4 Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
5 Last updated: Mon May 30 17:12:22 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.#
12 Bracketed items, "[]", have more detail.
14 This list contains all known PostgreSQL bugs and feature requests. If
15 you would like to work on an item, please read the Developer's FAQ
22 * Remove behavior of postmaster -o after making postmaster/postgres
24 * Allow limits on per-db/user connections
25 * Add group object ownership, so groups can rename/drop/grant on objects,
26 so we can implement roles
27 * Allow server log information to be output as INSERT statements
29 This would allow server log information to be easily loaded into
30 a database for analysis.
32 * Prevent default re-use of sysids for dropped users and groups
34 Currently, if a user is removed while he still owns objects, a new
35 user given might be given their user id and inherit the
36 previous users objects.
38 * Prevent dropping user that still owns objects, or auto-drop the objects
39 * Allow pooled connections to list all prepared queries
41 This would allow an application inheriting a pooled connection to know
42 the queries prepared in the current session.
44 * Allow major upgrades without dump/reload, perhaps using pg_upgrade
45 * Have SHOW ALL and pg_settings show descriptions for server-side variables
46 * Allow GRANT/REVOKE permissions to be applied to all schema objects with one
49 The proposed syntax is:
50 GRANT SELECT ON ALL TABLES IN public TO phpuser;
51 GRANT SELECT ON NEW TABLES IN public TO phpuser;
53 * Allow GRANT/REVOKE permissions to be inherited by objects based on
55 * Check for unreferenced table files created by transactions that were
56 in-progress when the server terminated abruptly
57 * Allow reporting of which objects are in which tablespaces
59 This item is difficult because a tablespace can contain objects from
60 multiple databases. There is a server-side function that returns the
61 databases which use a specific tablespace, so this requires a tool
62 that will call that function and connect to each database to find the
63 objects in each database for that tablespace.
65 * Allow a database in tablespace t1 with tables created in tablespace t2
66 to be used as a template for a new database created with default
69 All objects in the default database tablespace must have default tablespace
70 specifications. This is because new databases are created by copying
71 directories. If you mix default tablespace tables and tablespace-specified
72 tables in the same directory, creating a new database from such a mixed
73 directory would create a new database with tables that had incorrect
74 explicit tablespaces. To fix this would require modifying pg_class in the
75 newly copied database, which we don't currently do.
77 * Add a GUC variable to control the tablespace for temporary objects and
80 It could start with a random tablespace from a supplied list and cycle
83 * Add ability to monitor the use of temporary sort files
84 * Allow WAL replay of CREATE TABLESPACE to work when the directory
85 structure on the recovery computer is different from the original
86 * Add "include file" functionality in postgresql.conf
87 * -Add session start time and last statement time to pg_stat_activity
88 * Allow server logs to be remotely read using SQL commands
89 * Allow pg_hba.conf settings to be controlled via SQL
91 This would require a new global table that is dumped to flat file for
92 use by the postmaster. We do a similar thing for pg_shadow currently.
94 * Allow administrators to safely terminate individual sessions
96 Right now, SIGTERM will terminate a session, but it is treated as
97 though the postmaster has paniced and shared memory might not be
98 cleaned up properly. A new signal is needed for safe termination.
100 * Un-comment all variables in postgresql.conf
102 By not showing commented-out variables, we discourage people from
103 thinking that re-commenting a variable returns it to its default.
104 This has to address environment variables that are then overridden
105 by config file values. Another option is to allow commented values
106 to return to their default values.
108 * Allow point-in-time recovery to archive partially filled write-ahead
111 Currently only full WAL files are archived. This means that the most
112 recent transactions aren't available for recovery in case of a disk
113 failure. This could be triggered by a user command or a timer.
115 * Automatically force archiving of partially-filled WAL files when
116 pg_stop_backup() is called or the server is stopped
118 Doing this will allow administrators to know more easily when the
119 archive contins all the files needed for point-in-time recovery.
121 * Create dump tool for write-ahead logs for use in determining
122 transaction id for point-in-time recovery
123 * Set proper permissions on non-system schemas during db creation
125 Currently all schemas are owned by the super-user because they are
126 copied from the template1 database.
