4 Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
5 Last updated: Fri Feb 3 22:23:19 EST 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?
55 * Improve replication solutions
59 You can use any of the master/slave replication servers to use a
60 standby server for data warehousing. To allow read/write queries to
61 multiple servers, you need multi-master replication like pgcluster.
63 o Allow replication over unreliable or non-persistent links
68 o %Add "include file" functionality in postgresql.conf
69 o %Allow commenting of variables in postgresql.conf to restore them
72 Currently, if a variable is commented out, it keeps the
73 previous uncommented value until a server restarted.
75 o %Allow pg_hba.conf settings to be controlled via SQL
77 This would add a function to load the SQL table from
78 pg_hba.conf, and one to writes its contents to the flat file.
79 The table should have a line number that is a float so rows
80 can be inserted between existing rows, e.g. row 2.5 goes
81 between row 2 and row 3.
83 o %Allow postgresql.conf file values to be changed via an SQL
84 API, perhaps using SET GLOBAL
85 o Allow the server to be stopped/restarted via an SQL API
86 o -Issue a warning if a change-on-restart-only postgresql.conf value
87 is modified and the server config files are reloaded
88 o Mark change-on-restart-only values in postgresql.conf
93 * Allow a database in tablespace t1 with tables created in
94 tablespace t2 to be used as a template for a new database created
95 with default tablespace t2
97 All objects in the default database tablespace must have default
98 tablespace specifications. This is because new databases are
99 created by copying directories. If you mix default tablespace
100 tables and tablespace-specified tables in the same directory,
101 creating a new database from such a mixed directory would create a
102 new database with tables that had incorrect explicit tablespaces.
103 To fix this would require modifying pg_class in the newly copied
104 database, which we don't currently do.
106 * Allow reporting of which objects are in which tablespaces
108 This item is difficult because a tablespace can contain objects
109 from multiple databases. There is a server-side function that
110 returns the databases which use a specific tablespace, so this
111 requires a tool that will call that function and connect to each
112 database to find the objects in each database for that tablespace.
114 o %Add a GUC variable to control the tablespace for temporary objects
117 It could start with a random tablespace from a supplied list and
118 cycle through the list.
120 o Allow WAL replay of CREATE TABLESPACE to work when the directory
121 structure on the recovery computer is different from the original
123 o Allow per-tablespace quotas
126 * Point-In-Time Recovery (PITR)
128 o Allow point-in-time recovery to archive partially filled
129 write-ahead logs [pitr]
131 Currently only full WAL files are archived. This means that the
132 most recent transactions aren't available for recovery in case
133 of a disk failure. This could be triggered by a user command or
136 o Automatically force archiving of partially-filled WAL files when
137 pg_stop_backup() is called or the server is stopped
139 Doing this will allow administrators to know more easily when
140 the archive contains all the files needed for point-in-time
143 o %Create dump tool for write-ahead logs for use in determining
144 transaction id for point-in-time recovery
145 o Allow a warm standby system to also allow read-only statements
148 This is useful for checking PITR recovery.
150 o Allow the PITR process to be debugged and data examined
156 * Allow server log information to be output as INSERT statements
158 This would allow server log information to be easily loaded into
159 a database for analysis.
161 * %Add ability to monitor the use of temporary sort files
162 * Allow server logs to be remotely read and removed using SQL commands
163 * Allow protocol-level BIND parameter values to be logged
169 * Improve the MONEY data type
171 Change the MONEY data type to use DECIMAL internally, with special
172 locale-aware output formatting.
174 * Change NUMERIC to enforce the maximum precision
175 * Add NUMERIC division operator that doesn't round?
177 Currently NUMERIC _rounds_ the result to the specified precision.
178 This means division can return a result that multiplied by the
179 divisor is greater than the dividend, e.g. this returns a value > 10:
181 SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
183 The positive modulus result returned by NUMERICs might be considered
184 inaccurate, in one sense.
186 * %Disallow changing default expression of a SERIAL column?
187 * Fix data types where equality comparison isn't intuitive, e.g. box
188 * -Zero umasked bits in conversion from INET cast to CIDR
189 * -Prevent INET cast to CIDR from dropping netmask, SELECT '1.1.1.1'::inet::cidr
190 * Allow INET + INT4 to increment the host part of the address, or
191 throw an error on overflow
192 * %Add 'tid != tid ' operator for use in corruption recovery
193 * Allow user-defined types to specify a type modifier at table creation
199 o Allow infinite dates just like infinite timestamps
200 o Merge hardwired timezone names with the TZ database; allow either
201 kind everywhere a TZ name is currently taken
202 o Allow customization of the known set of TZ names (generalize the
203 present australian_timezones hack)
204 o Allow TIMESTAMP WITH TIME ZONE to store the original timezone
205 information, either zone name or offset from UTC [timezone]
207 If the TIMESTAMP value is stored with a time zone name, interval
208 computations should adjust based on the time zone rules.
210 o Fix SELECT '0.01 years'::interval, '0.01 months'::interval
211 o Fix SELECT INTERVAL '1' MONTH
212 o Add a GUC variable to allow output of interval values in ISO8601
214 o Improve timestamptz subtraction to be DST-aware
216 Currently, subtracting one date from another that crosses a
217 daylight savings time adjustment can return '1 day 1 hour', but
218 adding that back to the first date returns a time one hour in
219 the future. This is caused by the adjustment of '25 hours' to
220 '1 day 1 hour', and '1 day' is the same time the next day, even
221 if daylight savings adjustments are involved.
223 o Fix interval display to support values exceeding 2^31 hours
224 o Add overflow checking to timestamp and interval arithmetic
225 o Add ISO INTERVAL handling
226 o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
228 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
229 o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
230 '1', treat as ISO if there is a range specification clause,
231 and as PG if there no clause is present, e.g. interpret
232 '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
233 interpret '1:30' as '1 hour, 30 minutes'
234 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
235 INTERVAL MONTH), and this should return '12 months'
236 o Round or truncate values to the requested precision, e.g.
237 INTERVAL '11 months' AS YEAR should return one or zero
238 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
243 o -Allow NULLs in arrays
244 o Delay resolution of array expression's data type so assignment
245 coercion can be performed on empty array expressions
250 o Improve vacuum of large objects, like /contrib/vacuumlo?
251 o Add security checking for large objects
252 o Auto-delete large objects when referencing row is deleted
254 /contrib/lo offers this functionality.
256 o Allow read/write into TOAST values like large objects
258 This requires the TOAST column to be stored EXTERNAL.
264 * Allow INET subnet tests using non-constants to be indexed
265 * Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
268 Current CURRENT_TIMESTAMP returns the start time of the current
269 transaction, and gettimeofday() returns the wallclock time. This will
270 make time reporting more consistent and will allow reporting of
271 the statement start time.
273 * %Add pg_get_acldef(), pg_get_typedefault(), pg_get_attrdef(),
274 pg_get_tabledef(), pg_get_domaindef(), pg_get_functiondef()
275 * Allow to_char() to print localized month names
276 * Allow functions to have a schema search path specified at creation time
277 * Allow substring/replace() to get/set bit values
278 * Allow to_char() on interval values to accumulate the highest unit
281 Some special format flag would be required to request such
282 accumulation. Such functionality could also be added to EXTRACT.
283 Prevent accumulation that crosses the month/day boundary because of
284 the uneven number of days in a month.
286 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
287 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
288 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
289 o to_char(INTERVAL '3 years 5 months','MM') => 41
291 * -Add sleep() function, remove from regress.c
292 * Allow user-defined functions retuning a domain value to enforce domain
294 * Add SPI_gettypmod() to return the typemod for a TupleDesc
297 Multi-Language Support
298 ======================
300 * Add NCHAR (as distinguished from ordinary varchar),
301 * Allow locale to be set at database creation
303 Currently locale can only be set during initdb. No global tables have
304 locale-aware columns. However, the database template used during
305 database creation might have locale-aware indexes. The indexes would
306 need to be reindexed to match the new locale.
308 * Allow encoding on a per-column basis
310 Right now only one encoding is allowed per database.
312 * Support multiple simultaneous character sets, per SQL92
313 * Improve UTF8 combined character handling?
314 * Add octet_length_server() and octet_length_client()
315 * Make octet_length_client() the same as octet_length()?
316 * Fix problems with wrong runtime encoding conversion for NLS message files
322 * %Automatically create rules on views so they are updateable, per SQL99
324 We can only auto-create rules for simple views. For more complex
325 cases users will still have to write rules.
327 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
328 * Allow NOTIFY in rules involving conditionals
329 * Allow VIEW/RULE recompilation when the underlying tables change
331 Another issue is whether underlying table changes should be reflected
332 in the view, e.g. should SELECT * show additional columns if they
333 are added after the view is created.
339 * Change LIMIT/OFFSET and FETCH/MOVE to use int8
340 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
341 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
342 * %Allow SET CONSTRAINTS to be qualified by schema/table name
343 * %Allow TRUNCATE ... CASCADE/RESTRICT
345 This is like DELETE CASCADE, but truncates.
347 * %Add a separate TRUNCATE permission
349 Currently only the owner can TRUNCATE a table because triggers are not
350 called, and the table is locked in exclusive mode.
352 * Allow PREPARE of cursors
353 * Allow PREPARE to automatically determine parameter types based on the SQL
355 * Allow finer control over the caching of prepared query plans
357 Currently, queries prepared via the libpq API are planned on first
358 execute using the supplied parameters --- allow SQL PREPARE to do the
359 same. Also, allow control over replanning prepared queries either
360 manually or automatically when statistics for execute parameters
361 differ dramatically from those used during planning.
363 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
365 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
366 such information in memory would improve performance.
368 * Add optional textual message to NOTIFY
370 This would allow an informational message to be added to the notify
371 message, perhaps indicating the row modified or other custom
374 * Add a GUC variable to warn about non-standard SQL usage in queries
375 * Add SQL-standard MERGE command, typically used to merge two tables
378 This is similar to UPDATE, then for unmatched rows, INSERT.
379 Whether concurrent access allows modifications which could cause
380 row loss is implementation independent.
382 * Add REPLACE or UPSERT command that does UPDATE, or on failure, INSERT
385 To implement this cleanly requires that the table have a unique index
386 so duplicate checking can be easily performed. It is possible to
387 do it without a unique index if we require the user to LOCK the table
390 * Add NOVICE output level for helpful messages like automatic sequence/index
392 * %Add COMMENT ON for all cluster global objects (roles, databases
394 * -Make row-wise comparisons work per SQL spec
396 Right now, '(a, b) < (1, 2)' is processed as 'a < 1 and b < 2', but
397 the SQL standard requires it to be processed as a column-by-column
398 comparison, so the proper comparison is '(a < 1) OR (a = 1 AND b < 2)'.
400 * Add RESET CONNECTION command to reset all session state
402 This would include resetting of all variables (RESET ALL), dropping of
403 temporary tables, removing any NOTIFYs, cursors, open transactions,
404 prepared queries, currval()s, etc. This could be used for connection
405 pooling. We could also change RESET ALL to have this functionality.
406 The difficult of this features is allowing RESET ALL to not affect
407 changes made by the interface driver for its internal use. One idea
408 is for this to be a protocol-only feature. Another approach is to
409 notify the protocol when a RESET CONNECTION command is used.
411 * Add GUC to issue notice about statements that use unjoined tables
412 * Allow EXPLAIN to identify tables that were skipped because of
414 * Allow EXPLAIN output to be more easily processed by scripts
415 * Eventually enable escape_string_warning and standard_conforming_strings
416 * Simplify dropping roles that have objects in several databases
417 * Allow COMMENT ON to accept an expression rather than just a string
418 * Allow the count returned by SELECT, etc to be to represent as an int64
419 to allow a higher range of values
420 * Make CLUSTER preserve recently-dead tuples per MVCC requirements
421 * Add SQL99 WITH clause to SELECT
422 * Add SQL99 WITH RECURSIVE to SELECT
427 o Allow CREATE TABLE AS to determine column lengths for complex
428 expressions like SELECT col1 || col2
429 o Use more reliable method for CREATE DATABASE to get a consistent
431 o Add ON COMMIT capability to CREATE TABLE AS ... SELECT
435 o Allow UPDATE to handle complex aggregates [update]?
436 o -Allow an alias to be provided for the target table in
438 o Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple
444 o %Have ALTER TABLE RENAME rename SERIAL sequence names
445 o Add ALTER DOMAIN to modify the underlying data type
446 o %Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
447 o %Allow ALTER TABLE to change constraint deferrability and actions
448 o Add missing object types for ALTER ... SET SCHEMA
449 o Allow ALTER TABLESPACE to move to different directories
450 o Allow databases to be moved to different tablespaces
451 o Allow moving system tables to other tablespaces, where possible
453 Currently non-global system tables must be in the default database
454 tablespace. Global system tables can never be moved.
456 o %Disallow dropping of an inherited constraint
457 o %Prevent child tables from altering or dropping constraints
458 like CHECK that were inherited from the parent table
459 o Have ALTER INDEX update the name of a constraint using that index
460 o Add ALTER TABLE RENAME CONSTRAINT, update index name also
465 o Automatically maintain clustering on a table
467 This might require some background daemon to maintain clustering
468 during periods of low usage. It might also require tables to be only
469 partially filled for easier reorganization. Another idea would
470 be to create a merged heap/index data file so an index lookup would
471 automatically access the heap data too. A third idea would be to
472 store heap rows in hashed groups, perhaps using a user-supplied
475 o %Add default clustering to system tables
477 To do this, determine the ideal cluster index for each system
478 table and set the cluster setting during initdb.
483 o Allow COPY to report error lines and continue
485 This requires the use of a savepoint before each COPY line is
486 processed, with ROLLBACK on COPY failure.
488 o %Have COPY return the number of rows loaded/unloaded?
489 o Allow COPY on a newly-created table to skip WAL logging
491 On crash recovery, the table involved in the COPY would
492 be removed or have its heap and index files truncated. One
493 issue is that no other backend should be able to add to
494 the table at the same time, which is something that is
497 o Allow COPY to output from views
499 Another idea would be to allow actual SELECT statements in a COPY.
504 o Allow column-level privileges
505 o %Allow GRANT/REVOKE permissions to be applied to all schema objects
508 The proposed syntax is:
509 GRANT SELECT ON ALL TABLES IN public TO phpuser;
510 GRANT SELECT ON NEW TABLES IN public TO phpuser;
512 * Allow GRANT/REVOKE permissions to be inherited by objects based on
515 * Allow SERIAL sequences to inherit permissions from the base table?
520 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
522 This requires using the row ctid to map cursor rows back to the
523 original heap row. This become more complicated if WITH HOLD cursors
524 are to be supported because WITH HOLD cursors have a copy of the row
525 and no FOR UPDATE lock.
527 o Prevent DROP TABLE from dropping a row referenced by its own open
530 o -Allow pooled connections to list all open WITH HOLD cursors
532 Because WITH HOLD cursors exist outside transactions, this allows
533 them to be listed so they can be closed.
538 o Allow INSERT/UPDATE of the system-generated oid value for a row
539 o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
540 o Allow INSERT/UPDATE ... RETURNING new.col or old.col
542 This is useful for returning the auto-generated key for an INSERT.
543 One complication is how to handle rules that run as part of
549 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
551 o Add SET PATH for schemas?
553 This is basically the same as SET search_path.
556 * Server-Side Languages
558 o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
559 o Allow function parameters to be passed by name,
560 get_employee_salary(emp_id => 12345, tax_year => 2001)
561 o Add Oracle-style packages
562 o Add table function support to pltcl, plpython
563 o Add capability to create and call PROCEDURES
564 o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
565 o Allow function argument names to be statements from PL/PgSQL
566 o Add MOVE to PL/pgSQL
567 o Add support for polymorphic arguments and return types to
568 languages other than PL/PgSQL
569 o Add support for OUT and INOUT parameters to languages other
571 o Add single-step debugging of PL/PgSQL functions
572 o Allow PL/PgSQL to support WITH HOLD cursors
578 * -Have initdb set the input DateStyle (MDY or DMY) based on locale
579 * Have pg_ctl look at PGHOST in case it is a socket directory?
580 * Allow pg_ctl to work properly with configuration files located outside
583 pg_ctl can not read the pid file because it isn't located in the
584 config directory but in the PGDATA directory. The solution is to
585 allow pg_ctl to read and understand postgresql.conf to find the
586 data_directory value.
591 o Have psql show current values for a sequence
592 o Move psql backslash database information into the backend, use
593 mnemonic commands? [psql]
595 This would allow non-psql clients to pull the same information out
596 of the database as psql.
598 o Fix psql's display of schema information (Neil)
599 o Allow psql \pset boolean variables to set to fixed values, rather
601 o Consistently display privilege information for all objects in psql
602 o Improve psql's handling of multi-line statements
604 Currently, while \e saves a single statement as one entry, interactive
605 statements are saved one line at a time. Ideally all statements
606 would be saved like \e does.
608 o Allow multi-line column values to align in the proper columns
610 If the second output column value is 'a\nb', the 'b' should appear
611 in the second display column, rather than the first column as it
614 o Display IN, INOUT, and OUT parameters in \df+
616 It probably requires psql to output newlines in the proper
617 column, which is already on the TODO list.
619 o Add auto-expanded mode so expanded output is used if the row
620 length is wider than the screen width.
622 Consider using auto-expanded mode for backslash commands like \df+.
624 o Prevent tab completion of SET TRANSACTION from querying the
625 database and therefore preventing the transaction isolation
626 level from being set.
628 Currently, SET <tab> causes a database lookup to check all
629 supported session variables. This query causes problems
630 because setting the transaction isolation level must be the
631 first statement of a transaction.
636 o %Have pg_dump use multi-statement transactions for INSERT dumps
637 o %Allow pg_dump to use multiple -t and -n switches [pg_dump]
638 o %Add dumping of comments on index columns and composite type columns
639 o %Add full object name to the tag field. eg. for operators we need
640 '=(integer, integer)', instead of just '='.
641 o Add pg_dumpall custom format dumps?
642 o %Add CSV output format
643 o Update pg_dump and psql to use the new COPY libpq API (Christopher)
644 o Remove unnecessary function pointer abstractions in pg_dump source
646 o Allow selection of individual object(s) of all types, not just
648 o In a selective dump, allow dumping of an object and all its
650 o Add options like pg_restore -l and -L to pg_dump
651 o Stop dumping CASCADE on DROP TYPE commands in clean mode
652 o Allow pg_dump --clean to drop roles that own objects or have
654 o Add -f to pg_dumpall
661 Document differences between ecpg and the SQL standard and
662 information about the Informix-compatibility module.
664 o Solve cardinality > 1 for input descriptors / variables?
665 o Add a semantic check level, e.g. check if a table really exists
666 o fix handling of DB attributes that are arrays
667 o Use backend PREPARE/EXECUTE facility for ecpg where possible
669 o Fix nested C comments
670 o %sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
671 o Make SET CONNECTION thread-aware, non-standard?
672 o Allow multidimensional arrays
673 o Add internationalized message strings
678 o Add a function to support Parse/DescribeStatement capability
679 o Add PQescapeIdentifier()
680 o Prevent PQfnumber() from lowercasing unquoted the column name
682 PQfnumber() should never have been doing lowercasing, but
683 historically it has so we need a way to prevent it
685 o Allow statement results to be automatically batched to the client
687 Currently, all statement results are transferred to the libpq
688 client before libpq makes the results available to the
689 application. This feature would allow the application to make
690 use of the first result rows while the rest are transferred, or
691 held on the server waiting for them to be requested by libpq.
692 One complexity is that a statement like SELECT 1/col could error
693 out mid-way through the result set.
696 Referential Integrity
697 =====================
699 * Add MATCH PARTIAL referential integrity
700 * Add deferred trigger queue file
702 Right now all deferred trigger information is stored in backend
703 memory. This could exhaust memory for very large trigger queues.
704 This item involves dumping large queues into files.
706 * Change foreign key constraint for array -> element to mean element
708 * Allow DEFERRABLE UNIQUE constraints?
709 * Allow triggers to be disabled in only the current session.
711 This is currently possible by starting a multi-statement transaction,
712 modifying the system tables, performing the desired SQL, restoring the
713 system tables, and committing the transaction. ALTER TABLE ...
714 TRIGGER requires a table lock so it is not ideal for this usage.
716 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
718 If the dump is known to be valid, allow foreign keys to be added
719 without revalidating the data.
721 * Allow statement-level triggers to access modified rows
722 * Support triggers on columns (Greg Sabino Mullane)
723 * Enforce referential integrity for system tables
724 * Allow AFTER triggers on system tables
726 System tables are modified in many places in the backend without going
727 through the executor and therefore not causing triggers to fire. To
728 complete this item, the functions that modify system tables will have
735 * Flush cached query plans when the dependent objects change,
736 when the cardinality of parameters changes dramatically, or
737 when new ANALYZE statistics are available
739 A more complex solution would be to save multiple plans for different
740 cardinality and use the appropriate plan based on the EXECUTE values.
742 * Track dependencies in function bodies and recompile/invalidate
744 This is particularly important for references to temporary tables
745 in PL/PgSQL because PL/PgSQL caches query plans. The only workaround
746 in PL/PgSQL is to use EXECUTE. One complexity is that a function
747 might itself drop and recreate dependent tables, causing it to
748 invalidate its own query plan.
754 * Add pre-parsing phase that converts non-ISO syntax to supported
757 This could allow SQL written for other databases to run without
760 * Allow plug-in modules to emulate features from other databases
761 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
763 * Allow statements across databases or servers with transaction
766 This can be done using dblink and two-phase commit.
768 * Add the features of packages
770 o Make private objects accessible only to objects in the same schema
771 o Allow current_schema.objname to access current schema objects
772 o Add session variables
773 o Allow nested schemas
779 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
781 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
782 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
785 The main difficulty with this item is the problem of creating an index
786 that can span more than one table.
788 * Allow SELECT ... FOR UPDATE on inherited tables
789 * Add UNIQUE capability to non-btree indexes
790 * Prevent index uniqueness checks when UPDATE does not modify the column
792 Uniqueness (index) checks are done when updating a column even if the
793 column is not modified by the UPDATE.
795 * Allow the creation of on-disk bitmap indexes which can be quickly
796 combined with other bitmap indexes
798 Such indexes could be more compact if there are only a few distinct values.
799 Such indexes can also be compressed. Keeping such indexes updated can be
802 * Allow use of indexes to search for NULLs
804 One solution is to create a partial index on an IS NULL expression.
806 * Allow accurate statistics to be collected on indexes with more than
807 one column or expression indexes, perhaps using per-index statistics
808 * Add fillfactor to control reserved free space during index creation
809 * Allow the creation of indexes with mixed ascending/descending specifiers
810 * Allow constraint_exclusion to work for UNIONs like it does for
811 inheritance, allow it to work for UPDATE and DELETE statements, and allow
812 it to be used for all statements with little performance impact
813 * Allow CREATE INDEX to take an additional parameter for use with
815 * Consider compressing indexes by storing key values duplicated in
816 several rows as a single index entry
818 This is difficult because it requires datatype-specific knowledge.
823 o Add more GIST index support for geometric data types
824 o Allow GIST indexes to create certain complex index types, like
825 digital trees (see Aoki)
829 o Pack hash index buckets onto disk pages more efficiently
831 Currently only one hash bucket can be stored on a page. Ideally
832 several hash buckets could be stored on a single page and greater
833 granularity used for the hash algorithm.
835 o Consider sorting hash buckets so entries can be found using a
836 binary search, rather than a linear scan
838 o In hash indexes, consider storing the hash value with or instead
841 o Add WAL logging for crash recovery
842 o Allow multi-column hash indexes
848 * Improve commit_delay handling to reduce fsync()
849 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
851 Ideally this requires a separate test program that can be run
852 at initdb time or optionally later. Consider O_SYNC when
855 * %Add an option to sync() before fsync()'ing checkpoint files
856 * Add program to test if fsync has a delay compared to non-fsync
862 * Allow free-behind capability for large sequential scans, perhaps using
865 Posix_fadvise() can control both sequential/random file caching and
866 free-behind behavior, but it is unclear how the setting affects other
867 backends that also have the file open, and the feature is not supported
868 on all operating systems.
872 We could use a fixed row count and a +/- count to follow MVCC
873 visibility rules, or a single cached value could be used and
874 invalidated if anyone modifies the table. Another idea is to
875 get a count directly from a unique index, but for this to be
876 faster than a sequential scan it must avoid access to the heap
877 to obtain tuple visibility information.
879 * Add estimated_count(*) to return an estimate of COUNT(*)
881 This would use the planner ANALYZE statistics to return an estimated
884 * Allow data to be pulled directly from indexes
886 Currently indexes do not have enough tuple visibility information
887 to allow data to be pulled from the index without also accessing
888 the heap. One way to allow this is to set a bit on index tuples
889 to indicate if a tuple is currently visible to all transactions
890 when the first valid heap lookup happens. This bit would have to
891 be cleared when a heap tuple is expired.
893 Another idea is to maintain a bitmap of heap pages where all rows
894 are visible to all backends, and allow index lookups to reference
895 that bitmap to avoid heap lookups, perhaps the same bitmap we might
896 add someday to determine which heap pages need vacuuming. Frequently
897 accessed bitmaps would have to be stored in shared memory. One 8k
898 page of bitmaps could track 512MB of heap pages.
900 * Consider automatic caching of statements at various levels:
906 * Allow sequential scans to take advantage of other concurrent
907 sequential scans, also called "Synchronised Scanning"
909 One possible implementation is to start sequential scans from the lowest
910 numbered buffer in the shared cache, and when reaching the end wrap
911 around to the beginning, rather than always starting sequential scans
912 at the start of the table.
918 * Improve speed with indexes
920 For large table adjustments during VACUUM FULL, it is faster to
921 reindex rather than update the index.
923 * Reduce lock time during VACUUM FULL by moving tuples with read lock,
924 then write lock and truncate table
926 Moved tuples are invisible to other backends so they don't require a
927 write lock. However, the read lock promotion to write lock could lead
928 to deadlock situations.
930 * Auto-fill the free space map by scanning the buffer cache or by
931 checking pages written by the background writer
932 * Create a bitmap of pages that need vacuuming
934 Instead of sequentially scanning the entire table, have the background
935 writer or some other process record pages that have expired rows, then
936 VACUUM can look at just those pages rather than the entire table. In
937 the event of a system crash, the bitmap would probably be invalidated.
938 One complexity is that index entries still have to be vacuumed, and
939 doing this without an index scan (by using the heap values to find the
940 index entry) might be slow and unreliable, especially for user-defined
943 * %Add system view to show free space map contents
948 o Use free-space map information to guide refilling
949 o %Issue log message to suggest VACUUM FULL if a table is nearly
951 o Improve xid wraparound detection by recording per-table rather
953 o Consider logging activity either to the logs or a system view
959 * Fix priority ordering of read and write light-weight locks (Neil)
962 Startup Time Improvements
963 =========================
965 * Experiment with multi-threaded backend [thread]
967 This would prevent the overhead associated with process creation. Most
968 operating systems have trivial process creation time compared to
969 database startup overhead, but a few operating systems (Win32,
970 Solaris) might benefit from threading. Also explore the idea of
971 a single session using multiple threads to execute a statement faster.
973 * Add connection pooling
975 It is unclear if this should be done inside the backend code or done
976 by something external like pgpool. The passing of file descriptors to
977 existing backends is one of the difficulties with a backend approach.
983 * Eliminate need to write full pages to WAL before page modification [wal]
985 Currently, to protect against partial disk page writes, we write
986 full page images to WAL before they are modified so we can correct any
987 partial page writes during recovery. These pages can also be
988 eliminated from point-in-time archive files.
990 o When off, write CRC to WAL and check file system blocks
993 If CRC check fails during recovery, remember the page in case
994 a later CRC for that page properly matches.
996 o Write full pages during file system write and not when
997 the page is modified in the buffer cache
999 This allows most full page writes to happen in the background
1000 writer. It might cause problems for applying WAL on recovery
1001 into a partially-written page, but later the full page will be
1004 * Allow WAL traffic to be streamed to another server for stand-by
1006 * Reduce WAL traffic so only modified values are written rather than
1008 * Allow the pg_xlog directory location to be specified during initdb
1009 with a symlink back to the /data location
1010 * Allow WAL information to recover corrupted pg_controldata
1011 * Find a way to reduce rotational delay when repeatedly writing
1014 Currently fsync of WAL requires the disk platter to perform a full
1015 rotation to fsync again. One idea is to write the WAL to different
1016 offsets that might reduce the rotational delay.
1018 * Allow buffered WAL writes and fsync
1020 Instead of guaranteeing recovery of all committed transactions, this
1021 would provide improved performance by delaying WAL writes and fsync
1022 so an abrupt operating system restart might lose a few seconds of
1023 committed transactions but still be consistent. We could perhaps
1024 remove the 'fsync' parameter (which results in an an inconsistent
1025 database) in favor of this capability.
1027 * Allow control over which tables are WAL-logged [walcontrol]
1029 Allow tables to bypass WAL writes and just fsync() dirty pages on
1030 commit. To do this, only a single writer can modify the table, and
1031 writes must happen only on new pages. Readers can continue accessing
1032 the table. This would affect COPY, and perhaps INSERT/UPDATE too.
1033 Another option is to avoid transaction logging entirely and truncate
1034 or drop the table on crash recovery. These should be implemented
1035 using ALTER TABLE, e.g. ALTER TABLE PERSISTENCE [ DROP | TRUNCATE |
1036 STABLE | DEFAULT ]. Tables using non-default logging should not use
1037 referential integrity with default-logging tables, and tables using
1038 stable logging probably can not have indexes. One complexity is
1039 the handling of indexes on TOAST tables.
1042 Optimizer / Executor
1043 ====================
1045 * Improve selectivity functions for geometric operators
1046 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
1047 index using a sequential scan for highest/lowest values
1049 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
1050 all values to return the high/low value. Instead The idea is to do a
1051 sequential scan to find the high/low value, thus avoiding the sort.
1052 MIN/MAX already does this, but not for LIMIT > 1.
1054 * Precompile SQL functions to avoid overhead
1055 * Create utility to compute accurate random_page_cost value
1056 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
1057 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
1058 * Consider using hash buckets to do DISTINCT, rather than sorting
1060 This would be beneficial when there are few distinct values. This is
1061 already used by GROUP BY.
1063 * Log statements where the optimizer row estimates were dramatically
1064 different from the number of rows actually found?
1067 Miscellaneous Performance
1068 =========================
1070 * Do async I/O for faster random read-ahead of data
1072 Async I/O allows multiple I/O requests to be sent to the disk with
1073 results coming back asynchronously.
1075 * Use mmap() rather than SYSV shared memory or to write WAL files?
1077 This would remove the requirement for SYSV SHM but would introduce
1078 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
1079 to prevent I/O overhead.
1081 * Consider mmap()'ing files into a backend?
1083 Doing I/O to large tables would consume a lot of address space or
1084 require frequent mapping/unmapping. Extending the file also causes
1085 mapping problems that might require mapping only individual pages,
1086 leading to thousands of mappings. Another problem is that there is no
1087 way to _prevent_ I/O to disk from the dirty shared buffers so changes
1088 could hit disk before WAL is written.
1090 * Add a script to ask system configuration questions and tune postgresql.conf
1091 * Merge xmin/xmax/cmin/cmax back into three header fields
1093 Before subtransactions, there used to be only three fields needed to
1094 store these four values. This was possible because only the current
1095 transaction looks at the cmin/cmax values. If the current transaction
1096 created and expired the row the fields stored where xmin (same as
1097 xmax), cmin, cmax, and if the transaction was expiring a row from a
1098 another transaction, the fields stored were xmin (cmin was not
1099 needed), xmax, and cmax. Such a system worked because a transaction
1100 could only see rows from another completed transaction. However,
1101 subtransactions can see rows from outer transactions, and once the
1102 subtransaction completes, the outer transaction continues, requiring
1103 the storage of all four fields. With subtransactions, an outer
1104 transaction can create a row, a subtransaction expire it, and when the
1105 subtransaction completes, the outer transaction still has to have
1106 proper visibility of the row's cmin, for example, for cursors.
1108 One possible solution is to create a phantom cid which represents a
1109 cmin/cmax pair and is stored in local memory. Another idea is to
1110 store both cmin and cmax only in local memory.
1112 * Research storing disk pages with no alignment/padding
1118 * Add use of 'const' for variables in source tree
1119 * Rename some /contrib modules from pg* to pg_*
1120 * Move some things from /contrib into main tree
1121 * Move some /contrib modules out to their own project sites
1122 * %Remove warnings created by -Wcast-align
1123 * Move platform-specific ps status display info from ps_status.c to ports
1124 * Add optional CRC checksum to heap and index pages
1125 * Improve documentation to build only interfaces (Marc)
1126 * Remove or relicense modules that are not under the BSD license, if possible
1127 * %Remove memory/file descriptor freeing before ereport(ERROR)
1128 * Acquire lock on a relation before building a relcache entry for it
1129 * %Promote debug_query_string into a server-side function current_query()
1130 * %Allow the identifier length to be increased via a configure option
1131 * Allow cross-compiling by generating the zic database on the target system
1132 * Improve NLS maintenace of libpgport messages linked onto applications
1133 * Allow ecpg to work with MSVC and BCC
1134 * Add xpath_array() to /contrib/xml2 to return results as an array
1135 * Allow building in directories containing spaces
1137 This is probably not possible because 'gmake' and other compiler tools
1138 do not fully support quoting of paths with spaces.
1140 * -Allow installing to directories containing spaces
1142 This is possible if proper quoting is added to the makefiles for the
1143 install targets. Because PostgreSQL supports relocatable installs, it
1144 is already possible to install into a directory that doesn't contain
1145 spaces and then copy the install to a directory with spaces.
1147 * Fix sgmltools so PDFs can be generated with bookmarks
1148 * %Clean up compiler warnings (especially with gcc version 4)
1149 * Use UTF8 encoding for NLS messages so all server encodings can
1151 * Update Bonjour to work with newer cross-platform SDK
1152 * -Remove BeOS and QNX-specific code
1157 o Remove configure.in check for link failure when cause is found
1158 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1160 o Remove psql newline patch when we find out why mingw outputs an
1162 o Allow psql to use readline once non-US code pages work with
1164 o Re-enable timezone output on log_line_prefix '%t' when a
1165 shorter timezone string is available
1166 o Fix problem with shared memory on the Win32 Terminal Server
1167 o Improve signal handling,
1168 http://archives.postgresql.org/pgsql-patches/2005-06/msg00027.php
1169 o Add long file support for binary pg_dump output
1171 While Win32 supports 64-bit files, the MinGW API does not,
1172 meaning we have to build an fseeko replacement on top of the
1173 Win32 API, and we have to make sure MinGW handles it. Another
1174 option is to wait for the MinGW project to fix it, or use the
1175 code from the LibGW32C project as a guide.
1178 * Wire Protocol Changes
1180 o Allow dynamic character set handling
1181 o Add decoded type, length, precision
1183 o Update clients to use data types, typmod, schema.table.column names
1184 of result sets using new statement protocol
1187 ---------------------------------------------------------------------------
1190 Developers who have claimed items are:
1191 --------------------------------------
1192 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1193 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1194 * Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
1195 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1196 Family Health Network
1197 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1198 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1199 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1200 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1201 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1202 * Joe is Joe Conway <mail@joeconway.com>
1203 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1204 * Magnus is Magnus Hagander <mha@sollentuna.net>
1205 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1206 * Matthew T. O'Connor <matthew@zeut.net>
1207 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1208 * Neil is Neil Conway <neilc@samurai.com>
1209 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1210 * Peter is Peter Eisentraut <peter_e@gmx.net>
1211 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1212 * Rod is Rod Taylor <pg@rbt.ca>
1213 * Simon is Simon Riggs <simon@2ndquadrant.com>
1214 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1215 * Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
1216 * Teodor is Teodor Sigaev <teodor@sigaev.ru>
1217 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat