4 Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
5 Last updated: Tue Aug 9 13:14:41 EDT 2005
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 8.1 release.#
12 Bracketed items, "[]", have more detail.
14 This list contains all known PostgreSQL bugs and feature requests. If
15 you would like to work on an item, please read the Developer's FAQ
22 * Remove behavior of postmaster -o after making postmaster/postgres
24 * -Allow limits on per-db/role connections
25 * Allow pooled connections to list all prepared queries
27 This would allow an application inheriting a pooled connection to know
28 the queries prepared in the current session.
30 * Allow major upgrades without dump/reload, perhaps using pg_upgrade
32 * Check for unreferenced table files created by transactions that were
33 in-progress when the server terminated abruptly
34 * Allow administrators to safely terminate individual sessions either
35 via an SQL function or SIGTERM
37 Currently SIGTERM of a backend can lead to lock table corruption.
39 * -Prevent dropping user that still owns objects, or auto-drop the objects
40 * Set proper permissions on non-system schemas during db creation
42 Currently all schemas are owned by the super-user because they are
43 copied from the template1 database.
45 * -Add the client IP address and port to pg_stat_activity
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
51 * Improve replication solutions
55 You can use any of the master/slave replication servers to use a
56 standby server for data warehousing. To allow read/write queries to
57 multiple servers, you need multi-master replication like pgcluster.
59 o Allow replication over unreliable or non-persistent links
64 o Add "include file" functionality in postgresql.conf
65 o Allow postgresql.conf values to be set so they can not be changed
67 o Allow commenting of variables in postgresql.conf to restore them
69 o Allow pg_hba.conf settings to be controlled via SQL
71 This would add a function to load the SQL table from
72 pg_hba.conf, and one to writes its contents to the flat file.
73 The table should have a line number that is a float so rows
74 can be inserted between existing rows, e.g. row 2.5 goes
75 between row 2 and row 3.
77 o Allow postgresql.conf file values to be changed via an SQL
78 API, perhaps using SET GLOBAL
79 o Allow the server to be stopped/restarted via an SQL API
84 * Allow a database in tablespace t1 with tables created in
85 tablespace t2 to be used as a template for a new database created
86 with default tablespace t2
88 All objects in the default database tablespace must have default
89 tablespace specifications. This is because new databases are
90 created by copying directories. If you mix default tablespace
91 tables and tablespace-specified tables in the same directory,
92 creating a new database from such a mixed directory would create a
93 new database with tables that had incorrect explicit tablespaces.
94 To fix this would require modifying pg_class in the newly copied
95 database, which we don't currently do.
97 * Allow reporting of which objects are in which tablespaces
99 This item is difficult because a tablespace can contain objects
100 from multiple databases. There is a server-side function that
101 returns the databases which use a specific tablespace, so this
102 requires a tool that will call that function and connect to each
103 database to find the objects in each database for that tablespace.
105 o Add a GUC variable to control the tablespace for temporary objects
108 It could start with a random tablespace from a supplied list and
109 cycle through the list.
111 o Allow WAL replay of CREATE TABLESPACE to work when the directory
112 structure on the recovery computer is different from the original
114 o Allow per-tablespace quotas
117 * Point-In-Time Recovery (PITR)
119 o Allow point-in-time recovery to archive partially filled
120 write-ahead logs [pitr]
122 Currently only full WAL files are archived. This means that the
123 most recent transactions aren't available for recovery in case
124 of a disk failure. This could be triggered by a user command or
127 o Automatically force archiving of partially-filled WAL files when
128 pg_stop_backup() is called or the server is stopped
130 Doing this will allow administrators to know more easily when
131 the archive contins all the files needed for point-in-time
134 o Create dump tool for write-ahead logs for use in determining
135 transaction id for point-in-time recovery
136 o Allow a warm standby system to also allow read-only queries
139 This is useful for checking PITR recovery.
141 o Allow the PITR process to be debugged and data examined
147 * Allow server log information to be output as INSERT statements
149 This would allow server log information to be easily loaded into
150 a database for analysis.
152 * Add ability to monitor the use of temporary sort files
153 * -Add session start time and last statement time to pg_stat_activity
154 * -Add a function that returns the start time of the postmaster
155 * Allow server logs to be remotely read and removed using SQL commands
161 * Remove Money type, add money formatting for decimal type
162 * Change NUMERIC to enforce the maximum precision, and increase it
163 * Add NUMERIC division operator that doesn't round?
165 Currently NUMERIC _rounds_ the result to the specified precision.
166 This means division can return a result that multiplied by the
167 divisor is greater than the dividend, e.g. this returns a value > 10:
169 SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
171 The positive modulus result returned by NUMERICs might be considered
172 inaccurate, in one sense.
174 * Have sequence dependency track use of DEFAULT sequences,
176 * Disallow changing default expression of a SERIAL column?
177 * Fix data types where equality comparison isn't intuitive, e.g. box
178 * Prevent INET cast to CIDR if the unmasked bits are not zero, or
180 * Prevent INET cast to CIDR from droping netmask, SELECT '1.1.1.1'::inet::cidr
181 * Allow INET + INT4 to increment the host part of the address, or
182 throw an error on overflow
183 * Add 'tid != tid ' operator for use in corruption recovery
188 o Allow infinite dates just like infinite timestamps
189 o Add a GUC variable to allow output of interval values in ISO8601
191 o Merge hardwired timezone names with the TZ database; allow either
192 kind everywhere a TZ name is currently taken
193 o Allow customization of the known set of TZ names (generalize the
194 present australian_timezones hack)
195 o Allow TIMESTAMP WITH TIME ZONE to store the original timezone
196 information, either zone name or offset from UTC [timezone]
198 If the TIMESTAMP value is stored with a time zone name, interval
199 computations should adjust based on the time zone rules.
201 o Add ISO INTERVAL handling
202 o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
204 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
205 o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
206 '1', treat as ISO if there is a range specification clause,
207 and as PG if there no clause is present, e.g. interpret
208 '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
209 interpret '1:30' as '1 hour, 30 minutes'
210 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
211 INTERVAL MONTH), and this should return '12 months'
212 o Round or truncate values to the requested precision, e.g.
213 INTERVAL '11 months' AS YEAR should return one or zero
214 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
219 o Allow NULLs in arrays
220 o Allow MIN()/MAX() on arrays
221 o Delay resolution of array expression's data type so assignment
222 coercion can be performed on empty array expressions
223 o Modify array literal representation to handle array index lower bound
229 o Improve vacuum of large objects, like /contrib/vacuumlo?
230 o Add security checking for large objects
232 Currently large objects entries do not have owners. Permissions can
233 only be set at the pg_largeobject table level.
235 o Auto-delete large objects when referencing row is deleted
236 o Allow read/write into TOAST values like large objects
238 This requires the TOAST column to be stored EXTERNAL.
244 * -Add function to return compressed length of TOAST data values
245 * Allow INET subnet tests using non-constants to be indexed
246 * Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
249 Current CURRENT_TIMESTAMP returns the start time of the current
250 transaction, and gettimeofday() returns the wallclock time. This will
251 make time reporting more consistent and will allow reporting of
252 the statement start time.
254 * Add pg_get_acldef(), pg_get_typedefault(), and pg_get_attrdef()
255 * Allow to_char() to print localized month names
256 * Allow functions to have a schema search path specified at creation time
257 * Allow substring/replace() to get/set bit values
258 * Allow to_char() on interval values to accumulate the highest unit
261 Some special format flag would be required to request such
262 accumulation. Such functionality could also be added to EXTRACT.
263 Prevent accumulation that crosses the month/day boundary because of
264 the uneven number of days in a month.
266 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
267 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
268 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
269 o to_char(INTERVAL '3 years 5 months','MM') => 41
271 * Prevent to_char() on interval from returning meaningless values
273 For example, to_char('1 month', 'mon') is meaningless. Basically,
274 most date-related parameters to to_char() are meaningless for
275 intervals because interval is not anchored to a date.
278 Multi-Language Support
279 ======================
281 * Add NCHAR (as distinguished from ordinary varchar),
282 * Allow locale to be set at database creation
284 Currently locale can only be set during initdb. No global tables have
285 locale-aware columns. However, the database template used during
286 database creation might have locale-aware indexes. The indexes would
287 need to be reindexed to match the new locale.
289 * Allow encoding on a per-column basis
291 Right now only one encoding is allowed per database.
293 * Support multiple simultaneous character sets, per SQL92
294 * Improve UTF8 combined character handling?
295 * Add octet_length_server() and octet_length_client()
296 * Make octet_length_client() the same as octet_length()?
302 * Automatically create rules on views so they are updateable, per SQL99
304 We can only auto-create rules for simple views. For more complex
305 cases users will still have to write rules.
307 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
308 * Allow NOTIFY in rules involving conditionals
309 * Have views on temporary tables exist in the temporary namespace
310 * Allow temporary views on non-temporary tables
311 * Allow RULE recompilation
317 * -Add BETWEEN SYMMETRIC/ASYMMETRIC
318 * Change LIMIT/OFFSET and FETCH/MOVE to use int8
319 * -Add E'' escape string marker so eventually ordinary strings can treat
320 backslashes literally, for portability
322 * -Allow additional tables to be specified in DELETE for joins
324 UPDATE already allows this (UPDATE...FROM) but we need similar
325 functionality in DELETE. It's been agreed that the keyword should
326 be USING, to avoid anything as confusing as DELETE FROM a FROM b.
328 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
329 * -Allow REINDEX to rebuild all database indexes
330 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
331 * Allow SET CONSTRAINTS to be qualified by schema/table name
332 * Allow TRUNCATE ... CASCADE/RESTRICT
333 * Add a separate TRUNCATE permission
335 Currently only the owner can TRUNCATE a table because triggers are not
336 called, and the table is locked in exclusive mode.
338 * Allow PREPARE of cursors
339 * Allow PREPARE to automatically determine parameter types based on the SQL
341 * Allow finer control over the caching of prepared query plans
343 Currently, queries prepared via the libpq API are planned on first
344 execute using the supplied parameters --- allow SQL PREPARE to do the
345 same. Also, allow control over replanning prepared queries either
346 manually or automatically when statistics for execute parameters
347 differ dramatically from those used during planning.
349 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
351 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
352 such information in memory would improve performance.
354 * Add optional textual message to NOTIFY
356 This would allow an informational message to be added to the notify
357 message, perhaps indicating the row modified or other custom
360 * Add a GUC variable to warn about non-standard SQL usage in queries
361 * Add MERGE command that does UPDATE/DELETE, or on failure, INSERT (rules,
363 * Add NOVICE output level for helpful messages like automatic sequence/index
365 * Add COMMENT ON for all cluster global objects (roles, databases
367 * -Add an option to automatically use savepoints for each statement in a
368 multi-statement transaction.
370 When enabled, this would allow errors in multi-statement transactions
371 to be automatically ignored.
373 * Make row-wise comparisons work per SQL spec
374 * Add RESET CONNECTION command to reset all session state
376 This would include resetting of all variables (RESET ALL), dropping of
377 temporary tables, removing any NOTIFYs, cursors, open transactions,
378 prepared queries, currval()s, etc. This could be used for connection
379 pooling. We could also change RESET ALL to have this functionality.
380 The difficult of this features is allowing RESET ALL to not affect
381 changes made by the interface driver for its internal use. One idea
382 is for this to be a protocol-only feature. Another approach is to
383 notify the protocol when a RESET CONNECTION command is used.
385 * Add GUC to issue notice about queries that use unjoined tables
386 * Allow EXPLAIN to identify tables that were skipped because of
387 enable_constraint_exclusion
388 * Allow EXPLAIN output to be more easily processed by scripts
393 o Allow CREATE TABLE AS to determine column lengths for complex
394 expressions like SELECT col1 || col2
396 o Use more reliable method for CREATE DATABASE to get a consistent
399 o Currently the system uses the operating system COPY command to
400 create a new database. Add ON COMMIT capability to CREATE TABLE AS
405 o Allow UPDATE to handle complex aggregates [update]?
406 o Allow an alias to be provided for the target table in
409 This is not SQL-spec but many DBMSs allow it.
411 o Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple
413 o -Allow FOR UPDATE queries to do NOWAIT locks
418 o Have ALTER TABLE RENAME rename SERIAL sequence names
419 o Add ALTER DOMAIN TYPE
420 o Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
421 o Allow ALTER TABLE to change constraint deferrability and actions
422 o Disallow dropping of an inherited constraint
423 o -Allow objects to be moved to different schemas
424 o Allow ALTER TABLESPACE to move to different directories
425 o Allow databases to be moved to different tablespaces
426 o Allow moving system tables to other tablespaces, where possible
428 Currently non-global system tables must be in the default database
429 tablespace. Global system tables can never be moved.
431 o Prevent child tables from altering constraints like CHECK that were
432 inherited from the parent table
437 o Automatically maintain clustering on a table
439 This might require some background daemon to maintain clustering
440 during periods of low usage. It might also require tables to be only
441 paritally filled for easier reorganization. Another idea would
442 be to create a merged heap/index data file so an index lookup would
443 automatically access the heap data too. A third idea would be to
444 store heap rows in hashed groups, perhaps using a user-supplied
447 o Add default clustering to system tables
449 To do this, determine the ideal cluster index for each system
450 table and set the cluster setting during initdb.
455 o Allow COPY to report error lines and continue
457 This requires the use of a savepoint before each COPY line is
458 processed, with ROLLBACK on COPY failure.
460 o -Allow COPY to understand \x as a hex byte
461 o Have COPY return the number of rows loaded/unloaded?
462 o -Allow COPY to optionally include column headings in the first line
463 o -Allow COPY FROM ... CSV to interpret newlines and carriage
469 o Allow column-level privileges
470 o Allow GRANT/REVOKE permissions to be applied to all schema objects
473 The proposed syntax is:
474 GRANT SELECT ON ALL TABLES IN public TO phpuser;
475 GRANT SELECT ON NEW TABLES IN public TO phpuser;
477 * Allow GRANT/REVOKE permissions to be inherited by objects based on
483 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
485 This requires using the row ctid to map cursor rows back to the
486 original heap row. This become more complicated if WITH HOLD cursors
487 are to be supported because WITH HOLD cursors have a copy of the row
488 and no FOR UPDATE lock.
490 o Prevent DROP TABLE from dropping a row referenced by its own open
493 o Allow pooled connections to list all open WITH HOLD cursors
495 Because WITH HOLD cursors exist outside transactions, this allows
496 them to be listed so they can be closed.
501 o Allow INSERT/UPDATE of the system-generated oid value for a row
502 o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
503 o Allow INSERT/UPDATE ... RETURNING new.col or old.col
505 This is useful for returning the auto-generated key for an INSERT.
506 One complication is how to handle rules that run as part of
512 o -Have SHOW ALL show descriptions for server-side variables
513 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
515 o Add SET PATH for schemas?
517 This is basically the same as SET search_path.
520 * Server-Side Languages
522 o -Allow PL/PgSQL's RAISE function to take expressions
524 Currently only constants are supported.
526 o -Change PL/PgSQL to use palloc() instead of malloc()
527 o Handle references to temporary tables that are created, destroyed,
528 then recreated during a session, and EXECUTE is not used
530 This requires the cached PL/PgSQL byte code to be invalidated when
531 an object referenced in the function is changed.
533 o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
534 o Allow function parameters to be passed by name,
535 get_employee_salary(emp_id => 12345, tax_year => 2001)
536 o Add Oracle-style packages
537 o Add table function support to pltcl, plperl, plpython?
538 o Allow PL/pgSQL to name columns by ordinal position, e.g. rec.(3)
539 o -Allow PL/pgSQL EXECUTE query_var INTO record_var;
540 o Add capability to create and call PROCEDURES
541 o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
542 o Add MOVE to PL/pgSQL
543 o Pass arrays natively instead of as text between plperl and postgres
544 o Add support for polymorphic arguments and return types to plperl
550 * Add a libpq function to support Parse/DescribeStatement capability
551 * Prevent libpq's PQfnumber() from lowercasing the column name?
552 * Allow libpq to access SQLSTATE so pg_ctl can test for connection failure
554 This would be used for checking if the server is up.
556 * Add PQescapeIdentifier() to libpq
557 * Have initdb set DateStyle based on locale?
558 * Have pg_ctl look at PGHOST in case it is a socket directory?
559 * Add a schema option to createlang
560 * Allow pg_ctl to work properly with configuration files located outside
563 pg_ctl can not read the pid file because it isn't located in the
564 config directory but in the PGDATA directory. The solution is to
565 allow pg_ctl to read and understand postgresql.conf to find the
566 data_directory value.
571 o Have psql show current values for a sequence
572 o Move psql backslash database information into the backend, use
573 mnemonic commands? [psql]
575 This would allow non-psql clients to pull the same information out
576 of the database as psql.
578 o Fix psql's display of schema information (Neil)
579 o Allow psql \pset boolean variables to set to fixed values, rather
581 o Consistently display privilege information for all objects in psql
582 o Improve psql's handling of multi-line queries
587 o Have pg_dump use multi-statement transactions for INSERT dumps
588 o Allow pg_dump to use multiple -t and -n switches [pg_dump]
589 o Add dumping of comments on composite type columns
590 o Add dumping of comments on index columns
591 o Replace crude DELETE FROM method of pg_dumpall --clean for
592 cleaning of roles with separate DROP commands
593 o -Add dumping and restoring of LOB comments
594 o Stop dumping CASCADE on DROP TYPE commands in clean mode
595 o Add full object name to the tag field. eg. for operators we need
596 '=(integer, integer)', instead of just '='.
597 o Add pg_dumpall custom format dumps.
599 This is probably best done by combining pg_dump and pg_dumpall
600 into a single binary.
602 o Add CSV output format
603 o Update pg_dump and psql to use the new COPY libpq API (Christopher)
610 Document differences between ecpg and the SQL standard and
611 information about the Informix-compatibility module.
613 o Solve cardinality > 1 for input descriptors / variables?
614 o Add a semantic check level, e.g. check if a table really exists
615 o fix handling of DB attributes that are arrays
616 o Use backend PREPARE/EXECUTE facility for ecpg where possible
618 o Fix nested C comments
619 o sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
620 o Make SET CONNECTION thread-aware, non-standard?
621 o Allow multidimensional arrays
622 o Add internationalized message strings
625 Referential Integrity
626 =====================
628 * Add MATCH PARTIAL referential integrity
629 * Add deferred trigger queue file
631 Right now all deferred trigger information is stored in backend
632 memory. This could exhaust memory for very large trigger queues.
633 This item involves dumping large queues into files.
635 * -Implement shared row locks and use them in RI triggers
636 * Change foreign key constraint for array -> element to mean element
638 * Allow DEFERRABLE UNIQUE constraints?
639 * Allow triggers to be disabled [trigger]
641 Currently the only way to disable triggers is to modify the system
644 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
646 If the dump is known to be valid, allow foreign keys to be added
647 without revalidating the data.
649 * Allow statement-level triggers to access modified rows
650 * Support triggers on columns (Greg Sabino Mullane)
651 * Remove CREATE CONSTRAINT TRIGGER
653 This was used in older releases to dump referential integrity
656 * Enforce referential integrity for system tables
657 * Allow AFTER triggers on system tables
659 System tables are modified in many places in the backend without going
660 through the executor and therefore not causing triggers to fire. To
661 complete this item, the functions that modify system tables will have
668 * Flush cached query plans when the dependent objects change
669 * Track dependencies in function bodies and recompile/invalidate
675 * Add SQL99 WITH clause to SELECT
676 * Add SQL99 WITH RECURSIVE to SELECT
677 * Add pre-parsing phase that converts non-ISO syntax to supported
680 This could allow SQL written for other databases to run without
683 * Allow plug-in modules to emulate features from other databases
684 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
686 * Allow queries across databases or servers with transaction
689 This can be done using dblink and two-phase commit.
691 * -Add two-phase commit
694 * Add the features of packages
696 o Make private objects accessable only to objects in the same schema
697 o Allow current_schema.objname to access current schema objects
698 o Add session variables
699 o Allow nested schemas
705 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
707 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
708 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
711 The main difficulty with this item is the problem of creating an index
712 that can span more than one table.
714 * Allow SELECT ... FOR UPDATE on inherited tables
715 * -Prevent inherited tables from expanding temporary subtables of other
717 * Add UNIQUE capability to non-btree indexes
718 * -Use indexes for MIN() and MAX()
720 MIN/MAX queries can already be rewritten as SELECT col FROM tab ORDER
721 BY col {DESC} LIMIT 1. Completing this item involves doing this
722 transformation automatically.
724 * -Use index to restrict rows returned by multi-key index when used with
725 non-consecutive keys to reduce heap accesses
727 For an index on col1,col2,col3, and a WHERE clause of col1 = 5 and
728 col3 = 9, spin though the index checking for col1 and col3 matches,
729 rather than just col1; also called skip-scanning.
731 * Prevent index uniqueness checks when UPDATE does not modify the column
733 Uniqueness (index) checks are done when updating a column even if the
734 column is not modified by the UPDATE.
736 * Fetch heap pages matching index entries in sequential order
738 Rather than randomly accessing heap pages based on index entries, mark
739 heap pages needing access in a bitmap and do the lookups in sequential
740 order. Another method would be to sort heap ctids matching the index
741 before accessing the heap rows.
743 * -Allow non-bitmap indexes to be combined by creating bitmaps in memory
745 This feature allows separate indexes to be ANDed or ORed together. This
746 is particularly useful for data warehousing applications that need to
747 query the database in an many permutations. This feature scans an index
748 and creates an in-memory bitmap, and allows that bitmap to be combined
749 with other bitmap created in a similar way. The bitmap can either index
750 all TIDs, or be lossy, meaning it records just page numbers and each
751 page tuple has to be checked for validity in a separate pass.
753 * Allow the creation of on-disk bitmap indexes which can be quickly
754 combined with other bitmap indexes
756 Such indexes could be more compact if there are only a few distinct values.
757 Such indexes can also be compressed. Keeping such indexes updated can be
760 * Allow use of indexes to search for NULLs
762 One solution is to create a partial index on an IS NULL expression.
764 * Allow accurate statistics to be collected on indexes with more than
765 one column or expression indexes, perhaps using per-index statistics
766 * Add fillfactor to control reserved free space during index creation
767 * Allow the creation of indexes with mixed ascending/descending specifiers
768 * -Fix incorrect rtree results due to wrong assumptions about "over"
770 * Allow enable_constraint_exclusion to work for UNIONs like it does for
772 * Allow enable_constraint_exclusion to work for UPDATE and DELETE queries
777 o Add more GIST index support for geometric data types
778 o -Add concurrency to GIST
779 o Allow GIST indexes to create certain complex index types, like
780 digital trees (see Aoki)
784 o Pack hash index buckets onto disk pages more efficiently
786 Currently no only one hash bucket can be stored on a page. Ideally
787 several hash buckets could be stored on a single page and greater
788 granularity used for the hash algorithm.
790 o Consider sorting hash buckets so entries can be found using a
791 binary search, rather than a linear scan
793 o In hash indexes, consider storing the hash value with or instead
800 * Improve commit_delay handling to reduce fsync()
801 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
802 * -Allow multiple blocks to be written to WAL with one write()
803 * Add an option to sync() before fsync()'ing checkpoint files
804 * Add program to test if fsync has a delay compared to non-fsync
810 * Allow free-behind capability for large sequential scans, perhaps using
813 Posix_fadvise() can control both sequential/random file caching and
814 free-behind behavior, but it is unclear how the setting affects other
815 backends that also have the file open, and the feature is not supported
816 on all operating systems.
818 * -Consider use of open/fcntl(O_DIRECT) to minimize OS caching,
821 O_DIRECT doesn't have the same media write guarantees as fsync, so it
822 is in addition to the fsync method, not in place of it.
824 * -Cache last known per-tuple offsets to speed long tuple access
827 We could use a fixed row count and a +/- count to follow MVCC
828 visibility rules, or a single cached value could be used and
829 invalidated if anyone modifies the table. Another idea is to
830 get a count directly from a unique index, but for this to be
831 faster than a sequential scan it must avoid access to the heap
832 to obtain tuple visibility information.
834 * Allow data to be pulled directly from indexes
836 Currently indexes do not have enough tuple visibility information
837 to allow data to be pulled from the index without also accessing
838 the heap. One way to allow this is to set a bit to index tuples
839 to indicate if a tuple is currently visible to all transactions
840 when the first valid heap lookup happens. This bit would have to
841 be cleared when a heap tuple is expired.
844 * Consider automatic caching of queries at various levels:
850 * -Allow the size of the buffer cache used by temporary objects to be
851 specified as a GUC variable
853 Larger local buffer cache sizes requires more efficient handling of
856 * Improve the background writer
858 Allow the background writer to more efficiently write dirty buffers
859 from the end of the LRU cache and use a clock sweep algorithm to
860 write other dirty buffers to reduced checkpoint I/O
862 * Allow sequential scans to take advantage of other concurrent
863 sequentiqal scans, also called "Synchronised Scanning"
865 One possible implementation is to start sequential scans from the lowest
866 numbered buffer in the shared cache, and when reaching the end wrap
867 around to the beginning, rather than always starting sequential scans
868 at the start of the table.
874 * Improve speed with indexes
876 For large table adjustements during vacuum, it is faster to reindex
877 rather than update the index.
879 * Reduce lock time by moving tuples with read lock, then write
880 lock and truncate table
882 Moved tuples are invisible to other backends so they don't require a
883 write lock. However, the read lock promotion to write lock could lead
884 to deadlock situations.
886 * -Add a warning when the free space map is too small
887 * Maintain a map of recently-expired rows
889 This allows vacuum to target specific pages for possible free space
890 without requiring a sequential scan.
892 * Auto-fill the free space map by scanning the buffer cache or by
893 checking pages written by the background writer
894 * Create a bitmap of pages that need vacuuming
896 Instead of sequentially scanning the entire table, have the background
897 writer or some other process record pages that have expired rows, then
898 VACUUM can look at just those pages rather than the entire table. In
899 the event of a system crash, the bitmap would probably be invalidated.
901 * Add system view to show free space map contents
906 o -Move into the backend code
907 o Use free-space map information to guide refilling
908 o Do VACUUM FULL if table is nearly empty?
909 o Improve xid wraparound detection by recording per-table rather
916 * -Make locking of shared data structures more fine-grained
918 This requires that more locks be acquired but this would reduce lock
919 contention, improving concurrency.
921 * Add code to detect an SMP machine and handle spinlocks accordingly
922 from distributted.net, http://www1.distributed.net/source,
923 in client/common/cpucheck.cpp
925 On SMP machines, it is possible that locks might be released shortly,
926 while on non-SMP machines, the backend should sleep so the process
927 holding the lock can complete and release it.
929 * -Improve SMP performance on i386 machines
931 i386-based SMP machines can generate excessive context switching
932 caused by lock failure in high concurrency situations. This may be
933 caused by CPU cache line invalidation inefficiencies.
935 * Research use of sched_yield() for spinlock acquisition failure
936 * Fix priority ordering of read and write light-weight locks (Neil)
939 Startup Time Improvements
940 =========================
942 * Experiment with multi-threaded backend [thread]
944 This would prevent the overhead associated with process creation. Most
945 operating systems have trivial process creation time compared to
946 database startup overhead, but a few operating systems (WIn32,
947 Solaris) might benefit from threading. Also explore the idea of
948 a single session using multiple threads to execute a query faster.
950 * Add connection pooling
952 It is unclear if this should be done inside the backend code or done
953 by something external like pgpool. The passing of file descriptors to
954 existing backends is one of the difficulties with a backend approach.
960 * Eliminate need to write full pages to WAL before page modification [wal]
962 Currently, to protect against partial disk page writes, we write
963 full page images to WAL before they are modified so we can correct any
964 partial page writes during recovery. These pages can also be
965 eliminated from point-in-time archive files.
967 o -Add ability to turn off full page writes
968 o When off, write CRC to WAL and check file system blocks
971 If CRC check fails during recovery, remember the page in case
972 a later CRC for that page properly matches.
974 o Write full pages during file system write and not when
975 the page is modified in the buffer cache
977 This allows most full page writes to happen in the background
978 writer. It might cause problems for applying WAL on recovery
979 into a partially-written page, but later the full page will be
982 * Reduce WAL traffic so only modified values are written rather than
984 * Add WAL index reliability improvement to non-btree indexes
985 * Allow the pg_xlog directory location to be specified during initdb
986 with a symlink back to the /data location
987 * Allow WAL information to recover corrupted pg_controldata
988 * Find a way to reduce rotational delay when repeatedly writing
991 Currently fsync of WAL requires the disk platter to perform a full
992 rotation to fsync again. One idea is to write the WAL to different
993 offsets that might reduce the rotational delay.
995 * Allow buffered WAL writes and fsync
997 Instead of guaranteeing recovery of all committed transactions, this
998 would provide improved performance by delaying WAL writes and fsync
999 so an abrupt operating system restart might lose a few seconds of
1000 committed transactions but still be consistent. We could perhaps
1001 remove the 'fsync' parameter (which results in an an inconsistent
1002 database) in favor of this capability.
1004 * -Eliminate WAL logging for CREATE TABLE AS when not doing WAL archiving
1005 * -Change WAL to use 32-bit CRC, for performance reasons
1008 Optimizer / Executor
1009 ====================
1011 * Add missing optimizer selectivities for date, r-tree, etc
1012 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
1013 index using a sequential scan for highest/lowest values
1015 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
1016 all values to return the high/low value. Instead The idea is to do a
1017 sequential scan to find the high/low value, thus avoiding the sort.
1018 MIN/MAX already does this, but not for LIMIT > 1.
1020 * Precompile SQL functions to avoid overhead
1021 * Create utility to compute accurate random_page_cost value
1022 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
1023 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
1024 * -Use CHECK constraints to influence optimizer decisions
1026 CHECK constraints contain information about the distribution of values
1027 within the table. This is also useful for implementing subtables where
1028 a tables content is distributed across several subtables.
1030 * Consider using hash buckets to do DISTINCT, rather than sorting
1032 This would be beneficial when there are few distinct values.
1034 * ANALYZE should record a pg_statistic entry for an all-NULL column
1035 * Log queries where the optimizer row estimates were dramatically
1036 different from the number of rows actually found?
1039 Miscellaneous Performance
1040 =========================
1042 * Do async I/O for faster random read-ahead of data
1044 Async I/O allows multiple I/O requests to be sent to the disk with
1045 results coming back asynchronously.
1047 * Use mmap() rather than SYSV shared memory or to write WAL files?
1049 This would remove the requirement for SYSV SHM but would introduce
1050 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
1051 to prevent I/O overhead.
1053 * Consider mmap()'ing files into a backend?
1055 Doing I/O to large tables would consume a lot of address space or
1056 require frequent mapping/unmapping. Extending the file also causes
1057 mapping problems that might require mapping only individual pages,
1058 leading to thousands of mappings. Another problem is that there is no
1059 way to _prevent_ I/O to disk from the dirty shared buffers so changes
1060 could hit disk before WAL is written.
1062 * Add a script to ask system configuration questions and tune postgresql.conf
1063 * Use a phantom command counter for nested subtransactions to reduce
1065 * Research storing disk pages with no alignment/padding
1070 * Add use of 'const' for variables in source tree
1071 * Rename some /contrib modules from pg* to pg_*
1072 * Move some things from /contrib into main tree
1073 * Move some /contrib modules out to their own project sites
1074 * Remove warnings created by -Wcast-align
1075 * Move platform-specific ps status display info from ps_status.c to ports
1076 * Add optional CRC checksum to heap and index pages
1077 * Improve documentation to build only interfaces (Marc)
1078 * Remove or relicense modules that are not under the BSD license, if possible
1079 * Remove memory/file descriptor freeing before ereport(ERROR)
1080 * Acquire lock on a relation before building a relcache entry for it
1081 * Promote debug_query_string into a server-side function current_query()
1082 * Allow the identifier length to be increased via a configure option
1083 * Remove Win32 rename/unlink looping if unnecessary
1084 * -Remove kerberos4 from source tree
1085 * Allow cross-compiling by generating the zic database on the target system
1086 * Improve NLS maintenace of libpgport messages linked onto applications
1087 * Allow ecpg to work with MSVC and BCC
1088 * -Make src/port/snprintf.c thread-safe
1089 * Add xpath_array() to /contrib/xml2 to return results as an array
1090 * Allow building in directories containing spaces
1092 This is probably not possible because 'gmake' and other compiler tools
1093 do not fully support quoting of paths with spaces.
1095 * Allow installing to directories containing spaces
1097 This is possible if proper quoting is added to the makefiles for the
1098 install targets. Because PostgreSQL supports relocatable installs, it
1099 is already possible to install into a directory that doesn't contain
1100 spaces and then copy the install to a directory with spaces.
1102 * Fix cross-compiling of time zone database via 'zic'
1103 * Fix sgmltools so PDFs can be generated with bookmarks
1104 * -Add C code on Unix to copy directories for use in creating new databases
1109 o Remove configure.in check for link failure when cause is found
1110 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1112 o Remove psql newline patch when we find out why mingw outputs an
1114 o Allow psql to use readline once non-US code pages work with
1116 o Re-enable timezone output on log_line_prefix '%t' when a
1117 shorter timezone string is available
1118 o Improve dlerror() reporting string
1119 o Fix problem with shared memory on the Win32 Terminal Server
1120 o Add support for Unicode
1122 To fix this, the data needs to be converted to/from UTF16/UTF8
1123 so the Win32 wcscoll() can be used, and perhaps other functions
1124 like towupper(). However, UTF8 already works with normal
1125 locales but provides no ordering or character set classes.
1128 * Wire Protocol Changes
1130 o Allow dynamic character set handling
1131 o Add decoded type, length, precision
1133 o Update clients to use data types, typmod, schema.table.column names
1134 of result sets using new query protocol
1137 ---------------------------------------------------------------------------
1140 Developers who have claimed items are:
1141 --------------------------------------
1142 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1143 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1144 * Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
1145 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1146 Family Health Network
1147 * Claudio is Claudio Natoli <claudio.natoli@memetrics.com>
1148 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1149 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1150 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1151 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1152 * Hiroshi is Hiroshi Inoue <Inoue@tpf.co.jp>
1153 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1154 * Joe is Joe Conway <mail@joeconway.com>
1155 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1156 * Magnus is Magnus Hagander <mha@sollentuna.net>
1157 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1158 * Matthew T. O'Connor <matthew@zeut.net>
1159 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1160 * Neil is Neil Conway <neilc@samurai.com>
1161 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1162 * Peter is Peter Eisentraut <peter_e@gmx.net>
1163 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1164 * Rod is Rod Taylor <pg@rbt.ca>
1165 * Simon is Simon Riggs <simon@2ndquadrant.com>
1166 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1167 * Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
1168 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat