E.26. Release 9.5
Release date: 2016-01-07
Major enhancements in PostgreSQL 9.5 include:
INSERTs that would generate constraint conflicts to be turned into
UPDATEs or ignored
Create mechanisms for tracking the progress of replication, including methods for identifying the origin of individual changes during logical replication
Add Block Range Indexes (BRIN)
Substantial performance improvements for sorting
Substantial performance improvements for multi-CPU machines
The above items are explained in more detail in the sections below.
E.26.2. Migration to Version 9.5
Version 9.5 contains a number of changes that may affect compatibility with previous releases. Observe the following incompatibilities:
Adjust operator precedence to match the SQL standard (Tom Lane)
The precedence of
<>has been reduced to match that of
=. The precedence of
IS NULL) has been reduced to be just below these six comparison operators. Also, multi-keyword operators beginning with
NOTnow have the precedence of their base operator (for example,
NOT BETWEENnow has the same precedence as
BETWEEN) whereas before they had inconsistent precedence, behaving like
NOTwith respect to their left operand but like their base operator with respect to their right operand. The new configuration parameter operator_precedence_warning can be enabled to warn about queries in which these precedence changes result in different parsing choices.
Change pg_ctl's default shutdown mode from
This means the default behavior will be to forcibly cancel existing database sessions, not simply wait for them to exit.
Use assignment cast behavior for data type conversions in PL/pgSQL assignments, rather than converting to and from text (Tom Lane)
This change causes conversions of Booleans to strings to produce
f. Other type conversions may succeed in more cases than before; for example, assigning a numeric value
3.9to an integer variable will now assign 4 rather than failing. If no assignment-grade cast is defined for the particular source and destination types, PL/pgSQL will fall back to its old I/O conversion behavior.
Allow characters in server command-line options to be escaped with a backslash (Andres Freund)
Formerly, spaces in the options string always separated options, so there was no way to include a space in an option value. Including a backslash in an option value now requires writing
Change the default value of the GSSAPI
include_realmparameter to 1, so that by default the realm is not removed from a GSS or SSPI principal name (Stephen Frost)
If you previously adjusted
checkpoint_segments, the following formula will give you an approximately equivalent setting:
max_wal_size = (3 * checkpoint_segments) * 16MB
Note that the default setting for
max_wal_sizeis much higher than the default
checkpoint_segmentsused to be, so adjusting it might no longer be necessary.
Decommission server configuration parameter
ssl_renegotiation_limit, which was deprecated in earlier releases (Andres Freund)
While SSL renegotiation is a good idea in theory, it has caused enough bugs to be considered a net negative in practice, and it is due to be removed from future versions of the relevant standards. We have therefore removed support for it from PostgreSQL. The
ssl_renegotiation_limitparameter still exists, but cannot be set to anything but zero (disabled). It's not documented anymore, either.
Remove server configuration parameter
autocommit, which was already deprecated and non-operational (Tom Lane)
rolcatupdatefield, as it had no usefulness (Adam Brightwell)
sentfield is now NULL, not zero, when it has no valid value (Magnus Hagander)
jsonbarray extraction operators to accept negative subscripts, which count from the end of JSON arrays (Peter Geoghegan, Andrew Dunstan)
Previously, these operators returned
NULLfor negative subscripts.
Below you will find a detailed account of the changes between PostgreSQL 9.5 and the previous major release.
Add Block Range Indexes (BRIN) (Álvaro Herrera)
BRIN indexes store only summary data (such as minimum and maximum values) for ranges of heap blocks. They are therefore very compact and cheap to update; but if the data is naturally clustered, they can still provide substantial speedup of searches.
Allow queries to perform accurate distance filtering of bounding-box-indexed objects (polygons, circles) using GiST indexes (Alexander Korotkov, Heikki Linnakangas)
Previously, to exploit such an index a subquery had to be used to select a large number of rows ordered by bounding-box distance, and the result then had to be filtered further with a more accurate distance calculation.
Allow GiST indexes to perform index-only scans (Anastasia Lubennikova, Heikki Linnakangas, Andreas Karlsson)
Add configuration parameter gin_pending_list_limit to control the size of GIN pending lists (Fujii Masao)
This value can also be set on a per-index basis as an index storage parameter. Previously the pending-list size was controlled by work_mem, which was awkward because appropriate values for
work_memare often much too large for this purpose.
Issue a warning during the creation of hash indexes because they are not crash-safe (Bruce Momjian)
E.126.96.36.199. General Performance
Improve the speed of sorting of
numericfields via “abbreviated” keys (Peter Geoghegan, Andrew Gierth, Robert Haas)
Extend the infrastructure that allows sorting to be performed by inlined, non-SQL-callable comparison functions to cover
Improve performance of hash joins (Tomas Vondra, Robert Haas)
Improve concurrency of shared buffer replacement (Robert Haas, Amit Kapila, Andres Freund)
Reduce the number of page locks and pins during index scans (Kevin Grittner)
The primary benefit of this is to allow index vacuums to be blocked less often.
Make per-backend tracking of buffer pins more memory-efficient (Andres Freund)
Improve lock scalability (Andres Freund)
This particularly addresses scalability problems when running on systems with multiple CPU sockets.
Allow the optimizer to remove unnecessary references to left-joined subqueries (David Rowley)
Allow pushdown of query restrictions into subqueries with window functions, where appropriate (David Rowley)
Allow a non-leakproof function to be pushed down into a security barrier view if the function does not receive any view output columns (Dean Rasheed)
Teach the planner to use statistics obtained from an expression index on a boolean-returning function, when a matching function call appears in
ANALYZEcompute basic statistics (null fraction and average column width) even for columns whose data type lacks an equality function (Oleksandr Shulgin)
Speed up CRC (cyclic redundancy check) computations and switch to CRC-32C (Abhijit Menon-Sen, Heikki Linnakangas)
Improve bitmap index scan performance (Teodor Sigaev, Tom Lane)
CREATE INDEXby avoiding unnecessary memory copies (Robert Haas)
Increase the number of buffer mapping partitions (Amit Kapila, Andres Freund, Robert Haas)
This improves performance for highly concurrent workloads.
Add per-table autovacuum logging control via new
log_autovacuum_min_durationstorage parameter (Michael Paquier)
Add new configuration parameter cluster_name (Thomas Munro)
This string, typically set in
postgresql.conf, allows clients to identify the cluster. This name also appears in the process title of all server processes, allowing for easier identification of processes belonging to the same cluster.
Prevent non-superusers from changing log_disconnections on connection startup (Fujii Masao)
Check “Subject Alternative Names” in SSL server certificates, if present (Alexey Klyukin)
When they are present, this replaces checks against the certificate's “Common Name”.
Add system view
pg_stat_sslto report SSL connection information (Magnus Hagander)
Add libpq functions to return SSL information in an implementation-independent way (Heikki Linnakangas)
PQgetssl()can still be used to call OpenSSL functions, it is now considered deprecated because future versions of libpq might support other SSL implementations. When possible, use the new functions
PQsslInUse()to obtain SSL information in an SSL-implementation-independent way.
Make libpq honor any OpenSSL thread callbacks (Jan Urbanski)
Previously they were overwritten.
E.188.8.131.52. Server Settings
This change allows the allocation of a large number of WAL files without keeping them after they are no longer needed. Therefore the default for
max_wal_sizehas been set to
1GB, much larger than the old default for
checkpoint_segments. Also note that standby servers perform restartpoints to try to limit their WAL space consumption to
max_wal_size; previously they did not pay any attention to
The previous OOM control infrastructure involved compile-time options
LINUX_OOM_ADJ, which are no longer supported. The new behavior is available in all builds.
Allow recording of transaction commit time stamps when configuration parameter track_commit_timestamp is enabled (Álvaro Herrera, Petr Jelínek)
Time stamp information can be accessed using functions
Allow local_preload_libraries to be set by
ALTER ROLE SET(Peter Eisentraut, Kyotaro Horiguchi)
Allow autovacuum workers to respond to configuration parameter changes during a run (Michael Paquier)
Make configuration parameter debug_assertions read-only (Andres Freund)
This means that assertions can no longer be turned off if they were enabled at compile time, allowing for more efficient code optimization. This change also removes the postgres
Allow setting effective_io_concurrency on systems where it has no effect (Peter Eisentraut)
Add system view
pg_file_settingsto show the contents of the server's configuration files (Sawada Masahiko)
pending_restartto the system view
pg_settingsto indicate a change has been made but will not take effect until a database restart (Peter Eisentraut)
ALTER SYSTEMvalues to be reset with
ALTER SYSTEM RESET(Vik Fearing)
This command removes the specified setting from
E.26.3.2. Replication and Recovery
Create mechanisms for tracking the progress of replication, including methods for identifying the origin of individual changes during logical replication (Andres Freund)
This is helpful when implementing replication solutions.
Rework truncation of the multixact commit log to be properly WAL-logged (Andres Freund)
This makes things substantially simpler and more robust.
This replaces the old parameter
Add new archive_mode value
alwaysto allow standbys to always archive received WAL files (Fujii Masao)
Add configuration parameter wal_retrieve_retry_interval to control WAL read retry after failure (Alexey Vasiliev, Michael Paquier)
This is particularly helpful for warm standbys.
Allow compression of full-page images stored in WAL (Rahila Syed, Michael Paquier)
This feature reduces WAL volume, at the cost of more CPU time spent on WAL logging and WAL replay. It is controlled by a new configuration parameter wal_compression, which currently is off by default.
Archive WAL files with suffix
.partialduring standby promotion (Heikki Linnakangas)
Add configuration parameter log_replication_commands to log replication commands (Fujii Masao)
Report the processes holding replication slots in
The new output column is
primary_conninfosetting to use connection URIs, e.g.,
INSERTs that would generate constraint conflicts to be turned into
UPDATEs or ignored (Peter Geoghegan, Heikki Linnakangas, Andres Freund)
The syntax is
INSERT ... ON CONFLICT DO NOTHING/UPDATE. This is the Postgres implementation of the popular
Allow setting multiple target columns in an
UPDATEfrom the result of a single sub-SELECT (Tom Lane)
This is accomplished using the syntax
UPDATE tab SET (col1, col2, ...) = (SELECT ...).
SKIP LOCKEDto skip locked rows (Thomas Munro)
This does not throw an error for locked rows like
TABLESAMPLEto return a subset of a table (Petr Jelínek)
This feature supports the SQL-standard table sampling methods. In addition, there are provisions for user-defined table sampling methods.
Suggest possible matches for mistyped column names (Peter Geoghegan, Robert Haas)
E.26.3.4. Utility Commands
Add more details about sort ordering in
EXPLAINoutput (Marius Timmer, Lukas Kreft, Arne Scheffer)
VACUUMlog the number of pages skipped due to pins (Jim Nasby)
TRUNCATEproperly update the
pg_stat* tuple counters (Alexander Shulgin)
REINDEXto reindex an entire schema using the
SCHEMAoption (Sawada Masahiko)
SCHEMAfrom outputting object names, unless
VERBOSEis used (Simon Riggs)
E.26.3.5. Object Manipulation
Add row-level security control (Craig Ringer, KaiGai Kohei, Adam Brightwell, Dean Rasheed, Stephen Frost)
This feature allows row-by-row control over which users can add, modify, or even see rows in a table. This is controlled by new commands
ALTER TABLE ... ENABLE/DISABLE ROW SECURITY.
Allow changing of the WAL logging status of a table after creation with
ALTER TABLE ... SET LOGGED / UNLOGGED(Fabrízio de Royes Mello)
Add support for
ALTER TABLE ... RENAME CONSTRAINT(Bruce Momjian)
Allow some DDL commands to accept
SESSION_USER, meaning the current user or session user, in place of a specific user name (Kyotaro Horiguchi, Álvaro Herrera)
Support comments on domain constraints (Álvaro Herrera)
Reduce lock levels of some create and alter trigger and foreign key commands (Simon Riggs, Andreas Karlsson)
LOCK TABLE ... ROW EXCLUSIVE MODEfor those with
INSERTprivileges on the target table (Stephen Frost)
Previously this command required
Apply table and domain
CHECKconstraints in order by name (Tom Lane)
The previous ordering was indeterminate.
This allows these per-database settings to be changed without manually modifying the
Add support for IMPORT FOREIGN SCHEMA (Ronan Dunklau, Michael Paquier, Tom Lane)
This command allows automatic creation of local foreign tables that match the structure of existing tables on a remote server.
CHECKconstraints to be placed on foreign tables (Shigeru Hanada, Etsuro Fujita)
Such constraints are assumed to be enforced on the remote server, and are not enforced locally. However, they are assumed to hold for purposes of query optimization, such as constraint exclusion.
Allow foreign tables to participate in inheritance (Shigeru Hanada, Etsuro Fujita)
To let this work naturally, foreign tables are now allowed to have check constraints marked as not valid, and to set storage and
OIDcharacteristics, even though these operations are effectively no-ops for a foreign table.
Allow foreign data wrappers and custom scans to implement join pushdown (KaiGai Kohei)
ddl_command_endevent trigger is installed, capture details of DDL activity for it to inspect (Álvaro Herrera)
This information is available through a set-returning function
pg_event_trigger_ddl_commands(), or by inspection of C data structures if that function doesn't provide enough detail.
Allow event triggers on table rewrites caused by
ALTER TABLE(Dimitri Fontaine)
Add columns to the output of
This allows simpler processing of delete operations.
E.26.3.6. Data Types
xmldata type to accept empty or all-whitespace content values (Peter Eisentraut)
This is required by the SQL/XML specification.
macaddrinput using the format
Disallow non-SQL-standard syntax for
intervalwith both precision and field specifications (Bruce Momjian)
Per the standard, such type specifications should be written as, for example,
INTERVAL MINUTE TO SECOND(2). PostgreSQL formerly allowed this to be written as
INTERVAL(2) MINUTE TO SECOND, but it must now be written in the standard way.
Equivalent functions already existed for type
integervalues to be subtracted from
jsonbdocuments (Dmitry Dolgov, Andrew Dunstan)
||operator (Dmitry Dolgov, Andrew Dunstan)
numericvalues (Plato Malugin)
ARRAY()to take arrays as inputs (Ali Akbar, Tom Lane)
polygondistance operator <-> (Alexander Korotkov)
Previously, only a single-byte character was allowed as an escape.
width_bucket()variant that supports any sortable data type and non-uniform bucket widths (Petr Jelínek)
Add an optional
pg_read_file()and related functions (Michael Paquier, Heikki Linnakangas)
=>to specify named parameters in function calls (Pavel Stehule)
:=could be used. This requires removing the possibility for
=>to be a user-defined operator. Creation of user-defined
=>operators has been issuing warnings since PostgreSQL 9.0.
Add POSIX-compliant rounding for platforms that use PostgreSQL-supplied rounding functions (Pedro Gimeno Fortea)
E.184.108.40.206. System Information Functions and Views
Previously, only the specific role owning the target session could perform these operations; now membership in that role is sufficient.
pg_stat_get_snapshot_timestamp()to output the time stamp of the statistics snapshot (Matt Kelly)
This represents the last time the snapshot file was written to the file system.
mxid_age()to compute multi-xid age (Bruce Momjian)
E.26.3.8. Server-Side Languages
Improve support for composite types in PL/Python (Ed Behn, Ronan Dunklau)
This allows PL/Python functions to return arrays of composite types.
Reduce lossiness of PL/Python floating-point value conversions (Marko Kreen)
Allow specification of conversion routines between SQL data types and data types of procedural languages (Peter Eisentraut)
E.26.3.9. Client Applications
This should result in these programs being installed by default in most installations.
Add pg_rewind, which allows re-synchronizing a master server after failback (Heikki Linnakangas)
Allow pg_receivexlog to manage physical replication slots (Michael Paquier)
This is controlled via new
Allow pg_receivexlog to synchronously flush WAL to storage using new
--synchronousoption (Furuya Osamu, Fujii Masao)
Without this, WAL files are fsync'ed only on close.
Allow vacuumdb to vacuum in parallel using new
--jobsoption (Dilip Kumar)
In vacuumdb, do not prompt for the same password repeatedly when multiple connections are necessary (Haribabu Kommi, Michael Paquier)
--verboseoption to reindexdb (Sawada Masahiko)
Make pg_basebackup use a tablespace mapping file when using tar format, to support symbolic links and file paths of 100+ characters in length on MS Windows (Amit Kapila)
Add pg_xlogdump option
--statsto display summary statistics (Abhijit Menon-Sen)
Allow psql to produce AsciiDoc output (Szymon Guz)
errorsmode that displays only failed commands to psql's
ECHOvariable (Pavel Stehule)
This behavior can also be selected with psql's
Provide separate column, header, and border linestyle control in psql's unicode linestyle (Pavel Stehule)
Single or double lines are supported; the default is
Add new option
PROMPTvariables to display the current multiline statement line number (Sawada Masahiko)
pager_min_linesto control pager invocation (Andrew Dunstan)
Improve psql line counting used when deciding to invoke the pager (Andrew Dunstan)
psql now fails if the file specified by an
--log-fileswitch cannot be written (Tom Lane, Daniel Vérité)
Previously, it effectively ignored the switch in such cases.
Add psql tab completion when setting the search_path variable (Jeff Janes)
Currently only the first schema can be tab-completed.
Improve psql's tab completion for triggers and rules (Andreas Karlsson)
\? variablesshows psql's special variables and
\? optionsshows the command-line options.
\? commandsshows the meta-commands, which is the traditional output and remains the default. These help displays can also be obtained with the command-line option
Show tablespace size in psql's
\db+(Fabrízio de Royes Mello)
Show data type owners in psql's
\timinginformation (Fujii Masao)
\watchqueries, since that is generally unwanted.
Improve psql tab completion for
:variablenames (Pavel Stehule)
Allow tab completion of role names in psql
\ccommands (Ian Barwick)
Allow pg_dump to share a snapshot taken by another session using
--snapshot(Simon Riggs, Michael Paquier)
The remote snapshot must have been exported by
pg_export_snapshot()or logical replication slot creation. This can be used to share a consistent snapshot across multiple pg_dump processes.
Support table sizes exceeding 8GB in tar archive format (Tom Lane)
The POSIX standard for tar format does not allow elements of a tar archive to exceed 8GB, but most modern implementations of tar support an extension that does allow it. Use the extension format when necessary, rather than failing.
Make pg_dump always print the server and pg_dump versions (Jing Wang)
Previously, version information was only printed in
Remove the long-ignored
--ignore-versionoption from pg_dump, pg_dumpall, and pg_restore (Fujii Masao)
Support multiple pg_ctl
-ooptions, concatenating their values (Bruce Momjian)
Allow control of pg_ctl's event source logging on MS Windows (MauMau)
This only controls pg_ctl, not the server, which has separate settings in
If the server's listen address is set to a wildcard value (
0.0.0.0in IPv4 or
::in IPv6), connect via the loopback address rather than trying to use the wildcard address literally (Kondo Yuta)
This fix primarily affects Windows, since on other platforms pg_ctl will prefer to use a Unix-domain socket.
Move pg_upgrade from
In connection with this change, the functionality previously provided by the pg_upgrade_support module has been moved into the core server.
Support multiple pg_upgrade
-Ooptions, concatenating their values (Bruce Momjian)
Improve database collation comparisons in pg_upgrade (Heikki Linnakangas)
Remove support for upgrading from 8.3 clusters (Bruce Momjian)
Move pgbench from
Fix calculation of TPS number “excluding connections establishing” (Tatsuo Ishii, Fabien Coelho)
The overhead for connection establishment was miscalculated whenever the number of pgbench threads was less than the number of client connections. Although this is clearly a bug, we won't back-patch it into pre-9.5 branches since it makes TPS numbers not comparable to previous results.
Allow counting of pgbench transactions that take over a specified amount of time (Fabien Coelho)
This is controlled by a new
Allow pgbench to generate Gaussian/exponential distributions using
\setrandom(Kondo Mitsumasa, Fabien Coelho)
\setcommand to handle arithmetic expressions containing more than one operator, and add
%(modulo) to the set of operators it supports (Robert Haas, Fabien Coelho)
E.26.3.10. Source Code
Simplify WAL record format (Heikki Linnakangas)
This allows external tools to more easily track what blocks are modified.
Improve the representation of transaction commit and abort WAL records (Andres Freund)
Add atomic memory operations API (Andres Freund)
Allow custom path and scan methods (KaiGai Kohei, Tom Lane)
This allows extensions greater control over the optimizer and executor.
Allow foreign data wrappers to do post-filter locking (Etsuro Fujita)
Foreign tables can now take part in
INSERT ... ON CONFLICT DO NOTHINGqueries (Peter Geoghegan, Heikki Linnakangas, Andres Freund)
Foreign data wrappers must be modified to handle this.
INSERT ... ON CONFLICT DO UPDATEis not supported on foreign tables.
hash_create()'s API for selecting simple-binary-key hash functions (Teodor Sigaev, Tom Lane)
Improve parallel execution infrastructure (Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke)
Remove Alpha (CPU) and Tru64 (OS) ports (Andres Freund)
Remove swap-byte-based spinlock implementation for ARMv5 and earlier CPUs (Robert Haas)
ARMv5's weak memory ordering made this locking implementation unsafe. Spinlock support is still possible on newer gcc implementations with atomics support.
Generate an error when excessively long (100+ character) file paths are written to tar files (Peter Eisentraut)
Tar does not support such overly-long paths.
This avoids possible problems with these indexes when different databases of a cluster have different default collations.
Change the spinlock primitives to function as compiler barriers (Robert Haas)
E.220.127.116.11. MS Windows
Allow higher-precision time stamp resolution on Windows 8, Windows Server 2012, and later Windows systems (Craig Ringer)
Install shared libraries to
binin MS Windows (Peter Eisentraut, Michael Paquier)
contribon MSVC builds (Michael Paquier)
--with-extra-versionoption to be honored by the MSVC build (Michael Paquier)
PGFILEDESCinto MSVC contrib builds (Michael Paquier)
Add icons to all MSVC-built binaries and version information to all MS Windows binaries (Noah Misch)
MinGW already had such icons.
Add optional-argument support to the internal
getopt_long()implementation (Michael Paquier, Andres Freund)
This is used by the MSVC build.
E.26.3.11. Additional Modules
Add statistics for minimum, maximum, mean, and standard deviation times to pg_stat_statements (Mitsumasa Kondo, Andrew Dunstan)
Add pgcrypto function
pgp_armor_headers()to extract PGP armor headers (Marko Tiikkaja, Heikki Linnakangas)
Allow empty replacement strings in unaccent (Mohammad Alhashash)
This is useful in languages where diacritic signs are represented as separate characters.
Allow multicharacter source strings in unaccent (Tom Lane)
This could be useful in languages where diacritic signs are represented as separate characters. It also allows more complex unaccent dictionaries.
Add information about buffer pins to pg_buffercache display (Andres Freund)
Allow pgstattuple to report approximate answers with less overhead using
Move dummy_seclabel, test_shm_mq, test_parser, and worker_spi from
These modules are only meant for server testing, so they do not need to be built or installed when packaging PostgreSQL.