4 Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
5 Last updated: Fri Sep 23 16:32:41 EDT 2005
7 The most recent version of this document can be viewed at
8 http://www.postgresql.org/docs/faqs.TODO.html.
10 #A hyphen, "-", marks changes that will appear in the upcoming 8.1 release.#
11 #A percent sign, "%", marks items that are easier to implement.#
13 Bracketed items, "[]", have more detail.
15 This list contains all known PostgreSQL bugs and feature requests. If
16 you would like to work on an item, please read the Developer's FAQ
23 * %Remove behavior of postmaster -o after making postmaster/postgres
25 * %Allow pooled connections to list all prepared queries
27 This would allow an application inheriting a pooled connection to know
28 the queries prepared in the current session.
30 * Allow major upgrades without dump/reload, perhaps using pg_upgrade
32 * Check for unreferenced table files created by transactions that were
33 in-progress when the server terminated abruptly
34 * Allow administrators to safely terminate individual sessions either
35 via an SQL function or SIGTERM
37 Lock table corruption following SIGTERM of an individual backend
38 has been reported in 8.0. A possible cause was fixed in 8.1, but
39 it is unknown whether other problems exist. This item mostly
40 requires additional testing rather than of writing any new code.
42 * %Set proper permissions on non-system schemas during db creation
44 Currently all schemas are owned by the super-user because they are
45 copied from the template1 database.
47 * Support table partitioning that allows a single table to be stored
48 in subtables that are partitioned based on the primary key or a WHERE
50 * Add function to report the time of the most recent server reload
53 * Improve replication solutions
57 You can use any of the master/slave replication servers to use a
58 standby server for data warehousing. To allow read/write queries to
59 multiple servers, you need multi-master replication like pgcluster.
61 o Allow replication over unreliable or non-persistent links
66 o %Add "include file" functionality in postgresql.conf
67 o %Allow commenting of variables in postgresql.conf to restore them
70 Currently, if a variable is commented out, it keeps the
71 previous uncommented value until a server restarted.
73 o %Allow pg_hba.conf settings to be controlled via SQL
75 This would add a function to load the SQL table from
76 pg_hba.conf, and one to writes its contents to the flat file.
77 The table should have a line number that is a float so rows
78 can be inserted between existing rows, e.g. row 2.5 goes
79 between row 2 and row 3.
81 o %Allow postgresql.conf file values to be changed via an SQL
82 API, perhaps using SET GLOBAL
83 o Allow the server to be stopped/restarted via an SQL API
84 o Issue a warning if a change-on-restart-only postgresql.conf value
85 is modified and the server config files are reloaded
86 o Mark change-on-restart-only values in postgresql.conf
91 * Allow a database in tablespace t1 with tables created in
92 tablespace t2 to be used as a template for a new database created
93 with default tablespace t2
95 All objects in the default database tablespace must have default
96 tablespace specifications. This is because new databases are
97 created by copying directories. If you mix default tablespace
98 tables and tablespace-specified tables in the same directory,
99 creating a new database from such a mixed directory would create a
100 new database with tables that had incorrect explicit tablespaces.
101 To fix this would require modifying pg_class in the newly copied
102 database, which we don't currently do.
104 * Allow reporting of which objects are in which tablespaces
106 This item is difficult because a tablespace can contain objects
107 from multiple databases. There is a server-side function that
108 returns the databases which use a specific tablespace, so this
109 requires a tool that will call that function and connect to each
110 database to find the objects in each database for that tablespace.
112 o %Add a GUC variable to control the tablespace for temporary objects
115 It could start with a random tablespace from a supplied list and
116 cycle through the list.
118 o Allow WAL replay of CREATE TABLESPACE to work when the directory
119 structure on the recovery computer is different from the original
121 o Allow per-tablespace quotas
124 * Point-In-Time Recovery (PITR)
126 o Allow point-in-time recovery to archive partially filled
127 write-ahead logs [pitr]
129 Currently only full WAL files are archived. This means that the
130 most recent transactions aren't available for recovery in case
131 of a disk failure. This could be triggered by a user command or
134 o Automatically force archiving of partially-filled WAL files when
135 pg_stop_backup() is called or the server is stopped
137 Doing this will allow administrators to know more easily when
138 the archive contins all the files needed for point-in-time
141 o %Create dump tool for write-ahead logs for use in determining
142 transaction id for point-in-time recovery
143 o Allow a warm standby system to also allow read-only queries
146 This is useful for checking PITR recovery.
148 o Allow the PITR process to be debugged and data examined
154 * Allow server log information to be output as INSERT statements
156 This would allow server log information to be easily loaded into
157 a database for analysis.