128 * Add a function that returns the 'uptime' of the postmaster
129 * Allow a warm standby system to also allow read-only queries
131 This is useful for checking PITR recovery.
133 * Allow the PITR process to be debugged and data examined
134 * -Add the client IP address and port to pg_stat_activity
135 * Improve replication solutions
138 You can use any of the master/slave replication servers to use a
139 standby server for data warehousing. To allow read/write queries to
140 multiple servers, you need multi-master replication like pgcluster.
142 o Allow replication over unreliable or non-persistent links
144 * Support table partitioning that allows a single table to be stored
145 in subtables that are partitioned based on the primary key or a WHERE
147 * Allow postgresql.conf values to be set so they can not be changed by
154 * Remove Money type, add money formatting for decimal type
155 * Change NUMERIC to enforce the maximum precision, and increase it
156 * Add function to return compressed length of TOAST data values
157 * Allow INET subnet tests using non-constants to be indexed
158 * Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
161 Current CURRENT_TIMESTAMP returns the start time of the current
162 transaction, and gettimeofday() returns the wallclock time. This will
163 make time reporting more consistent and will allow reporting of
164 the statement start time.
166 * Have sequence dependency track use of DEFAULT sequences,
168 * Disallow changing default expression of a SERIAL column (?)
169 * Allow infinite dates just like infinite timestamps
170 * Have initdb set DateStyle based on locale?
171 * Add pg_get_acldef(), pg_get_typedefault(), and pg_get_attrdef()
172 * Allow to_char() to print localized month names
173 * Allow functions to have a schema search path specified at creation time
174 * Allow substring/replace() to get/set bit values
175 * Add a GUC variable to allow output of interval values in ISO8601 format
176 * Fix data types where equality comparison isn't intuitive, e.g. box
177 * Merge hardwired timezone names with the TZ database; allow either kind
178 everywhere a TZ name is currently taken
179 * Allow customization of the known set of TZ names (generalize the
180 present australian_timezones hack)
181 * Allow TIMESTAMP WITH TIME ZONE to store the original timezone
182 information, either zone name or offset from UTC
184 If the TIMESTAMP value is stored with a time zone name, interval
185 computations should adjust based on the time zone rules, e.g. adding
186 24 hours to a timestamp would yield a different result from adding one
189 * Prevent INET cast to CIDR if the unmasked bits are not zero, or
191 * Prevent INET cast to CIDR from droping netmask, SELECT '1.1.1.1'::inet::cidr
192 * Allow INET + INT4 to increment the host part of the address, or
193 throw an error on overflow
194 * Add 'tid != tid ' operator for use in corruption recovery
195 * Prevent to_char() on interval from returning meaningless values
197 For example, to_char('1 month', 'mon') is meaningless. Basically,
198 most date-related parameters to to_char() are meaningless for
199 intervals because interval is not anchored to a date.
201 * Allow to_char() on interval values to accumulate the highest unit
204 Some special format flag would be required to request such
205 accumulation. Such functionality could also be added to EXTRACT.
206 Prevent accumulation that crosses the month/day boundary because of
207 the uneven number of days in a month.
209 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
210 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
211 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
212 o to_char(INTERVAL '3 years 5 months','MM') => 41
214 * Add ISO INTERVAL handling
215 o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO SECOND
216 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
217 o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
218 '1', treat as ISO if there is a range specification clause,
219 and as PG if there no clause is present, e.g. interpret '1:30'
220 MINUTE TO SECOND as '1 minute 30 seconds', and interpret '1:30'
221 as '1 hour, 30 minutes'
222 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
223 INTERVAL MONTH), and this should return '12 months'
224 o Round or truncate values to the requested precision, e.g.
225 INTERVAL '11 months' AS YEAR should return one or zero
226 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
229 o Allow NULLs in arrays
230 o Allow MIN()/MAX() on arrays
231 o Delay resolution of array expression's data type so assignment
232 coercion can be performed on empty array expressions
233 o Modify array literal representation to handle array index lower bound
238 o Improve vacuum of large objects, like /contrib/vacuumlo (?)
239 o Add security checking for large objects
241 Currently large objects entries do not have owners. Permissions can
242 only be set at the pg_largeobject table level.
244 o Auto-delete large objects when referencing row is deleted
246 o Allow read/write into TOAST values like large objects
248 This requires the TOAST column to be stored EXTERNAL.
251 Multi-Language Support
252 ======================
254 * Add NCHAR (as distinguished from ordinary varchar),
255 * Allow locale to be set at database creation
257 Currently locale can only be set during initdb. No global tables have
258 locale-aware columns. However, the database template used during
259 database creation might have locale-aware indexes. The indexes would
260 need to be reindexed to match the new locale.
262 * Allow encoding on a per-column basis
264 Right now only one encoding is allowed per database.
266 * Support multiple simultaneous character sets, per SQL92
267 * Improve UTF8 combined character handling (?)
268 * Add octet_length_server() and octet_length_client()
269 * Make octet_length_client() the same as octet_length()?
275 * Automatically create rules on views so they are updateable, per SQL99
277 We can only auto-create rules for simple views. For more complex
278 cases users will still have to write rules.
280 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
281 * Allow NOTIFY in rules involving conditionals
282 * Have views on temporary tables exist in the temporary namespace
283 * Allow temporary views on non-temporary tables
284 * Allow RULE recompilation
290 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
292 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
293 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
296 The main difficulty with this item is the problem of creating an index
297 that can span more than one table.
299 * Add UNIQUE capability to non-btree indexes
300 * Add rtree index support for line, lseg, path, point
301 * -Use indexes for MIN() and MAX()
303 MIN/MAX queries can already be rewritten as SELECT col FROM tab ORDER
304 BY col {DESC} LIMIT 1. Completing this item involves doing this
305 transformation automatically.
307 * Use index to restrict rows returned by multi-key index when used with
308 non-consecutive keys to reduce heap accesses
310 For an index on col1,col2,col3, and a WHERE clause of col1 = 5 and
311 col3 = 9, spin though the index checking for col1 and col3 matches,
312 rather than just col1; also called skip-scanning.
314 * Prevent index uniqueness checks when UPDATE does not modify the column
316 Uniqueness (index) checks are done when updating a column even if the
317 column is not modified by the UPDATE.
319 * Fetch heap pages matching index entries in sequential order
321 Rather than randomly accessing heap pages based on index entries, mark
322 heap pages needing access in a bitmap and do the lookups in sequential
323 order. Another method would be to sort heap ctids matching the index
324 before accessing the heap rows.
326 * -Allow non-bitmap indexes to be combined by creating bitmaps in memory
328 This feature allows separate indexes to be ANDed or ORed together. This
329 is particularly useful for data warehousing applications that need to
330 query the database in an many permutations. This feature scans an index
331 and creates an in-memory bitmap, and allows that bitmap to be combined
332 with other bitmap created in a similar way. The bitmap can either index
333 all TIDs, or be lossy, meaning it records just page numbers and each
334 page tuple has to be checked for validity in a separate pass.
336 * Allow the creation of on-disk bitmap indexes which can be quickly
337 combined with other bitmap indexes
339 Such indexes could be more compact if there are only a few distinct values.
340 Such indexes can also be compressed. Keeping such indexes updated can be
343 * Allow use of indexes to search for NULLs
345 One solution is to create a partial index on an IS NULL expression.
347 * Add concurrency to GIST
348 * Pack hash index buckets onto disk pages more efficiently
350 Currently no only one hash bucket can be stored on a page. Ideally
351 several hash buckets could be stored on a single page and greater
352 granularity used for the hash algorithm.
354 * Consider sorting hash buckets so entries can be found using a binary
355 search, rather than a linear scan
356 * In hash indexes, consider storing the hash value with or instead
358 * Allow accurate statistics to be collected on indexes with more than
359 one column or expression indexes, perhaps using per-index statistics
360 * Add fillfactor to control reserved free space during index creation
361 * Allow the creation of indexes with mixed ascending/descending specifiers
367 * Add BETWEEN ASYMMETRIC/SYMMETRIC
368 * Change LIMIT/OFFSET and FETCH/MOVE to use int8
369 * Allow CREATE TABLE AS to determine column lengths for complex
370 expressions like SELECT col1 || col2
371 * Allow UPDATE to handle complex aggregates [update] (?)
372 * Allow backslash handling in quoted strings to be disabled for portability
374 The use of C-style backslashes (.e.g. \n, \r) in quoted strings is not
375 SQL-spec compliant, so allow such handling to be disabled. However,
376 disabling backslashes could break many third-party applications and tools.
378 * Allow an alias to be provided for the target table in UPDATE/DELETE
380 This is not SQL-spec but many DBMSs allow it.
382 * -Allow additional tables to be specified in DELETE for joins
384 UPDATE already allows this (UPDATE...FROM) but we need similar
385 functionality in DELETE. It's been agreed that the keyword should
386 be USING, to avoid anything as confusing as DELETE FROM a FROM b.
388 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
389 * Allow REINDEX to rebuild all database indexes, remove /contrib/reindex
390 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
391 * Add a schema option to createlang
392 * Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple columns
393 * Allow SET CONSTRAINTS to be qualified by schema/table name
394 * Allow TRUNCATE ... CASCADE/RESTRICT
395 * Allow PREPARE of cursors
396 * Allow PREPARE to automatically determine parameter types based on the SQL
398 * Allow finer control over the caching of prepared query plans
400 Currently, queries prepared via the libpq API are planned on first
401 execute using the supplied parameters --- allow SQL PREPARE to do the
402 same. Also, allow control over replanning prepared queries either
403 manually or automatically when statistics for execute parameters
404 differ dramatically from those used during planning.
406 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
408 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
409 such information in memory would improve performance.
411 * Dump large object comments in custom dump format
412 * Add optional textual message to NOTIFY
414 This would allow an informational message to be added to the notify
415 message, perhaps indicating the row modified or other custom
418 * Use more reliable method for CREATE DATABASE to get a consistent copy
421 Currently the system uses the operating system COPY command to create
424 * Add C code to copy directories for use in creating new databases
425 * Have pg_ctl look at PGHOST in case it is a socket directory?
426 * Allow column-level GRANT/REVOKE privileges
427 * Add a GUC variable to warn about non-standard SQL usage in queries
428 * Add MERGE command that does UPDATE/DELETE, or on failure, INSERT (rules,
430 * Add ON COMMIT capability to CREATE TABLE AS SELECT
431 * Add NOVICE output level for helpful messages like automatic sequence/index
433 * Add COMMENT ON for all cluster global objects (users, groups, databases
435 * Add an option to automatically use savepoints for each statement in a
436 multi-statement transaction.
438 When enabled, this would allow errors in multi-statement transactions
439 to be automatically ignored.
441 * Make row-wise comparisons work per SQL spec
442 * Add RESET CONNECTION command to reset all session state
444 This would include resetting of all variables (RESET ALL), dropping of
445 all temporary tables, removal of any NOTIFYs, cursors, prepared
446 queries(?), currval()s, etc. This could be used for connection pooling.
447 We could also change RESET ALL to have this functionality.
449 * Allow FOR UPDATE queries to do NOWAIT locks
450 * Add GUC to issue notice about queries that use unjoined tables
453 o Have ALTER TABLE RENAME rename SERIAL sequence names
454 o Add ALTER DOMAIN TYPE
455 o Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
456 o Allow ALTER TABLE to change constraint deferrability and actions
457 o Disallow dropping of an inherited constraint
458 o Allow objects to be moved to different schemas
459 o Allow ALTER TABLESPACE to move to different directories
460 o Allow databases and schemas to be moved to different tablespaces
462 One complexity is whether moving a schema should move all existing
463 schema objects or just define the location for future object creation.
465 o Allow moving system tables to other tablespaces, where possible
467 Currently non-global system tables must be in the default database
468 schema. Global system tables can never be moved.
470 o Prevent child tables from altering constraints like CHECK that were
471 inherited from the parent table
475 o Automatically maintain clustering on a table
477 This might require some background daemon to maintain clustering
478 during periods of low usage. It might also require tables to be only
479 paritally filled for easier reorganization. Another idea would
480 be to create a merged heap/index data file so an index lookup would
481 automatically access the heap data too. A third idea would be to
482 store heap rows in hashed groups, perhaps using a user-supplied
485 o Add default clustering to system tables
487 To do this, determine the ideal cluster index for each system
488 table and set the cluster setting during initdb.
492 o Allow COPY to report error lines and continue
494 This requires the use of a savepoint before each COPY line is
495 processed, with ROLLBACK on COPY failure.
497 o Allow COPY to understand \x as a hex byte
498 o Have COPY return the number of rows loaded/unloaded (?)
499 o -Allow COPY to optionally include column headings in the first line
500 o -Allow COPY FROM ... CSV to interpret newlines and carriage
505 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
507 This requires using the row ctid to map cursor rows back to the
508 original heap row. This become more complicated if WITH HOLD cursors
509 are to be supported because WITH HOLD cursors have a copy of the row
510 and no FOR UPDATE lock.
512 o Prevent DROP TABLE from dropping a row referenced by its own open
515 o Allow pooled connections to list all open WITH HOLD cursors
517 Because WITH HOLD cursors exist outside transactions, this allows
518 them to be listed so they can be closed.
522 o Allow INSERT/UPDATE of the system-generated oid value for a row
523 o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
524 o Allow INSERT/UPDATE ... RETURNING new.col or old.col
526 This is useful for returning the auto-generated key for an INSERT.
527 One complication is how to handle rules that run as part of
532 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
534 o Add SET PATH for schemas (?)
536 This is basically the same as SET search_path.
539 * SERVER-SIDE LANGUAGES
540 o Allow PL/PgSQL's RAISE function to take expressions (?)
542 Currently only constants are supported.
544 o -Change PL/PgSQL to use palloc() instead of malloc()
545 o Handle references to temporary tables that are created, destroyed,
546 then recreated during a session, and EXECUTE is not used
548 This requires the cached PL/PgSQL byte code to be invalidated when
549 an object referenced in the function is changed.
551 o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
552 o Allow function parameters to be passed by name,
553 get_employee_salary(emp_id => 12345, tax_year => 2001)
554 o Add Oracle-style packages
555 o Add table function support to pltcl, plperl, plpython (?)
556 o Allow PL/pgSQL to name columns by ordinal position, e.g. rec.(3)
557 o Allow PL/pgSQL EXECUTE query_var INTO record_var;
558 o Add capability to create and call PROCEDURES
559 o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
560 o Add MOVE to PL/pgSQL
566 * Add a libpq function to support Parse/DescribeStatement capability
567 * Prevent libpq's PQfnumber() from lowercasing the column name (?)
568 * Allow libpq to access SQLSTATE so pg_ctl can test for connection failure
570 This would be used for checking if the server is up.
572 * Have psql show current values for a sequence
573 * Move psql backslash database information into the backend, use mnemonic
576 This would allow non-psql clients to pull the same information out of
577 the database as psql.
579 * Fix psql's display of schema information (Neil)
580 * Allow psql \pset boolean variables to set to fixed values, rather than toggle
581 * Consistently display privilege information for all objects in psql
582 * Improve psql's handling of multi-line queries
584 o Have pg_dump use multi-statement transactions for INSERT dumps
585 o Allow pg_dump to use multiple -t and -n switches
587 This should be done by allowing a '-t schema.table' syntax.
589 o Add dumping of comments on composite type columns
590 o Add dumping of comments on index columns
591 o Replace crude DELETE FROM method of pg_dumpall for cleaning of
592 users and groups with separate DROP commands
593 o Add dumping and restoring of LOB comments
594 o Stop dumping CASCADE on DROP TYPE commands in clean mode
595 o Add full object name to the tag field. eg. for operators we need
596 '=(integer, integer)', instead of just '='.
597 o Add pg_dumpall custom format dumps.
599 This is probably best done by combining pg_dump and pg_dumpall
600 into a single binary.
602 o Add CSV output format
603 o Update pg_dump and psql to use the new COPY libpq API (Christopher)
608 Document differences between ecpg and the SQL standard and
609 information about the Informix-compatibility module.
611 o Solve cardinality > 1 for input descriptors / variables (?)
612 o Add a semantic check level, e.g. check if a table really exists
613 o fix handling of DB attributes that are arrays
614 o Use backend PREPARE/EXECUTE facility for ecpg where possible
616 o Fix nested C comments
617 o sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
618 o Make SET CONNECTION thread-aware, non-standard?
619 o Allow multidimensional arrays
620 o Add internationalized message strings
623 Referential Integrity
624 =====================
626 * Add MATCH PARTIAL referential integrity
627 * Add deferred trigger queue file
629 Right now all deferred trigger information is stored in backend
630 memory. This could exhaust memory for very large trigger queues.
631 This item involves dumping large queues into files.
633 * -Implement shared row locks and use them in RI triggers
634 * Enforce referential integrity for system tables
635 * Change foreign key constraint for array -> element to mean element
637 * Allow DEFERRABLE UNIQUE constraints (?)
638 * Allow triggers to be disabled [trigger]
640 Currently the only way to disable triggers is to modify the system
643 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
645 If the dump is known to be valid, allow foreign keys to be added
646 without revalidating the data.
648 * Allow statement-level triggers to access modified rows
649 * Support triggers on columns (Greg Sabino Mullane)
650 * Remove CREATE CONSTRAINT TRIGGER
652 This was used in older releases to dump referential integrity
655 * Allow AFTER triggers on system tables
657 System tables are modified in many places in the backend without going
658 through the executor and therefore not causing triggers to fire. To
659 complete this item, the functions that modify system tables will have
666 * Flush cached query plans when the dependent objects change
667 * Track dependencies in function bodies and recompile/invalidate
673 * Add SQL99 WITH clause to SELECT
674 * Add SQL99 WITH RECURSIVE to SELECT
675 * Add pre-parsing phase that converts non-ISO syntax to supported
678 This could allow SQL written for other databases to run without
681 * Allow plug-in modules to emulate features from other databases
682 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
684 * Allow queries across databases or servers with transaction
687 Right now contrib/dblink can be used to issue such queries except it
688 does not have locking or transaction semantics. Two-phase commit is
689 needed to enable transaction semantics.
691 * Add two-phase commit
692 * Add the features of packages
693 o Make private objects accessable only to objects in the same schema
694 o Allow current_schema.objname to access current schema objects
695 o Add session variables
696 o Allow nested schemas
706 * Improve commit_delay handling to reduce fsync()
707 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
708 * Allow multiple blocks to be written to WAL with one write()
709 * Add an option to sync() before fsync()'ing checkpoint files
715 * Allow free-behind capability for large sequential scans, perhaps using
718 Posix_fadvise() can control both sequential/random file caching and
719 free-behind behavior, but it is unclear how the setting affects other
720 backends that also have the file open, and the feature is not supported
721 on all operating systems.
723 * Consider use of open/fcntl(O_DIRECT) to minimize OS caching,
724 especially for WAL writes
725 * -Cache last known per-tuple offsets to speed long tuple access
728 We could use a fixed row count and a +/- count to follow MVCC
729 visibility rules, or a single cached value could be used and
730 invalidated if anyone modifies the table. Another idea is to
731 get a count directly from a unique index, but for this to be
732 faster than a sequential scan it must avoid access to the heap
733 to obtain tuple visibility information.
735 * Allow data to be pulled directly from indexes
737 Currently indexes do not have enough tuple visibility information
738 to allow data to be pulled from the index without also accessing
739 the heap. One way to allow this is to set a bit to index tuples
740 to indicate if a tuple is currently visible to all transactions
741 when the first valid heap lookup happens. This bit would have to
742 be cleared when a heap tuple is expired.
744 * Consider automatic caching of queries at various levels:
749 * -Allow the size of the buffer cache used by temporary objects to be
750 specified as a GUC variable
752 Larger local buffer cache sizes requires more efficient handling of
755 * Improve the background writer
757 Allow the background writer to more efficiently write dirty buffers
758 from the end of the LRU cache and use a clock sweep algorithm to
759 write other dirty buffers to reduced checkpoint I/O
761 * Allow sequential scans to take advantage of other concurrent
762 sequentiqal scans, also called "Synchronised Scanning"
764 One possible implementation is to start sequential scans from the lowest
765 numbered buffer in the shared cache, and when reaching the end wrap
766 around to the beginning, rather than always starting sequential scans
767 at the start of the table.
772 * Improve speed with indexes
774 For large table adjustements during vacuum, it is faster to reindex
775 rather than update the index.
777 * Reduce lock time by moving tuples with read lock, then write
778 lock and truncate table
780 Moved tuples are invisible to other backends so they don't require a
781 write lock. However, the read lock promotion to write lock could lead
782 to deadlock situations.
784 * -Add a warning when the free space map is too small
785 * Maintain a map of recently-expired rows
787 This allows vacuum to target specific pages for possible free space
788 without requiring a sequential scan.
791 o Move into the backend code
792 o Scan the buffer cache to find free space or use background writer
793 o Use free-space map information to guide refilling
794 o Do VACUUM FULL if table is nearly empty?
800 * Make locking of shared data structures more fine-grained
802 This requires that more locks be acquired but this would reduce lock
803 contention, improving concurrency.
805 * Add code to detect an SMP machine and handle spinlocks accordingly
806 from distributted.net, http://www1.distributed.net/source,
807 in client/common/cpucheck.cpp
809 On SMP machines, it is possible that locks might be released shortly,
810 while on non-SMP machines, the backend should sleep so the process
811 holding the lock can complete and release it.
813 * -Improve SMP performance on i386 machines
815 i386-based SMP machines can generate excessive context switching
816 caused by lock failure in high concurrency situations. This may be
817 caused by CPU cache line invalidation inefficiencies.
819 * Research use of sched_yield() for spinlock acquisition failure
820 * Fix priority ordering of read and write light-weight locks (Neil)
826 * Experiment with multi-threaded backend [thread]
828 This would prevent the overhead associated with process creation. Most
829 operating systems have trivial process creation time compared to
830 database startup overhead, but a few operating systems (WIn32,
831 Solaris) might benefit from threading. Also explore the idea of
832 a single session using multiple threads to execute a query faster.
834 * Add connection pooling
836 It is unclear if this should be done inside the backend code or done
837 by something external like pgpool. The passing of file descriptors to
838 existing backends is one of the difficulties with a backend approach.
844 * Eliminate need to write full pages to WAL before page modification [wal]
846 Currently, to protect against partial disk page writes, we write the
847 full page images to WAL before they are modified so we can correct any
848 partial page writes during recovery. These pages can also be
849 eliminated from point-in-time archive files.
851 * Reduce WAL traffic so only modified values are written rather than
853 * Turn off after-change writes if fsync is disabled
855 If fsync is off, there is no purpose in writing full pages to WAL
857 * Add WAL index reliability improvement to non-btree indexes
858 * Allow the pg_xlog directory location to be specified during initdb
859 with a symlink back to the /data location
860 * Allow WAL information to recover corrupted pg_controldata
861 * Find a way to reduce rotational delay when repeatedly writing
864 Currently fsync of WAL requires the disk platter to perform a full
865 rotation to fsync again. One idea is to write the WAL to different
866 offsets that might reduce the rotational delay.
868 * Allow buffered WAL writes and fsync
870 Instead of guaranteeing recovery of all committed transactions, this
871 would provide improved performance by delaying WAL writes and fsync
872 so an abrupt operating system restart might lose a few seconds of
873 committed transactions but still be consistent. We could perhaps
874 remove the 'fsync' parameter (which results in an an inconsistent
875 database) in favor of this capability.
877 * Eliminate WAL logging for CREATE TABLE AS when not doing WAL archiving
878 * Compress WAL entries [wal]
879 * Change WAL to use 32-bit CRC, for performance reasons
885 * Add missing optimizer selectivities for date, r-tree, etc
886 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
887 index using a sequential scan for highest/lowest values
889 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
890 all values to return the high/low value. Instead The idea is to do a
891 sequential scan to find the high/low value, thus avoiding the sort.
892 MIN/MAX already does this, but not for LIMIT > 1.
894 * Precompile SQL functions to avoid overhead
895 * Create utility to compute accurate random_page_cost value
896 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
897 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
898 * Use CHECK constraints to influence optimizer decisions
900 CHECK constraints contain information about the distribution of values
901 within the table. This is also useful for implementing subtables where
902 a tables content is distributed across several subtables.
904 * Consider using hash buckets to do DISTINCT, rather than sorting
906 This would be beneficial when there are few distinct values.
908 * ANALYZE should record a pg_statistic entry for an all-NULL column
909 * Log queries where the optimizer row estimates were dramatically
910 different from the number of rows actually found (?)
916 * Do async I/O for faster random read-ahead of data
918 Async I/O allows multiple I/O requests to be sent to the disk with
919 results coming back asynchronously.
921 * Use mmap() rather than SYSV shared memory or to write WAL files (?)
923 This would remove the requirement for SYSV SHM but would introduce
924 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
925 to prevent I/O overhead.
927 * Consider mmap()'ing files into a backend?
929 Doing I/O to large tables would consume a lot of address space or
930 require frequent mapping/unmapping. Extending the file also causes
931 mapping problems that might require mapping only individual pages,
932 leading to thousands of mappings. Another problem is that there is no
933 way to _prevent_ I/O to disk from the dirty shared buffers so changes
934 could hit disk before WAL is written.
936 * Add a script to ask system configuration questions and tune postgresql.conf
937 * Use a phantom command counter for nested subtransactions to reduce
944 * Add use of 'const' for variables in source tree
945 * Rename some /contrib modules from pg* to pg_*
946 * Move some things from /contrib into main tree
947 * Move some /contrib modules out to their own project sites
948 * Remove warnings created by -Wcast-align
949 * Move platform-specific ps status display info from ps_status.c to ports
950 * Add optional CRC checksum to heap and index pages
951 * Improve documentation to build only interfaces (Marc)
952 * Remove or relicense modules that are not under the BSD license, if possible
953 * Remove memory/file descriptor freeing before ereport(ERROR)
954 * Acquire lock on a relation before building a relcache entry for it
955 * Promote debug_query_string into a server-side function current_query()
956 * Allow the identifier length to be increased via a configure option
957 * Remove Win32 rename/unlink looping if unnecessary
958 * Remove kerberos4 from source tree?
959 * Allow cross-compiling by generating the zic database on the target system
960 * Improve NLS maintenace of libpgport messages linked onto applications
961 * Allow ecpg to work with MSVC and BCC
962 * -Make src/port/snprintf.c thread-safe
963 * Add xpath_array() to /contrib/xml2 to return results as an array
964 * Allow building in directories containing spaces
966 This is probably not possible because 'gmake' and other compiler tools
967 do not fully support quoting of paths with spaces.
969 * Allow installing to directories containing spaces
971 This is possible if proper quoting is added to the makefiles for the
972 install targets. Because PostgreSQL supports relocatable installs, it
973 is already possible to install into a directory that doesn't contain
974 spaces and then copy the install to a directory with spaces.
976 * Fix cross-compiling of time zone database via 'zic'
977 * Fix sgmltools so PDFs can be generated with bookmarks
979 o Remove configure.in check for link failure when cause is found
980 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
982 o Remove psql newline patch when we find out why mingw outputs an
984 o Allow psql to use readline once non-US code pages work with
986 o Re-enable timezone output on log_line_prefix '%t' when a
987 shorter timezone string is available
988 o Improve dlerror() reporting string
989 o Fix problem with shared memory on the Win32 Terminal Server
990 o Add support for Unicode
992 To fix this, the data needs to be converted to/from UTF16/UTF8
993 so the Win32 wcscoll() can be used, and perhaps other functions
994 like towupper(). However, UTF8 already works with normal
995 locales but provides no ordering or character set classes.
997 * Wire Protocol Changes
998 o Allow dynamic character set handling
999 o Add decoded type, length, precision
1001 o Update clients to use data types, typmod, schema.table.column names
1002 of result sets using new query protocol
1005 ---------------------------------------------------------------------------
1008 Developers who have claimed items are:
1009 --------------------------------------
1010 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1011 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1012 * Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
1013 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1014 Family Health Network
1015 * Claudio is Claudio Natoli <claudio.natoli@memetrics.com>
1016 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1017 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1018 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1019 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1020 * Hiroshi is Hiroshi Inoue <Inoue@tpf.co.jp>
1021 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1022 * Joe is Joe Conway <mail@joeconway.com>
1023 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1024 * Magnus is Magnus Hagander <mha@sollentuna.net>
1025 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1026 * Matthew T. O'Connor <matthew@zeut.net>
1027 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1028 * Neil is Neil Conway <neilc@samurai.com>
1029 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1030 * Peter is Peter Eisentraut <peter_e@gmx.net>
1031 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1032 * Rod is Rod Taylor <pg@rbt.ca>
1033 * Simon is Simon Riggs <simon@2ndquadrant.com>
1034 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1035 * Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
1036 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat