postgresql/doc/TODO
Bruce Momjian 977530d8da Done:
> * -Add concurrency to GIST
2005-06-27 23:21:58 +00:00

1069 lines
41 KiB
Plaintext

PostgreSQL TODO List
====================
Current maintainer: Bruce Momjian (pgman@candle.pha.pa.us)
Last updated: Mon Jun 27 19:21:55 EDT 2005
The most recent version of this document can be viewed at
http://www.postgresql.org/docs/faqs.TODO.html.
#A hyphen, "-", marks changes that will appear in the upcoming 8.1 release.#
Bracketed items, "[]", have more detail.
This list contains all known PostgreSQL bugs and feature requests. If
you would like to work on an item, please read the Developer's FAQ
first.
Administration
==============
* Remove behavior of postmaster -o after making postmaster/postgres
flags unique
* Allow limits on per-db/user connections
* Add group object ownership, so groups can rename/drop/grant on objects,
so we can implement roles
* Allow server log information to be output as INSERT statements
This would allow server log information to be easily loaded into
a database for analysis.
* Prevent default re-use of sysids for dropped users and groups
Currently, if a user is removed while he still owns objects, a new
user given might be given their user id and inherit the
previous users objects.
* Prevent dropping user that still owns objects, or auto-drop the objects
* Allow pooled connections to list all prepared queries
This would allow an application inheriting a pooled connection to know
the queries prepared in the current session.
* Allow major upgrades without dump/reload, perhaps using pg_upgrade
[pg_upgrade]
* Allow GRANT/REVOKE permissions to be applied to all schema objects with one
command
The proposed syntax is:
GRANT SELECT ON ALL TABLES IN public TO phpuser;
GRANT SELECT ON NEW TABLES IN public TO phpuser;
* Allow GRANT/REVOKE permissions to be inherited by objects based on
schema permissions
* Check for unreferenced table files created by transactions that were
in-progress when the server terminated abruptly
* Allow reporting of which objects are in which tablespaces
This item is difficult because a tablespace can contain objects from
multiple databases. There is a server-side function that returns the
databases which use a specific tablespace, so this requires a tool
that will call that function and connect to each database to find the
objects in each database for that tablespace.
* Allow a database in tablespace t1 with tables created in tablespace t2
to be used as a template for a new database created with default
tablespace t2
All objects in the default database tablespace must have default tablespace
specifications. This is because new databases are created by copying
directories. If you mix default tablespace tables and tablespace-specified
tables in the same directory, creating a new database from such a mixed
directory would create a new database with tables that had incorrect
explicit tablespaces. To fix this would require modifying pg_class in the
newly copied database, which we don't currently do.
* Add a GUC variable to control the tablespace for temporary objects and
sort files
It could start with a random tablespace from a supplied list and cycle
through the list.
* Add ability to monitor the use of temporary sort files
* Allow WAL replay of CREATE TABLESPACE to work when the directory
structure on the recovery computer is different from the original
* Add "include file" functionality in postgresql.conf
* -Add session start time and last statement time to pg_stat_activity
* Allow server logs to be remotely read using SQL commands
* Allow pg_hba.conf settings to be controlled via SQL
This would require a new global table that is dumped to flat file for
use by the postmaster. We do a similar thing for pg_shadow currently.
* Allow administrators to safely terminate individual sessions either
via an SQL function or SIGTERM
Currently SIGTERM of a backend can lead to lock table corruption.
* Un-comment all variables in postgresql.conf
By not showing commented-out variables, we discourage people from
thinking that re-commenting a variable returns it to its default.
This has to address environment variables that are then overridden
by config file values. Another option is to allow commented values
to return to their default values.
* Allow point-in-time recovery to archive partially filled write-ahead
logs [pitr]
Currently only full WAL files are archived. This means that the most
recent transactions aren't available for recovery in case of a disk
failure. This could be triggered by a user command or a timer.
* Automatically force archiving of partially-filled WAL files when
pg_stop_backup() is called or the server is stopped
Doing this will allow administrators to know more easily when the
archive contins all the files needed for point-in-time recovery.
* Create dump tool for write-ahead logs for use in determining
transaction id for point-in-time recovery
* Set proper permissions on non-system schemas during db creation
Currently all schemas are owned by the super-user because they are
copied from the template1 database.
* Add a function that returns the 'uptime' of the postmaster
* Allow a warm standby system to also allow read-only queries [pitr]
This is useful for checking PITR recovery.
* Allow the PITR process to be debugged and data examined
* -Add the client IP address and port to pg_stat_activity
* Improve replication solutions
o Load balancing
You can use any of the master/slave replication servers to use a
standby server for data warehousing. To allow read/write queries to
multiple servers, you need multi-master replication like pgcluster.
o Allow replication over unreliable or non-persistent links
* Support table partitioning that allows a single table to be stored
in subtables that are partitioned based on the primary key or a WHERE
clause
* Allow postgresql.conf values to be set so they can not be changed by
the user
* Allow per-tablespace quotas
Data Types
==========
* Remove Money type, add money formatting for decimal type
* Change NUMERIC to enforce the maximum precision, and increase it
* Add NUMERIC division operator that doesn't round?
Currently NUMERIC _rounds_ the result to the specified precision.
This means division can return a result that multiplied by the
divisor is greater than the dividend, e.g. this returns a value > 10:
SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
The positive modulus result returned by NUMERICs might be considered
inaccurate, in one sense.
* Add function to return compressed length of TOAST data values
* Allow INET subnet tests using non-constants to be indexed
* Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
functionality
Current CURRENT_TIMESTAMP returns the start time of the current
transaction, and gettimeofday() returns the wallclock time. This will
make time reporting more consistent and will allow reporting of
the statement start time.
* Have sequence dependency track use of DEFAULT sequences,
seqname.nextval (?)
* Disallow changing default expression of a SERIAL column (?)
* Allow infinite dates just like infinite timestamps
* Have initdb set DateStyle based on locale?
* Add pg_get_acldef(), pg_get_typedefault(), and pg_get_attrdef()
* Allow to_char() to print localized month names
* Allow functions to have a schema search path specified at creation time
* Allow substring/replace() to get/set bit values
* Add a GUC variable to allow output of interval values in ISO8601 format
* Fix data types where equality comparison isn't intuitive, e.g. box
* Merge hardwired timezone names with the TZ database; allow either kind
everywhere a TZ name is currently taken
* Allow customization of the known set of TZ names (generalize the
present australian_timezones hack)
* Allow TIMESTAMP WITH TIME ZONE to store the original timezone
information, either zone name or offset from UTC [timezone]
If the TIMESTAMP value is stored with a time zone name, interval
computations should adjust based on the time zone rules, e.g. adding
24 hours to a timestamp would yield a different result from adding one
day.
* Prevent INET cast to CIDR if the unmasked bits are not zero, or
zero the bits
* Prevent INET cast to CIDR from droping netmask, SELECT '1.1.1.1'::inet::cidr
* Allow INET + INT4 to increment the host part of the address, or
throw an error on overflow
* Add 'tid != tid ' operator for use in corruption recovery
* Prevent to_char() on interval from returning meaningless values
For example, to_char('1 month', 'mon') is meaningless. Basically,
most date-related parameters to to_char() are meaningless for
intervals because interval is not anchored to a date.
* Allow to_char() on interval values to accumulate the highest unit
requested
Some special format flag would be required to request such
accumulation. Such functionality could also be added to EXTRACT.
Prevent accumulation that crosses the month/day boundary because of
the uneven number of days in a month.
o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
o to_char(INTERVAL '3 years 5 months','MM') => 41
* Add ISO INTERVAL handling
o Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO SECOND
o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
o For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
'1', treat as ISO if there is a range specification clause,
and as PG if there no clause is present, e.g. interpret '1:30'
MINUTE TO SECOND as '1 minute 30 seconds', and interpret '1:30'
as '1 hour, 30 minutes'
o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
INTERVAL MONTH), and this should return '12 months'
o Round or truncate values to the requested precision, e.g.
INTERVAL '11 months' AS YEAR should return one or zero
o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
* ARRAYS
o Allow NULLs in arrays
o Allow MIN()/MAX() on arrays
o Delay resolution of array expression's data type so assignment
coercion can be performed on empty array expressions
o Modify array literal representation to handle array index lower bound
of other than one
* BINARY DATA
o Improve vacuum of large objects, like /contrib/vacuumlo (?)
o Add security checking for large objects
Currently large objects entries do not have owners. Permissions can
only be set at the pg_largeobject table level.
o Auto-delete large objects when referencing row is deleted
o Allow read/write into TOAST values like large objects
This requires the TOAST column to be stored EXTERNAL.
Multi-Language Support
======================
* Add NCHAR (as distinguished from ordinary varchar),
* Allow locale to be set at database creation
Currently locale can only be set during initdb. No global tables have
locale-aware columns. However, the database template used during
database creation might have locale-aware indexes. The indexes would
need to be reindexed to match the new locale.
* Allow encoding on a per-column basis
Right now only one encoding is allowed per database.
* Support multiple simultaneous character sets, per SQL92
* Improve UTF8 combined character handling (?)
* Add octet_length_server() and octet_length_client()
* Make octet_length_client() the same as octet_length()?
Views / Rules
=============
* Automatically create rules on views so they are updateable, per SQL99
We can only auto-create rules for simple views. For more complex
cases users will still have to write rules.
* Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
* Allow NOTIFY in rules involving conditionals
* Have views on temporary tables exist in the temporary namespace
* Allow temporary views on non-temporary tables
* Allow RULE recompilation
Indexes
=======
* Allow inherited tables to inherit index, UNIQUE constraint, and primary
key, foreign key
* UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
(dup) should fail
The main difficulty with this item is the problem of creating an index
that can span more than one table.
* Add UNIQUE capability to non-btree indexes
* Add rtree index support for line, lseg, path, point
* -Use indexes for MIN() and MAX()
MIN/MAX queries can already be rewritten as SELECT col FROM tab ORDER
BY col {DESC} LIMIT 1. Completing this item involves doing this
transformation automatically.
* -Use index to restrict rows returned by multi-key index when used with
non-consecutive keys to reduce heap accesses
For an index on col1,col2,col3, and a WHERE clause of col1 = 5 and
col3 = 9, spin though the index checking for col1 and col3 matches,
rather than just col1; also called skip-scanning.
* Prevent index uniqueness checks when UPDATE does not modify the column
Uniqueness (index) checks are done when updating a column even if the
column is not modified by the UPDATE.
* Fetch heap pages matching index entries in sequential order
Rather than randomly accessing heap pages based on index entries, mark
heap pages needing access in a bitmap and do the lookups in sequential
order. Another method would be to sort heap ctids matching the index
before accessing the heap rows.
* -Allow non-bitmap indexes to be combined by creating bitmaps in memory
This feature allows separate indexes to be ANDed or ORed together. This
is particularly useful for data warehousing applications that need to
query the database in an many permutations. This feature scans an index
and creates an in-memory bitmap, and allows that bitmap to be combined
with other bitmap created in a similar way. The bitmap can either index
all TIDs, or be lossy, meaning it records just page numbers and each
page tuple has to be checked for validity in a separate pass.
* Allow the creation of on-disk bitmap indexes which can be quickly
combined with other bitmap indexes
Such indexes could be more compact if there are only a few distinct values.
Such indexes can also be compressed. Keeping such indexes updated can be
costly.
* Allow use of indexes to search for NULLs
One solution is to create a partial index on an IS NULL expression.
* -Add concurrency to GIST
* Pack hash index buckets onto disk pages more efficiently
Currently no only one hash bucket can be stored on a page. Ideally
several hash buckets could be stored on a single page and greater
granularity used for the hash algorithm.
* Consider sorting hash buckets so entries can be found using a binary
search, rather than a linear scan
* In hash indexes, consider storing the hash value with or instead
of the key itself
* Allow accurate statistics to be collected on indexes with more than
one column or expression indexes, perhaps using per-index statistics
* Add fillfactor to control reserved free space during index creation
* Allow the creation of indexes with mixed ascending/descending specifiers
* -Fix incorrect rtree results due to wrong assumptions about "over"
operator semantics
Commands
========
* -Add BETWEEN SYMMETRIC/ASYMMETRIC
* Change LIMIT/OFFSET and FETCH/MOVE to use int8
* Allow CREATE TABLE AS to determine column lengths for complex
expressions like SELECT col1 || col2
* Allow UPDATE to handle complex aggregates [update] (?)
* Allow backslash handling in quoted strings to be disabled for portability
The use of C-style backslashes (.e.g. \n, \r) in quoted strings is not
SQL-spec compliant, so allow such handling to be disabled. However,
disabling backslashes could break many third-party applications and tools.
* Allow an alias to be provided for the target table in UPDATE/DELETE
This is not SQL-spec but many DBMSs allow it.
* -Allow additional tables to be specified in DELETE for joins
UPDATE already allows this (UPDATE...FROM) but we need similar
functionality in DELETE. It's been agreed that the keyword should
be USING, to avoid anything as confusing as DELETE FROM a FROM b.
* Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
* -Allow REINDEX to rebuild all database indexes
* Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
* Add a schema option to createlang
* Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple columns
* Allow SET CONSTRAINTS to be qualified by schema/table name
* Allow TRUNCATE ... CASCADE/RESTRICT
* Allow PREPARE of cursors
* Allow PREPARE to automatically determine parameter types based on the SQL
statement
* Allow finer control over the caching of prepared query plans
Currently, queries prepared via the libpq API are planned on first
execute using the supplied parameters --- allow SQL PREPARE to do the
same. Also, allow control over replanning prepared queries either
manually or automatically when statistics for execute parameters
differ dramatically from those used during planning.
* Allow LISTEN/NOTIFY to store info in memory rather than tables?
Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
such information in memory would improve performance.
* Dump large object comments in custom dump format
* Add optional textual message to NOTIFY
This would allow an informational message to be added to the notify
message, perhaps indicating the row modified or other custom
information.
* Use more reliable method for CREATE DATABASE to get a consistent copy
of db?
Currently the system uses the operating system COPY command to create
a new database.
* Add C code to copy directories for use in creating new databases
* Have pg_ctl look at PGHOST in case it is a socket directory?
* Allow pg_ctl to work properly with configuration files located outside
the PGDATA directory
pg_ctl can not read the pid file because it isn't located in the
config directory but in the PGDATA directory. The solution is to
allow pg_ctl to read and understand postgresql.conf to find the
data_directory value.
* Allow column-level GRANT/REVOKE privileges
* Add a GUC variable to warn about non-standard SQL usage in queries
* Add MERGE command that does UPDATE/DELETE, or on failure, INSERT (rules,
triggers?)
* Add ON COMMIT capability to CREATE TABLE AS SELECT
* Add NOVICE output level for helpful messages like automatic sequence/index
creation
* Add COMMENT ON for all cluster global objects (users, groups, databases
and tablespaces)
* Add an option to automatically use savepoints for each statement in a
multi-statement transaction.
When enabled, this would allow errors in multi-statement transactions
to be automatically ignored.
* Make row-wise comparisons work per SQL spec
* Add RESET CONNECTION command to reset all session state
This would include resetting of all variables (RESET ALL), dropping of
temporary tables, removing any NOTIFYs, cursors, open transactions,
prepared queries, currval()s, etc. This could be used for connection
pooling. We could also change RESET ALL to have this functionality.
The difficult of this features is allowing RESET ALL to not affect
changes made by the interface driver for its internal use. One idea
is for this to be a protocol-only feature. Another approach is to
notify the protocol when a RESET CONNECTION command is used.
* Allow FOR UPDATE queries to do NOWAIT locks
* Add GUC to issue notice about queries that use unjoined tables
* ALTER
o Have ALTER TABLE RENAME rename SERIAL sequence names
o Add ALTER DOMAIN TYPE
o Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
o Allow ALTER TABLE to change constraint deferrability and actions
o Disallow dropping of an inherited constraint
o Allow objects to be moved to different schemas
o Allow ALTER TABLESPACE to move to different directories
o Allow databases to be moved to different tablespaces
o Allow moving system tables to other tablespaces, where possible
Currently non-global system tables must be in the default database
tablespace. Global system tables can never be moved.
o Prevent child tables from altering constraints like CHECK that were
inherited from the parent table
* CLUSTER
o Automatically maintain clustering on a table
This might require some background daemon to maintain clustering
during periods of low usage. It might also require tables to be only
paritally filled for easier reorganization. Another idea would
be to create a merged heap/index data file so an index lookup would
automatically access the heap data too. A third idea would be to
store heap rows in hashed groups, perhaps using a user-supplied
hash function.
o Add default clustering to system tables
To do this, determine the ideal cluster index for each system
table and set the cluster setting during initdb.
* COPY
o Allow COPY to report error lines and continue
This requires the use of a savepoint before each COPY line is
processed, with ROLLBACK on COPY failure.
o -Allow COPY to understand \x as a hex byte
o Have COPY return the number of rows loaded/unloaded (?)
o -Allow COPY to optionally include column headings in the first line
o -Allow COPY FROM ... CSV to interpret newlines and carriage
returns in data
* CURSOR
o Allow UPDATE/DELETE WHERE CURRENT OF cursor
This requires using the row ctid to map cursor rows back to the
original heap row. This become more complicated if WITH HOLD cursors
are to be supported because WITH HOLD cursors have a copy of the row
and no FOR UPDATE lock.
o Prevent DROP TABLE from dropping a row referenced by its own open
cursor (?)
o Allow pooled connections to list all open WITH HOLD cursors
Because WITH HOLD cursors exist outside transactions, this allows
them to be listed so they can be closed.
* INSERT
o Allow INSERT/UPDATE of the system-generated oid value for a row
o Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
o Allow INSERT/UPDATE ... RETURNING new.col or old.col
This is useful for returning the auto-generated key for an INSERT.
One complication is how to handle rules that run as part of
the insert.
* SHOW/SET
o -Have SHOW ALL show descriptions for server-side variables
o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
ANALYZE, and CLUSTER
o Add SET PATH for schemas (?)
This is basically the same as SET search_path.
* SERVER-SIDE LANGUAGES
o -Allow PL/PgSQL's RAISE function to take expressions
Currently only constants are supported.
o -Change PL/PgSQL to use palloc() instead of malloc()
o Handle references to temporary tables that are created, destroyed,
then recreated during a session, and EXECUTE is not used
This requires the cached PL/PgSQL byte code to be invalidated when
an object referenced in the function is changed.
o Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
o Allow function parameters to be passed by name,
get_employee_salary(emp_id => 12345, tax_year => 2001)
o Add Oracle-style packages
o Add table function support to pltcl, plperl, plpython (?)
o Allow PL/pgSQL to name columns by ordinal position, e.g. rec.(3)
o -Allow PL/pgSQL EXECUTE query_var INTO record_var;
o Add capability to create and call PROCEDURES
o Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[]
o Add MOVE to PL/pgSQL
o Pass arrays natively instead of as text between plperl and postgres
o Add support for polymorphic arguments and return types to plperl
Clients
=======
* Add a libpq function to support Parse/DescribeStatement capability
* Prevent libpq's PQfnumber() from lowercasing the column name (?)
* Allow libpq to access SQLSTATE so pg_ctl can test for connection failure
This would be used for checking if the server is up.
* Have psql show current values for a sequence
* Move psql backslash database information into the backend, use mnemonic
commands? [psql]
This would allow non-psql clients to pull the same information out of
the database as psql.
* Fix psql's display of schema information (Neil)
* Allow psql \pset boolean variables to set to fixed values, rather than toggle
* Consistently display privilege information for all objects in psql
* Improve psql's handling of multi-line queries
* pg_dump
o Have pg_dump use multi-statement transactions for INSERT dumps
o Allow pg_dump to use multiple -t and -n switches [pg_dump]
o Add dumping of comments on composite type columns
o Add dumping of comments on index columns
o Replace crude DELETE FROM method of pg_dumpall for cleaning of
users and groups with separate DROP commands
o Add dumping and restoring of LOB comments
o Stop dumping CASCADE on DROP TYPE commands in clean mode
o Add full object name to the tag field. eg. for operators we need
'=(integer, integer)', instead of just '='.
o Add pg_dumpall custom format dumps.
This is probably best done by combining pg_dump and pg_dumpall
into a single binary.
o Add CSV output format
o Update pg_dump and psql to use the new COPY libpq API (Christopher)
* ECPG
o Docs
Document differences between ecpg and the SQL standard and
information about the Informix-compatibility module.
o Solve cardinality > 1 for input descriptors / variables (?)
o Add a semantic check level, e.g. check if a table really exists
o fix handling of DB attributes that are arrays
o Use backend PREPARE/EXECUTE facility for ecpg where possible
o Implement SQLDA
o Fix nested C comments
o sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
o Make SET CONNECTION thread-aware, non-standard?
o Allow multidimensional arrays
o Add internationalized message strings
Referential Integrity
=====================
* Add MATCH PARTIAL referential integrity
* Add deferred trigger queue file
Right now all deferred trigger information is stored in backend
memory. This could exhaust memory for very large trigger queues.
This item involves dumping large queues into files.
* -Implement shared row locks and use them in RI triggers
* Enforce referential integrity for system tables
* Change foreign key constraint for array -> element to mean element
in array (?)
* Allow DEFERRABLE UNIQUE constraints (?)
* Allow triggers to be disabled [trigger]
Currently the only way to disable triggers is to modify the system
tables.
* With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
If the dump is known to be valid, allow foreign keys to be added
without revalidating the data.
* Allow statement-level triggers to access modified rows
* Support triggers on columns (Greg Sabino Mullane)
* Remove CREATE CONSTRAINT TRIGGER
This was used in older releases to dump referential integrity
constraints.
* Allow AFTER triggers on system tables
System tables are modified in many places in the backend without going
through the executor and therefore not causing triggers to fire. To
complete this item, the functions that modify system tables will have
to fire triggers.
Dependency Checking
===================
* Flush cached query plans when the dependent objects change
* Track dependencies in function bodies and recompile/invalidate
Exotic Features
===============
* Add SQL99 WITH clause to SELECT
* Add SQL99 WITH RECURSIVE to SELECT
* Add pre-parsing phase that converts non-ISO syntax to supported
syntax
This could allow SQL written for other databases to run without
modification.
* Allow plug-in modules to emulate features from other databases
* SQL*Net listener that makes PostgreSQL appear as an Oracle database
to clients
* Allow queries across databases or servers with transaction
semantics
Right now contrib/dblink can be used to issue such queries except it
does not have locking or transaction semantics. Two-phase commit is
needed to enable transaction semantics.
* -Add two-phase commit
* Add the features of packages
o Make private objects accessable only to objects in the same schema
o Allow current_schema.objname to access current schema objects
o Add session variables
o Allow nested schemas
PERFORMANCE
===========
Fsync
=====
* Improve commit_delay handling to reduce fsync()
* Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
* Allow multiple blocks to be written to WAL with one write()
* Add an option to sync() before fsync()'ing checkpoint files
* Add program to test if fsync has a delay compared to non-fsync
Cache
=====
* Allow free-behind capability for large sequential scans, perhaps using
posix_fadvise()
Posix_fadvise() can control both sequential/random file caching and
free-behind behavior, but it is unclear how the setting affects other
backends that also have the file open, and the feature is not supported
on all operating systems.
* Consider use of open/fcntl(O_DIRECT) to minimize OS caching,
especially for WAL writes
O_DIRECT doesn't have the same media write guarantees as fsync, so it
is in addition to the fsync method, not in place of it.
* -Cache last known per-tuple offsets to speed long tuple access
* Speed up COUNT(*)
We could use a fixed row count and a +/- count to follow MVCC
visibility rules, or a single cached value could be used and
invalidated if anyone modifies the table. Another idea is to
get a count directly from a unique index, but for this to be
faster than a sequential scan it must avoid access to the heap
to obtain tuple visibility information.
* Allow data to be pulled directly from indexes
Currently indexes do not have enough tuple visibility information
to allow data to be pulled from the index without also accessing
the heap. One way to allow this is to set a bit to index tuples
to indicate if a tuple is currently visible to all transactions
when the first valid heap lookup happens. This bit would have to
be cleared when a heap tuple is expired.
* Consider automatic caching of queries at various levels:
o Parsed query tree
o Query execute plan
o Query results
* -Allow the size of the buffer cache used by temporary objects to be
specified as a GUC variable
Larger local buffer cache sizes requires more efficient handling of
local cache lookups.
* Improve the background writer
Allow the background writer to more efficiently write dirty buffers
from the end of the LRU cache and use a clock sweep algorithm to
write other dirty buffers to reduced checkpoint I/O
* Allow sequential scans to take advantage of other concurrent
sequentiqal scans, also called "Synchronised Scanning"
One possible implementation is to start sequential scans from the lowest
numbered buffer in the shared cache, and when reaching the end wrap
around to the beginning, rather than always starting sequential scans
at the start of the table.
Vacuum
======
* Improve speed with indexes
For large table adjustements during vacuum, it is faster to reindex
rather than update the index.
* Reduce lock time by moving tuples with read lock, then write
lock and truncate table
Moved tuples are invisible to other backends so they don't require a
write lock. However, the read lock promotion to write lock could lead
to deadlock situations.
* -Add a warning when the free space map is too small
* Maintain a map of recently-expired rows
This allows vacuum to target specific pages for possible free space
without requiring a sequential scan.
* Auto-fill the free space map by scanning the buffer cache or by
checking pages written by the background writer
* Create a bitmap of pages that need vacuuming
Instead of sequentially scanning the entire table, have the background
writer or some other process record pages that have expired rows, then
VACUUM can look at just those pages rather than the entire table. In
the event of a system crash, the bitmap would probably be invalidated.
* Auto-vacuum
o Move into the backend code
o Use free-space map information to guide refilling
o Do VACUUM FULL if table is nearly empty?
Locking
=======
* Make locking of shared data structures more fine-grained
This requires that more locks be acquired but this would reduce lock
contention, improving concurrency.
* Add code to detect an SMP machine and handle spinlocks accordingly
from distributted.net, http://www1.distributed.net/source,
in client/common/cpucheck.cpp
On SMP machines, it is possible that locks might be released shortly,
while on non-SMP machines, the backend should sleep so the process
holding the lock can complete and release it.
* -Improve SMP performance on i386 machines
i386-based SMP machines can generate excessive context switching
caused by lock failure in high concurrency situations. This may be
caused by CPU cache line invalidation inefficiencies.
* Research use of sched_yield() for spinlock acquisition failure
* Fix priority ordering of read and write light-weight locks (Neil)
Startup Time
============
* Experiment with multi-threaded backend [thread]
This would prevent the overhead associated with process creation. Most
operating systems have trivial process creation time compared to
database startup overhead, but a few operating systems (WIn32,
Solaris) might benefit from threading. Also explore the idea of
a single session using multiple threads to execute a query faster.
* Add connection pooling
It is unclear if this should be done inside the backend code or done
by something external like pgpool. The passing of file descriptors to
existing backends is one of the difficulties with a backend approach.
Write-Ahead Log
===============
* Eliminate need to write full pages to WAL before page modification [wal]
Currently, to protect against partial disk page writes, we write the
full page images to WAL before they are modified so we can correct any
partial page writes during recovery. These pages can also be
eliminated from point-in-time archive files.
* Reduce WAL traffic so only modified values are written rather than
entire rows (?)
* Turn off after-change writes if fsync is disabled
If fsync is off, there is no purpose in writing full pages to WAL
* Add WAL index reliability improvement to non-btree indexes
* Allow the pg_xlog directory location to be specified during initdb
with a symlink back to the /data location
* Allow WAL information to recover corrupted pg_controldata
* Find a way to reduce rotational delay when repeatedly writing
last WAL page
Currently fsync of WAL requires the disk platter to perform a full
rotation to fsync again. One idea is to write the WAL to different
offsets that might reduce the rotational delay.
* Allow buffered WAL writes and fsync
Instead of guaranteeing recovery of all committed transactions, this
would provide improved performance by delaying WAL writes and fsync
so an abrupt operating system restart might lose a few seconds of
committed transactions but still be consistent. We could perhaps
remove the 'fsync' parameter (which results in an an inconsistent
database) in favor of this capability.
* Eliminate WAL logging for CREATE TABLE AS when not doing WAL archiving
* -Change WAL to use 32-bit CRC, for performance reasons
Optimizer / Executor
====================
* Add missing optimizer selectivities for date, r-tree, etc
* Allow ORDER BY ... LIMIT # to select high/low value without sort or
index using a sequential scan for highest/lowest values
Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
all values to return the high/low value. Instead The idea is to do a
sequential scan to find the high/low value, thus avoiding the sort.
MIN/MAX already does this, but not for LIMIT > 1.
* Precompile SQL functions to avoid overhead
* Create utility to compute accurate random_page_cost value
* Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
* Have EXPLAIN ANALYZE highlight poor optimizer estimates
* Use CHECK constraints to influence optimizer decisions
CHECK constraints contain information about the distribution of values
within the table. This is also useful for implementing subtables where
a tables content is distributed across several subtables.
* Consider using hash buckets to do DISTINCT, rather than sorting
This would be beneficial when there are few distinct values.
* ANALYZE should record a pg_statistic entry for an all-NULL column
* Log queries where the optimizer row estimates were dramatically
different from the number of rows actually found (?)
Miscellaneous
=============
* Do async I/O for faster random read-ahead of data
Async I/O allows multiple I/O requests to be sent to the disk with
results coming back asynchronously.
* Use mmap() rather than SYSV shared memory or to write WAL files (?)
This would remove the requirement for SYSV SHM but would introduce
portability issues. Anonymous mmap (or mmap to /dev/zero) is required
to prevent I/O overhead.
* Consider mmap()'ing files into a backend?
Doing I/O to large tables would consume a lot of address space or
require frequent mapping/unmapping. Extending the file also causes
mapping problems that might require mapping only individual pages,
leading to thousands of mappings. Another problem is that there is no
way to _prevent_ I/O to disk from the dirty shared buffers so changes
could hit disk before WAL is written.
* Add a script to ask system configuration questions and tune postgresql.conf
* Use a phantom command counter for nested subtransactions to reduce
per-tuple overhead
Source Code
===========
* Add use of 'const' for variables in source tree
* Rename some /contrib modules from pg* to pg_*
* Move some things from /contrib into main tree
* Move some /contrib modules out to their own project sites
* Remove warnings created by -Wcast-align
* Move platform-specific ps status display info from ps_status.c to ports
* Add optional CRC checksum to heap and index pages
* Improve documentation to build only interfaces (Marc)
* Remove or relicense modules that are not under the BSD license, if possible
* Remove memory/file descriptor freeing before ereport(ERROR)
* Acquire lock on a relation before building a relcache entry for it
* Promote debug_query_string into a server-side function current_query()
* Allow the identifier length to be increased via a configure option
* Remove Win32 rename/unlink looping if unnecessary
* -Remove kerberos4 from source tree
* Allow cross-compiling by generating the zic database on the target system
* Improve NLS maintenace of libpgport messages linked onto applications
* Allow ecpg to work with MSVC and BCC
* -Make src/port/snprintf.c thread-safe
* Add xpath_array() to /contrib/xml2 to return results as an array
* Allow building in directories containing spaces
This is probably not possible because 'gmake' and other compiler tools
do not fully support quoting of paths with spaces.
* Allow installing to directories containing spaces
This is possible if proper quoting is added to the makefiles for the
install targets. Because PostgreSQL supports relocatable installs, it
is already possible to install into a directory that doesn't contain
spaces and then copy the install to a directory with spaces.
* Fix cross-compiling of time zone database via 'zic'
* Fix sgmltools so PDFs can be generated with bookmarks
* Win32
o Remove configure.in check for link failure when cause is found
o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1.4 is released
o Remove psql newline patch when we find out why mingw outputs an
extra newline
o Allow psql to use readline once non-US code pages work with
backslashes
o Re-enable timezone output on log_line_prefix '%t' when a
shorter timezone string is available
o Improve dlerror() reporting string
o Fix problem with shared memory on the Win32 Terminal Server
o Add support for Unicode
To fix this, the data needs to be converted to/from UTF16/UTF8
so the Win32 wcscoll() can be used, and perhaps other functions
like towupper(). However, UTF8 already works with normal
locales but provides no ordering or character set classes.
* Wire Protocol Changes
o Allow dynamic character set handling
o Add decoded type, length, precision
o Use compression?
o Update clients to use data types, typmod, schema.table.column names
of result sets using new query protocol
---------------------------------------------------------------------------
Developers who have claimed items are:
--------------------------------------
* Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
* Andrew is Andrew Dunstan <andrew@dunslane.net>
* Bruce is Bruce Momjian <pgman@candle.pha.pa.us> of Software Research Assoc.
* Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
Family Health Network
* Claudio is Claudio Natoli <claudio.natoli@memetrics.com>
* D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
* Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
* Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
* Greg is Greg Sabino Mullane <greg@turnstep.com>
* Hiroshi is Hiroshi Inoue <Inoue@tpf.co.jp>
* Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
* Joe is Joe Conway <mail@joeconway.com>
* Karel is Karel Zak <zakkr@zf.jcu.cz>
* Magnus is Magnus Hagander <mha@sollentuna.net>
* Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
* Matthew T. O'Connor <matthew@zeut.net>
* Michael is Michael Meskes <meskes@postgresql.org> of Credativ
* Neil is Neil Conway <neilc@samurai.com>
* Oleg is Oleg Bartunov <oleg@sai.msu.su>
* Peter is Peter Eisentraut <peter_e@gmx.net>
* Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
* Rod is Rod Taylor <pg@rbt.ca>
* Simon is Simon Riggs <simon@2ndquadrant.com>
* Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
* Tatsuo is Tatsuo Ishii <t-ishii@sra.co.jp> of Software Research Assoc.
* Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat