4 Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
5 Last updated: Thu Jan 5 11:35:12 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
52 * Improve replication solutions
56 You can use any of the master/slave replication servers to use a
57 standby server for data warehousing. To allow read/write queries to
58 multiple servers, you need multi-master replication like pgcluster.
60 o Allow replication over unreliable or non-persistent links
65 o %Add "include file" functionality in postgresql.conf
66 o %Allow commenting of variables in postgresql.conf to restore them
69 Currently, if a variable is commented out, it keeps the
70 previous uncommented value until a server restarted.
72 o %Allow pg_hba.conf settings to be controlled via SQL
74 This would add a function to load the SQL table from
75 pg_hba.conf, and one to writes its contents to the flat file.
76 The table should have a line number that is a float so rows
77 can be inserted between existing rows, e.g. row 2.5 goes
78 between row 2 and row 3.
80 o %Allow postgresql.conf file values to be changed via an SQL
81 API, perhaps using SET GLOBAL
82 o Allow the server to be stopped/restarted via an SQL API
83 o Issue a warning if a change-on-restart-only postgresql.conf value
84 is modified and the server config files are reloaded
85 o Mark change-on-restart-only values in postgresql.conf
90 * Allow a database in tablespace t1 with tables created in
91 tablespace t2 to be used as a template for a new database created
92 with default tablespace t2
94 All objects in the default database tablespace must have default
95 tablespace specifications. This is because new databases are
96 created by copying directories. If you mix default tablespace
97 tables and tablespace-specified tables in the same directory,
98 creating a new database from such a mixed directory would create a
99 new database with tables that had incorrect explicit tablespaces.
100 To fix this would require modifying pg_class in the newly copied
101 database, which we don't currently do.
103 * Allow reporting of which objects are in which tablespaces
105 This item is difficult because a tablespace can contain objects
106 from multiple databases. There is a server-side function that
107 returns the databases which use a specific tablespace, so this
108 requires a tool that will call that function and connect to each
109 database to find the objects in each database for that tablespace.
111 o %Add a GUC variable to control the tablespace for temporary objects
114 It could start with a random tablespace from a supplied list and
115 cycle through the list.
117 o Allow WAL replay of CREATE TABLESPACE to work when the directory
118 structure on the recovery computer is different from the original
120 o Allow per-tablespace quotas
123 * Point-In-Time Recovery (PITR)
125 o Allow point-in-time recovery to archive partially filled
126 write-ahead logs [pitr]
128 Currently only full WAL files are archived. This means that the
129 most recent transactions aren't available for recovery in case
130 of a disk failure. This could be triggered by a user command or
133 o Automatically force archiving of partially-filled WAL files when
134 pg_stop_backup() is called or the server is stopped
136 Doing this will allow administrators to know more easily when
137 the archive contains all the files needed for point-in-time
140 o %Create dump tool for write-ahead logs for use in determining
141 transaction id for point-in-time recovery
142 o Allow a warm standby system to also allow read-only statements
145 This is useful for checking PITR recovery.
147 o Allow the PITR process to be debugged and data examined
153 * Allow server log information to be output as INSERT statements
155 This would allow server log information to be easily loaded into
156 a database for analysis.
158 * %Add ability to monitor the use of temporary sort files
159 * Allow server logs to be remotely read and removed using SQL commands
160 * Allow protocol-level BIND parameter values to be logged
166 * Improve the MONEY data type
168 Change the MONEY data type to use DECIMAL internally, with special
169 locale-aware output formatting.
171 * Change NUMERIC to enforce the maximum precision
172 * Add NUMERIC division operator that doesn't round?
174 Currently NUMERIC _rounds_ the result to the specified precision.
175 This means division can return a result that multiplied by the
176 divisor is greater than the dividend, e.g. this returns a value > 10:
178 SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
180 The positive modulus result returned by NUMERICs might be considered
181 inaccurate, in one sense.
183 * %Disallow changing default expression of a SERIAL column?
184 * Fix data types where equality comparison isn't intuitive, e.g. box
185 * %Prevent INET cast to CIDR if the unmasked bits are not zero, or
187 * %Prevent INET cast to CIDR from dropping netmask, SELECT '1.1.1.1'::inet::cidr
188 * Allow INET + INT4 to increment the host part of the address, or
189 throw an error on overflow
190 * %Add 'tid != tid ' operator for use in corruption recovery
191 * Allow user-defined types to specify a type modifier at table creation
197 o Allow infinite dates just like infinite timestamps
198 o Merge hardwired timezone names with the TZ database; allow either
199 kind everywhere a TZ name is currently taken
200 o Allow customization of the known set of TZ names (generalize the
201 present australian_timezones hack)
202 o Allow TIMESTAMP WITH TIME ZONE to store the original timezone
203 information, either zone name or offset from UTC [timezone]
205 If the TIMESTAMP value is stored with a time zone name, interval
206 computations should adjust based on the time zone rules.
208 o Fix SELECT '0.01 years'::interval, '0.01 months'::interval
209 o Fix SELECT INTERVAL '1' MONTH
210 o Add a GUC variable to allow output of interval values in ISO8601
212 o Improve timestamptz subtraction to be DST-aware
214 Currently, subtracting one date from another that crosses a
215 daylight savings time adjustment can return '1 day 1 hour', but
216 adding that back to the first date returns a time one hour in
217 the future. This is caused by the adjustment of '25 hours' to
218 '1 day 1 hour', and '1 day' is the same time the next day, even
219 if daylight savings adjustments are involved.
221 o Fix interval display to support values exceeding 2^31 hours
222 o Add overflow checking to timestamp and interval arithmetic
223 o Add ISO INTERVAL handling
224 o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
226 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
227 o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
228 '1', treat as ISO if there is a range specification clause,
229 and as PG if there no clause is present, e.g. interpret
230 '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
231 interpret '1:30' as '1 hour, 30 minutes'
232 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
233 INTERVAL MONTH), and this should return '12 months'
234 o Round or truncate values to the requested precision, e.g.
235 INTERVAL '11 months' AS YEAR should return one or zero
236 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
241 o -Allow NULLs in arrays
242 o Delay resolution of array expression's data type so assignment
243 coercion can be performed on empty array expressions
248 o Improve vacuum of large objects, like /contrib/vacuumlo?
249 o Add security checking for large objects
250 o Auto-delete large objects when referencing row is deleted
252 /contrib/lo offers this functionality.
254 o Allow read/write into TOAST values like large objects
256 This requires the TOAST column to be stored EXTERNAL.
262 * Allow INET subnet tests using non-constants to be indexed
263 * Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
266 Current CURRENT_TIMESTAMP returns the start time of the current
267 transaction, and gettimeofday() returns the wallclock time. This will
268 make time reporting more consistent and will allow reporting of
269 the statement start time.
271 * %Add pg_get_acldef(), pg_get_typedefault(), pg_get_attrdef(),
272 pg_get_tabledef(), pg_get_domaindef(), pg_get_functiondef()
273 * Allow to_char() to print localized month names
274 * Allow functions to have a schema search path specified at creation time
275 * Allow substring/replace() to get/set bit values
276 * Allow to_char() on interval values to accumulate the highest unit
279 Some special format flag would be required to request such
280 accumulation. Such functionality could also be added to EXTRACT.
281 Prevent accumulation that crosses the month/day boundary because of
282 the uneven number of days in a month.
284 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
285 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
286 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
287 o to_char(INTERVAL '3 years 5 months','MM') => 41
289 * Add sleep() function, remove from regress.c
290 * Allow user-defined functions retuning a domain value to enforce domain
292 * Add SPI_gettypmod() to return the typemod for a TupleDesc
295 Multi-Language Support
296 ======================
298 * Add NCHAR (as distinguished from ordinary varchar),
299 * Allow locale to be set at database creation
301 Currently locale can only be set during initdb. No global tables have
302 locale-aware columns. However, the database template used during
303 database creation might have locale-aware indexes. The indexes would
304 need to be reindexed to match the new locale.
306 * Allow encoding on a per-column basis
308 Right now only one encoding is allowed per database.
310 * Support multiple simultaneous character sets, per SQL92
311 * Improve UTF8 combined character handling?
312 * Add octet_length_server() and octet_length_client()
313 * Make octet_length_client() the same as octet_length()?
314 * Fix problems with wrong runtime encoding conversion for NLS message files
320 * %Automatically create rules on views so they are updateable, per SQL99
322 We can only auto-create rules for simple views. For more complex
323 cases users will still have to write rules.
325 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
326 * Allow NOTIFY in rules involving conditionals
327 * Allow VIEW/RULE recompilation when the underlying tables change
329 Another issue is whether underlying table changes should be reflected
330 in the view, e.g. should SELECT * show additional columns if they
331 are added after the view is created.
337 * Change LIMIT/OFFSET and FETCH/MOVE to use int8
338 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
339 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
340 * %Allow SET CONSTRAINTS to be qualified by schema/table name
341 * %Allow TRUNCATE ... CASCADE/RESTRICT
343 This is like DELETE CASCADE, but truncates.
345 * %Add a separate TRUNCATE permission
347 Currently only the owner can TRUNCATE a table because triggers are not
348 called, and the table is locked in exclusive mode.
350 * Allow PREPARE of cursors
351 * Allow PREPARE to automatically determine parameter types based on the SQL
353 * Allow finer control over the caching of prepared query plans
355 Currently, queries prepared via the libpq API are planned on first
356 execute using the supplied parameters --- allow SQL PREPARE to do the
357 same. Also, allow control over replanning prepared queries either
358 manually or automatically when statistics for execute parameters
359 differ dramatically from those used during planning.
361 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
363 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
364 such information in memory would improve performance.
366 * Add optional textual message to NOTIFY
368 This would allow an informational message to be added to the notify
369 message, perhaps indicating the row modified or other custom
372 * Add a GUC variable to warn about non-standard SQL usage in queries
373 * Add SQL-standard MERGE command, typically used to merge two tables
376 This is similar to UPDATE, then for unmatched rows, INSERT.
377 Whether concurrent access allows modifications which could cause
378 row loss is implementation independent.
380 * Add REPLACE or UPSERT command that does UPDATE, or on failure, INSERT
383 To implement this cleanly requires that the table have a unique index
384 so duplicate checking can be easily performed. It is possible to
385 do it without a unique index if we require the user to LOCK the table
388 * Add NOVICE output level for helpful messages like automatic sequence/index
390 * %Add COMMENT ON for all cluster global objects (roles, databases
392 * -Make row-wise comparisons work per SQL spec
394 Right now, '(a, b) < (1, 2)' is processed as 'a < 1 and b < 2', but
395 the SQL standard requires it to be processed as a column-by-column
396 comparison, so the proper comparison is '(a < 1) OR (a = 1 AND b < 2)'.
398 * Add RESET CONNECTION command to reset all session state
400 This would include resetting of all variables (RESET ALL), dropping of
401 temporary tables, removing any NOTIFYs, cursors, open transactions,
402 prepared queries, currval()s, etc. This could be used for connection
403 pooling. We could also change RESET ALL to have this functionality.
404 The difficult of this features is allowing RESET ALL to not affect
405 changes made by the interface driver for its internal use. One idea
406 is for this to be a protocol-only feature. Another approach is to
407 notify the protocol when a RESET CONNECTION command is used.
409 * Add GUC to issue notice about statements that use unjoined tables
410 * Allow EXPLAIN to identify tables that were skipped because of
412 * Allow EXPLAIN output to be more easily processed by scripts
413 * Eventually enable escape_string_warning and standard_conforming_strings
414 * Simplify dropping roles that have objects in several databases
415 * Allow COMMENT ON to accept an expression rather than just a string
416 * Allow the count returned by SELECT, etc to be to represent as an int64
417 to allow a higher range of values
418 * Make CLUSTER preserve recently-dead tuples per MVCC requirements
423 o Allow CREATE TABLE AS to determine column lengths for complex
424 expressions like SELECT col1 || col2
425 o Use more reliable method for CREATE DATABASE to get a consistent
427 o Add ON COMMIT capability to CREATE TABLE AS ... SELECT
431 o Allow UPDATE to handle complex aggregates [update]?
432 o Allow an alias to be provided for the target table in
435 This is not SQL-spec but many DBMSs allow it.
437 o Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple
443 o %Have ALTER TABLE RENAME rename SERIAL sequence names
444 o Add ALTER DOMAIN to modify the underlying data type
445 o %Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
446 o %Allow ALTER TABLE to change constraint deferrability and actions
447 o Add missing object types for ALTER ... SET SCHEMA
448 o Allow ALTER TABLESPACE to move to different directories
449 o Allow databases to be moved to different tablespaces
450 o Allow moving system tables to other tablespaces, where possible
452 Currently non-global system tables must be in the default database
453 tablespace. Global system tables can never be moved.
455 o %Disallow dropping of an inherited constraint
456 o %Prevent child tables from altering or dropping constraints
457 like CHECK that were inherited from the parent table
458 o Have ALTER INDEX update the name of a constraint using that index
459 o Add ALTER TABLE RENAME CONSTRAINT, update index name also
464 o Automatically maintain clustering on a table
466 This might require some background daemon to maintain clustering
467 during periods of low usage. It might also require tables to be only
468 partially filled for easier reorganization. Another idea would
469 be to create a merged heap/index data file so an index lookup would
470 automatically access the heap data too. A third idea would be to
471 store heap rows in hashed groups, perhaps using a user-supplied
474 o %Add default clustering to system tables
476 To do this, determine the ideal cluster index for each system
477 table and set the cluster setting during initdb.
482 o Allow COPY to report error lines and continue
484 This requires the use of a savepoint before each COPY line is
485 processed, with ROLLBACK on COPY failure.
487 o %Have COPY return the number of rows loaded/unloaded?
488 o Allow COPY on a newly-created table to skip WAL logging
490 On crash recovery, the table involved in the COPY would
491 be removed or have its heap and index files truncated. One
492 issue is that no other backend should be able to add to
493 the table at the same time, which is something that is
496 o Allow COPY to output from views
498 Another idea would be to allow actual SELECT statements in a COPY.
503 o Allow column-level privileges
504 o %Allow GRANT/REVOKE permissions to be applied to all schema objects
507 The proposed syntax is:
508 GRANT SELECT ON ALL TABLES IN public TO phpuser;
509 GRANT SELECT ON NEW TABLES IN public TO phpuser;
511 * Allow GRANT/REVOKE permissions to be inherited by objects based on
517 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
519 This requires using the row ctid to map cursor rows back to the
520 original heap row. This become more complicated if WITH HOLD cursors
521 are to be supported because WITH HOLD cursors have a copy of the row
522 and no FOR UPDATE lock.
524 o Prevent DROP TABLE from dropping a row referenced by its own open
527 o %Allow pooled connections to list all open WITH HOLD cursors
529 Because WITH HOLD cursors exist outside transactions, this allows
530 them to be listed so they can be closed.
535 o Allow INSERT/UPDATE of the system-generated oid value for a row
536 o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
537 o Allow INSERT/UPDATE ... RETURNING new.col or old.col
539 This is useful for returning the auto-generated key for an INSERT.
540 One complication is how to handle rules that run as part of
546 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
548 o Add SET PATH for schemas?
550 This is basically the same as SET search_path.
553 * Server-Side Languages
555 o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
556 o Allow function parameters to be passed by name,
557 get_employee_salary(emp_id => 12345, tax_year => 2001)
558 o Add Oracle-style packages
559 o Add table function support to pltcl, plpython
560 o Add capability to create and call PROCEDURES
561 o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
562 o Allow function argument names to be statements from PL/PgSQL
563 o Add MOVE to PL/pgSQL
564 o Add support for polymorphic arguments and return types to
565 languages other than PL/PgSQL
566 o Add support for OUT and INOUT parameters to languages other
568 o Add single-step debugging of PL/PgSQL functions
569 o Allow PL/PgSQL to support WITH HOLD cursors
575 * -Have initdb set the input DateStyle (MDY or DMY) based on locale
576 * Have pg_ctl look at PGHOST in case it is a socket directory?
577 * Allow pg_ctl to work properly with configuration files located outside
580 pg_ctl can not read the pid file because it isn't located in the
581 config directory but in the PGDATA directory. The solution is to
582 allow pg_ctl to read and understand postgresql.conf to find the
583 data_directory value.
588 o Have psql show current values for a sequence
589 o Move psql backslash database information into the backend, use
590 mnemonic commands? [psql]
592 This would allow non-psql clients to pull the same information out
593 of the database as psql.
595 o Fix psql's display of schema information (Neil)
596 o Allow psql \pset boolean variables to set to fixed values, rather
598 o Consistently display privilege information for all objects in psql
599 o Improve psql's handling of multi-line statements
601 Currently, while \e saves a single statement as one entry, interactive
602 statements are saved one line at a time. Ideally all statements
603 would be saved like \e does.
605 o Allow multi-line column values to align in the proper columns
607 If the second output column value is 'a\nb', the 'b' should appear
608 in the second display column, rather than the first column as it
611 o Display IN, INOUT, and OUT parameters in \df+
613 It probably requires psql to output newlines in the proper
614 column, which is already on the TODO list.
616 o Add auto-expanded mode so expanded output is used if the row
617 length is wider than the screen width.
619 Consider using auto-expanded mode for backslash commands like \df+.
624 o %Have pg_dump use multi-statement transactions for INSERT dumps
625 o %Allow pg_dump to use multiple -t and -n switches [pg_dump]
626 o %Add dumping of comments on index columns and composite type columns
627 o %Add full object name to the tag field. eg. for operators we need
628 '=(integer, integer)', instead of just '='.
629 o Add pg_dumpall custom format dumps?
630 o %Add CSV output format
631 o Update pg_dump and psql to use the new COPY libpq API (Christopher)
632 o Remove unnecessary function pointer abstractions in pg_dump source
634 o Allow selection of individual object(s) of all types, not just
636 o In a selective dump, allow dumping of an object and all its
638 o Add options like pg_restore -l and -L to pg_dump
639 o Stop dumping CASCADE on DROP TYPE commands in clean mode
640 o Allow pg_dump --clean to drop roles that own objects or have
642 o Add -f to pg_dumpall
649 Document differences between ecpg and the SQL standard and
650 information about the Informix-compatibility module.
652 o Solve cardinality > 1 for input descriptors / variables?
653 o Add a semantic check level, e.g. check if a table really exists
654 o fix handling of DB attributes that are arrays
655 o Use backend PREPARE/EXECUTE facility for ecpg where possible
657 o Fix nested C comments
658 o %sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
659 o Make SET CONNECTION thread-aware, non-standard?
660 o Allow multidimensional arrays
661 o Add internationalized message strings
666 o Add a function to support Parse/DescribeStatement capability
667 o Add PQescapeIdentifier()
668 o Prevent PQfnumber() from lowercasing unquoted the column name
670 PQfnumber() should never have been doing lowercasing, but
671 historically it has so we need a way to prevent it
673 o Allow statement results to be automatically batched to the client
675 Currently, all statement results are transferred to the libpq
676 client before libpq makes the results available to the
677 application. This feature would allow the application to make
678 use of the first result rows while the rest are transferred, or
679 held on the server waiting for them to be requested by libpq.
680 One complexity is that a statement like SELECT 1/col could error
681 out mid-way through the result set.
684 Referential Integrity
685 =====================
687 * Add MATCH PARTIAL referential integrity
688 * Add deferred trigger queue file
690 Right now all deferred trigger information is stored in backend
691 memory. This could exhaust memory for very large trigger queues.
692 This item involves dumping large queues into files.
694 * Change foreign key constraint for array -> element to mean element
696 * Allow DEFERRABLE UNIQUE constraints?
697 * Allow triggers to be disabled in only the current session.
699 This is currently possible by starting a multi-statement transaction,
700 modifying the system tables, performing the desired SQL, restoring the
701 system tables, and committing the transaction. ALTER TABLE ...
702 TRIGGER requires a table lock so it is not ideal for this usage.
704 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
706 If the dump is known to be valid, allow foreign keys to be added
707 without revalidating the data.
709 * Allow statement-level triggers to access modified rows
710 * Support triggers on columns (Greg Sabino Mullane)
711 * Enforce referential integrity for system tables
712 * Allow AFTER triggers on system tables
714 System tables are modified in many places in the backend without going
715 through the executor and therefore not causing triggers to fire. To
716 complete this item, the functions that modify system tables will have
723 * Flush cached query plans when the dependent objects change,
724 when the cardinality of parameters changes dramatically, or
725 when new ANALYZE statistics are available
727 A more complex solution would be to save multiple plans for different
728 cardinality and use the appropriate plan based on the EXECUTE values.
730 * Track dependencies in function bodies and recompile/invalidate
732 This is particularly important for references to temporary tables
733 in PL/PgSQL because PL/PgSQL caches query plans. The only workaround
734 in PL/PgSQL is to use EXECUTE. One complexity is that a function
735 might itself drop and recreate dependent tables, causing it to
736 invalidate its own query plan.
742 * Add SQL99 WITH clause to SELECT
743 * Add SQL99 WITH RECURSIVE to SELECT
744 * Add pre-parsing phase that converts non-ISO syntax to supported
747 This could allow SQL written for other databases to run without
750 * Allow plug-in modules to emulate features from other databases
751 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
753 * Allow statements across databases or servers with transaction
756 This can be done using dblink and two-phase commit.
758 * Add the features of packages
760 o Make private objects accessible only to objects in the same schema
761 o Allow current_schema.objname to access current schema objects
762 o Add session variables
763 o Allow nested schemas
769 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
771 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
772 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
775 The main difficulty with this item is the problem of creating an index
776 that can span more than one table.
778 * Allow SELECT ... FOR UPDATE on inherited tables
779 * Add UNIQUE capability to non-btree indexes
780 * Prevent index uniqueness checks when UPDATE does not modify the column
782 Uniqueness (index) checks are done when updating a column even if the
783 column is not modified by the UPDATE.
785 * Allow the creation of on-disk bitmap indexes which can be quickly
786 combined with other bitmap indexes
788 Such indexes could be more compact if there are only a few distinct values.
789 Such indexes can also be compressed. Keeping such indexes updated can be
792 * Allow use of indexes to search for NULLs
794 One solution is to create a partial index on an IS NULL expression.
796 * Allow accurate statistics to be collected on indexes with more than
797 one column or expression indexes, perhaps using per-index statistics
798 * Add fillfactor to control reserved free space during index creation
799 * Allow the creation of indexes with mixed ascending/descending specifiers
800 * Allow constraint_exclusion to work for UNIONs like it does for
801 inheritance, allow it to work for UPDATE and DELETE statements, and allow
802 it to be used for all statements with little performance impact
803 * Allow CREATE INDEX to take an additional parameter for use with
805 * Consider compressing indexes by storing key values duplicated in
806 several rows as a single index entry
808 This is difficult because it requires datatype-specific knowledge.
813 o Add more GIST index support for geometric data types
814 o Allow GIST indexes to create certain complex index types, like
815 digital trees (see Aoki)
819 o Pack hash index buckets onto disk pages more efficiently
821 Currently only one hash bucket can be stored on a page. Ideally
822 several hash buckets could be stored on a single page and greater
823 granularity used for the hash algorithm.
825 o Consider sorting hash buckets so entries can be found using a
826 binary search, rather than a linear scan
828 o In hash indexes, consider storing the hash value with or instead
831 o Add WAL logging for crash recovery
832 o Allow multi-column hash indexes
838 * Improve commit_delay handling to reduce fsync()
839 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
841 Ideally this requires a separate test program that can be run
842 at initdb time or optionally later. Consider O_SYNC when
845 * %Add an option to sync() before fsync()'ing checkpoint files
846 * Add program to test if fsync has a delay compared to non-fsync
852 * Allow free-behind capability for large sequential scans, perhaps using
855 Posix_fadvise() can control both sequential/random file caching and
856 free-behind behavior, but it is unclear how the setting affects other
857 backends that also have the file open, and the feature is not supported
858 on all operating systems.
862 We could use a fixed row count and a +/- count to follow MVCC
863 visibility rules, or a single cached value could be used and
864 invalidated if anyone modifies the table. Another idea is to
865 get a count directly from a unique index, but for this to be
866 faster than a sequential scan it must avoid access to the heap
867 to obtain tuple visibility information.
869 * Add estimated_count(*) to return an estimate of COUNT(*)
871 This would use the planner ANALYZE statistics to return an estimated
874 * Allow data to be pulled directly from indexes
876 Currently indexes do not have enough tuple visibility information
877 to allow data to be pulled from the index without also accessing
878 the heap. One way to allow this is to set a bit on index tuples
879 to indicate if a tuple is currently visible to all transactions
880 when the first valid heap lookup happens. This bit would have to
881 be cleared when a heap tuple is expired.
883 Another idea is to maintain a bitmap of heap pages where all rows
884 are visible to all backends, and allow index lookups to reference
885 that bitmap to avoid heap lookups, perhaps the same bitmap we might
886 add someday to determine which heap pages need vacuuming. Frequently
887 accessed bitmaps would have to be stored in shared memory. One 8k
888 page of bitmaps could track 512MB of heap pages.
890 * Consider automatic caching of statements at various levels:
896 * Allow sequential scans to take advantage of other concurrent
897 sequential scans, also called "Synchronised Scanning"
899 One possible implementation is to start sequential scans from the lowest
900 numbered buffer in the shared cache, and when reaching the end wrap
901 around to the beginning, rather than always starting sequential scans
902 at the start of the table.
908 * Improve speed with indexes
910 For large table adjustments during VACUUM FULL, it is faster to
911 reindex rather than update the index.
913 * Reduce lock time during VACUUM FULL by moving tuples with read lock,
914 then write lock and truncate table
916 Moved tuples are invisible to other backends so they don't require a
917 write lock. However, the read lock promotion to write lock could lead
918 to deadlock situations.
920 * Auto-fill the free space map by scanning the buffer cache or by
921 checking pages written by the background writer
922 * Create a bitmap of pages that need vacuuming
924 Instead of sequentially scanning the entire table, have the background
925 writer or some other process record pages that have expired rows, then
926 VACUUM can look at just those pages rather than the entire table. In
927 the event of a system crash, the bitmap would probably be invalidated.
928 One complexity is that index entries still have to be vacuumed, and
929 doing this without an index scan (by using the heap values to find the
930 index entry) might be slow and unreliable, especially for user-defined
933 * %Add system view to show free space map contents
938 o Use free-space map information to guide refilling
939 o %Issue log message to suggest VACUUM FULL if a table is nearly
941 o Improve xid wraparound detection by recording per-table rather
943 o Consider logging activity either to the logs or a system view
949 * Fix priority ordering of read and write light-weight locks (Neil)
952 Startup Time Improvements
953 =========================
955 * Experiment with multi-threaded backend [thread]
957 This would prevent the overhead associated with process creation. Most
958 operating systems have trivial process creation time compared to
959 database startup overhead, but a few operating systems (Win32,
960 Solaris) might benefit from threading. Also explore the idea of
961 a single session using multiple threads to execute a statement faster.
963 * Add connection pooling
965 It is unclear if this should be done inside the backend code or done
966 by something external like pgpool. The passing of file descriptors to
967 existing backends is one of the difficulties with a backend approach.
973 * Eliminate need to write full pages to WAL before page modification [wal]
975 Currently, to protect against partial disk page writes, we write
976 full page images to WAL before they are modified so we can correct any
977 partial page writes during recovery. These pages can also be
978 eliminated from point-in-time archive files.
980 o When off, write CRC to WAL and check file system blocks
983 If CRC check fails during recovery, remember the page in case
984 a later CRC for that page properly matches.
986 o Write full pages during file system write and not when
987 the page is modified in the buffer cache
989 This allows most full page writes to happen in the background
990 writer. It might cause problems for applying WAL on recovery
991 into a partially-written page, but later the full page will be
994 * Allow WAL traffic to be streamed to another server for stand-by
996 * Reduce WAL traffic so only modified values are written rather than
998 * Allow the pg_xlog directory location to be specified during initdb
999 with a symlink back to the /data location
1000 * Allow WAL information to recover corrupted pg_controldata
1001 * Find a way to reduce rotational delay when repeatedly writing
1004 Currently fsync of WAL requires the disk platter to perform a full
1005 rotation to fsync again. One idea is to write the WAL to different
1006 offsets that might reduce the rotational delay.
1008 * Allow buffered WAL writes and fsync
1010 Instead of guaranteeing recovery of all committed transactions, this
1011 would provide improved performance by delaying WAL writes and fsync
1012 so an abrupt operating system restart might lose a few seconds of
1013 committed transactions but still be consistent. We could perhaps
1014 remove the 'fsync' parameter (which results in an an inconsistent
1015 database) in favor of this capability.
1017 * Allow control over which tables are WAL-logged
1019 Allow tables to bypass WAL writes and just fsync() dirty pages on
1020 commit. To do this, only a single writer can modify the table, and
1021 writes must happen only on new pages. Readers can continue accessing
1022 the table. This would affect COPY, and perhaps INSERT/UPDATE too.
1023 Another option is to avoid transaction logging entirely and truncate
1024 or drop the table on crash recovery. These should be implemented
1025 using ALTER TABLE, e.g. ALTER TABLE PERSISTENCE [ DROP | TRUNCATE |
1026 STABLE | DEFAULT ]. Tables using non-default logging should not use
1027 referential integrity with default-logging tables, and tables using
1028 stable logging probably can not have indexes. [wallog]
1031 Optimizer / Executor
1032 ====================
1034 * Improve selectivity functions for geometric operators
1035 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
1036 index using a sequential scan for highest/lowest values
1038 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
1039 all values to return the high/low value. Instead The idea is to do a
1040 sequential scan to find the high/low value, thus avoiding the sort.
1041 MIN/MAX already does this, but not for LIMIT > 1.
1043 * Precompile SQL functions to avoid overhead
1044 * Create utility to compute accurate random_page_cost value
1045 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
1046 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
1047 * Consider using hash buckets to do DISTINCT, rather than sorting
1049 This would be beneficial when there are few distinct values. This is
1050 already used by GROUP BY.
1052 * Log statements where the optimizer row estimates were dramatically
1053 different from the number of rows actually found?
1056 Miscellaneous Performance
1057 =========================
1059 * Do async I/O for faster random read-ahead of data
1061 Async I/O allows multiple I/O requests to be sent to the disk with
1062 results coming back asynchronously.
1064 * Use mmap() rather than SYSV shared memory or to write WAL files?
1066 This would remove the requirement for SYSV SHM but would introduce
1067 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
1068 to prevent I/O overhead.
1070 * Consider mmap()'ing files into a backend?
1072 Doing I/O to large tables would consume a lot of address space or
1073 require frequent mapping/unmapping. Extending the file also causes
1074 mapping problems that might require mapping only individual pages,
1075 leading to thousands of mappings. Another problem is that there is no
1076 way to _prevent_ I/O to disk from the dirty shared buffers so changes
1077 could hit disk before WAL is written.
1079 * Add a script to ask system configuration questions and tune postgresql.conf
1080 * Merge xmin/xmax/cmin/cmax back into three header fields
1082 Before subtransactions, there used to be only three fields needed to
1083 store these four values. This was possible because only the current
1084 transaction looks at the cmin/cmax values. If the current transaction
1085 created and expired the row the fields stored where xmin (same as
1086 xmax), cmin, cmax, and if the transaction was expiring a row from a
1087 another transaction, the fields stored were xmin (cmin was not
1088 needed), xmax, and cmax. Such a system worked because a transaction
1089 could only see rows from another completed transaction. However,
1090 subtransactions can see rows from outer transactions, and once the
1091 subtransaction completes, the outer transaction continues, requiring
1092 the storage of all four fields. With subtransactions, an outer
1093 transaction can create a row, a subtransaction expire it, and when the
1094 subtransaction completes, the outer transaction still has to have
1095 proper visibility of the row's cmin, for example, for cursors.
1097 One possible solution is to create a phantom cid which represents a
1098 cmin/cmax pair and is stored in local memory. Another idea is to
1099 store both cmin and cmax only in local memory.
1101 * Research storing disk pages with no alignment/padding
1107 * Add use of 'const' for variables in source tree
1108 * Rename some /contrib modules from pg* to pg_*
1109 * Move some things from /contrib into main tree
1110 * Move some /contrib modules out to their own project sites
1111 * %Remove warnings created by -Wcast-align
1112 * Move platform-specific ps status display info from ps_status.c to ports
1113 * Add optional CRC checksum to heap and index pages
1114 * Improve documentation to build only interfaces (Marc)
1115 * Remove or relicense modules that are not under the BSD license, if possible
1116 * %Remove memory/file descriptor freeing before ereport(ERROR)
1117 * Acquire lock on a relation before building a relcache entry for it
1118 * %Promote debug_query_string into a server-side function current_query()
1119 * %Allow the identifier length to be increased via a configure option
1120 * Allow cross-compiling by generating the zic database on the target system
1121 * Improve NLS maintenace of libpgport messages linked onto applications
1122 * Allow ecpg to work with MSVC and BCC
1123 * Add xpath_array() to /contrib/xml2 to return results as an array
1124 * Allow building in directories containing spaces
1126 This is probably not possible because 'gmake' and other compiler tools
1127 do not fully support quoting of paths with spaces.
1129 * -Allow installing to directories containing spaces
1131 This is possible if proper quoting is added to the makefiles for the
1132 install targets. Because PostgreSQL supports relocatable installs, it
1133 is already possible to install into a directory that doesn't contain
1134 spaces and then copy the install to a directory with spaces.
1136 * Fix sgmltools so PDFs can be generated with bookmarks
1137 * %Clean up compiler warnings (especially with gcc version 4)
1138 * Use UTF8 encoding for NLS messages so all server encodings can
1140 * Update Bonjour to work with newer cross-platform SDK
1141 * -Remove BeOS and QNX-specific code
1146 o Remove configure.in check for link failure when cause is found
1147 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1149 o Remove psql newline patch when we find out why mingw outputs an
1151 o Allow psql to use readline once non-US code pages work with
1153 o Re-enable timezone output on log_line_prefix '%t' when a
1154 shorter timezone string is available
1155 o Fix problem with shared memory on the Win32 Terminal Server
1156 o Improve signal handling,
1157 http://archives.postgresql.org/pgsql-patches/2005-06/msg00027.php
1158 o Add long file support for binary pg_dump output
1160 While Win32 supports 64-bit files, the MinGW API does not,
1161 meaning we have to build an fseeko replacement on top of the
1162 Win32 API, and we have to make sure MinGW handles it. Another
1163 option is to wait for the MinGW project to fix it, or use the
1164 code from the LibGW32C project as a guide.
1167 * Wire Protocol Changes
1169 o Allow dynamic character set handling
1170 o Add decoded type, length, precision
1172 o Update clients to use data types, typmod, schema.table.column names
1173 of result sets using new statement protocol
1176 ---------------------------------------------------------------------------
1179 Developers who have claimed items are:
1180 --------------------------------------
1181 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1182 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1183 * Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
1184 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1185 Family Health Network
1186 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1187 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1188 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1189 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1190 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1191 * Joe is Joe Conway <mail@joeconway.com>
1192 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1193 * Magnus is Magnus Hagander <mha@sollentuna.net>
1194 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1195 * Matthew T. O'Connor <matthew@zeut.net>
1196 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1197 * Neil is Neil Conway <neilc@samurai.com>
1198 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1199 * Peter is Peter Eisentraut <peter_e@gmx.net>
1200 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1201 * Rod is Rod Taylor <pg@rbt.ca>
1202 * Simon is Simon Riggs <simon@2ndquadrant.com>
1203 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1204 * Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
1205 * Teodor is Teodor Sigaev <teodor@sigaev.ru>
1206 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat