4 Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
5 Last updated: Wed Aug 24 10:24:13 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.#
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 after making postmaster/postgres
25 * -Allow limits on per-db/role connections
26 * %Allow pooled connections to list all prepared queries
28 This would allow an application inheriting a pooled connection to know
29 the queries prepared in the current session.
31 * Allow major upgrades without dump/reload, perhaps using pg_upgrade
33 * Check for unreferenced table files created by transactions that were
34 in-progress when the server terminated abruptly
35 * Allow administrators to safely terminate individual sessions either
36 via an SQL function or SIGTERM
38 Currently SIGTERM of a backend can lead to lock table corruption.
40 * -Prevent dropping user that still owns objects, or auto-drop the objects
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 * -Add the client IP address and port to pg_stat_activity
47 * Support table partitioning that allows a single table to be stored
48 in subtables that are partitioned based on the primary key or a WHERE
52 * Improve replication solutions
56 You can use any of the master/slave replication servers to use a
57 standby server for data warehousing. To allow read/write queries to
58 multiple servers, you need multi-master replication like pgcluster.
60 o Allow replication over unreliable or non-persistent links
65 o %Add "include file" functionality in postgresql.conf
66 o Allow postgresql.conf values to be set so they can not be changed
68 o %Allow commenting of variables in postgresql.conf to restore them
71 Currently, if a variable is commented out, it keeps the
72 previous uncommented value until a server restarted.
74 o %Allow pg_hba.conf settings to be controlled via SQL
76 This would add a function to load the SQL table from
77 pg_hba.conf, and one to writes its contents to the flat file.
78 The table should have a line number that is a float so rows
79 can be inserted between existing rows, e.g. row 2.5 goes
80 between row 2 and row 3.
82 o %Allow postgresql.conf file values to be changed via an SQL
83 API, perhaps using SET GLOBAL
84 o Allow the server to be stopped/restarted via an SQL API
89 * Allow a database in tablespace t1 with tables created in
90 tablespace t2 to be used as a template for a new database created
91 with default tablespace t2
93 All objects in the default database tablespace must have default
94 tablespace specifications. This is because new databases are
95 created by copying directories. If you mix default tablespace
96 tables and tablespace-specified tables in the same directory,
97 creating a new database from such a mixed directory would create a
98 new database with tables that had incorrect explicit tablespaces.
99 To fix this would require modifying pg_class in the newly copied
100 database, which we don't currently do.
102 * Allow reporting of which objects are in which tablespaces
104 This item is difficult because a tablespace can contain objects
105 from multiple databases. There is a server-side function that
106 returns the databases which use a specific tablespace, so this
107 requires a tool that will call that function and connect to each
108 database to find the objects in each database for that tablespace.
110 o %Add a GUC variable to control the tablespace for temporary objects
113 It could start with a random tablespace from a supplied list and
114 cycle through the list.
116 o Allow WAL replay of CREATE TABLESPACE to work when the directory
117 structure on the recovery computer is different from the original
119 o Allow per-tablespace quotas
122 * Point-In-Time Recovery (PITR)
124 o Allow point-in-time recovery to archive partially filled
125 write-ahead logs [pitr]
127 Currently only full WAL files are archived. This means that the
128 most recent transactions aren't available for recovery in case
129 of a disk failure. This could be triggered by a user command or
132 o Automatically force archiving of partially-filled WAL files when
133 pg_stop_backup() is called or the server is stopped
135 Doing this will allow administrators to know more easily when
136 the archive contins all the files needed for point-in-time
139 o %Create dump tool for write-ahead logs for use in determining
140 transaction id for point-in-time recovery
141 o Allow a warm standby system to also allow read-only queries
144 This is useful for checking PITR recovery.
146 o Allow the PITR process to be debugged and data examined
152 * Allow server log information to be output as INSERT statements
154 This would allow server log information to be easily loaded into
155 a database for analysis.
157 * %Add ability to monitor the use of temporary sort files
158 * -Add session start time and last statement time to pg_stat_activity
159 * -Add a function that returns the start time of the postmaster
160 * Allow server logs to be remotely read and removed using SQL commands
166 * %Remove Money type, add money formatting for decimal type
167 * Change NUMERIC to enforce the maximum precision, and increase it
168 * Add NUMERIC division operator that doesn't round?
170 Currently NUMERIC _rounds_ the result to the specified precision.
171 This means division can return a result that multiplied by the
172 divisor is greater than the dividend, e.g. this returns a value > 10:
174 SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
176 The positive modulus result returned by NUMERICs might be considered
177 inaccurate, in one sense.
179 * Have sequence dependency track use of DEFAULT sequences,
181 * %Disallow changing default expression of a SERIAL column?
182 * Fix data types where equality comparison isn't intuitive, e.g. box
183 * %Prevent INET cast to CIDR if the unmasked bits are not zero, or
185 * %Prevent INET cast to CIDR from droping netmask, SELECT '1.1.1.1'::inet::cidr
186 * Allow INET + INT4 to increment the host part of the address, or
187 throw an error on overflow
188 * %Add 'tid != tid ' operator for use in corruption recovery
193 o Allow infinite dates just like infinite timestamps
194 o Add a GUC variable to allow output of interval values in ISO8601
196 o Merge hardwired timezone names with the TZ database; allow either
197 kind everywhere a TZ name is currently taken
198 o Allow customization of the known set of TZ names (generalize the
199 present australian_timezones hack)
200 o Allow TIMESTAMP WITH TIME ZONE to store the original timezone
201 information, either zone name or offset from UTC [timezone]
203 If the TIMESTAMP value is stored with a time zone name, interval
204 computations should adjust based on the time zone rules.
206 o Add ISO INTERVAL handling
207 o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
209 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
210 o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
211 '1', treat as ISO if there is a range specification clause,
212 and as PG if there no clause is present, e.g. interpret
213 '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
214 interpret '1:30' as '1 hour, 30 minutes'
215 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
216 INTERVAL MONTH), and this should return '12 months'
217 o Round or truncate values to the requested precision, e.g.
218 INTERVAL '11 months' AS YEAR should return one or zero
219 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
224 o Allow NULLs in arrays
225 o %Allow MIN()/MAX() on arrays
226 o Delay resolution of array expression's data type so assignment
227 coercion can be performed on empty array expressions
228 o Modify array literal representation to handle array index lower bound
234 o Improve vacuum of large objects, like /contrib/vacuumlo?
235 o Add security checking for large objects
237 Currently large objects entries do not have owners. Permissions can
238 only be set at the pg_largeobject table level.
240 o Auto-delete large objects when referencing row is deleted
241 o Allow read/write into TOAST values like large objects
243 This requires the TOAST column to be stored EXTERNAL.
249 * -Add function to return compressed length of TOAST data values
250 * Allow INET subnet tests using non-constants to be indexed
251 * Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
254 Current CURRENT_TIMESTAMP returns the start time of the current
255 transaction, and gettimeofday() returns the wallclock time. This will
256 make time reporting more consistent and will allow reporting of
257 the statement start time.
259 * %Add pg_get_acldef(), pg_get_typedefault(), and pg_get_attrdef()
260 * Allow to_char() to print localized month names
261 * Allow functions to have a schema search path specified at creation time
262 * Allow substring/replace() to get/set bit values
263 * Allow to_char() on interval values to accumulate the highest unit
266 Some special format flag would be required to request such
267 accumulation. Such functionality could also be added to EXTRACT.
268 Prevent accumulation that crosses the month/day boundary because of
269 the uneven number of days in a month.
271 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
272 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
273 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
274 o to_char(INTERVAL '3 years 5 months','MM') => 41
276 * -Prevent to_char() on interval from returning meaningless values
278 For example, to_char('1 month', 'mon') is meaningless. Basically,
279 most date-related parameters to to_char() are meaningless for
280 intervals because interval is not anchored to a date.
283 Multi-Language Support
284 ======================
286 * Add NCHAR (as distinguished from ordinary varchar),
287 * Allow locale to be set at database creation
289 Currently locale can only be set during initdb. No global tables have
290 locale-aware columns. However, the database template used during
291 database creation might have locale-aware indexes. The indexes would
292 need to be reindexed to match the new locale.
294 * Allow encoding on a per-column basis
296 Right now only one encoding is allowed per database.
298 * Support multiple simultaneous character sets, per SQL92
299 * Improve UTF8 combined character handling?
300 * Add octet_length_server() and octet_length_client()
301 * Make octet_length_client() the same as octet_length()?
302 * Fix problems with wrong runtime encoding conversion for NLS message files
308 * %Automatically create rules on views so they are updateable, per SQL99
310 We can only auto-create rules for simple views. For more complex
311 cases users will still have to write rules.
313 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
314 * Allow NOTIFY in rules involving conditionals
315 * %Have views on temporary tables exist in the temporary namespace
316 * Allow temporary views on non-temporary tables
317 * %Allow RULE recompilation
323 * -Add BETWEEN SYMMETRIC/ASYMMETRIC
324 * Change LIMIT/OFFSET and FETCH/MOVE to use int8
325 * -Add E'' escape string marker so eventually ordinary strings can treat
326 backslashes literally, for portability
328 * -Allow additional tables to be specified in DELETE for joins
330 UPDATE already allows this (UPDATE...FROM) but we need similar
331 functionality in DELETE. It's been agreed that the keyword should
332 be USING, to avoid anything as confusing as DELETE FROM a FROM b.
334 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
335 * -Allow REINDEX to rebuild all database indexes
336 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
337 * %Allow SET CONSTRAINTS to be qualified by schema/table name
338 * %Allow TRUNCATE ... CASCADE/RESTRICT
339 * %Add a separate TRUNCATE permission
341 Currently only the owner can TRUNCATE a table because triggers are not
342 called, and the table is locked in exclusive mode.
344 * Allow PREPARE of cursors
345 * Allow PREPARE to automatically determine parameter types based on the SQL
347 * Allow finer control over the caching of prepared query plans
349 Currently, queries prepared via the libpq API are planned on first
350 execute using the supplied parameters --- allow SQL PREPARE to do the
351 same. Also, allow control over replanning prepared queries either
352 manually or automatically when statistics for execute parameters
353 differ dramatically from those used during planning.
355 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
357 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
358 such information in memory would improve performance.
360 * Add optional textual message to NOTIFY
362 This would allow an informational message to be added to the notify
363 message, perhaps indicating the row modified or other custom
366 * Add a GUC variable to warn about non-standard SQL usage in queries
367 * Add MERGE command that does UPDATE/DELETE, or on failure, INSERT (rules,
369 * Add NOVICE output level for helpful messages like automatic sequence/index
371 * %Add COMMENT ON for all cluster global objects (roles, databases
373 * -Add an option to automatically use savepoints for each statement in a
374 multi-statement transaction.
376 When enabled, this would allow errors in multi-statement transactions
377 to be automatically ignored.
379 * Make row-wise comparisons work per SQL spec
380 * Add RESET CONNECTION command to reset all session state
382 This would include resetting of all variables (RESET ALL), dropping of
383 temporary tables, removing any NOTIFYs, cursors, open transactions,
384 prepared queries, currval()s, etc. This could be used for connection
385 pooling. We could also change RESET ALL to have this functionality.
386 The difficult of this features is allowing RESET ALL to not affect
387 changes made by the interface driver for its internal use. One idea
388 is for this to be a protocol-only feature. Another approach is to
389 notify the protocol when a RESET CONNECTION command is used.
391 * Add GUC to issue notice about queries that use unjoined tables
392 * Allow EXPLAIN to identify tables that were skipped because of
394 * Allow EXPLAIN output to be more easily processed by scripts
399 o Allow CREATE TABLE AS to determine column lengths for complex
400 expressions like SELECT col1 || col2
402 o Use more reliable method for CREATE DATABASE to get a consistent
405 o Currently the system uses the operating system COPY command to
406 create a new database. Add ON COMMIT capability to CREATE TABLE AS
411 o Allow UPDATE to handle complex aggregates [update]?
412 o Allow an alias to be provided for the target table in
415 This is not SQL-spec but many DBMSs allow it.
417 o Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple
419 o -Allow FOR UPDATE queries to do NOWAIT locks
424 o %Have ALTER TABLE RENAME rename SERIAL sequence names
425 o %Add ALTER DOMAIN TYPE
426 o %Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
427 o %Allow ALTER TABLE to change constraint deferrability and actions
428 o %Disallow dropping of an inherited constraint
429 o -Allow objects to be moved to different schemas
430 o Allow ALTER TABLESPACE to move to different directories
431 o Allow databases to be moved to different tablespaces
432 o Allow moving system tables to other tablespaces, where possible
434 Currently non-global system tables must be in the default database
435 tablespace. Global system tables can never be moved.
437 o %Prevent child tables from altering constraints like CHECK that were
438 inherited from the parent table
443 o Automatically maintain clustering on a table
445 This might require some background daemon to maintain clustering
446 during periods of low usage. It might also require tables to be only
447 paritally filled for easier reorganization. Another idea would
448 be to create a merged heap/index data file so an index lookup would
449 automatically access the heap data too. A third idea would be to
450 store heap rows in hashed groups, perhaps using a user-supplied
453 o %Add default clustering to system tables
455 To do this, determine the ideal cluster index for each system
456 table and set the cluster setting during initdb.
461 o Allow COPY to report error lines and continue
463 This requires the use of a savepoint before each COPY line is
464 processed, with ROLLBACK on COPY failure.
466 o -Allow COPY to understand \x as a hex byte
467 o %Have COPY return the number of rows loaded/unloaded?
468 o -Allow COPY to optionally include column headings in the first line
469 o -Allow COPY FROM ... CSV to interpret newlines and carriage
475 o Allow column-level privileges
476 o %Allow GRANT/REVOKE permissions to be applied to all schema objects
479 The proposed syntax is:
480 GRANT SELECT ON ALL TABLES IN public TO phpuser;
481 GRANT SELECT ON NEW TABLES IN public TO phpuser;
483 * Allow GRANT/REVOKE permissions to be inherited by objects based on
489 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
491 This requires using the row ctid to map cursor rows back to the
492 original heap row. This become more complicated if WITH HOLD cursors
493 are to be supported because WITH HOLD cursors have a copy of the row
494 and no FOR UPDATE lock.
496 o Prevent DROP TABLE from dropping a row referenced by its own open
499 o %Allow pooled connections to list all open WITH HOLD cursors
501 Because WITH HOLD cursors exist outside transactions, this allows
502 them to be listed so they can be closed.
507 o Allow INSERT/UPDATE of the system-generated oid value for a row
508 o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
509 o Allow INSERT/UPDATE ... RETURNING new.col or old.col
511 This is useful for returning the auto-generated key for an INSERT.
512 One complication is how to handle rules that run as part of
518 o -Have SHOW ALL show descriptions for server-side variables
519 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
521 o Add SET PATH for schemas?
523 This is basically the same as SET search_path.
526 * Server-Side Languages
528 o -Allow PL/PgSQL's RAISE function to take expressions
530 Currently only constants are supported.
532 o -Change PL/PgSQL to use palloc() instead of malloc()
533 o Handle references to temporary tables that are created, destroyed,
534 then recreated during a session, and EXECUTE is not used
536 This requires the cached PL/PgSQL byte code to be invalidated when
537 an object referenced in the function is changed.
539 o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
540 o Allow function parameters to be passed by name,
541 get_employee_salary(emp_id => 12345, tax_year => 2001)
542 o Add Oracle-style packages
543 o Add table function support to pltcl, plperl, plpython?
544 o Allow PL/pgSQL to name columns by ordinal position, e.g. rec.(3)
545 o -Allow PL/pgSQL EXECUTE query_var INTO record_var;
546 o Add capability to create and call PROCEDURES
547 o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
548 o Add MOVE to PL/pgSQL
549 o Pass arrays natively instead of as text between plperl and postgres
550 o Add support for polymorphic arguments and return types to plperl
556 * Add a libpq function to support Parse/DescribeStatement capability
557 * Prevent libpq's PQfnumber() from lowercasing the column name?
558 * Allow libpq to access SQLSTATE so pg_ctl can test for connection failure
560 This would be used for checking if the server is up.
562 * Add PQescapeIdentifier() to libpq
563 * Have initdb set DateStyle based on locale?
564 * Have pg_ctl look at PGHOST in case it is a socket directory?
565 * Add a schema option to createlang
566 * Allow pg_ctl to work properly with configuration files located outside
569 pg_ctl can not read the pid file because it isn't located in the
570 config directory but in the PGDATA directory. The solution is to
571 allow pg_ctl to read and understand postgresql.conf to find the
572 data_directory value.
577 o Have psql show current values for a sequence
578 o Move psql backslash database information into the backend, use
579 mnemonic commands? [psql]
581 This would allow non-psql clients to pull the same information out
582 of the database as psql.
584 o Fix psql's display of schema information (Neil)
585 o Allow psql \pset boolean variables to set to fixed values, rather
587 o Consistently display privilege information for all objects in psql
588 o Improve psql's handling of multi-line queries
593 o %Have pg_dump use multi-statement transactions for INSERT dumps
594 o %Allow pg_dump to use multiple -t and -n switches [pg_dump]
595 o %Add dumping of comments on composite type columns
596 o %Add dumping of comments on index columns
597 o %Replace crude DELETE FROM method of pg_dumpall --clean for
598 cleaning of roles with separate DROP commands
599 o -Add dumping and restoring of LOB comments
600 o Stop dumping CASCADE on DROP TYPE commands in clean mode
601 o %Add full object name to the tag field. eg. for operators we need
602 '=(integer, integer)', instead of just '='.
603 o Add pg_dumpall custom format dumps.
605 This is probably best done by combining pg_dump and pg_dumpall
606 into a single binary.
608 o %Add CSV output format
609 o Update pg_dump and psql to use the new COPY libpq API (Christopher)
610 o Remove unnecessary abstractions in pg_dump source code
617 Document differences between ecpg and the SQL standard and
618 information about the Informix-compatibility module.
620 o Solve cardinality > 1 for input descriptors / variables?
621 o Add a semantic check level, e.g. check if a table really exists
622 o fix handling of DB attributes that are arrays
623 o Use backend PREPARE/EXECUTE facility for ecpg where possible
625 o Fix nested C comments
626 o %sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
627 o Make SET CONNECTION thread-aware, non-standard?
628 o Allow multidimensional arrays
629 o Add internationalized message strings
632 Referential Integrity
633 =====================
635 * Add MATCH PARTIAL referential integrity
636 * Add deferred trigger queue file
638 Right now all deferred trigger information is stored in backend
639 memory. This could exhaust memory for very large trigger queues.
640 This item involves dumping large queues into files.
642 * -Implement shared row locks and use them in RI triggers
643 * Change foreign key constraint for array -> element to mean element
645 * Allow DEFERRABLE UNIQUE constraints?
646 * -Allow triggers to be disabled [trigger]
647 * Allow triggers to be disabled in only the current session.
649 This is currently possible by starting a multi-statement transaction,
650 modifying the system tables, performing the desired SQL, restoring the
651 system tables, and committing the transaction. ALTER TABLE ...
652 TRIGGER requires a table lock so it is not ideal for this usage.
654 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
656 If the dump is known to be valid, allow foreign keys to be added
657 without revalidating the data.
659 * Allow statement-level triggers to access modified rows
660 * Support triggers on columns (Greg Sabino Mullane)
661 * %Remove CREATE CONSTRAINT TRIGGER
663 This was used in older releases to dump referential integrity
666 * Enforce referential integrity for system tables
667 * Allow AFTER triggers on system tables
669 System tables are modified in many places in the backend without going
670 through the executor and therefore not causing triggers to fire. To
671 complete this item, the functions that modify system tables will have
678 * Flush cached query plans when the dependent objects change
679 * Track dependencies in function bodies and recompile/invalidate
685 * Add SQL99 WITH clause to SELECT
686 * Add SQL99 WITH RECURSIVE to SELECT
687 * Add pre-parsing phase that converts non-ISO syntax to supported
690 This could allow SQL written for other databases to run without
693 * Allow plug-in modules to emulate features from other databases
694 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
696 * Allow queries across databases or servers with transaction
699 This can be done using dblink and two-phase commit.
701 * -Add two-phase commit
704 * Add the features of packages
706 o Make private objects accessable only to objects in the same schema
707 o Allow current_schema.objname to access current schema objects
708 o Add session variables
709 o Allow nested schemas
715 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
717 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
718 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
721 The main difficulty with this item is the problem of creating an index
722 that can span more than one table.
724 * Allow SELECT ... FOR UPDATE on inherited tables
725 * -Prevent inherited tables from expanding temporary subtables of other
727 * Add UNIQUE capability to non-btree indexes
728 * -Use indexes for MIN() and MAX()
730 MIN/MAX queries can already be rewritten as SELECT col FROM tab ORDER
731 BY col {DESC} LIMIT 1. Completing this item involves doing this
732 transformation automatically.
734 * -Use index to restrict rows returned by multi-key index when used with
735 non-consecutive keys to reduce heap accesses
737 For an index on col1,col2,col3, and a WHERE clause of col1 = 5 and
738 col3 = 9, spin though the index checking for col1 and col3 matches,
739 rather than just col1; also called skip-scanning.
741 * Prevent index uniqueness checks when UPDATE does not modify the column
743 Uniqueness (index) checks are done when updating a column even if the
744 column is not modified by the UPDATE.
746 * Fetch heap pages matching index entries in sequential order
748 Rather than randomly accessing heap pages based on index entries, mark
749 heap pages needing access in a bitmap and do the lookups in sequential
750 order. Another method would be to sort heap ctids matching the index
751 before accessing the heap rows.
753 * -Allow non-bitmap indexes to be combined by creating bitmaps in memory
755 This feature allows separate indexes to be ANDed or ORed together. This
756 is particularly useful for data warehousing applications that need to
757 query the database in an many permutations. This feature scans an index
758 and creates an in-memory bitmap, and allows that bitmap to be combined
759 with other bitmap created in a similar way. The bitmap can either index
760 all TIDs, or be lossy, meaning it records just page numbers and each
761 page tuple has to be checked for validity in a separate pass.
763 * Allow the creation of on-disk bitmap indexes which can be quickly
764 combined with other bitmap indexes
766 Such indexes could be more compact if there are only a few distinct values.
767 Such indexes can also be compressed. Keeping such indexes updated can be
770 * Allow use of indexes to search for NULLs
772 One solution is to create a partial index on an IS NULL expression.
774 * Allow accurate statistics to be collected on indexes with more than
775 one column or expression indexes, perhaps using per-index statistics
776 * Add fillfactor to control reserved free space during index creation
777 * Allow the creation of indexes with mixed ascending/descending specifiers
778 * -Fix incorrect rtree results due to wrong assumptions about "over"
780 * Allow constraint_exclusion to work for UNIONs like it does for
781 inheritance, and allow it to work for UPDATE and DELETE queries
786 o Add more GIST index support for geometric data types
787 o -Add concurrency to GIST
788 o Allow GIST indexes to create certain complex index types, like
789 digital trees (see Aoki)
793 o Pack hash index buckets onto disk pages more efficiently
795 Currently no only one hash bucket can be stored on a page. Ideally
796 several hash buckets could be stored on a single page and greater
797 granularity used for the hash algorithm.
799 o Consider sorting hash buckets so entries can be found using a
800 binary search, rather than a linear scan
802 o In hash indexes, consider storing the hash value with or instead
809 * Improve commit_delay handling to reduce fsync()
810 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
811 * -Allow multiple blocks to be written to WAL with one write()
812 * %Add an option to sync() before fsync()'ing checkpoint files
813 * Add program to test if fsync has a delay compared to non-fsync
819 * Allow free-behind capability for large sequential scans, perhaps using
822 Posix_fadvise() can control both sequential/random file caching and
823 free-behind behavior, but it is unclear how the setting affects other
824 backends that also have the file open, and the feature is not supported
825 on all operating systems.
827 * -Consider use of open/fcntl(O_DIRECT) to minimize OS caching,
830 O_DIRECT doesn't have the same media write guarantees as fsync, so it
831 is in addition to the fsync method, not in place of it.
833 * -Cache last known per-tuple offsets to speed long tuple access
836 We could use a fixed row count and a +/- count to follow MVCC
837 visibility rules, or a single cached value could be used and
838 invalidated if anyone modifies the table. Another idea is to
839 get a count directly from a unique index, but for this to be
840 faster than a sequential scan it must avoid access to the heap
841 to obtain tuple visibility information.
843 * Allow data to be pulled directly from indexes
845 Currently indexes do not have enough tuple visibility information
846 to allow data to be pulled from the index without also accessing
847 the heap. One way to allow this is to set a bit to index tuples
848 to indicate if a tuple is currently visible to all transactions
849 when the first valid heap lookup happens. This bit would have to
850 be cleared when a heap tuple is expired.
853 * Consider automatic caching of queries at various levels:
859 * -Allow the size of the buffer cache used by temporary objects to be
860 specified as a GUC variable
862 Larger local buffer cache sizes requires more efficient handling of
865 * Improve the background writer
867 Allow the background writer to more efficiently write dirty buffers
868 from the end of the LRU cache and use a clock sweep algorithm to
869 write other dirty buffers to reduced checkpoint I/O
871 * Allow sequential scans to take advantage of other concurrent
872 sequentiqal scans, also called "Synchronised Scanning"
874 One possible implementation is to start sequential scans from the lowest
875 numbered buffer in the shared cache, and when reaching the end wrap
876 around to the beginning, rather than always starting sequential scans
877 at the start of the table.
883 * Improve speed with indexes
885 For large table adjustements during vacuum, it is faster to reindex
886 rather than update the index.
888 * Reduce lock time by moving tuples with read lock, then write
889 lock and truncate table
891 Moved tuples are invisible to other backends so they don't require a
892 write lock. However, the read lock promotion to write lock could lead
893 to deadlock situations.
895 * -Add a warning when the free space map is too small
896 * Maintain a map of recently-expired rows
898 This allows vacuum to target specific pages for possible free space
899 without requiring a sequential scan.
901 * Auto-fill the free space map by scanning the buffer cache or by
902 checking pages written by the background writer
903 * Create a bitmap of pages that need vacuuming
905 Instead of sequentially scanning the entire table, have the background
906 writer or some other process record pages that have expired rows, then
907 VACUUM can look at just those pages rather than the entire table. In
908 the event of a system crash, the bitmap would probably be invalidated.
910 * %Add system view to show free space map contents
915 o -Move into the backend code
916 o Use free-space map information to guide refilling
917 o %Suggest VACUUM FULL if a table is nearly empty
918 o Improve xid wraparound detection by recording per-table rather
925 * -Make locking of shared data structures more fine-grained
927 This requires that more locks be acquired but this would reduce lock
928 contention, improving concurrency.
930 * Add code to detect an SMP machine and handle spinlocks accordingly
931 from distributted.net, http://www1.distributed.net/source,
932 in client/common/cpucheck.cpp
934 On SMP machines, it is possible that locks might be released shortly,
935 while on non-SMP machines, the backend should sleep so the process
936 holding the lock can complete and release it.
938 * -Improve SMP performance on i386 machines
940 i386-based SMP machines can generate excessive context switching
941 caused by lock failure in high concurrency situations. This may be
942 caused by CPU cache line invalidation inefficiencies.
944 * Research use of sched_yield() for spinlock acquisition failure
945 * Fix priority ordering of read and write light-weight locks (Neil)
948 Startup Time Improvements
949 =========================
951 * Experiment with multi-threaded backend [thread]
953 This would prevent the overhead associated with process creation. Most
954 operating systems have trivial process creation time compared to
955 database startup overhead, but a few operating systems (WIn32,
956 Solaris) might benefit from threading. Also explore the idea of
957 a single session using multiple threads to execute a query faster.
959 * Add connection pooling
961 It is unclear if this should be done inside the backend code or done
962 by something external like pgpool. The passing of file descriptors to
963 existing backends is one of the difficulties with a backend approach.
969 * Eliminate need to write full pages to WAL before page modification [wal]
971 Currently, to protect against partial disk page writes, we write
972 full page images to WAL before they are modified so we can correct any
973 partial page writes during recovery. These pages can also be
974 eliminated from point-in-time archive files.
976 o -Add ability to turn off full page writes
977 o When off, write CRC to WAL and check file system blocks
980 If CRC check fails during recovery, remember the page in case
981 a later CRC for that page properly matches.
983 o Write full pages during file system write and not when
984 the page is modified in the buffer cache
986 This allows most full page writes to happen in the background
987 writer. It might cause problems for applying WAL on recovery
988 into a partially-written page, but later the full page will be
991 * Reduce WAL traffic so only modified values are written rather than
993 * Add WAL index reliability improvement to non-btree indexes
994 * Allow the pg_xlog directory location to be specified during initdb
995 with a symlink back to the /data location
996 * Allow WAL information to recover corrupted pg_controldata
997 * Find a way to reduce rotational delay when repeatedly writing
1000 Currently fsync of WAL requires the disk platter to perform a full
1001 rotation to fsync again. One idea is to write the WAL to different
1002 offsets that might reduce the rotational delay.
1004 * Allow buffered WAL writes and fsync
1006 Instead of guaranteeing recovery of all committed transactions, this
1007 would provide improved performance by delaying WAL writes and fsync
1008 so an abrupt operating system restart might lose a few seconds of
1009 committed transactions but still be consistent. We could perhaps
1010 remove the 'fsync' parameter (which results in an an inconsistent
1011 database) in favor of this capability.
1013 * -Eliminate WAL logging for CREATE TABLE AS when not doing WAL archiving
1014 * -Change WAL to use 32-bit CRC, for performance reasons
1017 Optimizer / Executor
1018 ====================
1020 * Add missing optimizer selectivities for date, r-tree, etc
1021 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
1022 index using a sequential scan for highest/lowest values
1024 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
1025 all values to return the high/low value. Instead The idea is to do a
1026 sequential scan to find the high/low value, thus avoiding the sort.
1027 MIN/MAX already does this, but not for LIMIT > 1.
1029 * Precompile SQL functions to avoid overhead
1030 * Create utility to compute accurate random_page_cost value
1031 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
1032 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
1033 * -Use CHECK constraints to influence optimizer decisions
1035 CHECK constraints contain information about the distribution of values
1036 within the table. This is also useful for implementing subtables where
1037 a tables content is distributed across several subtables.
1039 * Consider using hash buckets to do DISTINCT, rather than sorting
1041 This would be beneficial when there are few distinct values.
1043 * ANALYZE should record a pg_statistic entry for an all-NULL column
1044 * Log queries where the optimizer row estimates were dramatically
1045 different from the number of rows actually found?
1048 Miscellaneous Performance
1049 =========================
1051 * Do async I/O for faster random read-ahead of data
1053 Async I/O allows multiple I/O requests to be sent to the disk with
1054 results coming back asynchronously.
1056 * Use mmap() rather than SYSV shared memory or to write WAL files?
1058 This would remove the requirement for SYSV SHM but would introduce
1059 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
1060 to prevent I/O overhead.
1062 * Consider mmap()'ing files into a backend?
1064 Doing I/O to large tables would consume a lot of address space or
1065 require frequent mapping/unmapping. Extending the file also causes
1066 mapping problems that might require mapping only individual pages,
1067 leading to thousands of mappings. Another problem is that there is no
1068 way to _prevent_ I/O to disk from the dirty shared buffers so changes
1069 could hit disk before WAL is written.
1071 * Add a script to ask system configuration questions and tune postgresql.conf
1072 * Use a phantom command counter for nested subtransactions to reduce
1074 * Research storing disk pages with no alignment/padding
1079 * Add use of 'const' for variables in source tree
1080 * Rename some /contrib modules from pg* to pg_*
1081 * Move some things from /contrib into main tree
1082 * Move some /contrib modules out to their own project sites
1083 * %Remove warnings created by -Wcast-align
1084 * Move platform-specific ps status display info from ps_status.c to ports
1085 * Add optional CRC checksum to heap and index pages
1086 * Improve documentation to build only interfaces (Marc)
1087 * Remove or relicense modules that are not under the BSD license, if possible
1088 * Remove memory/file descriptor freeing before ereport(ERROR)
1089 * Acquire lock on a relation before building a relcache entry for it
1090 * Promote debug_query_string into a server-side function current_query()
1091 * Allow the identifier length to be increased via a configure option
1092 * Remove Win32 rename/unlink looping if unnecessary
1093 * -Remove kerberos4 from source tree
1094 * Allow cross-compiling by generating the zic database on the target system
1095 * Improve NLS maintenace of libpgport messages linked onto applications
1096 * Allow ecpg to work with MSVC and BCC
1097 * -Make src/port/snprintf.c thread-safe
1098 * Add xpath_array() to /contrib/xml2 to return results as an array
1099 * Allow building in directories containing spaces
1101 This is probably not possible because 'gmake' and other compiler tools
1102 do not fully support quoting of paths with spaces.
1104 * Allow installing to directories containing spaces
1106 This is possible if proper quoting is added to the makefiles for the
1107 install targets. Because PostgreSQL supports relocatable installs, it
1108 is already possible to install into a directory that doesn't contain
1109 spaces and then copy the install to a directory with spaces.
1111 * Fix cross-compiling of time zone database via 'zic'
1112 * Fix sgmltools so PDFs can be generated with bookmarks
1113 * -Add C code on Unix to copy directories for use in creating new databases
1114 * %Clean up compiler warnings (especially with gcc version 4)
1119 o Remove configure.in check for link failure when cause is found
1120 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1122 o Remove psql newline patch when we find out why mingw outputs an
1124 o Allow psql to use readline once non-US code pages work with
1126 o Re-enable timezone output on log_line_prefix '%t' when a
1127 shorter timezone string is available
1128 o Improve dlerror() reporting string
1129 o Fix problem with shared memory on the Win32 Terminal Server
1130 o Add support for Unicode
1132 To fix this, the data needs to be converted to/from UTF16/UTF8
1133 so the Win32 wcscoll() can be used, and perhaps other functions
1134 like towupper(). However, UTF8 already works with normal
1135 locales but provides no ordering or character set classes.
1138 * Wire Protocol Changes
1140 o Allow dynamic character set handling
1141 o Add decoded type, length, precision
1143 o Update clients to use data types, typmod, schema.table.column names
1144 of result sets using new query protocol
1147 ---------------------------------------------------------------------------
1150 Developers who have claimed items are:
1151 --------------------------------------
1152 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1153 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1154 * Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
1155 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1156 Family Health Network
1157 * Claudio is Claudio Natoli <claudio.natoli@memetrics.com>
1158 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1159 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1160 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1161 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1162 * Hiroshi is Hiroshi Inoue <Inoue@tpf.co.jp>
1163 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1164 * Joe is Joe Conway <mail@joeconway.com>
1165 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1166 * Magnus is Magnus Hagander <mha@sollentuna.net>
1167 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1168 * Matthew T. O'Connor <matthew@zeut.net>
1169 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1170 * Neil is Neil Conway <neilc@samurai.com>
1171 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1172 * Peter is Peter Eisentraut <peter_e@gmx.net>
1173 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1174 * Rod is Rod Taylor <pg@rbt.ca>
1175 * Simon is Simon Riggs <simon@2ndquadrant.com>
1176 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1177 * Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
1178 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat