This release marks a major step in the development team's mastery of the source code we inherited from Berkeley. You will see we are now easily adding major features, thanks to the increasing size and experience of our world-wide development team.
Here is a brief summary of the more notable changes:
This removes our old table-level locking, and replaces it with a locking system that is superior to most commercial database systems. In a traditional system, each row that is modified is locked until committed, preventing reads by other users. MVCC uses the natural multi-version nature of PostgreSQL to allow readers to continue reading consistent data during writer activity. Writers continue to use the compact pg_log transaction system. This is all performed without having to allocate a lock for every row like traditional database systems. So, basically, we no longer are restricted by simple table-level locking; we have something better than row-level locking.
pg_dump takes advantage of the new MVCC features to give a consistant database dump/backup while the database stays online and available for queries.
We now have a true numeric data type, with user-specified precision.
Temporary tables are guaranteed to have unique names within a database session, and are destroyed on session exit.
We now have CASE, INTERSECT, and EXCEPT statement support. We have new LIMIT/OFFSET, SET TRANSACTION ISOLATION LEVEL, SELECT ... FOR UPDATE, and an improved LOCK TABLE command.
We continue to speed up PostgreSQL, thanks to the variety of talents within our team. We have sped up memory allocation, optimization, table joins, and row transfer routines.
We continue to expand our port list, this time including WinNT/ix86 and NetBSD/arm32.
Most interfaces have new versions, and existing functionality has been improved.
New and updated material is present throughout the documentation. New FAQs have been contributed for SGI and AIX platforms. The Tutorial has introductory information on SQL from Stefan Simkovics. For the User's Guide, there are reference pages covering the postmaster and more utility programs, and a new appendix contains details on date/time behavior. The Administrator's Guide has a new chapter on troubleshooting from Tom Lane. And the Programmer's Guide has a description of query processing, also from Stefan, and details on obtaining the Postgres source tree via anonymous CVS and CVSup.
A dump/restore using pg_dump is required for those wishing to migrate data from any previous release of Postgres. pg_upgrade can not be used to upgrade to this release because the on-disk structure of the tables has changed compared to previous releases.
The new Multi-Version Concurrency Control (MVCC) features can give somewhat different behaviors in multi-user environments. Read and understand the following section to ensure that your existing applications will give you the behavior you need.
Because readers in 6.5 don't lock data, regardless of transaction isolation level, data read by one transaction can be overwritten by another. In other words, if a row is returned by SELECT it doesn't mean that this row really exists at the time it is returned (i.e. sometime after the statement or transaction began) nor that the row is protected from being deleted or updated by concurrent transactions before the current transaction does a commit or rollback.
To ensure the actual existence of a row and protect it against concurrent updates one must use SELECT FOR UPDATE or an appropriate LOCK TABLE statement. This should be taken into account when porting applications from previous releases of Postgres and other environments.
Keep the above in mind if you are using contrib/refint.* triggers for referential integrity. Additional technics are required now. One way is to use LOCK parent_table IN SHARE ROW EXCLUSIVE MODE command if a transaction is going to update/delete a primary key and use LOCK parent_table IN SHARE MODE command if a transaction is going to update/insert a foreign key.
Note: Note that if you run a transaction in SERIALIZABLE mode then you must execute the LOCK commands above before execution of any DML statement (SELECT/INSERT/DELETE/UPDATE/FETCH/COPY_TO) in the transaction.
These inconveniences will disappear in the future when the ability to read dirty (uncommitted) data (regardless of isolation level) and true referential integrity will be implemented.
Bug Fixes --------- Fix text<->float8 and text<->float4 conversion functions(Thomas) Fix for creating tables with mixed-case constraints(Billy) Change exp()/pow() behavior to generate error on underflow/overflow(Jan) Fix bug in pg_dump -z Memory overrun cleanups(Tatsuo) Fix for lo_import crash(Tatsuo) Adjust handling of data type names to suppress double quotes(Thomas) Use type coersion for matching columns and DEFAULT(Thomas) Fix deadlock so it only checks once after one second of sleep(Bruce) Fixes for aggregates and PL/pgsql(Hiroshi) Fix for subquery crash(Vadim) Fix for libpq function PQfnumber and case-insensitive names(Bahman Rafatjoo) Fix for large object write-in-middle, no extra block, memory consumption(Tatsuo) Fix for pg_dump -d or -D and quote special characters in INSERT Repair serious problems with dynahash(Tom) Fix INET/CIDR portability problems Fix problem with selectivity error in ALTER TABLE ADD COLUMN(Bruce) Fix executor so mergejoin of different column types works(Tom) Fix for Alpha OR selectivity bug Fix OR index selectivity problem(Bruce) Fix so \d shows proper length for char()/varchar()(Ryan) Fix tutorial code(Clark) Improve destroyuser checking(Oliver) Fix for Kerberos(Rodney McDuff) Fix for dropping database while dirty buffers(Bruce) Fix so sequence nextval() can be case-sensitive(Bruce) Fix !!= operator Drop buffers before destroying database files(Bruce) Fix case where executor evaluates functions twice(Tatsuo) Allow sequence nextval actions to be case-sensitive(Bruce) Fix optimizer indexing not working for negative numbers(Bruce) Fix for memory leak in executor with fjIsNull Fix for aggregate memory leaks(Erik Riedel) Allow username containing a dash GRANT permissions Cleanup of NULL in inet types Clean up system table bugs(Tom) Fix problems of PAGER and \? command(Masaaki Sakaida) Reduce default multi-segment file size limit to 1GB(Peter) Fix for dumping of CREATE OPERATOR(Tom) Fix for backward scanning of cursors(Hiroshi Inoue) Fix for COPY FROM STDIN when using \i(Tom) Fix for subselect is compared inside an expression(Jan) Fix handling of error reporting while returning rows(Tom) Fix problems with reference to array types(Tom,Jan) Prevent UPDATE SET oid(Jan) Fix pg_dump so -t option can handle case-sensitive tablenames Fixes for GROUP BY in special cases(Tom, Jan) Fix for memory leak in failed queries(Tom) DEFAULT now supports mixed-case identifiers(Tom) Fix for multi-segment uses of DROP/RENAME table, indexes(Ole Gjerde) Disable use of pg_dump with both -o and -d options(Bruce) Allow pg_dump to properly dump GROUP permissions(Bruce) Fix GROUP BY in INSERT INTO table SELECT * FROM table2(Jan) Fix for computations in views(Jan) Fix for aggregates on array indexes(Tom) Fix for DEFAULT handles single quotes in value requiring too many quotes Fix security problem with non-super users importing/exporting large objects(Tom) Rollback of transaction that creates table cleaned up properly(Tom) Fix to allow long table and column names to generate proper serial names(Tom) Enhancements ------------ Add "vacuumdb" utility Speed up libpq by allocating memory better(Tom) EXPLAIN all indices used(Tom) Implement CASE, COALESCE, NULLIF expression(Thomas) New pg_dump table output format(Constantin) Add string min()/max() functions(Thomas) Extend new type coersion techniques to aggregates(Thomas) New moddatetime contrib(Terry) Update to pgaccess 0.96(Constantin) Add routines for single-byte "char" type(Thomas) Improved substr() function(Thomas) Improved multi-byte handling(Tatsuo) Multi-version concurrency control/MVCC(Vadim) New Serialized mode(Vadim) Fix for tables over 2gigs(Peter) New SET TRANSACTION ISOLATION LEVEL(Vadim) New LOCK TABLE IN ... MODE(Vadim) Update ODBC driver(Byron) New NUMERIC data type(Jan) New SELECT FOR UPDATE(Vadim) Handle "NaN" and "Infinity" for input values(Jan) Improved date/year handling(Thomas) Improved handling of backend connections(Magnus) New options ELOG_TIMESTAMPS and USE_SYSLOG options for log files(Massimo) New TCL_ARRAYS option(Massimo) New INTERSECT and EXCEPT(Stefan) New pg_index.indisprimary for primary key tracking(D'Arcy) New pg_dump option to allow dropping of tables before creation(Brook) Speedup of row output routines(Tom) New READ COMMITTED isolation level(Vadim) New TEMP tables/indexes(Bruce) Prevent sorting if result is already sorted(Jan) New memory allocation optimization(Jan) Allow psql to do \p\g(Bruce) Allow multiple rule actions(Jan) Added LIMIT/OFFSET functionality(Jan) Improve optimizer when joining a large number of tables(Bruce) New intro to SQL from S. Simkovics' Master's Thesis (Stefan, Thomas) New intro to backend processing from S. Simkovics' Master's Thesis (Stefan) Improved int8 support(Ryan Bradetich, Thomas, Tom) New routines to convert between int8 and text/varchar types(Thomas) New bushy plans, where meta-tables are joined(Bruce) Enable right-hand queries by default(Bruce) Allow reliable maximum number of backends to be set at configure time (--with-maxbackends and postmaster switch (-N backends))(Tom) GEQO default now 10 tables because of optimizer speedups(Tom) Allow NULL=Var for MS-SQL portability(Michael, Bruce) Modify contrib check_primary_key() so either "automatic" or "dependent"(Anand) Allow psql \d on a view show query(Ryan) Speedup for LIKE(Bruce) Ecpg fixes/features, see src/interfaces/ecpg/ChangeLog file(Michael) JDBC fixes/features, see src/interfaces/jdbc/CHANGELOG(Peter) Make % operator have precedence like /(Bruce) Add new postgres -O option to allow system table structure changes(Bruce) Update contrib/pginterface/findoidjoins script(Tom) Major speedup in vacuum of deleted rows with indexes(Vadim) Allow non-SQL functions to run different versions based on arguments(Tom) Add -E option that shows actual queries sent by \dt and friends(Masaaki Sakaida) Add version number in startup banners for psql(Masaaki Sakaida) New contrib/vacuumlo removes large objects not referenced(Peter) New initialization for table sizes so non-vacuumed tables perform better(Tom) Improve error messages when a connection is rejected(Tom) Support for arrays of char() and varchar() fields(Massimo) Overhaul of hash code to increase reliability and performance(Tom) Update to PyGreSQL 2.4(D'Arcy) Changed debug options so -d4 and -d5 produce different node displays(Jan) New pg_options: pretty_plan, pretty_parse, pretty_rewritten(Jan) Better optimization statistics for system table access(Tom) Better handling of non-default block sizes(Massimo) Improve GEQO optimizer memory consumption(Tom) UNION now suppports ORDER BY of columns not in target list(Jan) Major libpq++ improvements(Vince Vielhaber) pg_dump now uses -z(ACL's) as default(Bruce) backend cache, memory speedups(Tom) have pg_dump do everything in one snapshot transaction(Vadim) fix for large object memory leakage, fix for pg_dumping(Tom) INET type now respects netmask for comparisons Make VACUUM ANALYZE only use a readlock(Vadim) Allow VIEWs on UNIONS(Jan) pg_dump now can generate consistent snapshots on active databases(Vadim) Source Tree Changes ------------------- Improve port matching(Tom) Portability fixes for SunOS Add NT/Win32 backend port and enable dynamic loading(Magnus and Daniel Horak) New port to Cobalt Qube(Mips) running Linux(Tatsuo) Port to NetBSD/m68k(Mr. Mutsuki Nakajima) Port to NetBSD/sun3(Mr. Mutsuki Nakajima) Port to NetBSD/macppc(Toshimi Aoki) Fix for tcl/tk configuration(Vince) Removed CURRENT keyword for rule queries(Jan) NT dynamic loading now works(Daniel Horak) Add ARM32 support(Andrew McMurry) Better support for HPUX 11 and Unixware Improve file handling to be more uniform, prevent file descriptor leak(Tom) New install commands for plpgsql(Jan)