159 * %Add ability to monitor the use of temporary sort files
160 * Allow server logs to be remotely read and removed using SQL commands
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, and increase it
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 * Have sequence dependency track use of DEFAULT sequences,
185 * %Disallow changing default expression of a SERIAL column?
186 * Fix data types where equality comparison isn't intuitive, e.g. box
187 * %Prevent INET cast to CIDR if the unmasked bits are not zero, or
189 * %Prevent INET cast to CIDR from droping netmask, SELECT '1.1.1.1'::inet::cidr
190 * Allow INET + INT4 to increment the host part of the address, or
191 throw an error on overflow
192 * %Add 'tid != tid ' operator for use in corruption recovery
193 * Allow user-defined types to specify a type modifier at table creation
199 o Allow infinite dates just like infinite timestamps
200 o Add a GUC variable to allow output of interval values in ISO8601
202 o Merge hardwired timezone names with the TZ database; allow either
203 kind everywhere a TZ name is currently taken
204 o Allow customization of the known set of TZ names (generalize the
205 present australian_timezones hack)
206 o Allow TIMESTAMP WITH TIME ZONE to store the original timezone
207 information, either zone name or offset from UTC [timezone]
209 If the TIMESTAMP value is stored with a time zone name, interval
210 computations should adjust based on the time zone rules.
212 o Fix SELECT '0.01 years'::interval, '0.01 months'::interval
213 o Fix SELECT INTERVAL '1' MONTH
214 o Add ISO INTERVAL handling
215 o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
217 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
218 o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
219 '1', treat as ISO if there is a range specification clause,
220 and as PG if there no clause is present, e.g. interpret
221 '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
222 interpret '1:30' as '1 hour, 30 minutes'
223 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
224 INTERVAL MONTH), and this should return '12 months'
225 o Round or truncate values to the requested precision, e.g.
226 INTERVAL '11 months' AS YEAR should return one or zero
227 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
232 o Allow NULLs in arrays
233 o Delay resolution of array expression's data type so assignment
234 coercion can be performed on empty array expressions
239 o Improve vacuum of large objects, like /contrib/vacuumlo?
240 o Add security checking for large objects
241 o Auto-delete large objects when referencing row is deleted
243 /contrib/lo offers this functionality.
245 o Allow read/write into TOAST values like large objects
247 This requires the TOAST column to be stored EXTERNAL.
253 * Allow INET subnet tests using non-constants to be indexed
254 * Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
257 Current CURRENT_TIMESTAMP returns the start time of the current
258 transaction, and gettimeofday() returns the wallclock time. This will
259 make time reporting more consistent and will allow reporting of
260 the statement start time.
262 * %Add pg_get_acldef(), pg_get_typedefault(), and pg_get_attrdef()
263 * Allow to_char() to print localized month names
264 * Allow functions to have a schema search path specified at creation time
265 * Allow substring/replace() to get/set bit values
266 * Allow to_char() on interval values to accumulate the highest unit
269 Some special format flag would be required to request such
270 accumulation. Such functionality could also be added to EXTRACT.
271 Prevent accumulation that crosses the month/day boundary because of
272 the uneven number of days in a month.
274 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
275 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
276 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
277 o to_char(INTERVAL '3 years 5 months','MM') => 41
279 * Add sleep() function, remove from regress.c
282 Multi-Language Support
283 ======================
285 * Add NCHAR (as distinguished from ordinary varchar),
286 * Allow locale to be set at database creation
288 Currently locale can only be set during initdb. No global tables have
289 locale-aware columns. However, the database template used during
290 database creation might have locale-aware indexes. The indexes would
291 need to be reindexed to match the new locale.
293 * Allow encoding on a per-column basis
295 Right now only one encoding is allowed per database.
297 * Support multiple simultaneous character sets, per SQL92
298 * Improve UTF8 combined character handling?
299 * Add octet_length_server() and octet_length_client()
300 * Make octet_length_client() the same as octet_length()?
301 * Fix problems with wrong runtime encoding conversion for NLS message files
307 * %Automatically create rules on views so they are updateable, per SQL99
309 We can only auto-create rules for simple views. For more complex
310 cases users will still have to write rules.
312 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
313 * Allow NOTIFY in rules involving conditionals
314 * Allow VIEW/RULE recompilation when the underlying tables change
316 Another issue is whether underlying table changes should be reflected
317 in the view, e.g. should SELECT * show additional columns if they
318 are added after the view is created.
324 * Change LIMIT/OFFSET and FETCH/MOVE to use int8
325 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
326 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
327 * %Allow SET CONSTRAINTS to be qualified by schema/table name
328 * %Allow TRUNCATE ... CASCADE/RESTRICT
330 This is like DELETE CASCADE, but truncates.
332 * %Add a separate TRUNCATE permission
334 Currently only the owner can TRUNCATE a table because triggers are not
335 called, and the table is locked in exclusive mode.
337 * Allow PREPARE of cursors
338 * Allow PREPARE to automatically determine parameter types based on the SQL
340 * Allow finer control over the caching of prepared query plans
342 Currently, queries prepared via the libpq API are planned on first
343 execute using the supplied parameters --- allow SQL PREPARE to do the
344 same. Also, allow control over replanning prepared queries either
345 manually or automatically when statistics for execute parameters
346 differ dramatically from those used during planning.
348 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
350 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
351 such information in memory would improve performance.
353 * Add optional textual message to NOTIFY
355 This would allow an informational message to be added to the notify
356 message, perhaps indicating the row modified or other custom
359 * Add a GUC variable to warn about non-standard SQL usage in queries
360 * Add MERGE command that does UPDATE/DELETE, or on failure, INSERT (rules,
362 * Add NOVICE output level for helpful messages like automatic sequence/index
364 * %Add COMMENT ON for all cluster global objects (roles, databases
366 * %Make row-wise comparisons work per SQL spec
367 * Add RESET CONNECTION command to reset all session state
369 This would include resetting of all variables (RESET ALL), dropping of
370 temporary tables, removing any NOTIFYs, cursors, open transactions,
371 prepared queries, currval()s, etc. This could be used for connection
372 pooling. We could also change RESET ALL to have this functionality.
373 The difficult of this features is allowing RESET ALL to not affect
374 changes made by the interface driver for its internal use. One idea
375 is for this to be a protocol-only feature. Another approach is to
376 notify the protocol when a RESET CONNECTION command is used.
378 * Add GUC to issue notice about queries that use unjoined tables
379 * Allow EXPLAIN to identify tables that were skipped because of
381 * Allow EXPLAIN output to be more easily processed by scripts
382 * Eventually enable escape_string_warning and standard_conforming_strings
387 o Allow CREATE TABLE AS to determine column lengths for complex
388 expressions like SELECT col1 || col2
389 o Use more reliable method for CREATE DATABASE to get a consistent
391 o Add ON COMMIT capability to CREATE TABLE AS ... SELECT
395 o Allow UPDATE to handle complex aggregates [update]?
396 o Allow an alias to be provided for the target table in
399 This is not SQL-spec but many DBMSs allow it.
401 o Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple
407 o %Have ALTER TABLE RENAME rename SERIAL sequence names
408 o Add ALTER DOMAIN to modify the underlying data type
409 o %Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
410 o %Allow ALTER TABLE to change constraint deferrability and actions
411 o Add missing object types for ALTER ... SET SCHEMA
412 o Allow ALTER TABLESPACE to move to different directories
413 o Allow databases to be moved to different tablespaces
414 o Allow moving system tables to other tablespaces, where possible
416 Currently non-global system tables must be in the default database
417 tablespace. Global system tables can never be moved.
419 o %Disallow dropping of an inherited constraint
420 o %Prevent child tables from altering or dropping constraints
421 like CHECK that were inherited from the parent table
426 o Automatically maintain clustering on a table
428 This might require some background daemon to maintain clustering
429 during periods of low usage. It might also require tables to be only
430 paritally filled for easier reorganization. Another idea would
431 be to create a merged heap/index data file so an index lookup would
432 automatically access the heap data too. A third idea would be to
433 store heap rows in hashed groups, perhaps using a user-supplied
436 o %Add default clustering to system tables
438 To do this, determine the ideal cluster index for each system
439 table and set the cluster setting during initdb.
444 o Allow COPY to report error lines and continue
446 This requires the use of a savepoint before each COPY line is
447 processed, with ROLLBACK on COPY failure.
449 o %Have COPY return the number of rows loaded/unloaded?
450 o Allow COPY on a newly-created table to skip WAL logging
452 On crash recovery, the table involved in the COPY would
453 be removed or have its heap and index files truncated. One
454 issue is that no other backend should be able to add to
455 the table at the same time, which is something that is
458 o Allow COPY to output from views
460 Another idea would be to allow actual SELECT queries in a COPY.
465 o Allow column-level privileges
466 o %Allow GRANT/REVOKE permissions to be applied to all schema objects
469 The proposed syntax is:
470 GRANT SELECT ON ALL TABLES IN public TO phpuser;
471 GRANT SELECT ON NEW TABLES IN public TO phpuser;
473 * Allow GRANT/REVOKE permissions to be inherited by objects based on
479 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
481 This requires using the row ctid to map cursor rows back to the
482 original heap row. This become more complicated if WITH HOLD cursors
483 are to be supported because WITH HOLD cursors have a copy of the row
484 and no FOR UPDATE lock.
486 o Prevent DROP TABLE from dropping a row referenced by its own open
489 o %Allow pooled connections to list all open WITH HOLD cursors
491 Because WITH HOLD cursors exist outside transactions, this allows
492 them to be listed so they can be closed.
497 o Allow INSERT/UPDATE of the system-generated oid value for a row
498 o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
499 o Allow INSERT/UPDATE ... RETURNING new.col or old.col
501 This is useful for returning the auto-generated key for an INSERT.
502 One complication is how to handle rules that run as part of
508 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
510 o Add SET PATH for schemas?
512 This is basically the same as SET search_path.
515 * Server-Side Languages
517 o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
518 o Allow function parameters to be passed by name,
519 get_employee_salary(emp_id => 12345, tax_year => 2001)
520 o Add Oracle-style packages
521 o Add table function support to pltcl, plpython
522 o Add capability to create and call PROCEDURES
523 o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
524 o Allow function argument names to be queries from PL/PgSQL
525 o Add MOVE to PL/pgSQL
526 o Add support for polymorphic arguments and return types to
527 languages other than PL/PgSQL
528 o Add support for OUT and INOUT parameters to languages other
530 o Add single-step debugging of PL/PgSQL functions
536 * Add a libpq function to support Parse/DescribeStatement capability
537 * Prevent libpq's PQfnumber() from lowercasing the column name?
538 * Add PQescapeIdentifier() to libpq
539 * Have initdb set the input DateStyle (MDY or DMY) based on locale?
540 * Have pg_ctl look at PGHOST in case it is a socket directory?
541 * Allow pg_ctl to work properly with configuration files located outside
544 pg_ctl can not read the pid file because it isn't located in the
545 config directory but in the PGDATA directory. The solution is to
546 allow pg_ctl to read and understand postgresql.conf to find the
547 data_directory value.
552 o Have psql show current values for a sequence
553 o Move psql backslash database information into the backend, use
554 mnemonic commands? [psql]
556 This would allow non-psql clients to pull the same information out
557 of the database as psql.
559 o Fix psql's display of schema information (Neil)
560 o Allow psql \pset boolean variables to set to fixed values, rather
562 o Consistently display privilege information for all objects in psql
563 o Improve psql's handling of multi-line queries
565 Currently, while \e saves a single query as one entry, interactive
566 queries are saved one line at a time. Ideally all queries
567 whould be saved like \e does.
569 o Allow multi-line column values to align in the proper columns
571 If the second output column value is 'a\nb', the 'b' should appear
572 in the second display column, rather than the first column as it
575 o Display IN, INOUT, and OUT parameters in \df+
577 It probably requires psql to output newlines in the proper
578 column, which is already on the TODO list.
583 o %Have pg_dump use multi-statement transactions for INSERT dumps
584 o %Allow pg_dump to use multiple -t and -n switches [pg_dump]
585 o %Add dumping of comments on composite type columns
586 o %Add dumping of comments on index columns
587 o %Replace crude DELETE FROM method of pg_dumpall --clean for
588 cleaning of roles with separate DROP commands
589 o Stop dumping CASCADE on DROP TYPE commands in clean mode
590 o %Add full object name to the tag field. eg. for operators we need
591 '=(integer, integer)', instead of just '='.
592 o Add pg_dumpall custom format dumps?
593 o %Add CSV output format
594 o Update pg_dump and psql to use the new COPY libpq API (Christopher)
595 o Remove unnecessary function pointer abstractions in pg_dump source
603 Document differences between ecpg and the SQL standard and
604 information about the Informix-compatibility module.
606 o Solve cardinality > 1 for input descriptors / variables?
607 o Add a semantic check level, e.g. check if a table really exists
608 o fix handling of DB attributes that are arrays
609 o Use backend PREPARE/EXECUTE facility for ecpg where possible
611 o Fix nested C comments
612 o %sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
613 o Make SET CONNECTION thread-aware, non-standard?
614 o Allow multidimensional arrays
615 o Add internationalized message strings
618 Referential Integrity
619 =====================
621 * Add MATCH PARTIAL referential integrity
622 * Add deferred trigger queue file
624 Right now all deferred trigger information is stored in backend
625 memory. This could exhaust memory for very large trigger queues.
626 This item involves dumping large queues into files.
628 * Change foreign key constraint for array -> element to mean element
630 * Allow DEFERRABLE UNIQUE constraints?
631 * Allow triggers to be disabled in only the current session.
633 This is currently possible by starting a multi-statement transaction,
634 modifying the system tables, performing the desired SQL, restoring the
635 system tables, and committing the transaction. ALTER TABLE ...
636 TRIGGER requires a table lock so it is not ideal for this usage.
638 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
640 If the dump is known to be valid, allow foreign keys to be added
641 without revalidating the data.
643 * Allow statement-level triggers to access modified rows
644 * Support triggers on columns (Greg Sabino Mullane)
645 * Enforce referential integrity for system tables
646 * Allow AFTER triggers on system tables
648 System tables are modified in many places in the backend without going
649 through the executor and therefore not causing triggers to fire. To
650 complete this item, the functions that modify system tables will have
657 * Flush cached query plans when the dependent objects change
658 * Track dependencies in function bodies and recompile/invalidate
660 This is particularly important for references to temporary tables
661 in PL/PgSQL because PL/PgSQL caches query plans. The only workaround
662 in PL/PgSQL is to use EXECUTE. One complexity is that a function
663 might itself drop and recreate dependent tables, causing it to
664 invalidate its own query plan.
670 * Add SQL99 WITH clause to SELECT
671 * Add SQL99 WITH RECURSIVE to SELECT
672 * Add pre-parsing phase that converts non-ISO syntax to supported
675 This could allow SQL written for other databases to run without
678 * Allow plug-in modules to emulate features from other databases
679 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
681 * Allow queries across databases or servers with transaction
684 This can be done using dblink and two-phase commit.
686 * Add the features of packages
688 o Make private objects accessable only to objects in the same schema
689 o Allow current_schema.objname to access current schema objects
690 o Add session variables
691 o Allow nested schemas
697 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
699 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
700 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
703 The main difficulty with this item is the problem of creating an index
704 that can span more than one table.
706 * Allow SELECT ... FOR UPDATE on inherited tables
707 * Add UNIQUE capability to non-btree indexes
708 * Prevent index uniqueness checks when UPDATE does not modify the column
710 Uniqueness (index) checks are done when updating a column even if the
711 column is not modified by the UPDATE.
713 * Allow the creation of on-disk bitmap indexes which can be quickly
714 combined with other bitmap indexes
716 Such indexes could be more compact if there are only a few distinct values.
717 Such indexes can also be compressed. Keeping such indexes updated can be
720 * Allow use of indexes to search for NULLs
722 One solution is to create a partial index on an IS NULL expression.
724 * Allow accurate statistics to be collected on indexes with more than
725 one column or expression indexes, perhaps using per-index statistics
726 * Add fillfactor to control reserved free space during index creation
727 * Allow the creation of indexes with mixed ascending/descending specifiers
728 * Allow constraint_exclusion to work for UNIONs like it does for
729 inheritance, allow it to work for UPDATE and DELETE queries, and allow
730 it to be used for all queries with little performance impact
731 * Allow CREATE INDEX to take an additional parameter for use with
738 o Add more GIST index support for geometric data types
739 o Allow GIST indexes to create certain complex index types, like
740 digital trees (see Aoki)
744 o Pack hash index buckets onto disk pages more efficiently
746 Currently only one hash bucket can be stored on a page. Ideally
747 several hash buckets could be stored on a single page and greater
748 granularity used for the hash algorithm.
750 o Consider sorting hash buckets so entries can be found using a
751 binary search, rather than a linear scan
753 o In hash indexes, consider storing the hash value with or instead
756 o Add WAL logging for crash recovery
757 o Allow multi-column hash indexes
763 * Improve commit_delay handling to reduce fsync()
764 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
766 Ideally this requires a separate test program that can be run
767 at initdb time or optionally later.
769 * %Add an option to sync() before fsync()'ing checkpoint files
770 * Add program to test if fsync has a delay compared to non-fsync
776 * Allow free-behind capability for large sequential scans, perhaps using
779 Posix_fadvise() can control both sequential/random file caching and
780 free-behind behavior, but it is unclear how the setting affects other
781 backends that also have the file open, and the feature is not supported
782 on all operating systems.
786 We could use a fixed row count and a +/- count to follow MVCC
787 visibility rules, or a single cached value could be used and
788 invalidated if anyone modifies the table. Another idea is to
789 get a count directly from a unique index, but for this to be
790 faster than a sequential scan it must avoid access to the heap
791 to obtain tuple visibility information.
793 * Allow data to be pulled directly from indexes
795 Currently indexes do not have enough tuple visibility information
796 to allow data to be pulled from the index without also accessing
797 the heap. One way to allow this is to set a bit to index tuples
798 to indicate if a tuple is currently visible to all transactions
799 when the first valid heap lookup happens. This bit would have to
800 be cleared when a heap tuple is expired.
803 * Consider automatic caching of queries at various levels:
809 * Allow sequential scans to take advantage of other concurrent
810 sequentiqal scans, also called "Synchronised Scanning"
812 One possible implementation is to start sequential scans from the lowest
813 numbered buffer in the shared cache, and when reaching the end wrap
814 around to the beginning, rather than always starting sequential scans
815 at the start of the table.
821 * Improve speed with indexes
823 For large table adjustements during VACUUM FULL, it is faster to
824 reindex rather than update the index.
826 * Reduce lock time during VACUUM FULL by moving tuples with read lock,
827 then write lock and truncate table
829 Moved tuples are invisible to other backends so they don't require a
830 write lock. However, the read lock promotion to write lock could lead
831 to deadlock situations.
833 * Auto-fill the free space map by scanning the buffer cache or by
834 checking pages written by the background writer
835 * Create a bitmap of pages that need vacuuming
837 Instead of sequentially scanning the entire table, have the background
838 writer or some other process record pages that have expired rows, then
839 VACUUM can look at just those pages rather than the entire table. In
840 the event of a system crash, the bitmap would probably be invalidated.
841 One complexity is that index entries still have to be vacuumed, and
842 doing this without an index scan (by using the heap values to find the
843 index entry) might be slow and unreliable, especially for user-defined
846 * %Add system view to show free space map contents
851 o Use free-space map information to guide refilling
852 o %Issue log message to suggest VACUUM FULL if a table is nearly
854 o Improve xid wraparound detection by recording per-table rather
861 * Add code to detect an SMP machine and handle spinlocks accordingly
862 from distributted.net, http://www1.distributed.net/source,
863 in client/common/cpucheck.cpp
865 On SMP machines, it is possible that locks might be released shortly,
866 while on non-SMP machines, the backend should sleep so the process
867 holding the lock can complete and release it.
869 * Fix priority ordering of read and write light-weight locks (Neil)
872 Startup Time Improvements
873 =========================
875 * Experiment with multi-threaded backend [thread]
877 This would prevent the overhead associated with process creation. Most
878 operating systems have trivial process creation time compared to
879 database startup overhead, but a few operating systems (WIn32,
880 Solaris) might benefit from threading. Also explore the idea of
881 a single session using multiple threads to execute a query faster.
883 * Add connection pooling
885 It is unclear if this should be done inside the backend code or done
886 by something external like pgpool. The passing of file descriptors to
887 existing backends is one of the difficulties with a backend approach.
893 * Eliminate need to write full pages to WAL before page modification [wal]
895 Currently, to protect against partial disk page writes, we write
896 full page images to WAL before they are modified so we can correct any
897 partial page writes during recovery. These pages can also be
898 eliminated from point-in-time archive files.
900 o When off, write CRC to WAL and check file system blocks
903 If CRC check fails during recovery, remember the page in case
904 a later CRC for that page properly matches.
906 o Write full pages during file system write and not when
907 the page is modified in the buffer cache
909 This allows most full page writes to happen in the background
910 writer. It might cause problems for applying WAL on recovery
911 into a partially-written page, but later the full page will be
914 * Allow WAL traffic to be streamed to another server for stand-by
916 * Reduce WAL traffic so only modified values are written rather than
918 * Allow the pg_xlog directory location to be specified during initdb
919 with a symlink back to the /data location
920 * Allow WAL information to recover corrupted pg_controldata
921 * Find a way to reduce rotational delay when repeatedly writing
924 Currently fsync of WAL requires the disk platter to perform a full
925 rotation to fsync again. One idea is to write the WAL to different
926 offsets that might reduce the rotational delay.
928 * Allow buffered WAL writes and fsync
930 Instead of guaranteeing recovery of all committed transactions, this
931 would provide improved performance by delaying WAL writes and fsync
932 so an abrupt operating system restart might lose a few seconds of
933 committed transactions but still be consistent. We could perhaps
934 remove the 'fsync' parameter (which results in an an inconsistent
935 database) in favor of this capability.
941 * Add missing optimizer selectivities for date, r-tree, etc
942 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
943 index using a sequential scan for highest/lowest values
945 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
946 all values to return the high/low value. Instead The idea is to do a
947 sequential scan to find the high/low value, thus avoiding the sort.
948 MIN/MAX already does this, but not for LIMIT > 1.
950 * Precompile SQL functions to avoid overhead
951 * Create utility to compute accurate random_page_cost value
952 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
953 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
954 * Consider using hash buckets to do DISTINCT, rather than sorting
956 This would be beneficial when there are few distinct values. This is
957 already used by GROUP BY.
959 * Log queries where the optimizer row estimates were dramatically
960 different from the number of rows actually found?
963 Miscellaneous Performance
964 =========================
966 * Do async I/O for faster random read-ahead of data
968 Async I/O allows multiple I/O requests to be sent to the disk with
969 results coming back asynchronously.
971 * Use mmap() rather than SYSV shared memory or to write WAL files?
973 This would remove the requirement for SYSV SHM but would introduce
974 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
975 to prevent I/O overhead.
977 * Consider mmap()'ing files into a backend?
979 Doing I/O to large tables would consume a lot of address space or
980 require frequent mapping/unmapping. Extending the file also causes
981 mapping problems that might require mapping only individual pages,
982 leading to thousands of mappings. Another problem is that there is no
983 way to _prevent_ I/O to disk from the dirty shared buffers so changes
984 could hit disk before WAL is written.
986 * Add a script to ask system configuration questions and tune postgresql.conf
987 * Merge xmin/xmax/cmin/cmax back into three header fields
989 Before subtransactions, there used to be only three fields needed to
990 store these four values. This was possible because only the current
991 transaction looks at the cmin/cmax values. If the current transaction
992 created and expired the row the fields stored where xmin (same as
993 xmax), cmin, cmax, and if the transaction was expiring a row from a
994 another transaction, the fields stored were xmin (cmin was not
995 needed), xmax, and cmax. Such a system worked because a transaction
996 could only see rows from another completed transaction. However,
997 subtransactions can see rows from outer transactions, and once the
998 subtransaction completes, the outer transaction continues, requiring
999 the storage of all four fields. With subtransactions, an outer
1000 transaction can create a row, a subtransaction expire it, and when the
1001 subtransaction completes, the outer transaction still has to have
1002 proper visibility of the row's cmin, for example, for cursors.
1004 One possible solution is to create a phantom cid which represents a
1005 cmin/cmax pair and is stored in local memory. Another idea is to
1006 store both cmin and cmax only in local memory.
1008 * Research storing disk pages with no alignment/padding
1014 * Add use of 'const' for variables in source tree
1015 * Rename some /contrib modules from pg* to pg_*
1016 * Move some things from /contrib into main tree
1017 * Move some /contrib modules out to their own project sites
1018 * %Remove warnings created by -Wcast-align
1019 * Move platform-specific ps status display info from ps_status.c to ports
1020 * Add optional CRC checksum to heap and index pages
1021 * Improve documentation to build only interfaces (Marc)
1022 * Remove or relicense modules that are not under the BSD license, if possible
1023 * %Remove memory/file descriptor freeing before ereport(ERROR)
1024 * Acquire lock on a relation before building a relcache entry for it
1025 * %Promote debug_query_string into a server-side function current_query()
1026 * %Allow the identifier length to be increased via a configure option
1027 * Remove Win32 rename/unlink looping if unnecessary
1028 * Allow cross-compiling by generating the zic database on the target system
1029 * Improve NLS maintenace of libpgport messages linked onto applications
1030 * Allow ecpg to work with MSVC and BCC
1031 * Add xpath_array() to /contrib/xml2 to return results as an array
1032 * Allow building in directories containing spaces
1034 This is probably not possible because 'gmake' and other compiler tools
1035 do not fully support quoting of paths with spaces.
1037 * Allow installing to directories containing spaces
1039 This is possible if proper quoting is added to the makefiles for the
1040 install targets. Because PostgreSQL supports relocatable installs, it
1041 is already possible to install into a directory that doesn't contain
1042 spaces and then copy the install to a directory with spaces.
1044 * Fix sgmltools so PDFs can be generated with bookmarks
1045 * %Clean up compiler warnings (especially with gcc version 4)
1046 * Add function to return the thread safety status of libpq and ecpg
1047 * Use UTF8 encoding for NLS messages so all server encodings can
1049 * Update Bonjour to work with newer cross-platform SDK
1050 * Remove BeOS and QNX-specific code
1051 * Add options to pg_config to show the share_dir, sysconfdir,
1052 pkgincludedir, and localedir
1057 o Remove configure.in check for link failure when cause is found
1058 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1060 o Remove psql newline patch when we find out why mingw outputs an
1062 o Allow psql to use readline once non-US code pages work with
1064 o Re-enable timezone output on log_line_prefix '%t' when a
1065 shorter timezone string is available
1066 o Fix problem with shared memory on the Win32 Terminal Server
1069 * Wire Protocol Changes
1071 o Allow dynamic character set handling
1072 o Add decoded type, length, precision
1074 o Update clients to use data types, typmod, schema.table.column names
1075 of result sets using new query protocol
1078 ---------------------------------------------------------------------------
1081 Developers who have claimed items are:
1082 --------------------------------------
1083 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1084 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1085 * Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
1086 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1087 Family Health Network
1088 * Claudio is Claudio Natoli <claudio.natoli@memetrics.com>
1089 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1090 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1091 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1092 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1093 * Hiroshi is Hiroshi Inoue <Inoue@tpf.co.jp>
1094 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1095 * Joe is Joe Conway <mail@joeconway.com>
1096 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1097 * Magnus is Magnus Hagander <mha@sollentuna.net>
1098 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1099 * Matthew T. O'Connor <matthew@zeut.net>
1100 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1101 * Neil is Neil Conway <neilc@samurai.com>
1102 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1103 * Peter is Peter Eisentraut <peter_e@gmx.net>
1104 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1105 * Rod is Rod Taylor <pg@rbt.ca>
1106 * Simon is Simon Riggs <simon@2ndquadrant.com>
1107 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1108 * Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
1109 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat