4 Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
5 Last updated: Sat Apr 22 14:17:50 EDT 2006
7 The most recent version of this document can be viewed at
8 http://www.postgresql.org/docs/faqs.TODO.html.
10 #A hyphen, "-", marks changes that will appear in the upcoming 8.2 release.#
11 #A percent sign, "%", marks items that are easier to implement.#
13 Bracketed items, "[]", have more detail.
15 This list contains all known PostgreSQL bugs and feature requests. If
16 you would like to work on an item, please read the Developer's FAQ
23 * %Remove behavior of postmaster -o
24 * -%Allow pooled connections to list all prepared statements
26 This would allow an application inheriting a pooled connection to know
27 the statements prepared in the current session.
29 * Allow major upgrades without dump/reload, perhaps using pg_upgrade
31 * Check for unreferenced table files created by transactions that were
32 in-progress when the server terminated abruptly
33 * Allow administrators to safely terminate individual sessions either
34 via an SQL function or SIGTERM
36 Lock table corruption following SIGTERM of an individual backend
37 has been reported in 8.0. A possible cause was fixed in 8.1, but
38 it is unknown whether other problems exist. This item mostly
39 requires additional testing rather than of writing any new code.
41 * %Set proper permissions on non-system schemas during db creation
43 Currently all schemas are owned by the super-user because they are
44 copied from the template1 database.
46 * Support table partitioning that allows a single table to be stored
47 in subtables that are partitioned based on the primary key or a WHERE
49 * Add function to report the time of the most recent server reload
50 * Allow statistics collector information to be pulled from the collector
51 process directly, rather than requiring the collector to write a
52 filesystem file twice a second?
53 * Allow log_min_messages to be specified on a per-module basis
55 This would allow administrators to see more detailed information from
56 specific sections of the backend, e.g. checkpoints, autovacuum, etc.
58 * -Re-enable the GUC full_page_writes in 8.2 when reliability issues have
62 * Improve replication solutions
66 You can use any of the master/slave replication servers to use a
67 standby server for data warehousing. To allow read/write queries to
68 multiple servers, you need multi-master replication like pgcluster.
70 o Allow replication over unreliable or non-persistent links
75 o -Add "include file" functionality in postgresql.conf
76 o %Allow commenting of variables in postgresql.conf to restore them
79 Currently, if a variable is commented out, it keeps the
80 previous uncommented value until a server restarted.
82 o %Allow per-database permissions to be set via GRANT
84 Allow database connection checks based on GRANT rules in
85 addition to the existing access checks in pg_hba.conf.
87 o Allow pg_hba.conf to specify host names along with IP addresses
89 Host name lookup could occur when the postmaster reads the
90 pg_hba.conf file, or when the backend starts. Another
91 solution would be to reverse lookup the connection IP and
92 check that hostname against the host names in pg_hba.conf.
93 We could also then check that the host name maps to the IP
96 o %Allow postgresql.conf file values to be changed via an SQL
97 API, perhaps using SET GLOBAL
98 o Allow the server to be stopped/restarted via an SQL API
99 o -Issue a warning if a change-on-restart-only postgresql.conf value
100 is modified and the server config files are reloaded
101 o Mark change-on-restart-only values in postgresql.conf
106 * Allow a database in tablespace t1 with tables created in
107 tablespace t2 to be used as a template for a new database created
108 with default tablespace t2
110 All objects in the default database tablespace must have default
111 tablespace specifications. This is because new databases are
112 created by copying directories. If you mix default tablespace
113 tables and tablespace-specified tables in the same directory,
114 creating a new database from such a mixed directory would create a
115 new database with tables that had incorrect explicit tablespaces.
116 To fix this would require modifying pg_class in the newly copied
117 database, which we don't currently do.
119 * Allow reporting of which objects are in which tablespaces
121 This item is difficult because a tablespace can contain objects
122 from multiple databases. There is a server-side function that
123 returns the databases which use a specific tablespace, so this
124 requires a tool that will call that function and connect to each
125 database to find the objects in each database for that tablespace.
127 o %Add a GUC variable to control the tablespace for temporary objects
130 It could start with a random tablespace from a supplied list and
131 cycle through the list.
133 o Allow WAL replay of CREATE TABLESPACE to work when the directory
134 structure on the recovery computer is different from the original
136 o Allow per-tablespace quotas
139 * Point-In-Time Recovery (PITR)
141 o Allow point-in-time recovery to archive partially filled
142 write-ahead logs [pitr]
144 Currently only full WAL files are archived. This means that the
145 most recent transactions aren't available for recovery in case
146 of a disk failure. This could be triggered by a user command or
149 o Automatically force archiving of partially-filled WAL files when
150 pg_stop_backup() is called or the server is stopped
152 Doing this will allow administrators to know more easily when
153 the archive contains all the files needed for point-in-time
156 o %Create dump tool for write-ahead logs for use in determining
157 transaction id for point-in-time recovery
158 o Allow a warm standby system to also allow read-only statements
161 This is useful for checking PITR recovery.
163 o Allow the PITR process to be debugged and data examined
164 o Add reporting of the current WAL file, perhaps as part of
165 partial log file archiving
171 * Allow server log information to be output as INSERT statements
173 This would allow server log information to be easily loaded into
174 a database for analysis.
176 * %Add ability to monitor the use of temporary sort files
177 * Allow server logs to be remotely read and removed using SQL commands
178 * Allow protocol-level BIND parameter values to be logged
184 * Improve the MONEY data type
186 Change the MONEY data type to use DECIMAL internally, with special
187 locale-aware output formatting.
189 * Change NUMERIC to enforce the maximum precision
190 * Add NUMERIC division operator that doesn't round?
192 Currently NUMERIC _rounds_ the result to the specified precision.
193 This means division can return a result that multiplied by the
194 divisor is greater than the dividend, e.g. this returns a value > 10:
196 SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
198 The positive modulus result returned by NUMERICs might be considered
199 inaccurate, in one sense.
201 * %Disallow changing default expression of a SERIAL column
202 * %Disallow ALTER SEQUENCE changes for SERIAL sequences because pg_dump
203 does not dump the changes
204 * Fix data types where equality comparison isn't intuitive, e.g. box
205 * -Zero umasked bits in conversion from INET cast to CIDR
206 * -Prevent INET cast to CIDR from dropping netmask, SELECT '1.1.1.1'::inet::cidr
207 * -Allow INET + INT8 to increment the host part of the address or
208 throw an error on overflow
209 * -Add 'tid != tid ' operator for use in corruption recovery
210 * Allow user-defined types to specify a type modifier at table creation
216 o Allow infinite dates and intervals just like infinite timestamps
217 o Merge hardwired timezone names with the TZ database; allow either
218 kind everywhere a TZ name is currently taken
219 o Allow customization of the known set of TZ names (generalize the
220 present australian_timezones hack)
221 o Allow TIMESTAMP WITH TIME ZONE to store the original timezone
222 information, either zone name or offset from UTC [timezone]
224 If the TIMESTAMP value is stored with a time zone name, interval
225 computations should adjust based on the time zone rules.
227 o Fix SELECT '0.01 years'::interval, '0.01 months'::interval
228 o Fix SELECT INTERVAL '1' MONTH
229 o Add a GUC variable to allow output of interval values in ISO8601
231 o Improve timestamptz subtraction to be DST-aware
233 Currently, subtracting one date from another that crosses a
234 daylight savings time adjustment can return '1 day 1 hour', but
235 adding that back to the first date returns a time one hour in
236 the future. This is caused by the adjustment of '25 hours' to
237 '1 day 1 hour', and '1 day' is the same time the next day, even
238 if daylight savings adjustments are involved.
240 o Fix interval display to support values exceeding 2^31 hours
241 o Add overflow checking to timestamp and interval arithmetic
242 o Add ISO INTERVAL handling
243 o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
245 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
246 o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
247 '1', treat as ISO if there is a range specification clause,
248 and as PG if there no clause is present, e.g. interpret
249 '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
250 interpret '1:30' as '1 hour, 30 minutes'
251 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
252 INTERVAL MONTH), and this should return '12 months'
253 o Round or truncate values to the requested precision, e.g.
254 INTERVAL '11 months' AS YEAR should return one or zero
255 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
260 o -Allow NULLs in arrays
261 o Delay resolution of array expression's data type so assignment
262 coercion can be performed on empty array expressions
267 o Improve vacuum of large objects, like /contrib/vacuumlo?
268 o Add security checking for large objects
269 o Auto-delete large objects when referencing row is deleted
271 /contrib/lo offers this functionality.
273 o Allow read/write into TOAST values like large objects
275 This requires the TOAST column to be stored EXTERNAL.
281 * Allow INET subnet tests using non-constants to be indexed
282 * Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
285 Current CURRENT_TIMESTAMP returns the start time of the current
286 transaction, and gettimeofday() returns the wallclock time. This will
287 make time reporting more consistent and will allow reporting of
288 the statement start time.
290 * %Add pg_get_acldef(), pg_get_typedefault(), pg_get_attrdef(),
291 pg_get_tabledef(), pg_get_domaindef(), pg_get_functiondef()
292 * -Allow to_char() to print localized month names
293 * Allow to_date() and to_timestamp() accept localized month names
294 * Add missing parameter handling in to_char()
296 http://archives.postgresql.org/pgsql-hackers/2005-12/msg00948.php
298 * Allow functions to have a schema search path specified at creation time
299 * Allow substring/replace() to get/set bit values
300 * Allow to_char() on interval values to accumulate the highest unit
303 Some special format flag would be required to request such
304 accumulation. Such functionality could also be added to EXTRACT.
305 Prevent accumulation that crosses the month/day boundary because of
306 the uneven number of days in a month.
308 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
309 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
310 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
311 o to_char(INTERVAL '3 years 5 months','MM') => 41
313 * -Add sleep() function, remove from regress.c
314 * -Allow user-defined functions retuning a domain value to enforce domain
316 * Add SPI_gettypmod() to return the typemod for a TupleDesc
319 Multi-Language Support
320 ======================
322 * Add NCHAR (as distinguished from ordinary varchar),
323 * Allow locale to be set at database creation
325 Currently locale can only be set during initdb. No global tables have
326 locale-aware columns. However, the database template used during
327 database creation might have locale-aware indexes. The indexes would
328 need to be reindexed to match the new locale.
330 * Allow encoding on a per-column basis
332 Right now only one encoding is allowed per database.
334 * Support multiple simultaneous character sets, per SQL92
335 * Improve UTF8 combined character handling?
336 * Add octet_length_server() and octet_length_client()
337 * Make octet_length_client() the same as octet_length()?
338 * Fix problems with wrong runtime encoding conversion for NLS message files
344 * Automatically create rules on views so they are updateable, per SQL99
346 We can only auto-create rules for simple views. For more complex
347 cases users will still have to write rules manually.
349 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
350 * Allow NOTIFY in rules involving conditionals
351 * Allow VIEW/RULE recompilation when the underlying tables change
353 Another issue is whether underlying table changes should be reflected
354 in the view, e.g. should SELECT * show additional columns if they
355 are added after the view is created.
361 * Change LIMIT/OFFSET and FETCH/MOVE to use int8
362 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
363 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
364 * %Allow SET CONSTRAINTS to be qualified by schema/table name
365 * -Allow TRUNCATE ... CASCADE/RESTRICT
367 This is like DELETE CASCADE, but truncates.
369 * %Add a separate TRUNCATE permission
371 Currently only the owner can TRUNCATE a table because triggers are not
372 called, and the table is locked in exclusive mode.
374 * Allow PREPARE of cursors
375 * -Allow PREPARE to automatically determine parameter types based on the SQL
377 * Allow finer control over the caching of prepared query plans
379 Currently, queries prepared via the libpq API are planned on first
380 execute using the supplied parameters --- allow SQL PREPARE to do the
381 same. Also, allow control over replanning prepared queries either
382 manually or automatically when statistics for execute parameters
383 differ dramatically from those used during planning.
385 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
387 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
388 such information in memory would improve performance.
390 * Add optional textual message to NOTIFY
392 This would allow an informational message to be added to the notify
393 message, perhaps indicating the row modified or other custom
396 * Add a GUC variable to warn about non-standard SQL usage in queries
397 * Add SQL-standard MERGE command, typically used to merge two tables
400 This is similar to UPDATE, then for unmatched rows, INSERT.
401 Whether concurrent access allows modifications which could cause
402 row loss is implementation independent.
404 * Add REPLACE or UPSERT command that does UPDATE, or on failure, INSERT
407 To implement this cleanly requires that the table have a unique index
408 so duplicate checking can be easily performed. It is possible to
409 do it without a unique index if we require the user to LOCK the table
412 * Add NOVICE output level for helpful messages like automatic sequence/index
414 * -Add COMMENT ON for all cluster global objects (roles, databases
416 * -Make row-wise comparisons work per SQL spec
418 Right now, '(a, b) < (1, 2)' is processed as 'a < 1 and b < 2', but
419 the SQL standard requires it to be processed as a column-by-column
420 comparison, so the proper comparison is '(a < 1) OR (a = 1 AND b < 2)'.
422 * Add RESET CONNECTION command to reset all session state
424 This would include resetting of all variables (RESET ALL), dropping of
425 temporary tables, removing any NOTIFYs, cursors, open transactions,
426 prepared queries, currval()s, etc. This could be used for connection
427 pooling. We could also change RESET ALL to have this functionality.
428 The difficult of this features is allowing RESET ALL to not affect
429 changes made by the interface driver for its internal use. One idea
430 is for this to be a protocol-only feature. Another approach is to
431 notify the protocol when a RESET CONNECTION command is used.
433 * Add GUC to issue notice about statements that use unjoined tables
434 * Allow EXPLAIN to identify tables that were skipped because of
436 * Allow EXPLAIN output to be more easily processed by scripts
437 * -Eventually enable escape_string_warning and standard_conforming_strings
438 * Simplify dropping roles that have objects in several databases
439 * Allow COMMENT ON to accept an expression rather than just a string
440 * Allow the count returned by SELECT, etc to be to represent as an int64
441 to allow a higher range of values
442 * Make CLUSTER preserve recently-dead tuples per MVCC requirements
443 * Add SQL99 WITH clause to SELECT
444 * Add SQL99 WITH RECURSIVE to SELECT
449 o Allow CREATE TABLE AS to determine column lengths for complex
450 expressions like SELECT col1 || col2
451 o Use more reliable method for CREATE DATABASE to get a consistent
453 o -Add ON COMMIT capability to CREATE TABLE AS ... SELECT
457 o Allow UPDATE to handle complex aggregates [update]?
458 o -Allow an alias to be provided for the target table in
460 o Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple
466 o %Have ALTER TABLE RENAME rename SERIAL sequence names
467 o Add ALTER DOMAIN to modify the underlying data type
468 o %Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
469 o %Allow ALTER TABLE to change constraint deferrability and actions
470 o Add missing object types for ALTER ... SET SCHEMA
471 o Allow ALTER TABLESPACE to move to different directories
472 o Allow databases to be moved to different tablespaces
473 o Allow moving system tables to other tablespaces, where possible
475 Currently non-global system tables must be in the default database
476 tablespace. Global system tables can never be moved.
478 o %Prevent child tables from altering or dropping constraints
479 like CHECK that were inherited from the parent table
480 o Prevent parent tables from altering or dropping constraints
481 like CHECK that are inherited by child tables
483 Dropping constraints should only be possible with CASCADE.
485 o Have ALTER INDEX update the name of a constraint using that index
486 o Add ALTER TABLE RENAME CONSTRAINT, update index name also
491 o Automatically maintain clustering on a table
493 This might require some background daemon to maintain clustering
494 during periods of low usage. It might also require tables to be only
495 partially filled for easier reorganization. Another idea would
496 be to create a merged heap/index data file so an index lookup would
497 automatically access the heap data too. A third idea would be to
498 store heap rows in hashed groups, perhaps using a user-supplied
501 o %Add default clustering to system tables
503 To do this, determine the ideal cluster index for each system
504 table and set the cluster setting during initdb.
509 o Allow COPY to report error lines and continue
511 This requires the use of a savepoint before each COPY line is
512 processed, with ROLLBACK on COPY failure.
514 o -Have COPY return the number of rows loaded/unloaded?
515 o Allow COPY on a newly-created table to skip WAL logging
517 On crash recovery, the table involved in the COPY would
518 be removed or have its heap and index files truncated. One
519 issue is that no other backend should be able to add to
520 the table at the same time, which is something that is
523 o Allow COPY to output from views
525 Another idea would be to allow actual SELECT statements in a COPY.
530 o Allow column-level privileges
531 o %Allow GRANT/REVOKE permissions to be applied to all schema objects
534 The proposed syntax is:
535 GRANT SELECT ON ALL TABLES IN public TO phpuser;
536 GRANT SELECT ON NEW TABLES IN public TO phpuser;
538 * Allow GRANT/REVOKE permissions to be inherited by objects based on
541 * Allow SERIAL sequences to inherit permissions from the base table?
546 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
548 This requires using the row ctid to map cursor rows back to the
549 original heap row. This become more complicated if WITH HOLD cursors
550 are to be supported because WITH HOLD cursors have a copy of the row
551 and no FOR UPDATE lock.
553 o Prevent DROP TABLE from dropping a row referenced by its own open
556 o -Allow pooled connections to list all open WITH HOLD cursors
558 Because WITH HOLD cursors exist outside transactions, this allows
559 them to be listed so they can be closed.
564 o Allow INSERT/UPDATE of the system-generated oid value for a row
565 o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
566 o Allow INSERT/UPDATE ... RETURNING new.col or old.col
568 This is useful for returning the auto-generated key for an INSERT.
569 One complication is how to handle rules that run as part of
575 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
577 o Add SET PATH for schemas?
579 This is basically the same as SET search_path.
582 * Server-Side Languages
584 o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
585 o Allow function parameters to be passed by name,
586 get_employee_salary(emp_id => 12345, tax_year => 2001)
587 o Add Oracle-style packages
588 o Add table function support to pltcl, plpython
589 o Add capability to create and call PROCEDURES
590 o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
591 o Allow function argument names to be statements from PL/PgSQL
592 o Add MOVE to PL/pgSQL
593 o Add support for polymorphic arguments and return types to
594 languages other than PL/PgSQL
595 o Add support for OUT and INOUT parameters to languages other
597 o Add single-step debugging of PL/PgSQL functions
598 o Allow PL/PgSQL to support WITH HOLD cursors
604 * -Have initdb set the input DateStyle (MDY or DMY) based on locale
605 * Have pg_ctl look at PGHOST in case it is a socket directory?
606 * Allow pg_ctl to work properly with configuration files located outside
609 pg_ctl can not read the pid file because it isn't located in the
610 config directory but in the PGDATA directory. The solution is to
611 allow pg_ctl to read and understand postgresql.conf to find the
612 data_directory value.
617 o Have psql show current values for a sequence
618 o Move psql backslash database information into the backend, use
619 mnemonic commands? [psql]
621 This would allow non-psql clients to pull the same information out
622 of the database as psql.
624 o Fix psql's backslash commands more consistent
626 http://archives.postgresql.org/pgsql-hackers/2004-11/msg00014.php
627 http://archives.postgresql.org/pgsql-hackers/2004-11/msg00014.php
629 o Allow psql \pset boolean variables to set to fixed values, rather
631 o Consistently display privilege information for all objects in psql
632 o -Improve psql's handling of multi-line statements
634 Currently, while \e saves a single statement as one entry, interactive
635 statements are saved one line at a time. Ideally all statements
636 would be saved like \e does.
638 o -Allow multi-line column values to align in the proper columns
640 If the second output column value is 'a\nb', the 'b' should appear
641 in the second display column, rather than the first column as it
644 o Display IN, INOUT, and OUT parameters in \df+
646 It probably requires psql to output newlines in the proper
647 column, which is already on the TODO list.
649 o Add auto-expanded mode so expanded output is used if the row
650 length is wider than the screen width.
652 Consider using auto-expanded mode for backslash commands like \df+.
654 o Prevent tab completion of SET TRANSACTION from querying the
655 database and therefore preventing the transaction isolation
656 level from being set.
658 Currently, SET <tab> causes a database lookup to check all
659 supported session variables. This query causes problems
660 because setting the transaction isolation level must be the
661 first statement of a transaction.
666 o %Allow pg_dump to use multiple -t and -n switches [pg_dump]
667 o %Add dumping of comments on index columns and composite type columns
668 o %Add full object name to the tag field. eg. for operators we need
669 '=(integer, integer)', instead of just '='.
670 o Add pg_dumpall custom format dumps?
671 o %Add CSV output format
672 o -Update pg_dump and psql to use the new COPY libpq API (Christopher)
673 o Remove unnecessary function pointer abstractions in pg_dump source
675 o Allow selection of individual object(s) of all types, not just
677 o In a selective dump, allow dumping of an object and all its
679 o Add options like pg_restore -l and -L to pg_dump
680 o Stop dumping CASCADE on DROP TYPE commands in clean mode
681 o Allow pg_dump --clean to drop roles that own objects or have
683 o Add -f to pg_dumpall
690 Document differences between ecpg and the SQL standard and
691 information about the Informix-compatibility module.
693 o Solve cardinality > 1 for input descriptors / variables?
694 o Add a semantic check level, e.g. check if a table really exists
695 o fix handling of DB attributes that are arrays
696 o Use backend PREPARE/EXECUTE facility for ecpg where possible
698 o Fix nested C comments
699 o %sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
700 o Make SET CONNECTION thread-aware, non-standard?
701 o Allow multidimensional arrays
702 o Add internationalized message strings
703 o Add COPY TO STDIN / STDOUT handling
708 o Add a function to support Parse/DescribeStatement capability
709 o Add PQescapeIdentifier()
710 o Prevent PQfnumber() from lowercasing unquoted the column name
712 PQfnumber() should never have been doing lowercasing, but
713 historically it has so we need a way to prevent it
715 o Allow statement results to be automatically batched to the client
717 Currently, all statement results are transferred to the libpq
718 client before libpq makes the results available to the
719 application. This feature would allow the application to make
720 use of the first result rows while the rest are transferred, or
721 held on the server waiting for them to be requested by libpq.
722 One complexity is that a statement like SELECT 1/col could error
723 out mid-way through the result set.
725 o Add new version of PQescapeString() that doesn't double backslashes
726 that are part of a client-only multibyte sequence
728 Single-quote is not a valid byte in any supported client-only
729 encoding. This requires using mblen() to determine if the
730 backslash is inside or outside a multi-byte sequence.
732 o Add new version of PQescapeString() that doesn't double
733 backslashes when standard_conforming_strings is true and
734 non-E strings are used
737 Referential Integrity
738 =====================
740 * Add MATCH PARTIAL referential integrity
741 * Add deferred trigger queue file
743 Right now all deferred trigger information is stored in backend
744 memory. This could exhaust memory for very large trigger queues.
745 This item involves dumping large queues into files.
747 * Change foreign key constraint for array -> element to mean element
749 * Allow DEFERRABLE UNIQUE constraints?
750 * Allow triggers to be disabled in only the current session.
752 This is currently possible by starting a multi-statement transaction,
753 modifying the system tables, performing the desired SQL, restoring the
754 system tables, and committing the transaction. ALTER TABLE ...
755 TRIGGER requires a table lock so it is not ideal for this usage.
757 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
759 If the dump is known to be valid, allow foreign keys to be added
760 without revalidating the data.
762 * Allow statement-level triggers to access modified rows
763 * Support triggers on columns (Greg Sabino Mullane)
764 * Enforce referential integrity for system tables
765 * Allow AFTER triggers on system tables
767 System tables are modified in many places in the backend without going
768 through the executor and therefore not causing triggers to fire. To
769 complete this item, the functions that modify system tables will have
776 * Flush cached query plans when the dependent objects change,
777 when the cardinality of parameters changes dramatically, or
778 when new ANALYZE statistics are available
780 A more complex solution would be to save multiple plans for different
781 cardinality and use the appropriate plan based on the EXECUTE values.
783 * Track dependencies in function bodies and recompile/invalidate
785 This is particularly important for references to temporary tables
786 in PL/PgSQL because PL/PgSQL caches query plans. The only workaround
787 in PL/PgSQL is to use EXECUTE. One complexity is that a function
788 might itself drop and recreate dependent tables, causing it to
789 invalidate its own query plan.
795 * Add pre-parsing phase that converts non-ISO syntax to supported
798 This could allow SQL written for other databases to run without
801 * Allow plug-in modules to emulate features from other databases
802 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
804 * Allow statements across databases or servers with transaction
807 This can be done using dblink and two-phase commit.
809 * Add the features of packages
811 o Make private objects accessible only to objects in the same schema
812 o Allow current_schema.objname to access current schema objects
813 o Add session variables
814 o Allow nested schemas
820 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
822 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
823 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
826 The main difficulty with this item is the problem of creating an index
827 that can span more than one table.
829 * Allow SELECT ... FOR UPDATE on inherited tables
830 * Add UNIQUE capability to non-btree indexes
831 * Prevent index uniqueness checks when UPDATE does not modify the column
833 Uniqueness (index) checks are done when updating a column even if the
834 column is not modified by the UPDATE.
836 * Allow the creation of on-disk bitmap indexes which can be quickly
837 combined with other bitmap indexes
839 Such indexes could be more compact if there are only a few distinct values.
840 Such indexes can also be compressed. Keeping such indexes updated can be
843 * Allow use of indexes to search for NULLs
845 One solution is to create a partial index on an IS NULL expression.
847 * Allow accurate statistics to be collected on indexes with more than
848 one column or expression indexes, perhaps using per-index statistics
849 * Add fillfactor to control reserved free space during index creation
850 * Allow the creation of indexes with mixed ascending/descending
853 This is possible now by creating an operator class with reversed sort
854 operators. One complexity is that NULLs would then appear at the start
855 of the result set, and this might affect certain sort types, like
858 * Allow constraint_exclusion to work for UNIONs like it does for
859 inheritance, allow it to work for UPDATE and DELETE statements, and allow
860 it to be used for all statements with little performance impact
861 * Allow CREATE INDEX to take an additional parameter for use with
863 * Consider compressing indexes by storing key values duplicated in
864 several rows as a single index entry
866 This is difficult because it requires datatype-specific knowledge.
871 o Add more GIST index support for geometric data types
872 o Allow GIST indexes to create certain complex index types, like
873 digital trees (see Aoki)
877 o Pack hash index buckets onto disk pages more efficiently
879 Currently only one hash bucket can be stored on a page. Ideally
880 several hash buckets could be stored on a single page and greater
881 granularity used for the hash algorithm.
883 o Consider sorting hash buckets so entries can be found using a
884 binary search, rather than a linear scan
886 o In hash indexes, consider storing the hash value with or instead
889 o Add WAL logging for crash recovery
890 o Allow multi-column hash indexes
896 * Improve commit_delay handling to reduce fsync()
897 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
899 Ideally this requires a separate test program that can be run
900 at initdb time or optionally later. Consider O_SYNC when
903 * %Add an option to sync() before fsync()'ing checkpoint files
904 * Add program to test if fsync has a delay compared to non-fsync
910 * Allow free-behind capability for large sequential scans, perhaps using
913 Posix_fadvise() can control both sequential/random file caching and
914 free-behind behavior, but it is unclear how the setting affects other
915 backends that also have the file open, and the feature is not supported
916 on all operating systems.
920 We could use a fixed row count and a +/- count to follow MVCC
921 visibility rules, or a single cached value could be used and
922 invalidated if anyone modifies the table. Another idea is to
923 get a count directly from a unique index, but for this to be
924 faster than a sequential scan it must avoid access to the heap
925 to obtain tuple visibility information.
927 * Add estimated_count(*) to return an estimate of COUNT(*)
929 This would use the planner ANALYZE statistics to return an estimated
932 * Allow data to be pulled directly from indexes
934 Currently indexes do not have enough tuple visibility information
935 to allow data to be pulled from the index without also accessing
936 the heap. One way to allow this is to set a bit on index tuples
937 to indicate if a tuple is currently visible to all transactions
938 when the first valid heap lookup happens. This bit would have to
939 be cleared when a heap tuple is expired.
941 Another idea is to maintain a bitmap of heap pages where all rows
942 are visible to all backends, and allow index lookups to reference
943 that bitmap to avoid heap lookups, perhaps the same bitmap we might
944 add someday to determine which heap pages need vacuuming. Frequently
945 accessed bitmaps would have to be stored in shared memory. One 8k
946 page of bitmaps could track 512MB of heap pages.
948 * Consider automatic caching of statements at various levels:
954 * Allow sequential scans to take advantage of other concurrent
955 sequential scans, also called "Synchronised Scanning"
957 One possible implementation is to start sequential scans from the lowest
958 numbered buffer in the shared cache, and when reaching the end wrap
959 around to the beginning, rather than always starting sequential scans
960 at the start of the table.
966 * Improve speed with indexes
968 For large table adjustments during VACUUM FULL, it is faster to
969 reindex rather than update the index.
971 * Reduce lock time during VACUUM FULL by moving tuples with read lock,
972 then write lock and truncate table
974 Moved tuples are invisible to other backends so they don't require a
975 write lock. However, the read lock promotion to write lock could lead
976 to deadlock situations.
978 * Auto-fill the free space map by scanning the buffer cache or by
979 checking pages written by the background writer
980 * Create a bitmap of pages that need vacuuming
982 Instead of sequentially scanning the entire table, have the background
983 writer or some other process record pages that have expired rows, then
984 VACUUM can look at just those pages rather than the entire table. In
985 the event of a system crash, the bitmap would probably be invalidated.
986 One complexity is that index entries still have to be vacuumed, and
987 doing this without an index scan (by using the heap values to find the
988 index entry) might be slow and unreliable, especially for user-defined
991 * -Add system view to show free space map contents
992 * Allow FSM to return free space toward the beginning of the heap file,
993 in hopes that empty pages at the end can be truncated by VACUUM
994 * Allow FSM page return free space based on table clustering, to assist
995 in maintaining clustering?
1000 o Use free-space map information to guide refilling
1001 o %Issue log message to suggest VACUUM FULL if a table is nearly
1003 o Improve xid wraparound detection by recording per-table rather
1005 o Consider logging activity either to the logs or a system view
1011 * Fix priority ordering of read and write light-weight locks (Neil)
1014 Startup Time Improvements
1015 =========================
1017 * Experiment with multi-threaded backend for backend creation [thread]
1019 This would prevent the overhead associated with process creation. Most
1020 operating systems have trivial process creation time compared to
1021 database startup overhead, but a few operating systems (Win32,
1022 Solaris) might benefit from threading. Also explore the idea of
1023 a single session using multiple threads to execute a statement faster.
1025 * Experiment with multi-threaded backend better resource utilization
1027 This would allow a single query to make use of multiple CPU's or
1028 multiple I/O channels simultaneously. One idea is to create a
1029 background reader that can pre-fetch sequential and index scan
1030 pages needed by other backends. This could be expanded to allow
1031 concurrent reads from multiple devices in a partitioned table.
1033 * Add connection pooling
1035 It is unclear if this should be done inside the backend code or done
1036 by something external like pgpool. The passing of file descriptors to
1037 existing backends is one of the difficulties with a backend approach.
1043 * Eliminate need to write full pages to WAL before page modification [wal]
1045 Currently, to protect against partial disk page writes, we write
1046 full page images to WAL before they are modified so we can correct any
1047 partial page writes during recovery. These pages can also be
1048 eliminated from point-in-time archive files.
1050 o When off, write CRC to WAL and check file system blocks
1053 If CRC check fails during recovery, remember the page in case
1054 a later CRC for that page properly matches.
1056 o Write full pages during file system write and not when
1057 the page is modified in the buffer cache
1059 This allows most full page writes to happen in the background
1060 writer. It might cause problems for applying WAL on recovery
1061 into a partially-written page, but later the full page will be
1064 * Allow WAL traffic to be streamed to another server for stand-by
1066 * Reduce WAL traffic so only modified values are written rather than
1068 * Allow the pg_xlog directory location to be specified during initdb
1069 with a symlink back to the /data location
1070 * Allow WAL information to recover corrupted pg_controldata
1071 * Find a way to reduce rotational delay when repeatedly writing
1074 Currently fsync of WAL requires the disk platter to perform a full
1075 rotation to fsync again. One idea is to write the WAL to different
1076 offsets that might reduce the rotational delay.
1078 * Allow buffered WAL writes and fsync
1080 Instead of guaranteeing recovery of all committed transactions, this
1081 would provide improved performance by delaying WAL writes and fsync
1082 so an abrupt operating system restart might lose a few seconds of
1083 committed transactions but still be consistent. We could perhaps
1084 remove the 'fsync' parameter (which results in an an inconsistent
1085 database) in favor of this capability.
1087 * Allow WAL logging to be turned off for a table, but the table
1088 might be dropped or truncated during crash recovery [walcontrol]
1090 Allow tables to bypass WAL writes and just fsync() dirty pages on
1091 commit. This should be implemented using ALTER TABLE, e.g. ALTER
1092 TABLE PERSISTENCE [ DROP | TRUNCATE | DEFAULT ]. Tables using
1093 non-default logging should not use referential integrity with
1094 default-logging tables. A table without dirty buffers during a
1095 crash could perhaps avoid the drop/truncate.
1097 * Allow WAL logging to be turned off for a table, but the table would
1098 avoid being truncated/dropped [walcontrol]
1100 To do this, only a single writer can modify the table, and writes
1101 must happen only on new pages so the new pages can be removed during
1102 crash recovery. Readers can continue accessing the table. Such
1103 tables probably cannot have indexes. One complexity is the handling
1104 of indexes on TOAST tables.
1107 Optimizer / Executor
1108 ====================
1110 * Improve selectivity functions for geometric operators
1111 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
1112 index using a sequential scan for highest/lowest values
1114 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
1115 all values to return the high/low value. Instead The idea is to do a
1116 sequential scan to find the high/low value, thus avoiding the sort.
1117 MIN/MAX already does this, but not for LIMIT > 1.
1119 * Precompile SQL functions to avoid overhead
1120 * Create utility to compute accurate random_page_cost value
1121 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
1122 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
1123 * Consider using hash buckets to do DISTINCT, rather than sorting
1125 This would be beneficial when there are few distinct values. This is
1126 already used by GROUP BY.
1128 * Log statements where the optimizer row estimates were dramatically
1129 different from the number of rows actually found?
1132 Miscellaneous Performance
1133 =========================
1135 * Do async I/O for faster random read-ahead of data
1137 Async I/O allows multiple I/O requests to be sent to the disk with
1138 results coming back asynchronously.
1140 * Use mmap() rather than SYSV shared memory or to write WAL files?
1142 This would remove the requirement for SYSV SHM but would introduce
1143 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
1144 to prevent I/O overhead.
1146 * Consider mmap()'ing files into a backend?
1148 Doing I/O to large tables would consume a lot of address space or
1149 require frequent mapping/unmapping. Extending the file also causes
1150 mapping problems that might require mapping only individual pages,
1151 leading to thousands of mappings. Another problem is that there is no
1152 way to _prevent_ I/O to disk from the dirty shared buffers so changes
1153 could hit disk before WAL is written.
1155 * Add a script to ask system configuration questions and tune postgresql.conf
1156 * Merge xmin/xmax/cmin/cmax back into three header fields
1158 Before subtransactions, there used to be only three fields needed to
1159 store these four values. This was possible because only the current
1160 transaction looks at the cmin/cmax values. If the current transaction
1161 created and expired the row the fields stored where xmin (same as
1162 xmax), cmin, cmax, and if the transaction was expiring a row from a
1163 another transaction, the fields stored were xmin (cmin was not
1164 needed), xmax, and cmax. Such a system worked because a transaction
1165 could only see rows from another completed transaction. However,
1166 subtransactions can see rows from outer transactions, and once the
1167 subtransaction completes, the outer transaction continues, requiring
1168 the storage of all four fields. With subtransactions, an outer
1169 transaction can create a row, a subtransaction expire it, and when the
1170 subtransaction completes, the outer transaction still has to have
1171 proper visibility of the row's cmin, for example, for cursors.
1173 One possible solution is to create a phantom cid which represents a
1174 cmin/cmax pair and is stored in local memory. Another idea is to
1175 store both cmin and cmax only in local memory.
1177 * Research storing disk pages with no alignment/padding
1183 * Add use of 'const' for variables in source tree
1184 * Rename some /contrib modules from pg* to pg_*
1185 * Move some things from /contrib into main tree
1186 * Move some /contrib modules out to their own project sites
1187 * %Remove warnings created by -Wcast-align
1188 * Move platform-specific ps status display info from ps_status.c to ports
1189 * Add optional CRC checksum to heap and index pages
1190 * Improve documentation to build only interfaces (Marc)
1191 * Remove or relicense modules that are not under the BSD license, if possible
1192 * %Remove memory/file descriptor freeing before ereport(ERROR)
1193 * Acquire lock on a relation before building a relcache entry for it
1194 * %Promote debug_query_string into a server-side function current_query()
1195 * %Allow the identifier length to be increased via a configure option
1196 * Allow cross-compiling by generating the zic database on the target system
1197 * Improve NLS maintenance of libpgport messages linked onto applications
1198 * Allow ecpg to work with MSVC and BCC
1199 * Add xpath_array() to /contrib/xml2 to return results as an array
1200 * Allow building in directories containing spaces
1202 This is probably not possible because 'gmake' and other compiler tools
1203 do not fully support quoting of paths with spaces.
1205 * -Allow installing to directories containing spaces
1207 This is possible if proper quoting is added to the makefiles for the
1208 install targets. Because PostgreSQL supports relocatable installs, it
1209 is already possible to install into a directory that doesn't contain
1210 spaces and then copy the install to a directory with spaces.
1212 * Fix sgmltools so PDFs can be generated with bookmarks
1213 * -%Clean up compiler warnings (especially with gcc version 4)
1214 * Use UTF8 encoding for NLS messages so all server encodings can
1216 * Update Bonjour to work with newer cross-platform SDK
1217 * -Remove BeOS and QNX-specific code
1218 * Split out libpq pgpass and environment documentation sections to make
1219 it easier for non-developers to find
1220 * Improve port/qsort() to handle sorts with 50% unique and 50% duplicate
1223 This involves choosing better pivot points for the quicksort.
1228 o Remove configure.in check for link failure when cause is found
1229 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1231 o Remove psql newline patch when we find out why mingw outputs an
1233 o Allow psql to use readline once non-US code pages work with
1235 o Re-enable timezone output on log_line_prefix '%t' when a
1236 shorter timezone string is available
1237 o Fix problem with shared memory on the Win32 Terminal Server
1238 o Improve signal handling,
1239 http://archives.postgresql.org/pgsql-patches/2005-06/msg00027.php
1240 o Add long file support for binary pg_dump output
1242 While Win32 supports 64-bit files, the MinGW API does not,
1243 meaning we have to build an fseeko replacement on top of the
1244 Win32 API, and we have to make sure MinGW handles it. Another
1245 option is to wait for the MinGW project to fix it, or use the
1246 code from the LibGW32C project as a guide.
1249 o Check WSACancelBlockingCall() for interrupts (win32intr)
1252 * Wire Protocol Changes
1254 o Allow dynamic character set handling
1255 o Add decoded type, length, precision
1257 o Update clients to use data types, typmod, schema.table.column names
1258 of result sets using new statement protocol
1261 ---------------------------------------------------------------------------
1264 Developers who have claimed items are:
1265 --------------------------------------
1266 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1267 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1268 * Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
1269 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1270 Family Health Network
1271 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1272 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1273 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1274 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1275 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1276 * Joe is Joe Conway <mail@joeconway.com>
1277 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1278 * Magnus is Magnus Hagander <mha@sollentuna.net>
1279 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1280 * Matthew T. O'Connor <matthew@zeut.net>
1281 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1282 * Neil is Neil Conway <neilc@samurai.com>
1283 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1284 * Peter is Peter Eisentraut <peter_e@gmx.net>
1285 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1286 * Rod is Rod Taylor <pg@rbt.ca>
1287 * Simon is Simon Riggs <simon@2ndquadrant.com>
1288 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1289 * Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
1290 * Teodor is Teodor Sigaev <teodor@sigaev.ru>
1291 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat