RJDBC
About RJDBC
|
News/Changelog0.3-0* RJDBC now supports options that can be defined at driver or connection level which affect the defaults to some methods such as fetch() or dbGetQuery(). They can be set either as additional arguments to JDBC() or as a list passed in dbOptions= parameter to dbConnect(). The connections inherit driver parameters (if any). The options are in principle available to any further functions and can be checked using the new dbOption() generic which is akin to getOption() in R, but it is mostly used by methods in RJDBC. Options currently used by RJDBC: - autocommit - logical, if set during dbConnect() then auto-commit on the connection is explicitly enabled (TRUE) or disabled (FALSE) regardless of the initial state. If not set, the auto-commit state is determined by the driver and the default in JDBC is enabled. This option is provided as a last-resort override for drivers that do not support this option in the connection string and should be rarely used. Note that auto-commit is always disabled for the duration of transactions bracketed by dbBegin() and dbCommit()/dbRollback(). - fetch.block (default: 2048L) - see fetch(block=) - fetch.lossy (default: TRUE) - see fetch(lossy=) - fetch.posix.ts (default: TRUE) - see fetch(posix.ts=) - fetch.tz (default: "") - see fetch(tz=) For example, to avoid lossy numeric conversions for all connections of a given driver, it is possible to load the driver such as JDBC("org.postgresql.Driver", "postgresql-42.3.3.jar", fetch.lossy=FALSE) Similarly, it is possible to do so only for a particular connection, such as if the MySQL driver was loaded without such option then dbConnect(MySQL, "jdbc:mysql://localhost/foo", dbOptions=list(fetch.lossy=FALSE)) * Values which are representable as timestamps are now retrieved in POSIX form (seconds since the epoch, same as POSIXct in R). This behavior can be disabled using the posix.ts=FALSE argument (see below) to match RDJBC 0.2 which used string representation of date/time values. * BIGINT type (64-bit integer) is now correctly handled as non-representable (since it exceeds R's 53-bit double precision) and will be returned as charecter vector unless lossy=TRUE. Note that if non-lossy efficient storage is desired, it is it possible to use lossy=FALSE and bit64 package registered as a type map: dbSetTypeMaps(BIGINT=bit64::as.integer64) * fetch() (and by extension dbGetQuery() and dbReadTable()) gains new arguments: lossy: logical, if FALSE then result columns which contain numbers that may not be representable in IEEE double format are returned as strings. If TRUE then any numeric or decimal values are converted to doubles even if that may result in a loss of precision. (see also #92) The default is TRUE unless set by fetch.lossy dbOption (see above). posix.ts: logical, if TRUE (default) then result columns with date/time values are fetched as timestamps in POSIX form. This is most efficient and identical to POSIXct representation in R, but it relies on drivers and DBMS handling time zones properly. If FALSE then values are returned as strings which is less efficient, but may be used to uncover issues in date/time handling. tz: string, defaults to "" which denotes R's current time zone. Specifies the time zone in which the resulting POSIXct date should be displayed (same as using .POSIXct(x, tz=tz)). Note that this is only relevant for printing. The actual timestamp is always an absolute point in time, but may be shown in different time zones. * JDBC connections now allow auto-commit mode to be disabled either by options to the driver (connection string or corresponding property - see you driver's docummentation if supported) or by using the autocommit dbOption (see above). Previously, the auto-commit mode was required to be always enabled (see #97). Note: the auto-commit mode should NOT be changed by SQL commands, otherwise dbBegin()/dbCommit() won't know which state to restore. 0.2-11 * add field.types= argument to dbWriteTable() (#87) 0.2-10 2022-03-24 * dbSendQuery() and dbSendUpdate() now remove named arguments only from ... but not from list= to enable single-row updates from data frames along the lines of list=iris[1,]. * dbWriteTable() failed for data frames with exactly one row, because it used the form shown above for updates which will fail with JDBC ERROR: Parameter at position 1 is not set since all elements are named and thus none were set. * merging arguments from ... and list= was not always done correctly, in some cases list= would overwrite ... arguments. (Note: all of the above are 0.2-9 regressions related to the changes in handling of prepared statement arguments) * add dbIsValid() methods for JDBCConnection and JDBCResult (#89) 0.2-9 2022-03-16 * All Java-related errors are signalled by throwing error conditions. If a Java error occurred, the condition will be of the class JDBC.exception with the Java class names of the excption as subclasses. Such objects have following slots: - desc: string, description of the operation error - jex: Java object, reference to the Java exception object - statement: string (or NULL), executed statement This allows the programmatic use of tryCatch() to selectively catch exceptions and/or to use further Java utilities, such as e$jex$printStackTrace() * add schema= to dbExistsTable() (#84) * check precision of numeric types and revert to string representation if it is more than 15 to avoid loss of precision at conversion time. (#83) * NAs are supported in updates (#64) * add findDrivers() function which uses Java Service Provider mechanism (in Java 1.6 and higher) to find all JDBC driver classes on the class path. * dbConnection() used to always try DriverManager first (without properties) and only on failure it would use the driver. However, DriverManager is static and never updates the driver list, so it is inherently unreliable. Therefore we now prefer the driver if specified and DriverManager will only be used if the driver is NULL. (inspired by #45) * Use collect + join when fetch()ing results of unknown length. This should improve performance for fetching results with large number of rows. (mentioned in #50) * Named arguments are removed from dbSendUpdate and query calls when populating prepared statements to allow for future named arguments in the methods. * dbBegin() is implemented to disable auto-commit. Analogously, dbRollback() and dbCommit() perform the corresponding function and then re-enable auto-commit. (see also #58) * dbReadTable() passes ... through to dbGetQuery 0.2-8 2020-03-06 * re-use Java helper for subsequent fetch() calls on the same result set which should make them more efficient. * set has.completed only if the fetch() has encountered end of the result set (#73) * add use.label argument to fetch() - if set to FALSE then column names are used instead of labels (#74) * change both semantics and defaults for append=FALSE and overwrite=FALSE in dbWriteTable() to match what is documented in DBI: append=TRUE no longer requires the existence of the target table and the options are mutally exclusive (i.e. append=TRUE, overwrite=TRUE is illegal). * dbWriteTable() also gains force=TRUE option (disabled by default) which skips the existence check for the target and proceeds with removal and/or insertion regardless. This can be useful with append=TRUE where the target is not a table and thus would not pass the dbExistsTable() check (see #27). * dbRemoveTable() returns TRUE on success (#20). On failure it will fail with an error if silent=FALSE (as it always did so far) or return FALSE if silent=TRUE. * JDBC errors are reported such that the statement is shown at the end. Previously the statement preceded the JDBC error, so long statements caused errors to be truncated (#32). This also opens up the possibility of more structured error objects. * n, block and use.label are now passed through from dbGetQuery() to fetch() (#44) 0.2-7 2018-01-24 * cosmetic changes to appease CRAN 0.2-6 2018-01-05 * move java-src to java * add schema= argument to dbGetTables() and dbListTables() * support vectorized version of dbSendUpdate() which uses batch-inserts for more efficient loading. Note that only prepared statements work for now. * use column labels instead of names (#36) * dbColumnInfo() will return the column label in `name' and column name in `field.name' (which is optional, for compatbility). 0.2-5 2014-12-18 * fix bug in dbUnloadDriver() returning NULL instead of FALSE * added block argument to fetch() currently defaulting to 2048 which controls the fetch size. Note that 0.2-4 has set this to the stride which was much too big for some databases therefore it is now configurable. (issue #10) * recover if setFetchSize() fails since some drivers don't take it as a hint and fail regardless (issue #11) 0.2-4 2014-06-26 * set fetch size to match the capacity for better performance with some drivers. (thanks Jesse Rohland) * close DB objects explicitly (issue #4) to avoid Oracle DB running out of resources * add support for dbHasCompleted() even in older DBI versions 0.2-3 2013-12-12 * fix duplicate connection object (issue #1) 0.2-2 2013-12-03 * add dbGetTables() and dbGetFields() which are similar to dbListTable() and dbListFields() but return a full data frame as obtained from the corresponding JDBC calls. * add support for stored procedure call JDBC syntax. Currently only IN parameters are supported since DBI doesn't provide a way to retrieve OUT parameters. * JDBC now uses path.expand() on the classPath components 0.2-1 2012-11-30 * add support for dbWriteTable(..., append=TRUE) * pass any additional arguments to dbConnect() as properties to the connection * use prepared statements only if parameters are passed, otherwise use simple statements (this allows to work around bugs in drivers that do not support prepared statements properly) * dbGetQuery() explicitly closes the statement before it returns to work arround issues in the Teradata driver. 0.2-0 2011-05-16 * use Java code to fetch results -- it should result in much higher throughput in pulling result sets from the database. It is also more efficient to use fetch() with n specified (especially if you know it in advance) than relying on n = -1. The latter will try first run with n = 32k and then continue with n = 512k chunks and paste the results which is inherently slow (still much faster than the 0.1-x way which was fetching one record at a time). 0.1-6 2011-03-09 * fix a typo in dbClearResult * map NAs in query parameters into NULLs (thanks to Axel Klenk) * explicitly close statements right away (Oracle seems to need this) (thanks to Axel Klenk) * add "stat" slot (Statement object) to JDBCResult class to make sure the statement lives long enough to not close the result set * convert NULLs into NAs in numeric columns 0.1-5 2007-10-01 * instantiate driver's class and use a call to `connect' as a fall-back if the DriverManager fails to find a proper driver. * fix SQL syntax in dbWriteTable (superfluous semicolon) * fix a typo in prepared statement code (thanks to Carl Grant for the last two bugfixes) 0.1-4 2007-03-05 * added SQL error information to most JDBC errors * use .jfindClass to load drivers 0.1-3 2007-02-21 * remove beforeFirst() as some DBMS don't like it. This also allows a sequential use of fetch() with a limit to load data in chunks. * fix a bug that caused NULL entries in string columns to fail in fetch() 0.1-2 2007-01-23 * fix dbDisconnect return value and update manuals 0.1-1 2007-01-17 * added pre-compiled statements and proper dbWriteTable 0.1-0 (not released) * drop pgsql access, switch to DBI 0.0-3 2006-03-05 * add direct insert support for pgsql 0.0-2 2005-10-12 * added converson of tables to data.frames 0.0-1 2005-05-02 * added type conversion (int, double, string) 0.0-0 2004-11-02 * first public release |