- linkend="guc-wal-writer-flush-after"> control this behavior.
+ Formerly, anti-wraparound vacuum had to visit every page of
+ a table, even pages where there was nothing to do. Now, pages
+ containing only already-frozen tuples are identified in the table's
+ visibility map, and can be skipped by vacuum even when doing
+ transaction wraparound prevention. This should greatly reduce the
+ cost of maintaining large tables containing mostly-unchanged data.
+
+
+ If necessary, vacuum can be forced to process all-frozen
+ pages using the new DISABLE_PAGE_SKIPPING option.
+ Normally, this should never be needed but it might help in
+ recovering from visibility-map corruption.
- Perform checkpoint writes in sorted order (Fabien Coelho,
- Andres Freund)
+ Avoid useless heap-truncation attempts during VACUUM
+ (Jeff Janes, Tom Lane)
- Previously, checkpoints wrote out dirty pages in whatever order
- they happen to appear in shared buffers, which usually is nearly
- random. That performs poorly, especially on rotating media.
- This change causes checkpoint-driven writes to be done in order
- by file and block number, and to be balanced across tablespaces.
+ This change avoids taking an exclusive table lock in some cases
+ where no truncation is possible. The main benefit comes from
+ avoiding unnecessary query cancellations on standby servers.
+
+
+
+
+
+
General Performance
+
+
+
+ Avoid computing GROUP BY columns if they are
+ functionally dependent on other columns (David Rowley)
+
+
+ If a GROUP BY clause includes all columns of a
+ non-deferred primary key, as well as other columns of the same
+ table, those other columns are redundant and can be dropped
+ from the grouping. This saves computation in many common cases.
+
+
+
+
+
+ When appropriate, postpone evaluation of SELECT
+ output expressions until after an ORDER BY sort
+ (Konstantin Knizhnik)
+
+
+ This change ensures that volatile or expensive functions in the
+ output list are executed in the order suggested by ORDER
+ BY, and that they are not evaluated more times than required
+ when there is a LIMIT clause. Previously, these
+ properties held if the ordering was performed by an index scan or
+ pre-merge-join sort, but not if it was performed by a top-level
+ sort.
+
+
+
+
+
- Use foreign key relationships to infer selectivity for join
- predicates (Tomas Vondra, David Rowley)
+ Perform checkpoint writes in sorted order (Fabien Coelho,
+ Andres Freund)
- If a table t has a foreign key restriction, say
- (a,b) REFERENCES r (x,y), then a WHERE
- condition such as t.a = r.x AND t.b = r.y cannot
- select more than one r row per t row.
- The planner formerly considered AND conditions
- to be independent and would often drastically misestimate
- selectivity as a result. Now it compares the WHERE
- conditions to applicable foreign key constraints and produces
- better estimates.
+ Previously, checkpoints wrote out dirty pages in whatever order
+ they happen to appear in shared buffers, which usually is nearly
+ random. That performs poorly, especially on rotating media.
+ This change causes checkpoint-driven writes to be done in order
+ by file and block number, and to be balanced across tablespaces.
+
+
+
+
+
+ Where feasible, trigger kernel writeback after a configurable
+ number of writes, to prevent accumulation of dirty data in kernel
+ disk buffers (Fabien Coelho, Andres Freund)
+
+
+
PostgreSQL writes data to the kernel's disk cache,+ from where it will be flushed to physical storage in due time.
+ Many operating systems are not smart about managing this and allow
+ large amounts of dirty data to accumulate before deciding to flush
+ it all at once, leading to long delays for new I/O requests.
+ This change attempts to alleviate this problem by explicitly
+ requesting data flushes after a configurable interval.
+
+
+ On Linux, sync_file_range() is used for this purpose,
+ and the feature is on by default on Linux because that function has few
+ downsides. This sync capability is also available on other platforms
+ that have msync() or posix_fadvise(),
+ but those interfaces have some undesirable side-effects so the
+ feature is disabled by default on non-Linux platforms.
+
+
+ The new configuration parameters
+ linkend="guc-backend-flush-after">,
+ linkend="guc-bgwriter-flush-after">,
+ linkend="guc-checkpoint-flush-after">, and
+ linkend="guc-wal-writer-flush-after"> control this behavior.
- Improve ANALYZE's estimates for columns with many nulls
- (Tomas Vondra, Alex Shulgin)
-
-
- Previously ANALYZE tended to underestimate the number
- of non-NULL distinct values in a column with many
- NULLs, and was also inaccurate in computing the
- most-common values.
-
-
-
-
-
- Improve planner's estimate of the number of distinct values in
- a query result (Tomas Vondra)
-
-
-
-
-
- Improve sorting performance by using quicksort, not replacement
- selection sort, when performing external sort steps (Peter
- Geoghegan)
-
-
- The new approach makes better use of the
CPU cache- for typical cache sizes and data volumes. Where necessary,
- the behavior can be adjusted via the new configuration parameter
- .
-
-
-
-
-
- Speed up text sorts where the same string occurs multiple times
- (Peter Geoghegan)
-
-
-
-
-
- Speed up sorting of uuid, bytea, and
- char(n) fields by using abbreviated keys
- (Peter Geoghegan)
-
-
- Support for abbreviated keys has also been
- added to the non-default operator classes
- linkend="indexes-opclass">text_pattern_ops,
- varchar_pattern_ops, and
- bpchar_pattern_ops. Processing of ordered-set
- aggregates can also now exploit abbreviated keys.
-
-
-
-
-
- Speed up CREATE INDEX CONCURRENTLY by treating
-
TIDs as 64-bit integers during sorting (Peter- Geoghegan)
-
-
-
-
-
- Reduce contention for the ProcArrayLock (Amit Kapila,
- Robert Haas)
-
-
-
-
-
- Improve performance by moving buffer content locks into the buffer
- descriptors (Andres Freund, Simon Riggs)
-
-
-
-
-
- Replace shared-buffer header spinlocks with atomic operations to
- improve scalability (Alexander Korotkov, Andres Freund)
-
-
-
-
-
- Use atomic operations, rather than a spinlock, to protect an
- LWLock's wait queue (Andres Freund)
-
-
-
-
-
- Partition the shared hash table freelist to reduce contention on
- multi-
CPU-socket servers (Aleksander Alekseev)-
-
-
-
-
+ Add
+ linkend="functions-controldata">pg_control_system(),
+ pg_control_checkpoint(),
+ pg_control_recovery(), and
+ pg_control_init() functions to expose fields of
+
pg_control to SQL (Joe Conway, Michael+ Paquier)
+
+
+
+
+
- Add
- linkend="functions-controldata">pg_control_system(),
- pg_control_checkpoint(),
- pg_control_recovery(), and
- pg_control_init() functions to expose fields of
-
pg_control to SQL (Joe Conway, Michael- Paquier)
-
-
-
-
-
- Add configure option
- calling sd_notify() at server start and stop (Peter
- Eisentraut)
+ Allow sessions to be terminated automatically if they are in
+ idle-in-transaction state for too long (Vik Fearing)
- This allows the use of
systemd service units of- type notify, which greatly simplifies the management
- of
PostgreSQL under systemd.+ This behavior is controlled by the new configuration parameter
+ . It can
+ be useful to prevent forgotten transactions from holding locks
+ or preventing vacuum cleanup for too long.
- Allow sessions to be terminated automatically if they are in
- idle-in-transaction state for too long (Vik Fearing)
+ Add configure option
+ calling sd_notify() at server start and stop (Peter
+ Eisentraut)
- This behavior is controlled by the new configuration parameter
- . It can
- be useful to prevent forgotten transactions from holding locks
- or preventing vacuum cleanup for too long.
+ This allows the use of
systemd service units of+ type notify, which greatly simplifies the management
+ of
PostgreSQL under systemd.
+ Improve full-text search to support searching for phrases, that
+ is, lexemes appearing adjacent to each other in a specific order,
+ or with a specified distance between them (Teodor Sigaev, Oleg
+ Bartunov, Dmitry Ivanov)
+
+
+ A phrase-search query can be specified in tsquery
+ input using the new operators <-> and
+ <N>. The former means
+ that the lexemes before and after it must appear adjacent to
+ each other in that order. The latter means they must be exactly
+ N lexemes apart.
+
+
+
+
+
- Improve full-text search to support searching for phrases, that
- is, lexemes appearing adjacent to each other in a specific order,
- or with a specified distance between them (Teodor Sigaev, Oleg
- Bartunov, Dmitry Ivanov)
-
-
- A phrase-search query can be specified in tsquery
- input using the new operators <-> and
- <N>. The former means
- that the lexemes before and after it must appear adjacent to
- each other in that order. The latter means they must be exactly
- N lexemes apart.
-
-
-
-
-
projects / postgresql.git / commitdiff