Last updated: Thu Jul 27 12:52:35 EDT 2000
Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
The most recent version of this document can be viewed at http://www.PostgreSQL.org/docs/faq-english.html.
Platform-specific questions are answered at http://www.PostgreSQL.org/docs/.
IN so slow?PostgreSQL is an enhancement of the POSTGRES database management system, a next-generation DBMS research prototype. While PostgreSQL retains the powerful data model and rich data types of POSTGRES, it replaces the PostQuel query language with an extended subset of SQL. PostgreSQL is free and the complete source is available.
PostgreSQL development is performed by a team of Internet developers who all subscribe to the PostgreSQL development mailing list. The current coordinator is Marc G. Fournier (scrappy@PostgreSQL.org). (See below on how to join). This team is now responsible for all development of PostgreSQL.
The authors of PostgreSQL 1.01 were Andrew Yu and Jolly Chen. Many others have contributed to the porting, testing, debugging and enhancement of the code. The original Postgres code, from which PostgreSQL is derived, was the effort of many graduate students, undergraduate students, and staff programmers working under the direction of Professor Michael Stonebraker at the University of California, Berkeley.
The original name of the software at Berkeley was Postgres. When SQL functionality was added in 1995, its name was changed to Postgres95. The name was changed at the end of 1996 to PostgreSQL.
It is pronounced Post-Gres-Q-L.
PostgreSQL is subject to the following COPYRIGHT.
PostgreSQL Data Base Management System
Portions copyright (c) 1996-2000, PostgreSQL, Inc Portions Copyright (c) 1994-6 Regents of the University of California
Permission to use, copy, modify, and distribute this software and its documentation for any purpose, without fee, and without a written agreement is hereby granted, provided that the above copyright notice and this paragraph and the following two paragraphs appear in all copies.
IN NO EVENT SHALL THE UNIVERSITY OF CALIFORNIA BE LIABLE TO ANY PARTY FOR DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES, INCLUDING LOST PROFITS, ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN IF THE UNIVERSITY OF CALIFORNIA HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
THE UNIVERSITY OF CALIFORNIA SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS ON AN "AS IS" BASIS, AND THE UNIVERSITY OF CALIFORNIA HAS NO OBLIGATIONS TO PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
The authors have compiled and tested PostgreSQL on the following platforms (some of these compiles require gcc):
It is possible to compile the libpq C library, psql, and other interfaces and binaries to run on MS Windows platforms. In this case, the client is running on MS Windows, and communicates via TCP/IP to a server running on one of our supported Unix platforms.
A file win31.mak is included in the distribution for making a Win32 libpq library and psql.
The database server is now working on Windows NT using the Cygnus Unix/NT porting library. See pgsql/doc/FAQ_NT in the distribution.
The primary anonymous ftp site for PostgreSQL is ftp://ftp.PostgreSQL.org/pub. For mirror sites, see our main web site.
There is no support for PostgreSQL from the University of California, Berkeley. It is maintained through volunteer effort.
The main mailing list is: pgsql-general@PostgreSQL.org. It is available for discussion of matters pertaining to PostgreSQL. To subscribe, send mail with the following lines in the body (not the subject line)
subscribe end
to pgsql-general-request@PostgreSQL.org.
There is also a digest list available. To subscribe to this list, send email to: pgsql-general-digest-request@PostgreSQL.org with a body of:
subscribe endDigests are sent out to members of this list whenever the main list has received around 30k of messages.
The bugs mailing list is available. To subscribe to this list, send email to pgsql-bugs-request@PostgreSQL.org with a body of:
subscribe endThere is also a developers discussion mailing list available. To subscribe to this list, send email to pgsql-hackers-request@PostgreSQL.org with a body of:
subscribe end
Additional mailing lists and information about PostgreSQL can be found via the PostgreSQL WWW home page at:
http://www.PostgreSQL.org
There is also an IRC channel on EFNet, channel #PostgreSQL.
I use the unix command irc -c '#PostgreSQL' "$USER"
irc.phoenix.net.
Commercial support for PostgreSQL is available at http://www.pgsql.com/.
The latest release of PostgreSQL is version 7.0.2.
We plan to have major releases every four months.
Several manuals, manual pages, and some small test examples are included in the distribution. See the /doc directory. You can also browse the manual on-line at http://www.PostgreSQL.org/docs/postgres.
There is a PostgreSQL book available at http://www.PostgreSQL.org/docs/awbook.html.
psql has some nice \d commands to show information about types, operators, functions, aggregates, etc.
Our web site contains even more documentation.
PostgreSQL supports an extended subset of SQL-92. See our TODO list for known bugs, missing features, and future plans.
The PostgreSQL book at http://www.PostgreSQL.org/docs/awbook.html teaches SQL. There is a nice tutorial at http://w3.one.net/~jhoffman/sqltut.htm and at http://ourworld.compuserve.com/homepages/graeme_birchall/HTM_COOK.HTM.
Another one is "Teach Yourself SQL in 21 Days, Second Edition" at http://members.tripod.com/er4ebus/sql/index.htm
Many of our users like The Practical SQL Handbook, Bowman et al., Addison Wesley. Others like The Complete Reference SQL, Groff et al., McGraw-Hill.
Yes, we easily handle dates past the year 2000AD, and before 2000BC.
First, download the latest source and read the PostgreSQL Developers documentation on our web site, or in the distribution. Second, subscribe to the pgsql-hackers and pgsql-patches mailing lists. Third, submit high-quality patches to pgsql-patches.
There are about a dozen people who have commit privileges to the PostgreSQL CVS archive. They each have submitted so many high-quality patches that it was impossible for the existing committers to keep up, and we had confidence that patches they committed were of high quality.
Fill out the "bug-template" file and send it to: pgsql-bugs@PostgreSQL.org
Also check out our ftp site ftp://ftp.PostgreSQL.org/pub to see if there is a more recent PostgreSQL version or patches.
There are several ways of measuring software: features, performance, reliability, support, and price.
There are two ODBC drivers available, PsqlODBC and OpenLink ODBC.
PsqlODBC is included in the distribution. More information about it can be gotten from ftp://ftp.PostgreSQL.org/pub/odbc/.
OpenLink ODBC can be gotten from http://www.openlinksw.com. It works with their standard ODBC client software so you'll have PostgreSQL ODBC available on every client platform they support (Win, Mac, Unix, VMS).
They will probably be selling this product to people who need commercial-quality support, but a freeware version will always be available. Questions to postgres95@openlink.co.uk.
See also the ODBC chapter of the Programmer's Guide.
A nice introduction to Database-backed Web pages can be seen at: http://www.webtools.com
There is also one at http://www.phone.net/home/mwm/hotlist/.
For web integration, PHP is an excellent interface. It is at http://www.php.net
For complex cases, many use the perl interface and CGI.pm.
A WWW gateway based on WDB using perl can be downloaded from http://www.eol.ists.ca/~dunlop/wdb-p95
We have a nice graphical user interface called pgaccess, which is shipped as part of the distribution. Pgaccess also has a report generator. The web page is http://www.flex.ro/pgaccess
We also include ecpg, which is an embedded SQL query language interface for C.
We have:
Try these:
The simplest way is to specify the --prefix option when running configure. If you forgot to do that, you can edit Makefile.global and change POSTGRESDIR accordingly, or create a Makefile.custom and define POSTGRESDIR there.
It could be a variety of problems, but first check to see that you have System V extensions installed in your kernel. PostgreSQL requires kernel support for shared memory and semaphores.
You either do not have shared memory configured properly in your kernel or you need to enlarge the shared memory available in the kernel. The exact amount you need depends on your architecture and how many buffers and backend processes you configure for the postmaster. For most systems, with default numbers of buffers and processes, you need a minimum of ~1MB.
If the error message is IpcSemaphoreCreate: semget failed (No space left on device) then your kernel is not configured with enough semaphores. Postgres needs one semaphore per potential backend process. A temporary solution is to start the postmaster with a smaller limit on the number of backend processes. Use -N with a parameter less than the default of 32. A more permanent solution is to increase your kernel's SEMMNS and SEMMNI parameters.
If the error message is something else, you might not have semaphore support configured in your kernel at all.
By default, PostgreSQL only allows connections from the local machine using Unix domain sockets. Other machines will not be able to connect unless you add the -i flag to the postmaster, and enable host-based authentication by modifying the file $PGDATA/pg_hba.conf accordingly. This will allow TCP/IP connections.
The default configuration allows only unix domain socket connections from the local machine. To enable TCP/IP connections, make sure the postmaster has been started with the -i option, and add an appropriate host entry to the file pgsql/data/pg_hba.conf.
You should not create database users with user id 0 (root). They will be unable to access the database. This is a security precaution because of the ability of users to dynamically link object modules into the database engine.
This problem can be caused by a kernel that is not configured to support semaphores.
Certainly, indices can speed up queries. The EXPLAIN command allows you to see how PostgreSQL is interpreting your query, and which indices are being used.
If you are doing a lot of INSERTs, consider doing them in a large batch using the COPY command. This is much faster than individual INSERTS. Second, statements not in a BEGIN WORK/COMMIT transaction block are considered to be in their own transaction. Consider performing several statements in a single transaction block. This reduces the transaction overhead. Also consider dropping and recreating indices when making large data changes.
There are several tuning options. You can disable fsync() by starting the postmaster with a -o -F option. This will prevent fsync()'s from flushing to disk after every transaction.
You can also use the postmaster -B option to increase the number of shared memory buffers used by the backend processes. If you make this parameter too high, the postmaster may not start because you've exceeded your kernel's limit on shared memory space. Each buffer is 8K and the default is 64 buffers.
You can also use the backend -S option to increase the maximum amount of memory used by the backend process for temporary sorts. The -S value is measured in kilobytes, and the default is 512 (ie, 512K).
You can also use the CLUSTER command to group data in tables to match an index. See the CLUSTER manual page for more details.
PostgreSQL has several features that report status information that can be valuable for debugging purposes.
First, by running configure with the --enable-cassert option, many assert()'s monitor the progress of the backend and halt the program when something unexpected occurs.
Both postmaster and postgres have several debug options available. First, whenever you start the postmaster, make sure you send the standard output and error to a log file, like:
cd /usr/local/pgsql ./bin/postmaster >server.log 2>&1 &
This will put a server.log file in the top-level PostgreSQL directory. This file contains useful information about problems or errors encountered by the server. Postmaster has a -d option that allows even more detailed information to be reported. The -d option takes a number that specifies the debug level. Be warned that high debug level values generate large log files.
If the postmaster is not running, you can actually run the postgres backend from the command line, and type your SQL statement directly. This is recommended only for debugging purposes. Note that a newline terminates the query, not a semicolon. If you have compiled with debugging symbols, you can use a debugger to see what is happening. Because the backend was not started from the postmaster, it is not running in an identical environment and locking/backend interaction problems may not be duplicated.
If the postmaster is running, start psql in one window, then find the PID of the postgres process used by psql. Use a debugger to attach to the postgres PID. You can set breakpoints in the debugger and issue queries from psql. If you are debugging postgres startup, you can set PGOPTIONS="-W n", then start psql. This will cause startup to delay for n seconds so you can attach with the debugger and trace through the startup sequence.
The postgres program has -s, -A, and -t options that can be very useful for debugging and performance measurements.
You can also compile with profiling to see what functions are taking execution time. The backend profile files will be deposited in the pgsql/data/base/dbname directory. The client profile file will be put in the client's current directory.
You need to increase the postmaster's limit on how many concurrent backend processes it can start.
In PostgreSQL 6.5 and up, the default limit is 32 processes. You can increase it by restarting the postmaster with a suitable -N value. With the default configuration you can set -N as large as 1024. If you need more, increase MAXBACKENDS in include/config.h and rebuild. You can set the default value of -N at configuration time, if you like, using configure's --with-maxbackends switch.
Note that if you make -N larger than 32, you must also increase -B beyond its default of 64; -B must be at least twice -N, and probably should be more than that for best performance. For large numbers of backend processes, you are also likely to find that you need to increase various Unix kernel configuration parameters. Things to check include the maximum size of shared memory blocks, SHMMAX, the maximum number of semaphores, SEMMNS and SEMMNI, the maximum number of processes, NPROC, the maximum number of processes per user, MAXUPRC, and the maximum number of open files, NFILE and NINODE. The reason that PostgreSQL has a limit on the number of allowed backend processes is so your system won't run out of resources.
In PostgreSQL versions prior to 6.5, the maximum number of backends was 64, and changing it required a rebuild after altering the MaxBackendId constant in include/storage/sinvaladt.h.
They are temporary files generated by the query executor. For example, if a sort needs to be done to satisfy an ORDER BY, and the sort requires more space than the backend's -S parameter allows, then temporary files are created to hold the extra data.
The temporary files should be deleted automatically, but might not if a backend crashes during a sort. If you have no backends running at the time, it is safe to delete the pg_tempNNN.NN files.
Check your locale configuration. PostgreSQL uses the locale setting of the user that ran the postmaster process. There are postgres and psql SET commands to control the date format. Set those accordingly for your operating environment.
See the DECLARE manual page for a description.
See the FETCH manual page, or use SELECT ... LIMIT....
The entire query may have to be evaluated, even if you only want the first few rows. Consider a query that has an ORDER BY. If there is an index that matches the ORDER BY, PostgreSQL may be able to evaluate only the first few records requested, or the entire query may have to be evaluated until the desired rows have been generated.
You can read the source code for psql in file pgsql/src/bin/psql/psql.c. It contains SQL commands that generate the output for psql's backslash commands. You can also start psql with the -E option so it will print out the queries it uses to execute the commands you give.
We do not support ALTER TABLE DROP COLUMN, but do this:
SELECT ... -- select all columns but the one you want to remove INTO TABLE new_table FROM old_table; DROP TABLE old_table; ALTER TABLE new_table RENAME TO old_table;
These are the limits:
Maximum size for a database? unlimited (60GB databases exist) Maximum size for a table? unlimited on all operating systems Maximum size for a row? 8k, configurable to 32k Maximum number of rows in a table? unlimited Maximum number of columns table? unlimited Maximum number of indexes on a table? unlimitedOf course, these are not actually unlimited, but limited to available disk space.
To change the maximum row size, edit include/config.h and change BLCKSZ. To use attributes larger than 8K, you can also use the large object interface.
The row length limit will be removed in 7.1.
A PostgreSQL database may need six and a half times the disk space required to store the data in a flat file.
Consider a file of 300,000 lines with two integers on each line. The flat file is 2.4MB. The size of the PostgreSQL database file containing this data can be estimated at 14MB:
36 bytes: each row header (approximate)
+ 8 bytes: two int fields @ 4 bytes each
+ 4 bytes: pointer on page to tuple
----------------------------------------
48 bytes per row
The data page size in PostgreSQL is 8192 bytes (8 KB), so:
8192 bytes per page
------------------- = 171 rows per database page (rounded up)
48 bytes per row
300000 data rows
-------------------- = 1755 database pages
171 rows per page
1755 database pages * 8192 bytes per page = 14,376,960 bytes (14MB)
Indexes do not require as much overhead, but do contain the data that is
being indexed, so they can be large also.
psql has a variety of backslash commands to show such information. Use \? to see them.
Also try the file pgsql/src/tutorial/syscat.source. It illustrates many of the SELECTs needed to get information from the database system tables.
PostgreSQL does not automatically maintain statistics. VACUUM must be run to update the statistics. After statistics are updated, the optimizer knows how many rows in the table, and can better decide if it should use indices. Note that the optimizer does not use indices in cases when the table is small because a sequential scan would be faster.
For column-specific optimization statistics, use VACUUM ANALYZE. VACUUM ANALYZE is important for complex multi-join queries, so the optimizer can estimate the number of rows returned from each table, and choose the proper join order. The backend does not keep track of column statistics on its own, so VACUUM ANALYZE must be run to collect them periodically.
Indexes are usually not used for ORDER BY operations: a sequential scan followed by an explicit sort is faster than an indexscan of all tuples of a large table, because it takes fewer disk accesses.
When using wild-card operators such as LIKE or ~, indices can only be used if the beginning of the search is anchored to the start of the string. So, to use indices, LIKE searches should not begin with %, and ~(regular expression searches) should start with ^.
See the EXPLAIN manual page.
An R-tree index is used for indexing spatial data. A hash index can't handle range searches. A B-tree index only handles range searches in a single dimension. R-tree's can handle multi-dimensional data. For example, if an R-tree index can be built on an attribute of type point, the system can more efficient answer queries like select all points within a bounding rectangle.
The canonical paper that describes the original R-Tree design is:
Guttman, A. "R-Trees: A Dynamic Index Structure for Spatial Searching." Proc of the 1984 ACM SIGMOD Int'l Conf on Mgmt of Data, 45-57.
You can also find this paper in Stonebraker's "Readings in Database Systems"
Builtin R-Trees can handle polygons and boxes. In theory, R-trees can be extended to handle higher number of dimensions. In practice, extending R-trees require a bit of work and we don't currently have any documentation on how to do it.
The GEQO module speeds query optimization when joining many tables by means of a Genetic Algorithm (GA). It allows the handling of large join queries through non-exhaustive search.
The ~ operator does regular-expression matching, and ~* does case-insensitive regular-expression matching. There is no case-insensitive variant of the LIKE operator, but you can get the effect of case-insensitive LIKE with this:
WHERE lower(textfield) LIKE lower(pattern)
You test the column with IS NULL and IS NOT NULL.
Type Internal Name Notes -------------------------------------------------- "char" char 1 character CHAR(#) bpchar blank padded to the specified fixed length VARCHAR(#) varchar size specifies maximum length, no padding TEXT text length limited only by maximum row length BYTEA bytea variable-length array of bytes
You will see the internal name when examining system catalogs and in some error messages.
The last four types above are "varlena" types (i.e. the first four bytes are the length, followed by the data). char(#) allocates the maximum number of bytes no matter how much data is stored in the field. text, varchar(#), and bytea all have variable length on the disk, and because of this, there is a small performance penalty for using them. Specifically, the penalty is for access to all columns after the first column of this type.
PostgreSQL supports a SERIAL data type. It auto-creates a sequence and index on the column. For example, this:
CREATE TABLE person ( id SERIAL, name TEXT );is automatically translated into this:
CREATE SEQUENCE person_id_seq;
CREATE TABLE person (
id INT4 NOT NULL DEFAULT nextval('person_id_seq'),
name TEXT
);
CREATE UNIQUE INDEX person_id_key ON person ( id );
See the create_sequence manual page for more information about sequences.
You can also use each row's OID field as a unique value. However, if
you need to dump and reload the database, you need to use pg_dump's -o
option or COPY WITH OIDS option to preserve the OIDs.One approach is to to retrieve the next SERIAL value from the sequence object with the nextval() function before inserting and then insert it explicitly. Using the example table in 4.16.1, that might look like this:
$newSerialID = nextval('person_id_seq');
INSERT INTO person (id, name) VALUES ($newSerialID, 'Blaise Pascal');
You would then also have the new value stored in $newSerialID for use in other queries (e.g., as a foreign key to the person table). Note that the name of the automatically-created SEQUENCE object will be named <table>_<serialcolumn>_seq, where table and serialcolumn are the names of your table and your SERIAL column, respectively.
Alternatively, you could retrieve the just-assigned SERIAL value with the currval() function after it was inserted by default, e.g.,
INSERT INTO person (name) VALUES ('Blaise Pascal');
$newID = currval('person_id_seq');
Finally, you could use the OID returned from the
INSERT statement to lookup the default value, though this is probably
the least portable approach. In perl, using DBI with Edmund Mergl's
DBD::Pg module, the oid value is made available via
$sth->{pg_oid_status} after $sth->execute().
No. This is handled by the backends.
OIDs are PostgreSQL's answer to unique row ids. Every row that is created in PostgreSQL gets a unique OID. All OIDs generated during initdb are less than 16384 (from backend/access/transam.h). All user-created OIDs are equal or greater that this. By default, all these OIDs are unique not only within a table, or database, but unique within the entire PostgreSQL installation.
PostgreSQL uses OIDs in its internal system tables to link rows between tables. These OIDs can be used to identify specific user rows and used in joins. It is recommended you use column type OID to store OID values. You can create an index on the OID field for faster access.
Oids are assigned to all new rows from a central area that is used by all databases. If you want to change the OID to something else, or if you want to make a copy of the table, with the original OID's, there is no reason you can't do it:
CREATE TABLE new_table(old_oid oid, mycol int);
SELECT old_oid, mycol INTO new FROM old;
COPY new TO '/tmp/pgtable';
DELETE FROM new;
COPY new WITH OIDS FROM '/tmp/pgtable';
OIDs are stored as 4-byte integers, and will overflow at 4 billion. No one has reported this ever happening, and we plan to have the limit removed before anyone does.
TIDs are used to identify specific physical rows with block and offset values. Tids change after rows are modified or reloaded. They are used by index entries to point to physical rows.
Some of the source code and older documentation use terms that have more common usage. Here are some:
It is possible you have run out of virtual memory on your system, or your kernel has a low limit for certain resources. Try this before starting the postmaster:
ulimit -d 65536 limit datasize 64mDepending on your shell, only one of these may succeed, but it will set your process data segment limit much higher and perhaps allow the query to complete. This command applies to the current process, and all subprocesses created after the command is run. If you are having a problem with the SQL client because the backend is returning too much data, try it before starting the client.
From psql, type select version();
You need to put BEGIN WORK and COMMIT
around any use of a large object handle, that is,
surrounding lo_open ... lo_close.
Currently PostgreSQL enforces the rule by closing large object handles at transaction commit. So the first attempt to do anything with the handle will draw invalid large obj descriptor. So code that used to work (at least most of the time) will now generate that error message if you fail to use a transaction.
If you are using a client interface like ODBC you may need to set
auto-commit off.
Use now():
CREATE TABLE test (x int, modtime timestamp default now() );
IN so slow?
Currently, we join subqueries to outer queries by sequential scanning
the result of the subquery for each row of the outer query. A workaround
is to replace IN with EXISTS:
SELECT *
FROM tab
WHERE col1 IN (SELECT col2 FROM TAB2)
SELECT *
FROM tab
WHERE EXISTS (SELECT col2 FROM TAB2 WHERE col1 = col2)
PostgreSQL does not support outer joins in the current release. They can be simulated using UNION and NOT IN. For example, when joining tab1 and tab2, the following query does an outer join of the two tables:
SELECT tab1.col1, tab2.col2 FROM tab1, tab2 WHERE tab1.col1 = tab2.col1 UNION ALL SELECT tab1.col1, NULL FROM tab1 WHERE tab1.col1 NOT IN (SELECT tab2.col1 FROM tab2) ORDER BY tab1.col1
The problem could be a number of things. Try testing your user-defined function in a stand alone test program first.
You are pfree'ing something that was not palloc'ed. Beware of mixing malloc/free and palloc/pfree.
Send your extensions to the pgsql-hackers mailing list, and they will eventually end up in the contrib/ subdirectory.
This requires wizardry so extreme that the authors have never tried it, though in principle it can be done.
The Makefiles do not have the proper dependencies for include files. You have to do a make clean and then another make.