4 Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
5 Last updated: Mon Apr 10 19:15:36 EDT 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?
53 * Allow log_min_messages to be specified on a per-module basis
55 This would allow administrators to see more detailed information from
56 specific sections of the backend, e.g. checkpoints, autovacuum, etc.
58 * Re-enable the GUC full_page_writes in 8.2 when reliability issues have
62 * Improve replication solutions
66 You can use any of the master/slave replication servers to use a
67 standby server for data warehousing. To allow read/write queries to
68 multiple servers, you need multi-master replication like pgcluster.
70 o Allow replication over unreliable or non-persistent links
75 o -Add "include file" functionality in postgresql.conf
76 o %Allow commenting of variables in postgresql.conf to restore them
79 Currently, if a variable is commented out, it keeps the
80 previous uncommented value until a server restarted.
82 o %Allow pg_hba.conf settings to be controlled via SQL
84 This would add a function to load the SQL table from
85 pg_hba.conf, and one to writes its contents to the flat file.
86 The table should have a line number that is a float so rows
87 can be inserted between existing rows, e.g. row 2.5 goes
88 between row 2 and row 3.
90 o Allow pg_hba.conf to specify host names along with IP addresses
92 Host name lookup could occur when the postmaster reads the
93 pg_hba.conf file, or when the backend starts. Another
94 solution would be to reverse lookup the connection IP and
95 check that hostname against the host names in pg_hba.conf.
96 We could also then check that the host name maps to the IP
99 o %Allow postgresql.conf file values to be changed via an SQL
100 API, perhaps using SET GLOBAL
101 o Allow the server to be stopped/restarted via an SQL API
102 o -Issue a warning if a change-on-restart-only postgresql.conf value
103 is modified and the server config files are reloaded
104 o Mark change-on-restart-only values in postgresql.conf
109 * Allow a database in tablespace t1 with tables created in
110 tablespace t2 to be used as a template for a new database created
111 with default tablespace t2
113 All objects in the default database tablespace must have default
114 tablespace specifications. This is because new databases are
115 created by copying directories. If you mix default tablespace
116 tables and tablespace-specified tables in the same directory,
117 creating a new database from such a mixed directory would create a
118 new database with tables that had incorrect explicit tablespaces.
119 To fix this would require modifying pg_class in the newly copied
120 database, which we don't currently do.
122 * Allow reporting of which objects are in which tablespaces
124 This item is difficult because a tablespace can contain objects
125 from multiple databases. There is a server-side function that
126 returns the databases which use a specific tablespace, so this
127 requires a tool that will call that function and connect to each
128 database to find the objects in each database for that tablespace.
130 o %Add a GUC variable to control the tablespace for temporary objects
133 It could start with a random tablespace from a supplied list and
134 cycle through the list.
136 o Allow WAL replay of CREATE TABLESPACE to work when the directory
137 structure on the recovery computer is different from the original
139 o Allow per-tablespace quotas
142 * Point-In-Time Recovery (PITR)
144 o Allow point-in-time recovery to archive partially filled
145 write-ahead logs [pitr]
147 Currently only full WAL files are archived. This means that the
148 most recent transactions aren't available for recovery in case
149 of a disk failure. This could be triggered by a user command or
152 o Automatically force archiving of partially-filled WAL files when
153 pg_stop_backup() is called or the server is stopped
155 Doing this will allow administrators to know more easily when
156 the archive contains all the files needed for point-in-time
159 o %Create dump tool for write-ahead logs for use in determining
160 transaction id for point-in-time recovery
161 o Allow a warm standby system to also allow read-only statements
164 This is useful for checking PITR recovery.
166 o Allow the PITR process to be debugged and data examined
172 * Allow server log information to be output as INSERT statements
174 This would allow server log information to be easily loaded into
175 a database for analysis.
177 * %Add ability to monitor the use of temporary sort files
178 * Allow server logs to be remotely read and removed using SQL commands
179 * Allow protocol-level BIND parameter values to be logged
185 * Improve the MONEY data type
187 Change the MONEY data type to use DECIMAL internally, with special
188 locale-aware output formatting.
190 * Change NUMERIC to enforce the maximum precision
191 * Add NUMERIC division operator that doesn't round?
193 Currently NUMERIC _rounds_ the result to the specified precision.
194 This means division can return a result that multiplied by the
195 divisor is greater than the dividend, e.g. this returns a value > 10:
197 SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
199 The positive modulus result returned by NUMERICs might be considered
200 inaccurate, in one sense.
202 * %Disallow changing default expression of a SERIAL column
203 * %Disallow ALTER SEQUENCE changes for SERIAL sequences because pg_dump
204 does not dump the changes
205 * Fix data types where equality comparison isn't intuitive, e.g. box
206 * -Zero umasked bits in conversion from INET cast to CIDR
207 * -Prevent INET cast to CIDR from dropping netmask, SELECT '1.1.1.1'::inet::cidr
208 * -Allow INET + INT8 to increment the host part of the address or
209 throw an error on overflow
210 * -Add 'tid != tid ' operator for use in corruption recovery
211 * Allow user-defined types to specify a type modifier at table creation
217 o Allow infinite dates and intervals just like infinite timestamps
218 o Merge hardwired timezone names with the TZ database; allow either
219 kind everywhere a TZ name is currently taken
220 o Allow customization of the known set of TZ names (generalize the
221 present australian_timezones hack)
222 o Allow TIMESTAMP WITH TIME ZONE to store the original timezone
223 information, either zone name or offset from UTC [timezone]
225 If the TIMESTAMP value is stored with a time zone name, interval
226 computations should adjust based on the time zone rules.
228 o Fix SELECT '0.01 years'::interval, '0.01 months'::interval
229 o Fix SELECT INTERVAL '1' MONTH
230 o Add a GUC variable to allow output of interval values in ISO8601
232 o Improve timestamptz subtraction to be DST-aware
234 Currently, subtracting one date from another that crosses a
235 daylight savings time adjustment can return '1 day 1 hour', but
236 adding that back to the first date returns a time one hour in
237 the future. This is caused by the adjustment of '25 hours' to
238 '1 day 1 hour', and '1 day' is the same time the next day, even
239 if daylight savings adjustments are involved.
241 o Fix interval display to support values exceeding 2^31 hours
242 o Add overflow checking to timestamp and interval arithmetic
243 o Add ISO INTERVAL handling
244 o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
246 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
247 o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
248 '1', treat as ISO if there is a range specification clause,
249 and as PG if there no clause is present, e.g. interpret
250 '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
251 interpret '1:30' as '1 hour, 30 minutes'
252 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
253 INTERVAL MONTH), and this should return '12 months'
254 o Round or truncate values to the requested precision, e.g.
255 INTERVAL '11 months' AS YEAR should return one or zero
256 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
261 o -Allow NULLs in arrays
262 o Delay resolution of array expression's data type so assignment
263 coercion can be performed on empty array expressions
268 o Improve vacuum of large objects, like /contrib/vacuumlo?
269 o Add security checking for large objects
270 o Auto-delete large objects when referencing row is deleted
272 /contrib/lo offers this functionality.
274 o Allow read/write into TOAST values like large objects
276 This requires the TOAST column to be stored EXTERNAL.
282 * Allow INET subnet tests using non-constants to be indexed
283 * Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
286 Current CURRENT_TIMESTAMP returns the start time of the current
287 transaction, and gettimeofday() returns the wallclock time. This will
288 make time reporting more consistent and will allow reporting of
289 the statement start time.
291 * %Add pg_get_acldef(), pg_get_typedefault(), pg_get_attrdef(),
292 pg_get_tabledef(), pg_get_domaindef(), pg_get_functiondef()
293 * -Allow to_char() to print localized month names
294 * Allow to_date() and to_timestamp() accept localized month names
295 * Add missing parameter handling in to_char()
297 http://archives.postgresql.org/pgsql-hackers/2005-12/msg00948.php
299 * Allow functions to have a schema search path specified at creation time
300 * Allow substring/replace() to get/set bit values
301 * Allow to_char() on interval values to accumulate the highest unit
304 Some special format flag would be required to request such
305 accumulation. Such functionality could also be added to EXTRACT.
306 Prevent accumulation that crosses the month/day boundary because of
307 the uneven number of days in a month.
309 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
310 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
311 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
312 o to_char(INTERVAL '3 years 5 months','MM') => 41
314 * -Add sleep() function, remove from regress.c
315 * Allow user-defined functions retuning a domain value to enforce domain
317 * Add SPI_gettypmod() to return the typemod for a TupleDesc
320 Multi-Language Support
321 ======================
323 * Add NCHAR (as distinguished from ordinary varchar),
324 * Allow locale to be set at database creation
326 Currently locale can only be set during initdb. No global tables have
327 locale-aware columns. However, the database template used during
328 database creation might have locale-aware indexes. The indexes would
329 need to be reindexed to match the new locale.
331 * Allow encoding on a per-column basis
333 Right now only one encoding is allowed per database.
335 * Support multiple simultaneous character sets, per SQL92
336 * Improve UTF8 combined character handling?
337 * Add octet_length_server() and octet_length_client()
338 * Make octet_length_client() the same as octet_length()?
339 * Fix problems with wrong runtime encoding conversion for NLS message files
345 * Automatically create rules on views so they are updateable, per SQL99
347 We can only auto-create rules for simple views. For more complex
348 cases users will still have to write rules manually.
350 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
351 * Allow NOTIFY in rules involving conditionals
352 * Allow VIEW/RULE recompilation when the underlying tables change
354 Another issue is whether underlying table changes should be reflected
355 in the view, e.g. should SELECT * show additional columns if they
356 are added after the view is created.
362 * Change LIMIT/OFFSET and FETCH/MOVE to use int8
363 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
364 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
365 * %Allow SET CONSTRAINTS to be qualified by schema/table name
366 * -Allow TRUNCATE ... CASCADE/RESTRICT
368 This is like DELETE CASCADE, but truncates.
370 * %Add a separate TRUNCATE permission
372 Currently only the owner can TRUNCATE a table because triggers are not
373 called, and the table is locked in exclusive mode.
375 * Allow PREPARE of cursors
376 * -Allow PREPARE to automatically determine parameter types based on the SQL
378 * Allow finer control over the caching of prepared query plans
380 Currently, queries prepared via the libpq API are planned on first
381 execute using the supplied parameters --- allow SQL PREPARE to do the
382 same. Also, allow control over replanning prepared queries either
383 manually or automatically when statistics for execute parameters
384 differ dramatically from those used during planning.
386 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
388 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
389 such information in memory would improve performance.
391 * Add optional textual message to NOTIFY
393 This would allow an informational message to be added to the notify
394 message, perhaps indicating the row modified or other custom
397 * Add a GUC variable to warn about non-standard SQL usage in queries
398 * Add SQL-standard MERGE command, typically used to merge two tables
401 This is similar to UPDATE, then for unmatched rows, INSERT.
402 Whether concurrent access allows modifications which could cause
403 row loss is implementation independent.
405 * Add REPLACE or UPSERT command that does UPDATE, or on failure, INSERT
408 To implement this cleanly requires that the table have a unique index
409 so duplicate checking can be easily performed. It is possible to
410 do it without a unique index if we require the user to LOCK the table
413 * Add NOVICE output level for helpful messages like automatic sequence/index
415 * -Add COMMENT ON for all cluster global objects (roles, databases
417 * -Make row-wise comparisons work per SQL spec
419 Right now, '(a, b) < (1, 2)' is processed as 'a < 1 and b < 2', but
420 the SQL standard requires it to be processed as a column-by-column
421 comparison, so the proper comparison is '(a < 1) OR (a = 1 AND b < 2)'.
423 * Add RESET CONNECTION command to reset all session state
425 This would include resetting of all variables (RESET ALL), dropping of
426 temporary tables, removing any NOTIFYs, cursors, open transactions,
427 prepared queries, currval()s, etc. This could be used for connection
428 pooling. We could also change RESET ALL to have this functionality.
429 The difficult of this features is allowing RESET ALL to not affect
430 changes made by the interface driver for its internal use. One idea
431 is for this to be a protocol-only feature. Another approach is to
432 notify the protocol when a RESET CONNECTION command is used.
434 * Add GUC to issue notice about statements that use unjoined tables
435 * Allow EXPLAIN to identify tables that were skipped because of
437 * Allow EXPLAIN output to be more easily processed by scripts
438 * -Eventually enable escape_string_warning and standard_conforming_strings
439 * Simplify dropping roles that have objects in several databases
440 * Allow COMMENT ON to accept an expression rather than just a string
441 * Allow the count returned by SELECT, etc to be to represent as an int64
442 to allow a higher range of values
443 * Make CLUSTER preserve recently-dead tuples per MVCC requirements
444 * Add SQL99 WITH clause to SELECT
445 * Add SQL99 WITH RECURSIVE to SELECT
450 o Allow CREATE TABLE AS to determine column lengths for complex
451 expressions like SELECT col1 || col2
452 o Use more reliable method for CREATE DATABASE to get a consistent
454 o -Add ON COMMIT capability to CREATE TABLE AS ... SELECT
458 o Allow UPDATE to handle complex aggregates [update]?
459 o -Allow an alias to be provided for the target table in
461 o Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple
467 o %Have ALTER TABLE RENAME rename SERIAL sequence names
468 o Add ALTER DOMAIN to modify the underlying data type
469 o %Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
470 o %Allow ALTER TABLE to change constraint deferrability and actions
471 o Add missing object types for ALTER ... SET SCHEMA
472 o Allow ALTER TABLESPACE to move to different directories
473 o Allow databases to be moved to different tablespaces
474 o Allow moving system tables to other tablespaces, where possible
476 Currently non-global system tables must be in the default database
477 tablespace. Global system tables can never be moved.
479 o %Prevent child tables from altering or dropping constraints
480 like CHECK that were inherited from the parent table
481 o Prevent parent tables from altering or dropping constraints
482 like CHECK that are inherited by child tables
484 Dropping constraints should only be possible with CASCADE.
486 o Have ALTER INDEX update the name of a constraint using that index
487 o Add ALTER TABLE RENAME CONSTRAINT, update index name also
492 o Automatically maintain clustering on a table
494 This might require some background daemon to maintain clustering
495 during periods of low usage. It might also require tables to be only
496 partially filled for easier reorganization. Another idea would
497 be to create a merged heap/index data file so an index lookup would
498 automatically access the heap data too. A third idea would be to
499 store heap rows in hashed groups, perhaps using a user-supplied
502 o %Add default clustering to system tables
504 To do this, determine the ideal cluster index for each system
505 table and set the cluster setting during initdb.
510 o Allow COPY to report error lines and continue
512 This requires the use of a savepoint before each COPY line is
513 processed, with ROLLBACK on COPY failure.
515 o -Have COPY return the number of rows loaded/unloaded?
516 o Allow COPY on a newly-created table to skip WAL logging
518 On crash recovery, the table involved in the COPY would
519 be removed or have its heap and index files truncated. One
520 issue is that no other backend should be able to add to
521 the table at the same time, which is something that is
524 o Allow COPY to output from views
526 Another idea would be to allow actual SELECT statements in a COPY.
531 o Allow column-level privileges
532 o %Allow GRANT/REVOKE permissions to be applied to all schema objects
535 The proposed syntax is:
536 GRANT SELECT ON ALL TABLES IN public TO phpuser;
537 GRANT SELECT ON NEW TABLES IN public TO phpuser;
539 * Allow GRANT/REVOKE permissions to be inherited by objects based on
542 * Allow SERIAL sequences to inherit permissions from the base table?
547 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
549 This requires using the row ctid to map cursor rows back to the
550 original heap row. This become more complicated if WITH HOLD cursors
551 are to be supported because WITH HOLD cursors have a copy of the row
552 and no FOR UPDATE lock.
554 o Prevent DROP TABLE from dropping a row referenced by its own open
557 o -Allow pooled connections to list all open WITH HOLD cursors
559 Because WITH HOLD cursors exist outside transactions, this allows
560 them to be listed so they can be closed.
565 o Allow INSERT/UPDATE of the system-generated oid value for a row
566 o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
567 o Allow INSERT/UPDATE ... RETURNING new.col or old.col
569 This is useful for returning the auto-generated key for an INSERT.
570 One complication is how to handle rules that run as part of
576 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
578 o Add SET PATH for schemas?
580 This is basically the same as SET search_path.
583 * Server-Side Languages
585 o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
586 o Allow function parameters to be passed by name,
587 get_employee_salary(emp_id => 12345, tax_year => 2001)
588 o Add Oracle-style packages
589 o Add table function support to pltcl, plpython
590 o Add capability to create and call PROCEDURES
591 o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
592 o Allow function argument names to be statements from PL/PgSQL
593 o Add MOVE to PL/pgSQL
594 o Add support for polymorphic arguments and return types to
595 languages other than PL/PgSQL
596 o Add support for OUT and INOUT parameters to languages other
598 o Add single-step debugging of PL/PgSQL functions
599 o Allow PL/PgSQL to support WITH HOLD cursors
605 * -Have initdb set the input DateStyle (MDY or DMY) based on locale
606 * Have pg_ctl look at PGHOST in case it is a socket directory?
607 * Allow pg_ctl to work properly with configuration files located outside
610 pg_ctl can not read the pid file because it isn't located in the
611 config directory but in the PGDATA directory. The solution is to
612 allow pg_ctl to read and understand postgresql.conf to find the
613 data_directory value.
618 o Have psql show current values for a sequence
619 o Move psql backslash database information into the backend, use
620 mnemonic commands? [psql]
622 This would allow non-psql clients to pull the same information out
623 of the database as psql.
625 o Fix psql's display of schema information (Neil)
626 o Allow psql \pset boolean variables to set to fixed values, rather
628 o Consistently display privilege information for all objects in psql
629 o -Improve psql's handling of multi-line statements
631 Currently, while \e saves a single statement as one entry, interactive
632 statements are saved one line at a time. Ideally all statements
633 would be saved like \e does.
635 o -Allow multi-line column values to align in the proper columns
637 If the second output column value is 'a\nb', the 'b' should appear
638 in the second display column, rather than the first column as it
641 o Display IN, INOUT, and OUT parameters in \df+
643 It probably requires psql to output newlines in the proper
644 column, which is already on the TODO list.
646 o Add auto-expanded mode so expanded output is used if the row
647 length is wider than the screen width.
649 Consider using auto-expanded mode for backslash commands like \df+.
651 o Prevent tab completion of SET TRANSACTION from querying the
652 database and therefore preventing the transaction isolation
653 level from being set.
655 Currently, SET <tab> causes a database lookup to check all
656 supported session variables. This query causes problems
657 because setting the transaction isolation level must be the
658 first statement of a transaction.
663 o %Have pg_dump use multi-statement transactions for INSERT dumps
664 o %Allow pg_dump to use multiple -t and -n switches [pg_dump]
665 o %Add dumping of comments on index columns and composite type columns
666 o %Add full object name to the tag field. eg. for operators we need
667 '=(integer, integer)', instead of just '='.
668 o Add pg_dumpall custom format dumps?
669 o %Add CSV output format
670 o Update pg_dump and psql to use the new COPY libpq API (Christopher)
671 o Remove unnecessary function pointer abstractions in pg_dump source
673 o Allow selection of individual object(s) of all types, not just
675 o In a selective dump, allow dumping of an object and all its
677 o Add options like pg_restore -l and -L to pg_dump
678 o Stop dumping CASCADE on DROP TYPE commands in clean mode
679 o Allow pg_dump --clean to drop roles that own objects or have
681 o Add -f to pg_dumpall
688 Document differences between ecpg and the SQL standard and
689 information about the Informix-compatibility module.
691 o Solve cardinality > 1 for input descriptors / variables?
692 o Add a semantic check level, e.g. check if a table really exists
693 o fix handling of DB attributes that are arrays
694 o Use backend PREPARE/EXECUTE facility for ecpg where possible
696 o Fix nested C comments
697 o %sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
698 o Make SET CONNECTION thread-aware, non-standard?
699 o Allow multidimensional arrays
700 o Add internationalized message strings
701 o Add COPY TO STDIN / STDOUT handling
706 o Add a function to support Parse/DescribeStatement capability
707 o Add PQescapeIdentifier()
708 o Prevent PQfnumber() from lowercasing unquoted the column name
710 PQfnumber() should never have been doing lowercasing, but
711 historically it has so we need a way to prevent it
713 o Allow statement results to be automatically batched to the client
715 Currently, all statement results are transferred to the libpq
716 client before libpq makes the results available to the
717 application. This feature would allow the application to make
718 use of the first result rows while the rest are transferred, or
719 held on the server waiting for them to be requested by libpq.
720 One complexity is that a statement like SELECT 1/col could error
721 out mid-way through the result set.
724 Referential Integrity
725 =====================
727 * Add MATCH PARTIAL referential integrity
728 * Add deferred trigger queue file
730 Right now all deferred trigger information is stored in backend
731 memory. This could exhaust memory for very large trigger queues.
732 This item involves dumping large queues into files.
734 * Change foreign key constraint for array -> element to mean element
736 * Allow DEFERRABLE UNIQUE constraints?
737 * Allow triggers to be disabled in only the current session.
739 This is currently possible by starting a multi-statement transaction,
740 modifying the system tables, performing the desired SQL, restoring the
741 system tables, and committing the transaction. ALTER TABLE ...
742 TRIGGER requires a table lock so it is not ideal for this usage.
744 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
746 If the dump is known to be valid, allow foreign keys to be added
747 without revalidating the data.
749 * Allow statement-level triggers to access modified rows
750 * Support triggers on columns (Greg Sabino Mullane)
751 * Enforce referential integrity for system tables
752 * Allow AFTER triggers on system tables
754 System tables are modified in many places in the backend without going
755 through the executor and therefore not causing triggers to fire. To
756 complete this item, the functions that modify system tables will have
763 * Flush cached query plans when the dependent objects change,
764 when the cardinality of parameters changes dramatically, or
765 when new ANALYZE statistics are available
767 A more complex solution would be to save multiple plans for different
768 cardinality and use the appropriate plan based on the EXECUTE values.
770 * Track dependencies in function bodies and recompile/invalidate
772 This is particularly important for references to temporary tables
773 in PL/PgSQL because PL/PgSQL caches query plans. The only workaround
774 in PL/PgSQL is to use EXECUTE. One complexity is that a function
775 might itself drop and recreate dependent tables, causing it to
776 invalidate its own query plan.
782 * Add pre-parsing phase that converts non-ISO syntax to supported
785 This could allow SQL written for other databases to run without
788 * Allow plug-in modules to emulate features from other databases
789 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
791 * Allow statements across databases or servers with transaction
794 This can be done using dblink and two-phase commit.
796 * Add the features of packages
798 o Make private objects accessible only to objects in the same schema
799 o Allow current_schema.objname to access current schema objects
800 o Add session variables
801 o Allow nested schemas
807 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
809 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
810 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
813 The main difficulty with this item is the problem of creating an index
814 that can span more than one table.
816 * Allow SELECT ... FOR UPDATE on inherited tables
817 * Add UNIQUE capability to non-btree indexes
818 * Prevent index uniqueness checks when UPDATE does not modify the column
820 Uniqueness (index) checks are done when updating a column even if the
821 column is not modified by the UPDATE.
823 * Allow the creation of on-disk bitmap indexes which can be quickly
824 combined with other bitmap indexes
826 Such indexes could be more compact if there are only a few distinct values.
827 Such indexes can also be compressed. Keeping such indexes updated can be
830 * Allow use of indexes to search for NULLs
832 One solution is to create a partial index on an IS NULL expression.
834 * Allow accurate statistics to be collected on indexes with more than
835 one column or expression indexes, perhaps using per-index statistics
836 * Add fillfactor to control reserved free space during index creation
837 * Allow the creation of indexes with mixed ascending/descending
840 This is possible now by creating an operator class with reversed sort
841 operators. One complexity is that NULLs would then appear at the start
842 of the result set, and this might affect certain sort types, like
845 * Allow constraint_exclusion to work for UNIONs like it does for
846 inheritance, allow it to work for UPDATE and DELETE statements, and allow
847 it to be used for all statements with little performance impact
848 * Allow CREATE INDEX to take an additional parameter for use with
850 * Consider compressing indexes by storing key values duplicated in
851 several rows as a single index entry
853 This is difficult because it requires datatype-specific knowledge.
858 o Add more GIST index support for geometric data types
859 o Allow GIST indexes to create certain complex index types, like
860 digital trees (see Aoki)
864 o Pack hash index buckets onto disk pages more efficiently
866 Currently only one hash bucket can be stored on a page. Ideally
867 several hash buckets could be stored on a single page and greater
868 granularity used for the hash algorithm.
870 o Consider sorting hash buckets so entries can be found using a
871 binary search, rather than a linear scan
873 o In hash indexes, consider storing the hash value with or instead
876 o Add WAL logging for crash recovery
877 o Allow multi-column hash indexes
883 * Improve commit_delay handling to reduce fsync()
884 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
886 Ideally this requires a separate test program that can be run
887 at initdb time or optionally later. Consider O_SYNC when
890 * %Add an option to sync() before fsync()'ing checkpoint files
891 * Add program to test if fsync has a delay compared to non-fsync
897 * Allow free-behind capability for large sequential scans, perhaps using
900 Posix_fadvise() can control both sequential/random file caching and
901 free-behind behavior, but it is unclear how the setting affects other
902 backends that also have the file open, and the feature is not supported
903 on all operating systems.
907 We could use a fixed row count and a +/- count to follow MVCC
908 visibility rules, or a single cached value could be used and
909 invalidated if anyone modifies the table. Another idea is to
910 get a count directly from a unique index, but for this to be
911 faster than a sequential scan it must avoid access to the heap
912 to obtain tuple visibility information.
914 * Add estimated_count(*) to return an estimate of COUNT(*)
916 This would use the planner ANALYZE statistics to return an estimated
919 * Allow data to be pulled directly from indexes
921 Currently indexes do not have enough tuple visibility information
922 to allow data to be pulled from the index without also accessing
923 the heap. One way to allow this is to set a bit on index tuples
924 to indicate if a tuple is currently visible to all transactions
925 when the first valid heap lookup happens. This bit would have to
926 be cleared when a heap tuple is expired.
928 Another idea is to maintain a bitmap of heap pages where all rows
929 are visible to all backends, and allow index lookups to reference
930 that bitmap to avoid heap lookups, perhaps the same bitmap we might
931 add someday to determine which heap pages need vacuuming. Frequently
932 accessed bitmaps would have to be stored in shared memory. One 8k
933 page of bitmaps could track 512MB of heap pages.
935 * Consider automatic caching of statements at various levels:
941 * Allow sequential scans to take advantage of other concurrent
942 sequential scans, also called "Synchronised Scanning"
944 One possible implementation is to start sequential scans from the lowest
945 numbered buffer in the shared cache, and when reaching the end wrap
946 around to the beginning, rather than always starting sequential scans
947 at the start of the table.
953 * Improve speed with indexes
955 For large table adjustments during VACUUM FULL, it is faster to
956 reindex rather than update the index.
958 * Reduce lock time during VACUUM FULL by moving tuples with read lock,
959 then write lock and truncate table
961 Moved tuples are invisible to other backends so they don't require a
962 write lock. However, the read lock promotion to write lock could lead
963 to deadlock situations.
965 * Auto-fill the free space map by scanning the buffer cache or by
966 checking pages written by the background writer
967 * Create a bitmap of pages that need vacuuming
969 Instead of sequentially scanning the entire table, have the background
970 writer or some other process record pages that have expired rows, then
971 VACUUM can look at just those pages rather than the entire table. In
972 the event of a system crash, the bitmap would probably be invalidated.
973 One complexity is that index entries still have to be vacuumed, and
974 doing this without an index scan (by using the heap values to find the
975 index entry) might be slow and unreliable, especially for user-defined
978 * -Add system view to show free space map contents
979 * Allow FSM to return free space toward the beginning of the heap file,
980 in hopes that empty pages at the end can be truncated by VACUUM
981 * Allow FSM page return free space based on table clustering, to assist
982 in maintaining clustering?
987 o Use free-space map information to guide refilling
988 o %Issue log message to suggest VACUUM FULL if a table is nearly
990 o Improve xid wraparound detection by recording per-table rather
992 o Consider logging activity either to the logs or a system view
998 * Fix priority ordering of read and write light-weight locks (Neil)
1001 Startup Time Improvements
1002 =========================
1004 * Experiment with multi-threaded backend for backend creation [thread]
1006 This would prevent the overhead associated with process creation. Most
1007 operating systems have trivial process creation time compared to
1008 database startup overhead, but a few operating systems (Win32,
1009 Solaris) might benefit from threading. Also explore the idea of
1010 a single session using multiple threads to execute a statement faster.
1012 * Experiment with multi-threaded backend better resource utilization
1014 This would allow a single query to make use of multiple CPU's or
1015 multiple I/O channels simultaneously.
1017 * Add connection pooling
1019 It is unclear if this should be done inside the backend code or done
1020 by something external like pgpool. The passing of file descriptors to
1021 existing backends is one of the difficulties with a backend approach.
1027 * Eliminate need to write full pages to WAL before page modification [wal]
1029 Currently, to protect against partial disk page writes, we write
1030 full page images to WAL before they are modified so we can correct any
1031 partial page writes during recovery. These pages can also be
1032 eliminated from point-in-time archive files.
1034 o When off, write CRC to WAL and check file system blocks
1037 If CRC check fails during recovery, remember the page in case
1038 a later CRC for that page properly matches.
1040 o Write full pages during file system write and not when
1041 the page is modified in the buffer cache
1043 This allows most full page writes to happen in the background
1044 writer. It might cause problems for applying WAL on recovery
1045 into a partially-written page, but later the full page will be
1048 * Allow WAL traffic to be streamed to another server for stand-by
1050 * Reduce WAL traffic so only modified values are written rather than
1052 * Allow the pg_xlog directory location to be specified during initdb
1053 with a symlink back to the /data location
1054 * Allow WAL information to recover corrupted pg_controldata
1055 * Find a way to reduce rotational delay when repeatedly writing
1058 Currently fsync of WAL requires the disk platter to perform a full
1059 rotation to fsync again. One idea is to write the WAL to different
1060 offsets that might reduce the rotational delay.
1062 * Allow buffered WAL writes and fsync
1064 Instead of guaranteeing recovery of all committed transactions, this
1065 would provide improved performance by delaying WAL writes and fsync
1066 so an abrupt operating system restart might lose a few seconds of
1067 committed transactions but still be consistent. We could perhaps
1068 remove the 'fsync' parameter (which results in an an inconsistent
1069 database) in favor of this capability.
1071 * Allow WAL logging to be turned off for a table, but the table
1072 might be dropped or truncated during crash recovery [walcontrol]
1074 Allow tables to bypass WAL writes and just fsync() dirty pages on
1075 commit. This should be implemented using ALTER TABLE, e.g. ALTER
1076 TABLE PERSISTENCE [ DROP | TRUNCATE | DEFAULT ]. Tables using
1077 non-default logging should not use referential integrity with
1078 default-logging tables. A table without dirty buffers during a
1079 crash could perhaps avoid the drop/truncate.
1081 * Allow WAL logging to be turned off for a table, but the table would
1082 avoid being truncated/dropped [walcontrol]
1084 To do this, only a single writer can modify the table, and writes
1085 must happen only on new pages so the new pages can be removed during
1086 crash recovery. Readers can continue accessing the table. Such
1087 tables probably cannot have indexes. One complexity is the handling
1088 of indexes on TOAST tables.
1091 Optimizer / Executor
1092 ====================
1094 * Improve selectivity functions for geometric operators
1095 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
1096 index using a sequential scan for highest/lowest values
1098 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
1099 all values to return the high/low value. Instead The idea is to do a
1100 sequential scan to find the high/low value, thus avoiding the sort.
1101 MIN/MAX already does this, but not for LIMIT > 1.
1103 * Precompile SQL functions to avoid overhead
1104 * Create utility to compute accurate random_page_cost value
1105 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
1106 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
1107 * Consider using hash buckets to do DISTINCT, rather than sorting
1109 This would be beneficial when there are few distinct values. This is
1110 already used by GROUP BY.
1112 * Log statements where the optimizer row estimates were dramatically
1113 different from the number of rows actually found?
1116 Miscellaneous Performance
1117 =========================
1119 * Do async I/O for faster random read-ahead of data
1121 Async I/O allows multiple I/O requests to be sent to the disk with
1122 results coming back asynchronously.
1124 * Use mmap() rather than SYSV shared memory or to write WAL files?
1126 This would remove the requirement for SYSV SHM but would introduce
1127 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
1128 to prevent I/O overhead.
1130 * Consider mmap()'ing files into a backend?
1132 Doing I/O to large tables would consume a lot of address space or
1133 require frequent mapping/unmapping. Extending the file also causes
1134 mapping problems that might require mapping only individual pages,
1135 leading to thousands of mappings. Another problem is that there is no
1136 way to _prevent_ I/O to disk from the dirty shared buffers so changes
1137 could hit disk before WAL is written.
1139 * Add a script to ask system configuration questions and tune postgresql.conf
1140 * Merge xmin/xmax/cmin/cmax back into three header fields
1142 Before subtransactions, there used to be only three fields needed to
1143 store these four values. This was possible because only the current
1144 transaction looks at the cmin/cmax values. If the current transaction
1145 created and expired the row the fields stored where xmin (same as
1146 xmax), cmin, cmax, and if the transaction was expiring a row from a
1147 another transaction, the fields stored were xmin (cmin was not
1148 needed), xmax, and cmax. Such a system worked because a transaction
1149 could only see rows from another completed transaction. However,
1150 subtransactions can see rows from outer transactions, and once the
1151 subtransaction completes, the outer transaction continues, requiring
1152 the storage of all four fields. With subtransactions, an outer
1153 transaction can create a row, a subtransaction expire it, and when the
1154 subtransaction completes, the outer transaction still has to have
1155 proper visibility of the row's cmin, for example, for cursors.
1157 One possible solution is to create a phantom cid which represents a
1158 cmin/cmax pair and is stored in local memory. Another idea is to
1159 store both cmin and cmax only in local memory.
1161 * Research storing disk pages with no alignment/padding
1167 * Add use of 'const' for variables in source tree
1168 * Rename some /contrib modules from pg* to pg_*
1169 * Move some things from /contrib into main tree
1170 * Move some /contrib modules out to their own project sites
1171 * %Remove warnings created by -Wcast-align
1172 * Move platform-specific ps status display info from ps_status.c to ports
1173 * Add optional CRC checksum to heap and index pages
1174 * Improve documentation to build only interfaces (Marc)
1175 * Remove or relicense modules that are not under the BSD license, if possible
1176 * %Remove memory/file descriptor freeing before ereport(ERROR)
1177 * Acquire lock on a relation before building a relcache entry for it
1178 * %Promote debug_query_string into a server-side function current_query()
1179 * %Allow the identifier length to be increased via a configure option
1180 * Allow cross-compiling by generating the zic database on the target system
1181 * Improve NLS maintenance of libpgport messages linked onto applications
1182 * Allow ecpg to work with MSVC and BCC
1183 * Add xpath_array() to /contrib/xml2 to return results as an array
1184 * Allow building in directories containing spaces
1186 This is probably not possible because 'gmake' and other compiler tools
1187 do not fully support quoting of paths with spaces.
1189 * -Allow installing to directories containing spaces
1191 This is possible if proper quoting is added to the makefiles for the
1192 install targets. Because PostgreSQL supports relocatable installs, it
1193 is already possible to install into a directory that doesn't contain
1194 spaces and then copy the install to a directory with spaces.
1196 * Fix sgmltools so PDFs can be generated with bookmarks
1197 * -%Clean up compiler warnings (especially with gcc version 4)
1198 * Use UTF8 encoding for NLS messages so all server encodings can
1200 * Update Bonjour to work with newer cross-platform SDK
1201 * -Remove BeOS and QNX-specific code
1202 * Split out libpq pgpass and environment documentation sections to make
1203 it easier for non-developers to find
1204 * Improve port/qsort() to handle sorts with 50% unique and 50% duplicate
1207 This involves choosing better pivot points for the quicksort.
1212 o Remove configure.in check for link failure when cause is found
1213 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1215 o Remove psql newline patch when we find out why mingw outputs an
1217 o Allow psql to use readline once non-US code pages work with
1219 o Re-enable timezone output on log_line_prefix '%t' when a
1220 shorter timezone string is available
1221 o Fix problem with shared memory on the Win32 Terminal Server
1222 o Improve signal handling,
1223 http://archives.postgresql.org/pgsql-patches/2005-06/msg00027.php
1224 o Add long file support for binary pg_dump output
1226 While Win32 supports 64-bit files, the MinGW API does not,
1227 meaning we have to build an fseeko replacement on top of the
1228 Win32 API, and we have to make sure MinGW handles it. Another
1229 option is to wait for the MinGW project to fix it, or use the
1230 code from the LibGW32C project as a guide.
1233 o Check WSACancelBlockingCall() for interrupts (win32intr)
1236 * Wire Protocol Changes
1238 o Allow dynamic character set handling
1239 o Add decoded type, length, precision
1241 o Update clients to use data types, typmod, schema.table.column names
1242 of result sets using new statement protocol
1245 ---------------------------------------------------------------------------
1248 Developers who have claimed items are:
1249 --------------------------------------
1250 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1251 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1252 * Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
1253 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1254 Family Health Network
1255 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1256 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1257 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1258 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1259 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1260 * Joe is Joe Conway <mail@joeconway.com>
1261 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1262 * Magnus is Magnus Hagander <mha@sollentuna.net>
1263 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1264 * Matthew T. O'Connor <matthew@zeut.net>
1265 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1266 * Neil is Neil Conway <neilc@samurai.com>
1267 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1268 * Peter is Peter Eisentraut <peter_e@gmx.net>
1269 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1270 * Rod is Rod Taylor <pg@rbt.ca>
1271 * Simon is Simon Riggs <simon@2ndquadrant.com>
1272 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1273 * Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
1274 * Teodor is Teodor Sigaev <teodor@sigaev.ru>
1275 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat