOD
array$insertOptionsOptions for the INSERT part of the query, see IDatabase::insert() for details. Also, one additional option is available: pass 'NO_AUTO_COLUMNS' to hint that the query does not use an auto-increment or sequence to determine any column values.
array$selectOptionsOptions for the SELECT part of the query, see IDatabase::select() for details.
array$selectJoinCondsJoin conditions for the SELECT part of the query, see IDatabase::select() for details.
Returns
bool Return true if no exception was thrown (deprecated since 1.33)
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ isReadOnly()

Wikimedia\Rdbms\IDatabase::isReadOnly ( )

Check if this DB server is marked as read-only according to load balancer info.

Note
LoadBalancer checks serverIsReadOnly() when setting the load balancer info array
Returns
bool
Since
1.27

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by LCStoreDB\startWrite().

◆ lastDoneWrites()

Wikimedia\Rdbms\IDatabase::lastDoneWrites ( )

Get the last time the connection may have been used for a write query.

Returns
int|float|false UNIX timestamp or false
Since
1.24

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ lock()

Wikimedia\Rdbms\IDatabase::lock ( $lockName,
$method,
$timeout = 5,
$flags = 0 )

Acquire a named lock.

Named locks are not related to transactions

Parameters
string$lockNameName of lock to acquire
string$methodName of the calling method
int$timeoutAcquisition timeout in seconds (0 means non-blocking)
int$flagsBit field of IDatabase::LOCK_* constants
Returns
bool|float|null Success (bool); acquisition time (float/null) if LOCK_TIMESTAMP
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by SqlBagOStuff\deleteObjectsExpiringBefore(), and SqlBagOStuff\doLock().

◆ lockForUpdate()

Wikimedia\Rdbms\IDatabase::lockForUpdate ( $table,
$conds = '',
$fname = __METHOD__,
$options = [],
$join_conds = [] )

Lock all rows meeting the given conditions/options FOR UPDATE.

Parameters
string | string[]$tableUnqualified name of table(s) (use an array for a join)
string|IExpression|array<string,?scalar|non-empty-array<?scalar>>|array<int,string|IExpression>$conds Condition in the format of IDatabase::select() conditions
string$fnameFunction name for profiling
array$optionsOptions for select ("FOR UPDATE" is added automatically)
array$join_condsJoin conditions
Returns
int Number of matching rows found (and locked)
Exceptions
DBErrorIf an error occurs, {
See also
query}
Since
1.32
Deprecated
since 1.43; Use SelectQueryBuilder::acquireRowLocks

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ lockIsFree()

Wikimedia\Rdbms\IDatabase::lockIsFree ( $lockName,
$method )

Check to see if a named lock is not locked by any thread (non-blocking)

Parameters
string$lockNameName of lock to poll
string$methodName of method calling us
Returns
bool
Exceptions
DBErrorIf an error occurs, {
See also
query}
Since
1.20

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ newDeleteQueryBuilder()

Wikimedia\Rdbms\IDatabase::newDeleteQueryBuilder ( )

Get an DeleteQueryBuilder bound to this connection.

This is overridden by DBConnRef.

Note
A new query builder must be created per query. Query builders should not be reused since this uses a fluent interface and the state of the builder changes during the query which may cause unexpected results.
Returns
DeleteQueryBuilder

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by SqlBagOStuff\deleteAll(), MediaWiki\Deferred\LinksUpdate\LinksTable\doWrites(), LCStoreDB\finishWrite(), MediaWiki\Title\Title\purgeExpiredRestrictions(), MediaWiki\SpecialPage\QueryPage\recache(), and WatchedItemStore\removeExpired().

◆ newInsertQueryBuilder()

Wikimedia\Rdbms\IDatabase::newInsertQueryBuilder ( )

Get an InsertQueryBuilder bound to this connection.

This is overridden by DBConnRef.

Note
A new query builder must be created per query. Query builders should not be reused since this uses a fluent interface and the state of the builder changes during the query which may cause unexpected results.
Returns
InsertQueryBuilder

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by MediaWiki\User\User\addToDatabase(), MediaWiki\CommentStore\CommentStore\createComment(), JobQueueDB\doBatchPushInternal(), MediaWiki\Deferred\SiteStatsUpdate\doUpdate(), MediaWiki\Deferred\LinksUpdate\LinksTable\doWrites(), LCStoreDB\finishWrite(), and MediaWiki\SpecialPage\QueryPage\recache().

◆ newReplaceQueryBuilder()

Wikimedia\Rdbms\IDatabase::newReplaceQueryBuilder ( )

Get an ReplaceQueryBuilder bound to this connection.

This is overridden by DBConnRef.

Note
A new query builder must be created per query. Query builders should not be reused since this uses a fluent interface and the state of the builder changes during the query which may cause unexpected results.
Returns
ReplaceQueryBuilder

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by WatchedItemStore\duplicateEntry().

◆ newUpdateQueryBuilder()

Wikimedia\Rdbms\IDatabase::newUpdateQueryBuilder ( )

Get an UpdateQueryBuilder bound to this connection.

This is overridden by DBConnRef.

Note
A new query builder must be created per query. Query builders should not be reused since this uses a fluent interface and the state of the builder changes during the query which may cause unexpected results.
Returns
UpdateQueryBuilder

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by MediaWiki\Deferred\SiteStatsUpdate\cacheUpdate(), LocalFileMoveBatch\doDBUpdates(), MediaWiki\Deferred\SiteStatsUpdate\doUpdate(), MediaWiki\Title\Title\invalidateCache(), PurgeJobUtils\invalidatePages(), DeleteLocalPasswords\processUsers(), and MediaWiki\User\User\saveSettings().

◆ nextSequenceValue()

Wikimedia\Rdbms\IDatabase::nextSequenceValue ( $seqName)

Deprecated method, calls should be removed.

This was formerly used for PostgreSQL to handle self::insertId() auto-incrementing fields. It is no longer necessary since DatabasePostgres::insertId() has been reimplemented using lastval()

Implementations should return null if inserting NULL into an auto-incrementing field works, otherwise it should return an instance of NextSequenceValue and filter it on calls to relevant methods.

Deprecated
since 1.30, no longer needed
Parameters
string$seqName
Returns
null|NextSequenceValue

Implemented in Wikimedia\Rdbms\Database, Wikimedia\Rdbms\DatabasePostgres, and Wikimedia\Rdbms\DBConnRef.

◆ onAtomicSectionCancel()

Wikimedia\Rdbms\IDatabase::onAtomicSectionCancel ( callable $callback,
$fname = __METHOD__ )

Run a callback when the atomic section is cancelled.

The callback is run just after the current atomic section, any outer atomic section, or the whole transaction is rolled back.

An error is thrown if no atomic section is pending. The atomic section need not have been created with the ATOMIC_CANCELABLE flag.

Queries in the function may be running in the context of an outer transaction or may be running in AUTOCOMMIT mode. The callback should use atomic sections if necessary.

Note
do not assume that other IDatabase instances will be AUTOCOMMIT mode

The callback takes the following arguments:

  • IDatabase::TRIGGER_CANCEL or IDatabase::TRIGGER_ROLLBACK
  • This IDatabase instance
Parameters
callable$callback
string$fnameCaller name
Since
1.34

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ onTransactionCommitOrIdle()

Wikimedia\Rdbms\IDatabase::onTransactionCommitOrIdle ( callable $callback,
$fname = __METHOD__ )

Run a callback when the current transaction commits or now if there is none.

If there is a transaction and it is rolled back, then the callback is cancelled.

When transaction round mode (DBO_TRX) is set, the callback will run at the end of the round, just after all peer transactions COMMIT. If the transaction round is rolled back, then the callback is cancelled.

This IDatabase instance will start off in auto-commit mode when the callback starts. The use of other IDatabase handles from the callback should be avoided unless they are known to be in auto-commit mode. Callbacks that create transactions via begin() or startAtomic() must have matching calls to commit()/endAtomic().

Use this method only for the following purposes:

  • (a) RDBMS updates, prone to lock timeouts/deadlocks, that do not require atomicity with respect to the updates in the current transaction (if any)
  • (b) Purges to lightweight cache services due to RDBMS updates
  • (c) Updates to secondary DBs/stores that must only commit once the updates in the current transaction (if any) are committed (e.g. insert user account row to DB1, then, initialize corresponding LDAP account)

The callback takes the following arguments:

  • How the transaction ended (IDatabase::TRIGGER_COMMIT or IDatabase::TRIGGER_IDLE)
  • This IDatabase instance (since 1.32)

Callbacks will execute in the order they were enqueued.

Parameters
callable$callback
string$fnameCaller name
Exceptions
DBErrorIf an error occurs, {
See also
query}
Exceptions
ExceptionIf the callback runs immediately and an error occurs in it
Since
1.32

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ onTransactionPreCommitOrIdle()

Wikimedia\Rdbms\IDatabase::onTransactionPreCommitOrIdle ( callable $callback,
$fname = __METHOD__ )

Run a callback before the current transaction commits or now if there is none.

If there is a transaction and it is rolled back, then the callback is cancelled.

When transaction round mode (DBO_TRX) is set, the callback will run at the end of the round, just after all peer transactions COMMIT. If the transaction round is rolled back, then the callback is cancelled.

If there is no current transaction, one will be created to wrap the callback. Callbacks cannot use begin()/commit() to manage transactions. The use of other IDatabase handles from the callback should be avoided.

Use this method only for the following purposes:

  • a) RDBMS updates, prone to lock timeouts/deadlocks, that require atomicity with respect to the updates in the current transaction (if any)
  • b) Purges to lightweight cache services due to RDBMS updates

The callback takes the one argument:

  • This IDatabase instance (since 1.32)

Callbacks will execute in the order they were enqueued.

Parameters
callable$callback
string$fnameCaller name
Exceptions
DBErrorIf an error occurs, {
See also
query}
Exceptions
ExceptionIf the callback runs immediately and an error occurs in it
Since
1.22

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ onTransactionResolution()

Wikimedia\Rdbms\IDatabase::onTransactionResolution ( callable $callback,
$fname = __METHOD__ )

Run a callback when the current transaction commits or rolls back.

An error is thrown if no transaction is pending.

When transaction round mode (DBO_TRX) is set, the callback will run at the end of the round, just after all peer transactions COMMIT/ROLLBACK.

This IDatabase instance will start off in auto-commit mode when the callback starts. The use of other IDatabase handles from the callback should be avoided unless they are known to be in auto-commit mode. Callbacks that create transactions via begin() or startAtomic() must have matching calls to commit()/endAtomic().

Use this method only for the following purposes:

  • (a) Release of cooperative locks on resources
  • (b) Cancellation of in-process deferred tasks

The callback takes the following arguments:

  • How the current atomic section (if any) or overall transaction (otherwise) ended (IDatabase::TRIGGER_COMMIT or IDatabase::TRIGGER_ROLLBACK)
  • This IDatabase instance (since 1.32)

Callbacks will execute in the order they were enqueued.

Note
Use onAtomicSectionCancel() to take action as soon as an atomic section is cancelled
Parameters
callable$callback
string$fnameCaller name
Exceptions
DBErrorIf an error occurs, {
See also
query}
Exceptions
ExceptionIf the callback runs immediately and an error occurs in it
Since
1.28

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by MediaWiki\Deferred\AtomicSectionUpdate\__construct(), and MediaWiki\Deferred\AutoCommitUpdate\__construct().

◆ pendingWriteCallers()

Wikimedia\Rdbms\IDatabase::pendingWriteCallers ( )

Get the list of method names that did write queries for this transaction.

Returns
array
Since
1.27

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ pendingWriteQueryDuration()

Wikimedia\Rdbms\IDatabase::pendingWriteQueryDuration ( $type = self::ESTIMATE_TOTAL)

Get the time spend running write queries for this transaction.

High values could be due to scanning, updates, locking, and such.

Parameters
string$typeIDatabase::ESTIMATE_* constant [default: ESTIMATE_ALL]
Returns
float|false Returns false if not transaction is active
Since
1.26

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ query()

Wikimedia\Rdbms\IDatabase::query ( $sql,
$fname = __METHOD__,
$flags = 0 )

Run an SQL query statement and return the result.

If a connection loss is detected, then an attempt to reconnect will be made. For queries that involve no larger transactions or locks, they will be re-issued for convenience, provided the connection was re-established.

In new code, the query wrappers select(), insert(), update(), delete(), etc. should be used where possible, since they give much better DBMS independence and automatically quote or validate user input in a variety of contexts. This function is generally only useful for queries which are explicitly DBMS-dependent and are unsupported by the query wrappers, such as CREATE TABLE.

However, the query wrappers themselves should call this function.

Callers should avoid the use of statements like BEGIN, COMMIT, and ROLLBACK. Methods like startAtomic(), endAtomic(), and cancelAtomic() can be used instead.

Parameters
string | Query$sqlSingle-statement SQL query
string$fnameCaller name; used for profiling/SHOW PROCESSLIST comments
int$flagsBit field of IDatabase::QUERY_* constants.
Returns
bool|IResultWrapper True for a successful write query, IResultWrapper object for a successful read query, or false on failure if QUERY_SILENCE_ERRORS is set.
Exceptions
DBQueryErrorIf the query is issued, fails, and QUERY_SILENCE_ERRORS is not set.
DBExpectedErrorIf the query is not, and cannot, be issued yet (non-DBQueryError)
DBErrorIf the query is inherently not allowed (non-DBExpectedError)

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by SqlBagOStuff\createTables(), Wikimedia\Rdbms\Replication\MysqlReplicationReporter\fetchSecondsSinceHeartbeat(), Wikimedia\Rdbms\Replication\MysqlReplicationReporter\getLagFromSlaveStatus(), Wikimedia\Rdbms\Replication\MysqlReplicationReporter\getServerGTIDs(), Wikimedia\Rdbms\Replication\MysqlReplicationReporter\getServerId(), Wikimedia\Rdbms\Replication\MysqlReplicationReporter\getServerRoleStatus(), Wikimedia\Rdbms\Replication\MysqlReplicationReporter\getServerUUID(), Wikimedia\Rdbms\Replication\MysqlReplicationReporter\primaryPosWait(), and MwSql\sqlDoQuery().

◆ replace()

Wikimedia\Rdbms\IDatabase::replace ( $table,
$uniqueKeys,
$rows,
$fname = __METHOD__ )

Insert row(s) into a table, in the provided order, while deleting conflicting rows.

Conflicts are determined by the provided unique indexes. Note that it is possible for the provided rows to conflict even among themselves; it is preferable for the caller to de-duplicate such input beforehand.

Note some important implications of the deletion semantics:

  • If the table has an AUTOINCREMENT column and $rows omit that column, then any conflicting existing rows will be replaced with newer having higher values for that column, even if nothing else changed.
  • There might be worse contention than upsert() due to the use of gap-locking. This does not apply to RDBMS types that use predicate locking nor those that just lock the whole table or databases anyway.

This operation will be seen by affectedRows()/insertId() as one query statement, regardless of how many statements are actually sent by the class implementation.

Access: internal
callers outside of rdbms library should use ReplaceQueryBuilder instead.
Parameters
string$tableUnqualified name of table
string | string[] | string[][]$uniqueKeysColumn name or non-empty list of column name lists that define all applicable unique keys on the table. There must only be one such key. Each unique key on the table is "applicable" unless either:
  • It involves an AUTOINCREMENT column for which no values are assigned in $rows
  • It involves a UUID column for which newly generated UUIDs are assigned in $rows
array | array[]$rowsRow(s) to insert, in the form of either:
  • A string-keyed map of (column name => value) defining a new row. Values are treated as literals and quoted appropriately; null is interpreted as NULL. Columns belonging to a key in $uniqueKeys must be defined here and non-null.
  • An integer-keyed list of such string-keyed maps, defining a list of new rows. The keys in each map must be identical to each other and in the same order. The rows must not collide with each other.
string$fnameCalling function name (use METHOD) for logs/profiling
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, Wikimedia\Rdbms\DatabaseMySQL, Wikimedia\Rdbms\DatabaseSqlite, and Wikimedia\Rdbms\DBConnRef.

◆ rollback()

Wikimedia\Rdbms\IDatabase::rollback ( $fname = __METHOD__,
$flush = self::FLUSHING_ONE )

Rollback a transaction previously started using begin()

Only call this from code with outer transaction scope. See https://www.mediawiki.org/wiki/Database_transactions for details. Nesting of transactions is not supported. If a serious unexpected error occurs, throwing an Exception is preferable, using a pre-installed error handler to trigger rollback (in any case, failure to issue COMMIT will cause rollback server-side).

Query, connection, and onTransaction* callback errors will be suppressed and logged.

Parameters
string$fnameCalling function name
string$flushFlush flag, set to a situationally valid IDatabase::FLUSHING_* constant to disable warnings about explicitly rolling back implicit transactions. This will silently break any ongoing explicit transaction. Only set the flush flag if you are sure that it is safe to ignore these warnings in your context.
Exceptions
DBErrorIf an error occurs, {
See also
query}
Since
1.23 Added $flush parameter

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by Wikimedia\Rdbms\LoadBalancer\rollbackPrimaryChanges(), and Maintenance\rollbackTransaction().

◆ serverIsReadOnly()

Wikimedia\Rdbms\IDatabase::serverIsReadOnly ( )
Returns
bool Whether this DB server is running in server-side read-only mode
Exceptions
DBErrorIf an error occurs, {
See also
query}
Since
1.28

Implemented in Wikimedia\Rdbms\Database, Wikimedia\Rdbms\DatabaseMySQL, Wikimedia\Rdbms\DatabasePostgres, Wikimedia\Rdbms\DatabaseSqlite, and Wikimedia\Rdbms\DBConnRef.

◆ setLBInfo()

Wikimedia\Rdbms\IDatabase::setLBInfo ( $nameOrArray,
$value = null )

Set the entire array or a particular key of the managing load balancer info array.

Keys matching the IDatabase::LB_* constants are also used internally by subclasses

Access: internal
should not be called outside of rdbms library.
Parameters
array | string$nameOrArrayThe new array or the name of a key to set
array | mixed | null$valueIf $nameOrArray is a string, the new key value (null to unset)

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ setSessionOptions()

Wikimedia\Rdbms\IDatabase::setSessionOptions ( array $options)

Override database's default behavior.

Not all options are supported on all database backends; unsupported options are silently ignored.

$options include:

  • 'connTimeout': Set the connection timeout value in seconds. May be useful for very long batch queries such as full-wiki dumps, where a single query reads out over hours or days. Only supported on MySQL and MariaDB.
  • 'groupConcatMaxLen': Maximum length of a GROUP_CONCAT() result. Only supported on MySQL and MariaDB.
Parameters
array$options
Returns
void
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, Wikimedia\Rdbms\DatabaseMySQL, and Wikimedia\Rdbms\DBConnRef.

Referenced by BackupDumper\backupDb().

◆ setTransactionListener()

Wikimedia\Rdbms\IDatabase::setTransactionListener ( $name,
callable $callback = null )

Run a callback after each time any transaction commits or rolls back.

The callback takes two arguments:

  • IDatabase::TRIGGER_COMMIT or IDatabase::TRIGGER_ROLLBACK
  • This IDatabase object Callbacks must commit any transactions that they begin.

Registering a callback here will not affect writesOrCallbacks() pending.

Since callbacks from this or onTransactionCommitOrIdle() can start and end transactions, a single call to IDatabase::commit might trigger multiple runs of the listener callbacks.

Parameters
string$nameCallback name
callable | null$callbackUse null to unset a listener
Since
1.28

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ startAtomic()

Wikimedia\Rdbms\IDatabase::startAtomic ( $fname = __METHOD__,
$cancelable = self::ATOMIC_NOT_CANCELABLE )

Begin an atomic section of SQL statements.

Start an implicit transaction if no transaction is already active, set a savepoint (if $cancelable is ATOMIC_CANCELABLE), and track the given section name to enforce that the transaction is not committed prematurely. The end of the section must be signified exactly once, either by endAtomic() or cancelAtomic(). Sections can have have layers of inner sections (sub-sections), but all sections must be ended in order of innermost to outermost. Transactions cannot be started or committed until all atomic sections are closed.

ATOMIC_CANCELABLE is useful when the caller needs to handle specific failure cases by discarding the section's writes. This should not be used for failures when:

  • upsert() could easily be used instead
  • insert() with IGNORE could easily be used instead
  • select() with FOR UPDATE could be checked before issuing writes instead
  • The failure is from code that runs after the first write but doesn't need to
  • The failures are from contention solvable via onTransactionPreCommitOrIdle()
  • The failures are deadlocks; the RDBMs usually discard the whole transaction
Note
callers must use additional measures for situations involving two or more (peer) transactions (e.g. updating two database servers at once). The transaction and savepoint logic of this method only applies to this specific IDatabase instance.

Example usage:

// Start a transaction if there isn't one already
$dbw->startAtomic( __METHOD__ );
// Serialize these thread table updates
$dbw->select( 'thread', '1', [ 'td_id' => $tid ], __METHOD__, 'FOR UPDATE' );
// Add a new comment for the thread
$dbw->insert( 'comment', $row, __METHOD__ );
$cid = $db->insertId();
// Update thread reference to last comment
$dbw->update( 'thread', [ 'td_latest' => $cid ], [ 'td_id' => $tid ], __METHOD__ );
// Demark the end of this conceptual unit of updates
$dbw->endAtomic( __METHOD__ );

Example usage (atomic changes that might have to be discarded):

// Start a transaction if there isn't one already
$sectionId = $dbw->startAtomic( __METHOD__, $dbw::ATOMIC_CANCELABLE );
// Create new record metadata row
$dbw->insert( 'records', $row, __METHOD__ );
// Figure out where to store the data based on the new row's ID
$path = $recordDirectory . '/' . $dbw->insertId();
// Write the record data to the storage system
$status = $fileBackend->create( [ 'dst' => $path, 'content' => $data ] );
if ( $status->isOK() ) {
// Try to cleanup files orphaned by transaction rollback
$dbw->onTransactionResolution(
function ( $type ) use ( $fileBackend, $path ) {
if ( $type === IDatabase::TRIGGER_ROLLBACK ) {
$fileBackend->delete( [ 'src' => $path ] );
}
},
__METHOD__
);
// Demark the end of this conceptual unit of updates
$dbw->endAtomic( __METHOD__ );
} else {
// Discard these writes from the transaction (preserving prior writes)
$dbw->cancelAtomic( __METHOD__, $sectionId );
}
Since
1.23
Parameters
string$fname
string$cancelablePass self::ATOMIC_CANCELABLE to use a savepoint and enable self::cancelAtomic() for this section.
Returns
AtomicSectionIdentifier section ID token
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\DBConnRef, and Wikimedia\Rdbms\Database.

Referenced by JobQueueDB\doBatchPushInternal(), and LCStoreDB\finishWrite().

◆ trxLevel()

Wikimedia\Rdbms\IDatabase::trxLevel ( )

Gets the current transaction level.

Historically, transactions were allowed to be "nested". This is no longer supported, so this function really only returns a boolean.

Returns
int The previous value

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by Wikimedia\Rdbms\Replication\ReplicationReporter\getRecordedTransactionLagStatus().

◆ trxTimestamp()

Wikimedia\Rdbms\IDatabase::trxTimestamp ( )

Get the UNIX timestamp of the time that the transaction was established.

This can be used to reason about the staleness of SELECT data in REPEATABLE-READ transaction isolation level. Callers can assume that if a view-snapshot isolation is used, then the data read by SQL queries is at least up to date to that point (possibly more up-to-date since the first SELECT defines the snapshot).

Returns
float|null Returns null if there is not active transaction
Since
1.25

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ unlock()

Wikimedia\Rdbms\IDatabase::unlock ( $lockName,
$method )

Release a lock.

Named locks are not related to transactions

Parameters
string$lockNameName of lock to release
string$methodName of the calling method
Returns
bool Success
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by SqlBagOStuff\deleteObjectsExpiringBefore(), and SqlBagOStuff\doUnlock().

◆ update()

Wikimedia\Rdbms\IDatabase::update ( $table,
$set,
$conds,
$fname = __METHOD__,
$options = [] )

Update all rows in a table that match a given condition.

This operation will be seen by affectedRows()/insertId() as one query statement, regardless of how many statements are actually sent by the class implementation.

Access: internal
callers outside of rdbms library should use UpdateQueryBuilder instead.
Parameters
string$tableUnqualified name of table
array$setCombination map/list where each string-keyed entry maps a column to a literal assigned value and each integer-keyed value is a SQL expression in the format of a column assignment within UPDATE...SET. The (column => value) entries are convenient due to automatic value quoting and conversion of null to NULL. The SQL assignment format is useful for updates like "column = column + X". All assignments have no defined execution order, so they should not depend on each other. Do not modify AUTOINCREMENT or UUID columns in assignments.
string|IExpression|array<string,?scalar|non-empty-array<?scalar>>|array<int,string|IExpression>$conds Condition in the format of IDatabase::select() conditions. In order to prevent possible performance or replication issues or damaging a data accidentally, an empty condition for 'update' queries isn't allowed. IDatabase::ALL_ROWS should be passed explicitly in order to update all rows.
string$fnameCalling function name (use METHOD) for logs/profiling
string | array$optionsCombination map/list where each string-keyed entry maps a non-boolean option to the option parameters and each integer-keyed value is the name of a boolean option. Supported options are:
  • IGNORE: Boolean: skip update of rows that would cause unique key conflicts. IDatabase::affectedRows() includes all matching rows, that includes also rows not updated due to key conflict.
Returns
bool Return true if no exception was thrown (deprecated since 1.33)
Exceptions
DBErrorIf an error occurs, {
See also
query}

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

◆ upsert()

Wikimedia\Rdbms\IDatabase::upsert ( $table,
array $rows,
$uniqueKeys,
array $set,
$fname = __METHOD__ )

Upsert row(s) into a table, in the provided order, while updating conflicting rows.

Conflicts are determined by the provided unique indexes. Note that it is possible for the provided rows to conflict even among themselves; it is preferable for the caller to de-duplicate such input beforehand.

This operation will be seen by affectedRows()/insertId() as one query statement, regardless of how many statements are actually sent by the class implementation.

Access: internal
callers outside of rdbms library should use InsertQueryBuilder instead.
Parameters
string$tableUnqualified name of table
array | array[]$rowsRow(s) to insert, in the form of either:
  • A string-keyed map of (column name => value) defining a new row. Values are treated as literals and quoted appropriately; null is interpreted as NULL. Columns belonging to a key in $uniqueKeys must be defined here and non-null.
  • An integer-keyed list of such string-keyed maps, defining a list of new rows. The keys in each map must be identical to each other and in the same order. The rows must not collide with each other.
string | string[] | string[][]$uniqueKeysColumn name or non-empty list of column name lists that define all applicable unique keys on the table. There must only be one such key. Each unique key on the table is "applicable" unless either:
  • It involves an AUTOINCREMENT column for which no values are assigned in $rows
  • It involves a UUID column for which newly generated UUIDs are assigned in $rows
array$setCombination map/list where each string-keyed entry maps a column to a literal assigned value and each integer-keyed value is a SQL assignment expression of the form "<unquoted alphanumeric column> = <SQL expression>". The (column => value) entries are convenient due to automatic value quoting and conversion of null to NULL. The SQL assignment entries are useful for updates like "column = column + X". All of the assignments have no defined execution order, so callers should make sure that they not depend on each other. Do not modify AUTOINCREMENT or UUID columns in assignments, even if they are just "secondary" unique keys. For multi-row upserts, use buildExcludedValue() to reference the value of a column from the corresponding row in $rows that conflicts with the current row.
string$fnameCalling function name (use METHOD) for logs/profiling
Exceptions
DBErrorIf an error occurs, {
See also
query}
Since
1.22

Implemented in Wikimedia\Rdbms\Database, Wikimedia\Rdbms\DatabaseMySQL, and Wikimedia\Rdbms\DBConnRef.

◆ writesOrCallbacksPending()

Wikimedia\Rdbms\IDatabase::writesOrCallbacksPending ( )

Whether there is a transaction open with either possible write queries or unresolved pre-commit/commit/resolution callbacks pending.

This does not count recurring callbacks, e.g. from setTransactionListener().

Returns
bool

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Referenced by Wikimedia\Rdbms\TransactionManager\trxCheckBeforeClose().

◆ writesPending()

Wikimedia\Rdbms\IDatabase::writesPending ( )
Returns
bool Whether there is a transaction open with possible write queries
Since
1.27

Implemented in Wikimedia\Rdbms\Database, and Wikimedia\Rdbms\DBConnRef.

Member Data Documentation

◆ LOCK_TIMESTAMP

const Wikimedia\Rdbms\IDatabase::LOCK_TIMESTAMP = 1

Flag to return the lock acquisition timestamp (null if not acquired)

Definition at line 69 of file IDatabase.php.


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