4 Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
5 Last updated: Mon Mar 6 17:51:00 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 pg_hba.conf to specify host names along with IP addresses
85 Host name lookup could occur when the postmaster reads the
86 pg_hba.conf file, or when the backend starts. Another
87 solution would be to reverse lookup the connection IP and
88 check that hostname against the host names in pg_hba.conf.
89 We could also then check that the host name maps to the IP
92 o %Allow postgresql.conf file values to be changed via an SQL
93 API, perhaps using SET GLOBAL
94 o Allow the server to be stopped/restarted via an SQL API
95 o -Issue a warning if a change-on-restart-only postgresql.conf value
96 is modified and the server config files are reloaded
97 o Mark change-on-restart-only values in postgresql.conf
102 * Allow a database in tablespace t1 with tables created in
103 tablespace t2 to be used as a template for a new database created
104 with default tablespace t2
106 All objects in the default database tablespace must have default
107 tablespace specifications. This is because new databases are
108 created by copying directories. If you mix default tablespace
109 tables and tablespace-specified tables in the same directory,
110 creating a new database from such a mixed directory would create a
111 new database with tables that had incorrect explicit tablespaces.
112 To fix this would require modifying pg_class in the newly copied
113 database, which we don't currently do.
115 * Allow reporting of which objects are in which tablespaces
117 This item is difficult because a tablespace can contain objects
118 from multiple databases. There is a server-side function that
119 returns the databases which use a specific tablespace, so this
120 requires a tool that will call that function and connect to each
121 database to find the objects in each database for that tablespace.
123 o %Add a GUC variable to control the tablespace for temporary objects
126 It could start with a random tablespace from a supplied list and
127 cycle through the list.
129 o Allow WAL replay of CREATE TABLESPACE to work when the directory
130 structure on the recovery computer is different from the original
132 o Allow per-tablespace quotas
135 * Point-In-Time Recovery (PITR)
137 o Allow point-in-time recovery to archive partially filled
138 write-ahead logs [pitr]
140 Currently only full WAL files are archived. This means that the
141 most recent transactions aren't available for recovery in case
142 of a disk failure. This could be triggered by a user command or
145 o Automatically force archiving of partially-filled WAL files when
146 pg_stop_backup() is called or the server is stopped
148 Doing this will allow administrators to know more easily when
149 the archive contains all the files needed for point-in-time
152 o %Create dump tool for write-ahead logs for use in determining
153 transaction id for point-in-time recovery
154 o Allow a warm standby system to also allow read-only statements
157 This is useful for checking PITR recovery.
159 o Allow the PITR process to be debugged and data examined
165 * Allow server log information to be output as INSERT statements
167 This would allow server log information to be easily loaded into
168 a database for analysis.
170 * %Add ability to monitor the use of temporary sort files
171 * Allow server logs to be remotely read and removed using SQL commands
172 * Allow protocol-level BIND parameter values to be logged
178 * Improve the MONEY data type
180 Change the MONEY data type to use DECIMAL internally, with special
181 locale-aware output formatting.
183 * Change NUMERIC to enforce the maximum precision
184 * Add NUMERIC division operator that doesn't round?
186 Currently NUMERIC _rounds_ the result to the specified precision.
187 This means division can return a result that multiplied by the
188 divisor is greater than the dividend, e.g. this returns a value > 10:
190 SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
192 The positive modulus result returned by NUMERICs might be considered
193 inaccurate, in one sense.
195 * %Disallow changing default expression of a SERIAL column
196 * %Disallow ALTER SEQUENCE changes for SERIAL sequences because pg_dump
197 does not dump the changes
198 * Fix data types where equality comparison isn't intuitive, e.g. box
199 * -Zero umasked bits in conversion from INET cast to CIDR
200 * -Prevent INET cast to CIDR from dropping netmask, SELECT '1.1.1.1'::inet::cidr
201 * -Allow INET + INT8 to increment the host part of the address or
202 throw an error on overflow
203 * -Add 'tid != tid ' operator for use in corruption recovery
204 * Allow user-defined types to specify a type modifier at table creation
210 o Allow infinite dates and intervals just like infinite timestamps
211 o Merge hardwired timezone names with the TZ database; allow either
212 kind everywhere a TZ name is currently taken
213 o Allow customization of the known set of TZ names (generalize the
214 present australian_timezones hack)
215 o Allow TIMESTAMP WITH TIME ZONE to store the original timezone
216 information, either zone name or offset from UTC [timezone]
218 If the TIMESTAMP value is stored with a time zone name, interval
219 computations should adjust based on the time zone rules.
221 o Fix SELECT '0.01 years'::interval, '0.01 months'::interval
222 o Fix SELECT INTERVAL '1' MONTH
223 o Add a GUC variable to allow output of interval values in ISO8601
225 o Improve timestamptz subtraction to be DST-aware
227 Currently, subtracting one date from another that crosses a
228 daylight savings time adjustment can return '1 day 1 hour', but
229 adding that back to the first date returns a time one hour in
230 the future. This is caused by the adjustment of '25 hours' to
231 '1 day 1 hour', and '1 day' is the same time the next day, even
232 if daylight savings adjustments are involved.
234 o Fix interval display to support values exceeding 2^31 hours
235 o Add overflow checking to timestamp and interval arithmetic
236 o Add ISO INTERVAL handling
237 o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
239 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
240 o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
241 '1', treat as ISO if there is a range specification clause,
242 and as PG if there no clause is present, e.g. interpret
243 '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
244 interpret '1:30' as '1 hour, 30 minutes'
245 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
246 INTERVAL MONTH), and this should return '12 months'
247 o Round or truncate values to the requested precision, e.g.
248 INTERVAL '11 months' AS YEAR should return one or zero
249 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
254 o -Allow NULLs in arrays
255 o Delay resolution of array expression's data type so assignment
256 coercion can be performed on empty array expressions
261 o Improve vacuum of large objects, like /contrib/vacuumlo?
262 o Add security checking for large objects
263 o Auto-delete large objects when referencing row is deleted
265 /contrib/lo offers this functionality.
267 o Allow read/write into TOAST values like large objects
269 This requires the TOAST column to be stored EXTERNAL.
275 * Allow INET subnet tests using non-constants to be indexed
276 * Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
279 Current CURRENT_TIMESTAMP returns the start time of the current
280 transaction, and gettimeofday() returns the wallclock time. This will
281 make time reporting more consistent and will allow reporting of
282 the statement start time.
284 * %Add pg_get_acldef(), pg_get_typedefault(), pg_get_attrdef(),
285 pg_get_tabledef(), pg_get_domaindef(), pg_get_functiondef()
286 * -Allow to_char() to print localized month names
287 * Allow to_date() and to_timestamp() accept localized month names
288 * Add missing parameter handling in to_char()
290 http://archives.postgresql.org/pgsql-hackers/2005-12/msg00948.php
292 * Allow functions to have a schema search path specified at creation time
293 * Allow substring/replace() to get/set bit values
294 * Allow to_char() on interval values to accumulate the highest unit
297 Some special format flag would be required to request such
298 accumulation. Such functionality could also be added to EXTRACT.
299 Prevent accumulation that crosses the month/day boundary because of
300 the uneven number of days in a month.
302 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
303 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
304 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
305 o to_char(INTERVAL '3 years 5 months','MM') => 41
307 * -Add sleep() function, remove from regress.c
308 * Allow user-defined functions retuning a domain value to enforce domain
310 * Add SPI_gettypmod() to return the typemod for a TupleDesc
313 Multi-Language Support
314 ======================
316 * Add NCHAR (as distinguished from ordinary varchar),
317 * Allow locale to be set at database creation
319 Currently locale can only be set during initdb. No global tables have
320 locale-aware columns. However, the database template used during
321 database creation might have locale-aware indexes. The indexes would
322 need to be reindexed to match the new locale.
324 * Allow encoding on a per-column basis
326 Right now only one encoding is allowed per database.
328 * Support multiple simultaneous character sets, per SQL92
329 * Improve UTF8 combined character handling?
330 * Add octet_length_server() and octet_length_client()
331 * Make octet_length_client() the same as octet_length()?
332 * Fix problems with wrong runtime encoding conversion for NLS message files
338 * Automatically create rules on views so they are updateable, per SQL99
340 We can only auto-create rules for simple views. For more complex
341 cases users will still have to write rules manually.
343 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
344 * Allow NOTIFY in rules involving conditionals
345 * Allow VIEW/RULE recompilation when the underlying tables change
347 Another issue is whether underlying table changes should be reflected
348 in the view, e.g. should SELECT * show additional columns if they
349 are added after the view is created.
355 * Change LIMIT/OFFSET and FETCH/MOVE to use int8
356 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
357 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
358 * %Allow SET CONSTRAINTS to be qualified by schema/table name
359 * -Allow TRUNCATE ... CASCADE/RESTRICT
361 This is like DELETE CASCADE, but truncates.
363 * %Add a separate TRUNCATE permission
365 Currently only the owner can TRUNCATE a table because triggers are not
366 called, and the table is locked in exclusive mode.
368 * Allow PREPARE of cursors
369 * -Allow PREPARE to automatically determine parameter types based on the SQL
371 * Allow finer control over the caching of prepared query plans
373 Currently, queries prepared via the libpq API are planned on first
374 execute using the supplied parameters --- allow SQL PREPARE to do the
375 same. Also, allow control over replanning prepared queries either
376 manually or automatically when statistics for execute parameters
377 differ dramatically from those used during planning.
379 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
381 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
382 such information in memory would improve performance.
384 * Add optional textual message to NOTIFY
386 This would allow an informational message to be added to the notify
387 message, perhaps indicating the row modified or other custom
390 * Add a GUC variable to warn about non-standard SQL usage in queries
391 * Add SQL-standard MERGE command, typically used to merge two tables
394 This is similar to UPDATE, then for unmatched rows, INSERT.
395 Whether concurrent access allows modifications which could cause
396 row loss is implementation independent.
398 * Add REPLACE or UPSERT command that does UPDATE, or on failure, INSERT
401 To implement this cleanly requires that the table have a unique index
402 so duplicate checking can be easily performed. It is possible to
403 do it without a unique index if we require the user to LOCK the table
406 * Add NOVICE output level for helpful messages like automatic sequence/index
408 * -Add COMMENT ON for all cluster global objects (roles, databases
410 * -Make row-wise comparisons work per SQL spec
412 Right now, '(a, b) < (1, 2)' is processed as 'a < 1 and b < 2', but
413 the SQL standard requires it to be processed as a column-by-column
414 comparison, so the proper comparison is '(a < 1) OR (a = 1 AND b < 2)'.
416 * Add RESET CONNECTION command to reset all session state
418 This would include resetting of all variables (RESET ALL), dropping of
419 temporary tables, removing any NOTIFYs, cursors, open transactions,
420 prepared queries, currval()s, etc. This could be used for connection
421 pooling. We could also change RESET ALL to have this functionality.
422 The difficult of this features is allowing RESET ALL to not affect
423 changes made by the interface driver for its internal use. One idea
424 is for this to be a protocol-only feature. Another approach is to
425 notify the protocol when a RESET CONNECTION command is used.
427 * Add GUC to issue notice about statements that use unjoined tables
428 * Allow EXPLAIN to identify tables that were skipped because of
430 * Allow EXPLAIN output to be more easily processed by scripts
431 * -Eventually enable escape_string_warning and standard_conforming_strings
432 * Simplify dropping roles that have objects in several databases
433 * Allow COMMENT ON to accept an expression rather than just a string
434 * Allow the count returned by SELECT, etc to be to represent as an int64
435 to allow a higher range of values
436 * Make CLUSTER preserve recently-dead tuples per MVCC requirements
437 * Add SQL99 WITH clause to SELECT
438 * Add SQL99 WITH RECURSIVE to SELECT
443 o Allow CREATE TABLE AS to determine column lengths for complex
444 expressions like SELECT col1 || col2
445 o Use more reliable method for CREATE DATABASE to get a consistent
447 o -Add ON COMMIT capability to CREATE TABLE AS ... SELECT
451 o Allow UPDATE to handle complex aggregates [update]?
452 o -Allow an alias to be provided for the target table in
454 o Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple
460 o %Have ALTER TABLE RENAME rename SERIAL sequence names
461 o Add ALTER DOMAIN to modify the underlying data type
462 o %Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
463 o %Allow ALTER TABLE to change constraint deferrability and actions
464 o Add missing object types for ALTER ... SET SCHEMA
465 o Allow ALTER TABLESPACE to move to different directories
466 o Allow databases to be moved to different tablespaces
467 o Allow moving system tables to other tablespaces, where possible
469 Currently non-global system tables must be in the default database
470 tablespace. Global system tables can never be moved.
472 o %Prevent child tables from altering or dropping constraints
473 like CHECK that were inherited from the parent table
474 o Prevent parent tables from altering or dropping constraints
475 like CHECK that are inherited by child tables
477 Dropping constraints should only be possible with CASCADE.
479 o Have ALTER INDEX update the name of a constraint using that index
480 o Add ALTER TABLE RENAME CONSTRAINT, update index name also
485 o Automatically maintain clustering on a table
487 This might require some background daemon to maintain clustering
488 during periods of low usage. It might also require tables to be only
489 partially filled for easier reorganization. Another idea would
490 be to create a merged heap/index data file so an index lookup would
491 automatically access the heap data too. A third idea would be to
492 store heap rows in hashed groups, perhaps using a user-supplied
495 o %Add default clustering to system tables
497 To do this, determine the ideal cluster index for each system
498 table and set the cluster setting during initdb.
503 o Allow COPY to report error lines and continue
505 This requires the use of a savepoint before each COPY line is
506 processed, with ROLLBACK on COPY failure.
508 o -Have COPY return the number of rows loaded/unloaded?
509 o Allow COPY on a newly-created table to skip WAL logging
511 On crash recovery, the table involved in the COPY would
512 be removed or have its heap and index files truncated. One
513 issue is that no other backend should be able to add to
514 the table at the same time, which is something that is
517 o Allow COPY to output from views
519 Another idea would be to allow actual SELECT statements in a COPY.
524 o Allow column-level privileges
525 o %Allow GRANT/REVOKE permissions to be applied to all schema objects
528 The proposed syntax is:
529 GRANT SELECT ON ALL TABLES IN public TO phpuser;
530 GRANT SELECT ON NEW TABLES IN public TO phpuser;
532 * Allow GRANT/REVOKE permissions to be inherited by objects based on
535 * Allow SERIAL sequences to inherit permissions from the base table?
540 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
542 This requires using the row ctid to map cursor rows back to the
543 original heap row. This become more complicated if WITH HOLD cursors
544 are to be supported because WITH HOLD cursors have a copy of the row
545 and no FOR UPDATE lock.
547 o Prevent DROP TABLE from dropping a row referenced by its own open
550 o -Allow pooled connections to list all open WITH HOLD cursors
552 Because WITH HOLD cursors exist outside transactions, this allows
553 them to be listed so they can be closed.
558 o Allow INSERT/UPDATE of the system-generated oid value for a row
559 o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
560 o Allow INSERT/UPDATE ... RETURNING new.col or old.col
562 This is useful for returning the auto-generated key for an INSERT.
563 One complication is how to handle rules that run as part of
569 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
571 o Add SET PATH for schemas?
573 This is basically the same as SET search_path.
576 * Server-Side Languages
578 o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
579 o Allow function parameters to be passed by name,
580 get_employee_salary(emp_id => 12345, tax_year => 2001)
581 o Add Oracle-style packages
582 o Add table function support to pltcl, plpython
583 o Add capability to create and call PROCEDURES
584 o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
585 o Allow function argument names to be statements from PL/PgSQL
586 o Add MOVE to PL/pgSQL
587 o Add support for polymorphic arguments and return types to
588 languages other than PL/PgSQL
589 o Add support for OUT and INOUT parameters to languages other
591 o Add single-step debugging of PL/PgSQL functions
592 o Allow PL/PgSQL to support WITH HOLD cursors
598 * -Have initdb set the input DateStyle (MDY or DMY) based on locale
599 * Have pg_ctl look at PGHOST in case it is a socket directory?
600 * Allow pg_ctl to work properly with configuration files located outside
603 pg_ctl can not read the pid file because it isn't located in the
604 config directory but in the PGDATA directory. The solution is to
605 allow pg_ctl to read and understand postgresql.conf to find the
606 data_directory value.
611 o Have psql show current values for a sequence
612 o Move psql backslash database information into the backend, use
613 mnemonic commands? [psql]
615 This would allow non-psql clients to pull the same information out
616 of the database as psql.
618 o Fix psql's display of schema information (Neil)
619 o Allow psql \pset boolean variables to set to fixed values, rather
621 o Consistently display privilege information for all objects in psql
622 o -Improve psql's handling of multi-line statements
624 Currently, while \e saves a single statement as one entry, interactive
625 statements are saved one line at a time. Ideally all statements
626 would be saved like \e does.
628 o -Allow multi-line column values to align in the proper columns
630 If the second output column value is 'a\nb', the 'b' should appear
631 in the second display column, rather than the first column as it
634 o Display IN, INOUT, and OUT parameters in \df+
636 It probably requires psql to output newlines in the proper
637 column, which is already on the TODO list.
639 o Add auto-expanded mode so expanded output is used if the row
640 length is wider than the screen width.
642 Consider using auto-expanded mode for backslash commands like \df+.
644 o Prevent tab completion of SET TRANSACTION from querying the
645 database and therefore preventing the transaction isolation
646 level from being set.
648 Currently, SET <tab> causes a database lookup to check all
649 supported session variables. This query causes problems
650 because setting the transaction isolation level must be the
651 first statement of a transaction.
656 o %Have pg_dump use multi-statement transactions for INSERT dumps
657 o %Allow pg_dump to use multiple -t and -n switches [pg_dump]
658 o %Add dumping of comments on index columns and composite type columns
659 o %Add full object name to the tag field. eg. for operators we need
660 '=(integer, integer)', instead of just '='.
661 o Add pg_dumpall custom format dumps?
662 o %Add CSV output format
663 o Update pg_dump and psql to use the new COPY libpq API (Christopher)
664 o Remove unnecessary function pointer abstractions in pg_dump source
666 o Allow selection of individual object(s) of all types, not just
668 o In a selective dump, allow dumping of an object and all its
670 o Add options like pg_restore -l and -L to pg_dump
671 o Stop dumping CASCADE on DROP TYPE commands in clean mode
672 o Allow pg_dump --clean to drop roles that own objects or have
674 o Add -f to pg_dumpall
681 Document differences between ecpg and the SQL standard and
682 information about the Informix-compatibility module.
684 o Solve cardinality > 1 for input descriptors / variables?
685 o Add a semantic check level, e.g. check if a table really exists
686 o fix handling of DB attributes that are arrays
687 o Use backend PREPARE/EXECUTE facility for ecpg where possible
689 o Fix nested C comments
690 o %sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
691 o Make SET CONNECTION thread-aware, non-standard?
692 o Allow multidimensional arrays
693 o Add internationalized message strings
694 o Add COPY TO STDIN / STDOUT handling
699 o Add a function to support Parse/DescribeStatement capability
700 o Add PQescapeIdentifier()
701 o Prevent PQfnumber() from lowercasing unquoted the column name
703 PQfnumber() should never have been doing lowercasing, but
704 historically it has so we need a way to prevent it
706 o Allow statement results to be automatically batched to the client
708 Currently, all statement results are transferred to the libpq
709 client before libpq makes the results available to the
710 application. This feature would allow the application to make
711 use of the first result rows while the rest are transferred, or
712 held on the server waiting for them to be requested by libpq.
713 One complexity is that a statement like SELECT 1/col could error
714 out mid-way through the result set.
717 Referential Integrity
718 =====================
720 * Add MATCH PARTIAL referential integrity
721 * Add deferred trigger queue file
723 Right now all deferred trigger information is stored in backend
724 memory. This could exhaust memory for very large trigger queues.
725 This item involves dumping large queues into files.
727 * Change foreign key constraint for array -> element to mean element
729 * Allow DEFERRABLE UNIQUE constraints?
730 * Allow triggers to be disabled in only the current session.
732 This is currently possible by starting a multi-statement transaction,
733 modifying the system tables, performing the desired SQL, restoring the
734 system tables, and committing the transaction. ALTER TABLE ...
735 TRIGGER requires a table lock so it is not ideal for this usage.
737 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
739 If the dump is known to be valid, allow foreign keys to be added
740 without revalidating the data.
742 * Allow statement-level triggers to access modified rows
743 * Support triggers on columns (Greg Sabino Mullane)
744 * Enforce referential integrity for system tables
745 * Allow AFTER triggers on system tables
747 System tables are modified in many places in the backend without going
748 through the executor and therefore not causing triggers to fire. To
749 complete this item, the functions that modify system tables will have
756 * Flush cached query plans when the dependent objects change,
757 when the cardinality of parameters changes dramatically, or
758 when new ANALYZE statistics are available
760 A more complex solution would be to save multiple plans for different
761 cardinality and use the appropriate plan based on the EXECUTE values.
763 * Track dependencies in function bodies and recompile/invalidate
765 This is particularly important for references to temporary tables
766 in PL/PgSQL because PL/PgSQL caches query plans. The only workaround
767 in PL/PgSQL is to use EXECUTE. One complexity is that a function
768 might itself drop and recreate dependent tables, causing it to
769 invalidate its own query plan.
775 * Add pre-parsing phase that converts non-ISO syntax to supported
778 This could allow SQL written for other databases to run without
781 * Allow plug-in modules to emulate features from other databases
782 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
784 * Allow statements across databases or servers with transaction
787 This can be done using dblink and two-phase commit.
789 * Add the features of packages
791 o Make private objects accessible only to objects in the same schema
792 o Allow current_schema.objname to access current schema objects
793 o Add session variables
794 o Allow nested schemas
800 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
802 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
803 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
806 The main difficulty with this item is the problem of creating an index
807 that can span more than one table.
809 * Allow SELECT ... FOR UPDATE on inherited tables
810 * Add UNIQUE capability to non-btree indexes
811 * Prevent index uniqueness checks when UPDATE does not modify the column
813 Uniqueness (index) checks are done when updating a column even if the
814 column is not modified by the UPDATE.
816 * Allow the creation of on-disk bitmap indexes which can be quickly
817 combined with other bitmap indexes
819 Such indexes could be more compact if there are only a few distinct values.
820 Such indexes can also be compressed. Keeping such indexes updated can be
823 * Allow use of indexes to search for NULLs
825 One solution is to create a partial index on an IS NULL expression.
827 * Allow accurate statistics to be collected on indexes with more than
828 one column or expression indexes, perhaps using per-index statistics
829 * Add fillfactor to control reserved free space during index creation
830 * Allow the creation of indexes with mixed ascending/descending specifiers
831 * Allow constraint_exclusion to work for UNIONs like it does for
832 inheritance, allow it to work for UPDATE and DELETE statements, and allow
833 it to be used for all statements with little performance impact
834 * Allow CREATE INDEX to take an additional parameter for use with
836 * Consider compressing indexes by storing key values duplicated in
837 several rows as a single index entry
839 This is difficult because it requires datatype-specific knowledge.
844 o Add more GIST index support for geometric data types
845 o Allow GIST indexes to create certain complex index types, like
846 digital trees (see Aoki)
850 o Pack hash index buckets onto disk pages more efficiently
852 Currently only one hash bucket can be stored on a page. Ideally
853 several hash buckets could be stored on a single page and greater
854 granularity used for the hash algorithm.
856 o Consider sorting hash buckets so entries can be found using a
857 binary search, rather than a linear scan
859 o In hash indexes, consider storing the hash value with or instead
862 o Add WAL logging for crash recovery
863 o Allow multi-column hash indexes
869 * Improve commit_delay handling to reduce fsync()
870 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
872 Ideally this requires a separate test program that can be run
873 at initdb time or optionally later. Consider O_SYNC when
876 * %Add an option to sync() before fsync()'ing checkpoint files
877 * Add program to test if fsync has a delay compared to non-fsync
883 * Allow free-behind capability for large sequential scans, perhaps using
886 Posix_fadvise() can control both sequential/random file caching and
887 free-behind behavior, but it is unclear how the setting affects other
888 backends that also have the file open, and the feature is not supported
889 on all operating systems.
893 We could use a fixed row count and a +/- count to follow MVCC
894 visibility rules, or a single cached value could be used and
895 invalidated if anyone modifies the table. Another idea is to
896 get a count directly from a unique index, but for this to be
897 faster than a sequential scan it must avoid access to the heap
898 to obtain tuple visibility information.
900 * Add estimated_count(*) to return an estimate of COUNT(*)
902 This would use the planner ANALYZE statistics to return an estimated
905 * Allow data to be pulled directly from indexes
907 Currently indexes do not have enough tuple visibility information
908 to allow data to be pulled from the index without also accessing
909 the heap. One way to allow this is to set a bit on index tuples
910 to indicate if a tuple is currently visible to all transactions
911 when the first valid heap lookup happens. This bit would have to
912 be cleared when a heap tuple is expired.
914 Another idea is to maintain a bitmap of heap pages where all rows
915 are visible to all backends, and allow index lookups to reference
916 that bitmap to avoid heap lookups, perhaps the same bitmap we might
917 add someday to determine which heap pages need vacuuming. Frequently
918 accessed bitmaps would have to be stored in shared memory. One 8k
919 page of bitmaps could track 512MB of heap pages.
921 * Consider automatic caching of statements at various levels:
927 * Allow sequential scans to take advantage of other concurrent
928 sequential scans, also called "Synchronised Scanning"
930 One possible implementation is to start sequential scans from the lowest
931 numbered buffer in the shared cache, and when reaching the end wrap
932 around to the beginning, rather than always starting sequential scans
933 at the start of the table.
939 * Improve speed with indexes
941 For large table adjustments during VACUUM FULL, it is faster to
942 reindex rather than update the index.
944 * Reduce lock time during VACUUM FULL by moving tuples with read lock,
945 then write lock and truncate table
947 Moved tuples are invisible to other backends so they don't require a
948 write lock. However, the read lock promotion to write lock could lead
949 to deadlock situations.
951 * Auto-fill the free space map by scanning the buffer cache or by
952 checking pages written by the background writer
953 * Create a bitmap of pages that need vacuuming
955 Instead of sequentially scanning the entire table, have the background
956 writer or some other process record pages that have expired rows, then
957 VACUUM can look at just those pages rather than the entire table. In
958 the event of a system crash, the bitmap would probably be invalidated.
959 One complexity is that index entries still have to be vacuumed, and
960 doing this without an index scan (by using the heap values to find the
961 index entry) might be slow and unreliable, especially for user-defined
964 * -Add system view to show free space map contents
965 * Allow FSM to return free space toward the beginning of the heap file,
966 in hopes that empty pages at the end can be truncated by VACUUM
967 * Allow FSM page return free space based on table clustering, to assist
968 in maintaining clustering?
973 o Use free-space map information to guide refilling
974 o %Issue log message to suggest VACUUM FULL if a table is nearly
976 o Improve xid wraparound detection by recording per-table rather
978 o Consider logging activity either to the logs or a system view
984 * Fix priority ordering of read and write light-weight locks (Neil)
987 Startup Time Improvements
988 =========================
990 * Experiment with multi-threaded backend [thread]
992 This would prevent the overhead associated with process creation. Most
993 operating systems have trivial process creation time compared to
994 database startup overhead, but a few operating systems (Win32,
995 Solaris) might benefit from threading. Also explore the idea of
996 a single session using multiple threads to execute a statement faster.
998 * Add connection pooling
1000 It is unclear if this should be done inside the backend code or done
1001 by something external like pgpool. The passing of file descriptors to
1002 existing backends is one of the difficulties with a backend approach.
1008 * Eliminate need to write full pages to WAL before page modification [wal]
1010 Currently, to protect against partial disk page writes, we write
1011 full page images to WAL before they are modified so we can correct any
1012 partial page writes during recovery. These pages can also be
1013 eliminated from point-in-time archive files.
1015 o When off, write CRC to WAL and check file system blocks
1018 If CRC check fails during recovery, remember the page in case
1019 a later CRC for that page properly matches.
1021 o Write full pages during file system write and not when
1022 the page is modified in the buffer cache
1024 This allows most full page writes to happen in the background
1025 writer. It might cause problems for applying WAL on recovery
1026 into a partially-written page, but later the full page will be
1029 * Allow WAL traffic to be streamed to another server for stand-by
1031 * Reduce WAL traffic so only modified values are written rather than
1033 * Allow the pg_xlog directory location to be specified during initdb
1034 with a symlink back to the /data location
1035 * Allow WAL information to recover corrupted pg_controldata
1036 * Find a way to reduce rotational delay when repeatedly writing
1039 Currently fsync of WAL requires the disk platter to perform a full
1040 rotation to fsync again. One idea is to write the WAL to different
1041 offsets that might reduce the rotational delay.
1043 * Allow buffered WAL writes and fsync
1045 Instead of guaranteeing recovery of all committed transactions, this
1046 would provide improved performance by delaying WAL writes and fsync
1047 so an abrupt operating system restart might lose a few seconds of
1048 committed transactions but still be consistent. We could perhaps
1049 remove the 'fsync' parameter (which results in an an inconsistent
1050 database) in favor of this capability.
1052 * Allow WAL logging to be turned off for a table, but the table
1053 might be dropped or truncated during crash recovery [walcontrol]
1055 Allow tables to bypass WAL writes and just fsync() dirty pages on
1056 commit. This should be implemented using ALTER TABLE, e.g. ALTER
1057 TABLE PERSISTENCE [ DROP | TRUNCATE | DEFAULT ]. Tables using
1058 non-default logging should not use referential integrity with
1059 default-logging tables. A table without dirty buffers during a
1060 crash could perhaps avoid the drop/truncate.
1062 * Allow WAL logging to be turned off for a table, but the table would
1063 avoid being truncated/dropped [walcontrol]
1065 To do this, only a single writer can modify the table, and writes
1066 must happen only on new pages so the new pages can be removed during
1067 crash recovery. Readers can continue accessing the table. Such
1068 tables probably cannot have indexes. One complexity is the handling
1069 of indexes on TOAST tables.
1072 Optimizer / Executor
1073 ====================
1075 * Improve selectivity functions for geometric operators
1076 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
1077 index using a sequential scan for highest/lowest values
1079 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
1080 all values to return the high/low value. Instead The idea is to do a
1081 sequential scan to find the high/low value, thus avoiding the sort.
1082 MIN/MAX already does this, but not for LIMIT > 1.
1084 * Precompile SQL functions to avoid overhead
1085 * Create utility to compute accurate random_page_cost value
1086 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
1087 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
1088 * Consider using hash buckets to do DISTINCT, rather than sorting
1090 This would be beneficial when there are few distinct values. This is
1091 already used by GROUP BY.
1093 * Log statements where the optimizer row estimates were dramatically
1094 different from the number of rows actually found?
1097 Miscellaneous Performance
1098 =========================
1100 * Do async I/O for faster random read-ahead of data
1102 Async I/O allows multiple I/O requests to be sent to the disk with
1103 results coming back asynchronously.
1105 * Use mmap() rather than SYSV shared memory or to write WAL files?
1107 This would remove the requirement for SYSV SHM but would introduce
1108 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
1109 to prevent I/O overhead.
1111 * Consider mmap()'ing files into a backend?
1113 Doing I/O to large tables would consume a lot of address space or
1114 require frequent mapping/unmapping. Extending the file also causes
1115 mapping problems that might require mapping only individual pages,
1116 leading to thousands of mappings. Another problem is that there is no
1117 way to _prevent_ I/O to disk from the dirty shared buffers so changes
1118 could hit disk before WAL is written.
1120 * Add a script to ask system configuration questions and tune postgresql.conf
1121 * Merge xmin/xmax/cmin/cmax back into three header fields
1123 Before subtransactions, there used to be only three fields needed to
1124 store these four values. This was possible because only the current
1125 transaction looks at the cmin/cmax values. If the current transaction
1126 created and expired the row the fields stored where xmin (same as
1127 xmax), cmin, cmax, and if the transaction was expiring a row from a
1128 another transaction, the fields stored were xmin (cmin was not
1129 needed), xmax, and cmax. Such a system worked because a transaction
1130 could only see rows from another completed transaction. However,
1131 subtransactions can see rows from outer transactions, and once the
1132 subtransaction completes, the outer transaction continues, requiring
1133 the storage of all four fields. With subtransactions, an outer
1134 transaction can create a row, a subtransaction expire it, and when the
1135 subtransaction completes, the outer transaction still has to have
1136 proper visibility of the row's cmin, for example, for cursors.
1138 One possible solution is to create a phantom cid which represents a
1139 cmin/cmax pair and is stored in local memory. Another idea is to
1140 store both cmin and cmax only in local memory.
1142 * Research storing disk pages with no alignment/padding
1148 * Add use of 'const' for variables in source tree
1149 * Rename some /contrib modules from pg* to pg_*
1150 * Move some things from /contrib into main tree
1151 * Move some /contrib modules out to their own project sites
1152 * %Remove warnings created by -Wcast-align
1153 * Move platform-specific ps status display info from ps_status.c to ports
1154 * Add optional CRC checksum to heap and index pages
1155 * Improve documentation to build only interfaces (Marc)
1156 * Remove or relicense modules that are not under the BSD license, if possible
1157 * %Remove memory/file descriptor freeing before ereport(ERROR)
1158 * Acquire lock on a relation before building a relcache entry for it
1159 * %Promote debug_query_string into a server-side function current_query()
1160 * %Allow the identifier length to be increased via a configure option
1161 * Allow cross-compiling by generating the zic database on the target system
1162 * Improve NLS maintenance of libpgport messages linked onto applications
1163 * Allow ecpg to work with MSVC and BCC
1164 * Add xpath_array() to /contrib/xml2 to return results as an array
1165 * Allow building in directories containing spaces
1167 This is probably not possible because 'gmake' and other compiler tools
1168 do not fully support quoting of paths with spaces.
1170 * -Allow installing to directories containing spaces
1172 This is possible if proper quoting is added to the makefiles for the
1173 install targets. Because PostgreSQL supports relocatable installs, it
1174 is already possible to install into a directory that doesn't contain
1175 spaces and then copy the install to a directory with spaces.
1177 * Fix sgmltools so PDFs can be generated with bookmarks
1178 * -%Clean up compiler warnings (especially with gcc version 4)
1179 * Use UTF8 encoding for NLS messages so all server encodings can
1181 * Update Bonjour to work with newer cross-platform SDK
1182 * -Remove BeOS and QNX-specific code
1183 * Split out libpq pgpass and environment documentation sections to make
1184 it easier for non-developers to find
1185 * Improve port/qsort() to handle sorts with 50% unique and 50% duplicate
1188 This involves choosing better pivot points for the quicksort.
1193 o Remove configure.in check for link failure when cause is found
1194 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1196 o Remove psql newline patch when we find out why mingw outputs an
1198 o Allow psql to use readline once non-US code pages work with
1200 o Re-enable timezone output on log_line_prefix '%t' when a
1201 shorter timezone string is available
1202 o Fix problem with shared memory on the Win32 Terminal Server
1203 o Improve signal handling,
1204 http://archives.postgresql.org/pgsql-patches/2005-06/msg00027.php
1205 o Add long file support for binary pg_dump output
1207 While Win32 supports 64-bit files, the MinGW API does not,
1208 meaning we have to build an fseeko replacement on top of the
1209 Win32 API, and we have to make sure MinGW handles it. Another
1210 option is to wait for the MinGW project to fix it, or use the
1211 code from the LibGW32C project as a guide.
1214 o Check WSACancelBlockingCall() for interrupts (win32intr)
1217 * Wire Protocol Changes
1219 o Allow dynamic character set handling
1220 o Add decoded type, length, precision
1222 o Update clients to use data types, typmod, schema.table.column names
1223 of result sets using new statement protocol
1226 ---------------------------------------------------------------------------
1229 Developers who have claimed items are:
1230 --------------------------------------
1231 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1232 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1233 * Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
1234 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1235 Family Health Network
1236 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1237 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1238 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1239 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1240 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1241 * Joe is Joe Conway <mail@joeconway.com>
1242 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1243 * Magnus is Magnus Hagander <mha@sollentuna.net>
1244 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1245 * Matthew T. O'Connor <matthew@zeut.net>
1246 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1247 * Neil is Neil Conway <neilc@samurai.com>
1248 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1249 * Peter is Peter Eisentraut <peter_e@gmx.net>
1250 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1251 * Rod is Rod Taylor <pg@rbt.ca>
1252 * Simon is Simon Riggs <simon@2ndquadrant.com>
1253 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1254 * Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
1255 * Teodor is Teodor Sigaev <teodor@sigaev.ru>
1256 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat