While beneficial, both for throughput and average/worst case latency, in
a significant number of workloads, there are other workloads in which
backend_flush_after can cause significant performance regressions in
comparison to < 9.6 releases. The regression is most likely when the hot
data set is bigger than shared buffers, but significantly smaller than
the operating system's page cache.
I personally think that the benefit of enabling backend flush control is
considerably bigger than the potential downsides, but a fair argument
can be made that not regressing is more important than improving
performance/latency. As the latter is the consensus, change the default
to 0.
The other settings introduced in 428b1d6b2 do not have the same
potential for regressions, so leave them enabled.
Benchmarks leading up to changing the default have been performed by
Mithun Cy, Ashutosh Sharma and Robert Haas.
Discussion: CAD__OuhPmc6XH=wYRm_+Q657yQE88DakN4=Ybh2oveFasHkoeA@mail.gmail.com
This terminology provoked widespread complaints. So, instead, rename
the GUC max_parallel_degree to max_parallel_workers_per_gather
(leaving room for a possible future GUC max_parallel_workers that acts
as a system-wide limit), and rename the parallel_degree reloption to
parallel_workers. Rename structure members to match.
These changes create a dump/restore hazard for users of PostgreSQL
9.6beta1 who have set the reloption (or applied the GUC using ALTER
USER or ALTER DATABASE).
Mostly these are just comments but there are a few in documentation
and a handful in code and tests. Hopefully this doesn't cause too much
unnecessary pain for backpatching. I relented from some of the most
common like "thru" for that reason. The rest don't seem numerous
enough to cause problems.
Thanks to Kevin Lyda's tool https://pypi.python.org/pypi/misspellings
We didn't have any real user documentation about how index-only scans
work or how to design indexes to exploit them. Remedy that.
Per gripe from David Johnston.
Hash indexes are not WAL-logged, and so do not maintain the LSN of
index pages. Since the "snapshot too old" feature counts on
detecting error conditions using the LSN of a table and all indexes
on it, this makes it impossible to safely do early vacuuming on any
table with a hash index, so add this to the tests for whether the
xid used to vacuum a table can be adjusted based on
old_snapshot_threshold.
While at it, add a paragraph to the docs for old_snapshot_threshold
which specifically mentions this and other aspects of the feature
which may otherwise surprise users.
Problem reported and patch reviewed by Amit Kapila
If we're not going to reject such setups entirely, throwing a WARNING in
check_synchronous_standby_names() is unhelpful, because it will cause the
warning to be logged again every time the postmaster receives SIGHUP.
Per discussion, just remove the warning.
In passing, improve the documentation for synchronous_commit, which had not
gotten the word that now there can be more than one synchronous standby.
Commit 989be0810d added a flex/bison lexer/parser to interpret
synchronous_standby_names. It was done in a pretty crufty way, though,
making assorted end-use sites responsible for calling the parser at the
right times. That was not only vulnerable to errors of omission, but made
it possible that lexer/parser errors occur at very undesirable times,
and created memory leakages even if there was no error.
Instead, perform the parsing once during check_synchronous_standby_names
and let guc.c manage the resulting data. To do that, we have to flatten
the parsed representation into a single hunk of malloc'd memory, but that
is not very hard.
While at it, work a little harder on making useful error reports for
parsing problems; the previous code felt that "synchronous_standby_names
parser returned 1" was an appropriate user-facing error message. (To
be fair, it did also log a syntax error message, but separately from the
GUC problem report, which is at best confusing.) It had some outright
bugs in the face of invalid input, too.
I (tgl) also concluded that we need to restrict unquoted names in
synchronous_standby_names to be just SQL identifiers. The previous coding
would accept darn near anything, which (1) makes the quoting convention
both nearly-unnecessary and formally ambiguous, (2) makes it very hard to
understand what is a syntax error and what is a creative interpretation of
the input as a standby name, and (3) makes it impossible to further extend
the syntax in future without a compatibility break. I presume that we're
intending future extensions of the syntax, else this parsing infrastructure
is massive overkill, so (3) is an important objection. Since we've taken
a compatibility hit for non-identifier names with this change anyway, we
might as well lock things down now and insist that users use double quotes
for standby names that aren't identifiers.
Kyotaro Horiguchi and Tom Lane
Change max_parallel_degree default from 0 to 2. It is possible that
this is not a good idea, or that we should go with 1 worker rather
than 2, but we won't find out without trying it. Along the way,
reword the documentation for max_parallel_degree a little bit to
hopefully make it more clear.
Discussion: 20160420174631.3qjjhpwsvvx5bau5@alap3.anarazel.de
Several issues:
1) checkpoint_flush_after doc and code disagreed about the default
2) new GUCs were missing from postgresql.conf.sample
3) Outdated source-code comment about bgwriter_flush_after's default
4) Sub-optimal categories assigned to new GUCs
5) Docs suggested backend_flush_after is PGC_SIGHUP, but it's PGC_USERSET.
6) Spell out int as integer in the docs, as done elsewhere
Reported-By: Magnus Hagander, Fujii Masao
Discussion: CAHGQGwETyTG5VYQQ5C_srwxWX7RXvFcD3dKROhvAWWhoSBdmZw@mail.gmail.com
This feature is controlled by a new old_snapshot_threshold GUC. A
value of -1 disables the feature, and that is the default. The
value of 0 is just intended for testing. Above that it is the
number of minutes a snapshot can reach before pruning and vacuum
are allowed to remove dead tuples which the snapshot would
otherwise protect. The xmin associated with a transaction ID does
still protect dead tuples. A connection which is using an "old"
snapshot does not get an error unless it accesses a page modified
recently enough that it might not be able to produce accurate
results.
This is similar to the Oracle feature, and we use the same SQLSTATE
and error message for compatibility.
We still use replacement selection for the first run of the sort only
and only when the number of tuples is relatively small. Otherwise,
the first run, and subsequent runs in all cases, are produced using
quicksort. This tends to be faster except perhaps for very small
amounts of working memory.
Peter Geoghegan, reviewed by Tomas Vondra, Jeff Janes, Mithun Cy,
Greg Stark, and me.
Previously synchronous replication offered only the ability to confirm
that all changes made by a transaction had been transferred to at most
one synchronous standby server.
This commit extends synchronous replication so that it supports multiple
synchronous standby servers. It enables users to consider one or more
standby servers as synchronous, and increase the level of transaction
durability by ensuring that transaction commits wait for replies from
all of those synchronous standbys.
Multiple synchronous standby servers are configured in
synchronous_standby_names which is extended to support new syntax of
'num_sync ( standby_name [ , ... ] )', where num_sync specifies
the number of synchronous standbys that transaction commits need to
wait for replies from and standby_name is the name of a standby
server.
The syntax of 'standby_name [ , ... ]' which was used in 9.5 or before
is also still supported. It's the same as new syntax with num_sync=1.
This commit doesn't include "quorum commit" feature which was discussed
in pgsql-hackers. Synchronous standbys are chosen based on their priorities.
synchronous_standby_names determines the priority of each standby for
being chosen as a synchronous standby. The standbys whose names appear
earlier in the list are given higher priority and will be considered as
synchronous. Other standby servers appearing later in this list
represent potential synchronous standbys.
The regression test for multiple synchronous standbys is not included
in this commit. It should come later.
Authors: Sawada Masahiko, Beena Emerson, Michael Paquier, Fujii Masao
Reviewed-By: Kyotaro Horiguchi, Amit Kapila, Robert Haas, Simon Riggs,
Amit Langote, Thomas Munro, Sameer Thakur, Suraj Kharage, Abhijit Menon-Sen,
Rajeev Rastogi
Many thanks to the various individuals who were involved in
discussing and developing this feature.
In this mode, the master waits for the transaction to be applied on
the remote side, not just written to disk. That means that you can
count on a transaction started on the standby to see all commits
previously acknowledged by the master.
To make this work, the standby sends a reply after replaying each
commit record generated with synchronous_commit >= 'remote_apply'.
This introduces a small inefficiency: the extra replies will be sent
even by standbys that aren't the current synchronous standby. But
previously-existing synchronous_commit levels make no attempt at all
to optimize which replies are sent based on what the primary cares
about, so this is no worse, and at least avoids any extra replies for
people not using the feature at all.
Thomas Munro, reviewed by Michael Paquier and by me. Some additional
tweaks by me.
The distinction between "archive" and "hot_standby" existed only because
at the time "hot_standby" was added, there was some uncertainty about
stability. This is now a long time ago. We would like to move forward
with simplifying the replication configuration, but this distinction is
in the way, because a primary server cannot tell (without asking a
standby or predicting the future) which one of these would be the
appropriate level.
Pick a new name for the combined setting to make it clearer that it
covers all (non-logical) backup and replication uses. The old values
are still accepted but are converted internally.
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Reviewed-by: David Steele <david@pgmasters.net>
Currently writes to the main data files of postgres all go through the
OS page cache. This means that some operating systems can end up
collecting a large number of dirty buffers in their respective page
caches. When these dirty buffers are flushed to storage rapidly, be it
because of fsync(), timeouts, or dirty ratios, latency for other reads
and writes can increase massively. This is the primary reason for
regular massive stalls observed in real world scenarios and artificial
benchmarks; on rotating disks stalls on the order of hundreds of seconds
have been observed.
On linux it is possible to control this by reducing the global dirty
limits significantly, reducing the above problem. But global
configuration is rather problematic because it'll affect other
applications; also PostgreSQL itself doesn't always generally want this
behavior, e.g. for temporary files it's undesirable.
Several operating systems allow some control over the kernel page
cache. Linux has sync_file_range(2), several posix systems have msync(2)
and posix_fadvise(2). sync_file_range(2) is preferable because it
requires no special setup, whereas msync() requires the to-be-flushed
range to be mmap'ed. For the purpose of flushing dirty data
posix_fadvise(2) is the worst alternative, as flushing dirty data is
just a side-effect of POSIX_FADV_DONTNEED, which also removes the pages
from the page cache. Thus the feature is enabled by default only on
linux, but can be enabled on all systems that have any of the above
APIs.
While desirable and likely possible this patch does not contain an
implementation for windows.
With the infrastructure added, writes made via checkpointer, bgwriter
and normal user backends can be flushed after a configurable number of
writes. Each of these sources of writes controlled by a separate GUC,
checkpointer_flush_after, bgwriter_flush_after and backend_flush_after
respectively; they're separate because the number of flushes that are
good are separate, and because the performance considerations of
controlled flushing for each of these are different.
A later patch will add checkpoint sorting - after that flushes from the
ckeckpoint will almost always be desirable. Bgwriter flushes are most of
the time going to be random, which are slow on lots of storage hardware.
Flushing in backends works well if the storage and bgwriter can keep up,
but if not it can have negative consequences. This patch is likely to
have negative performance consequences without checkpoint sorting, but
unfortunately so has sorting without flush control.
Discussion: alpine.DEB.2.10.1506011320000.28433@sto
Author: Fabien Coelho and Andres Freund
Commit a892234f83 gave us enough
infrastructure to avoid vacuuming pages where every tuple on the
page is already frozen. So, replace the notion of a scan_all or
whole-table vacuum with the less onerous notion of an "aggressive"
vacuum, which will pages that are all-visible, but still skip those
that are all-frozen.
This should greatly reduce the cost of anti-wraparound vacuuming
on large clusters where the majority of data is never touched
between one cycle and the next, because we'll no longer have to
read all of those pages only to find out that we don't need to
do anything with them.
Patch by me, reviewed by Masahiko Sawada.
Commit 4de82f7d7 increased the WAL flush rate, mainly to increase the
likelihood that hint bits can be set quickly. More quickly set hint bits
can reduce contention around the clog et al. But unfortunately the
increased flush rate can have a significant negative performance impact,
I have measured up to a factor of ~4. The reason for this slowdown is
that if there are independent writes to the underlying devices, for
example because shared buffers is a lot smaller than the hot data set,
or because a checkpoint is ongoing, the fdatasync() calls force cache
flushes to be emitted to the storage.
This is achieved by flushing WAL only if the last flush was longer than
wal_writer_delay ago, or if more than wal_writer_flush_after (new GUC)
unflushed blocks are pending. Based on some tests the default for
wal_writer_delay is 1MB, which seems to work well both on SSD and
rotational media.
To avoid negative performance impact due to 4de82f7d7 an earlier
commit (db76b1e) made SetHintBits() more likely to succeed; preventing
performance regressions in the pgbench tests I performed.
Discussion: 20160118163908.GW10941@awork2.anarazel.de
When force_parallel_mode = true, we enable the parallel mode restrictions
for all queries for which this is believed to be safe. For the subset of
those queries believed to be safe to run entirely within a worker, we spin
up a worker and run the query there instead of running it in the
original process. When force_parallel_mode = regress, make additional
changes to allow the regression tests to run cleanly even though parallel
workers have been injected under the hood.
Taken together, this facilitates both better user testing and better
regression testing of the parallelism code.
Robert Haas, with help from Amit Kapila and Rushabh Lathia.
These were discussed in three different sections of the manual, which
unsurprisingly had diverged over time; and the descriptions of individual
variables lacked stylistic consistency even within each section (and
frequently weren't in very good English anyway). Clean up the mess, and
remove some of the redundant information in hopes that future additions
will be less likely to re-introduce inconsistency. For instance I see
no need for maintenance.sgml to include its very own list of all the
autovacuum storage parameters, especially since that list was already
incomplete.
Specifically, make its effect independent from the row_security GUC, and
make it affect permission checks pertinent to views the BYPASSRLS role
owns. The row_security GUC thereby ceases to change successful-query
behavior; it can only make a query fail with an error. Back-patch to
9.5, where BYPASSRLS was introduced.
A Gather executor node runs any number of copies of a plan in an equal
number of workers and merges all of the results into a single tuple
stream. It can also run the plan itself, if the workers are
unavailable or haven't started up yet. It is intended to work with
the Partial Seq Scan node which will be added in future commits.
It could also be used to implement parallel query of a different sort
by itself, without help from Partial Seq Scan, if the single_copy mode
is used. In that mode, a worker executes the plan, and the parallel
leader does not, merely collecting the worker's results. So, a Gather
node could be inserted into a plan to split the execution of that plan
across two processes. Nested Gather nodes aren't currently supported,
but we might want to add support for that in the future.
There's nothing in the planner to actually generate Gather nodes yet,
so it's not quite time to break out the champagne. But we're getting
close.
Amit Kapila. Some designs suggestions were provided by me, and I also
reviewed the patch. Single-copy mode, documentation, and other minor
changes also by me.
Every query of a single ENABLE ROW SECURITY table has two meanings, with
the row_security GUC selecting between them. With row_security=force
available, every function author would have been advised to either set
the GUC locally or test both meanings. Non-compliance would have
threatened reliability and, for SECURITY DEFINER functions, security.
Authors already face an obligation to account for search_path, and we
should not mimic that example. With this change, only BYPASSRLS roles
need exercise the aforementioned care. Back-patch to 9.5, where the
row_security GUC was introduced.
Since this narrows the domain of pg_db_role_setting.setconfig and
pg_proc.proconfig, one might bump catversion. A row_security=force
setting in one of those columns will elicit a clear message, so don't.
Per discussion, nowadays it is possible to have tablespaces that have
wildly different I/O characteristics from others. Setting different
effective_io_concurrency parameters for those has been measured to
improve performance.
Author: Julien Rouhaud
Reviewed by: Andres Freund
While postgres' use of SSL renegotiation is a good idea in theory, it
turned out to not work well in practice. The specification and openssl's
implementation of it have lead to several security issues. Postgres' use
of renegotiation also had its share of bugs.
Additionally OpenSSL has a bunch of bugs around renegotiation, reported
and open for years, that regularly lead to connections breaking with
obscure error messages. We tried increasingly complex workarounds to get
around these bugs, but we didn't find anything complete.
Since these connection breakages often lead to hard to debug problems,
e.g. spuriously failing base backups and significant latency spikes when
synchronous replication is used, we have decided to change the default
setting for ssl renegotiation to 0 (disabled) in the released
backbranches and remove it entirely in 9.5 and master.
Author: Andres Freund
Discussion: 20150624144148.GQ4797@alap3.anarazel.de
Backpatch: 9.5 and master, 9.0-9.4 get a different patch
When enabling wal_compression, there is a risk to leak data similarly to
the BREACH and CRIME attacks on SSL where the compression ratio of
a full page image gives a hint of what is the existing data of this page.
This vulnerability is quite cumbersome to exploit in practice, but doable.
So this patch makes wal_compression PGC_SUSET in order to prevent
non-superusers from enabling it and exploiting the vulnerability while
DBA thinks the risk very seriously and disables it in postgresql.conf.
Back-patch to 9.5 where wal_compression was introduced.
As first committed, this view reported on the file contents as they were
at the last SIGHUP event. That's not as useful as reporting on the current
contents, and what's more, it didn't work right on Windows unless the
current session had serviced at least one SIGHUP. Therefore, arrange to
re-read the files when pg_show_all_settings() is called. This requires
only minor refactoring so that we can pass changeVal = false to
set_config_option() so that it won't actually apply any changes locally.
In addition, add error reporting so that errors that would prevent the
configuration files from being loaded, or would prevent individual settings
from being applied, are visible directly in the view. This makes the view
usable for pre-testing whether edits made in the config files will have the
desired effect, before one actually issues a SIGHUP.
I also added an "applied" column so that it's easy to identify entries that
are superseded by later entries; this was the main use-case for the original
design, but it seemed unnecessarily hard to use for that.
Also fix a 9.4.1 regression that allowed multiple entries for a
PGC_POSTMASTER variable to cause bogus complaints in the postmaster log.
(The issue here was that commit bf007a27ac unintentionally reverted
3e3f65973a, which suppressed any duplicate entries within
ParseConfigFp. However, since the original coding of the pg_file_settings
view depended on such suppression *not* happening, we couldn't have fixed
this issue now without first doing something with pg_file_settings.
Now we suppress duplicates by marking them "ignored" within
ProcessConfigFileInternal, which doesn't hide them in the view.)
Lesser changes include:
Drive the view directly off the ConfigVariable list, instead of making a
basically-equivalent second copy of the data. There's no longer any need
to hang onto the data permanently, anyway.
Convert show_all_file_settings() to do its work in one call and return a
tuplestore; this avoids risks associated with assuming that the GUC state
will hold still over the course of query execution. (I think there were
probably latent bugs here, though you might need something like a cursor
on the view to expose them.)
Arrange to run SIGHUP processing in a short-lived memory context, to
forestall process-lifespan memory leaks. (There is one known leak in this
code, in ProcessConfigDirectory; it seems minor enough to not be worth
back-patching a specific fix for.)
Remove mistaken assignment to ConfigFileLineno that caused line counting
after an include_dir directive to be completely wrong.
Add missed failure check in AlterSystemSetConfigFile(). We don't really
expect ParseConfigFp() to fail, but that's not an excuse for not checking.
This file isn't entirely consistent about whether "on" and "off"
should be marked up with <literal>, but it doesn't make much sense
to be inconsistent within a single sentence.
Etsuro Fujita
The introduction of min_wal_size & max_wal_size in 88e9823026 makes it
feasible to increase the default upper bound in checkpoint
size. Previously raising the default would lead to a increased disk
footprint, even if more segments weren't beneficial. The low default of
checkpoint size is one of common performance problem users have thus
increasing the default makes sense. Setups where the increase in
maximum disk usage is a problem will very likely have to run with a
modified configuration anyway.
Discussion: 54F4EFB8.40202@agliodbs.com,
CA+TgmoZEAgX5oMGJOHVj8L7XOkAe05Gnf45rP40m-K3FhZRVKg@mail.gmail.com
Author: Josh Berkus, after a discussion involving lots of people.
While the SQL standard is pretty vague on the overall topic of operator
precedence (because it never presents a unified BNF for all expressions),
it does seem reasonable to conclude from the spec for <boolean value
expression> that OR has the lowest precedence, then AND, then NOT, then IS
tests, then the six standard comparison operators, then everything else
(since any non-boolean operator in a WHERE clause would need to be an
argument of one of these).
We were only sort of on board with that: most notably, while "<" ">" and
"=" had properly low precedence, "<=" ">=" and "<>" were treated as generic
operators and so had significantly higher precedence. And "IS" tests were
even higher precedence than those, which is very clearly wrong per spec.
Another problem was that "foo NOT SOMETHING bar" constructs, such as
"x NOT LIKE y", were treated inconsistently because of a bison
implementation artifact: they had the documented precedence with respect
to operators to their right, but behaved like NOT (i.e., very low priority)
with respect to operators to their left.
Fixing the precedence issues is just a small matter of rearranging the
precedence declarations in gram.y, except for the NOT problem, which
requires adding an additional lookahead case in base_yylex() so that we
can attach a different token precedence to NOT LIKE and allied two-word
operators.
The bulk of this patch is not the bug fix per se, but adding logic to
parse_expr.c to allow giving warnings if an expression has changed meaning
because of these precedence changes. These warnings are off by default
and are enabled by the new GUC operator_precedence_warning. It's believed
that very few applications will be affected by these changes, but it was
agreed that a warning mechanism is essential to help debug any that are.
When newly-added GUC parameter, wal_compression, is on, the PostgreSQL server
compresses a full page image written to WAL when full_page_writes is on or
during a base backup. A compressed page image will be decompressed during WAL
replay. Turning this parameter on can reduce the WAL volume without increasing
the risk of unrecoverable data corruption, but at the cost of some extra CPU
spent on the compression during WAL logging and on the decompression during
WAL replay.
This commit changes the WAL format (so bumping WAL version number) so that
the one-byte flag indicating whether a full page image is compressed or not is
included in its header information. This means that the commit increases the
WAL volume one-byte per a full page image even if WAL compression is not used
at all. We can save that one-byte by borrowing one-bit from the existing field
like hole_offset in the header and using it as the flag, for example. But which
would reduce the code readability and the extensibility of the feature.
Per discussion, it's not worth paying those prices to save only one-byte, so we
decided to add the one-byte flag to the header.
This commit doesn't introduce any new compression algorithm like lz4.
Currently a full page image is compressed using the existing PGLZ algorithm.
Per discussion, we decided to use it at least in the first version of the
feature because there were no performance reports showing that its compression
ratio is unacceptably lower than that of other algorithm. Of course,
in the future, it's worth considering the support of other compression
algorithm for the better compression.
Rahila Syed and Michael Paquier, reviewed in various versions by myself,
Andres Freund, Robert Haas, Abhijit Menon-Sen and many others.
The introduction in the Shared Library Preloading section already
instructs the user to separate multiple library names with commas, so
just remove the fragment from here.
Author: Dagfinn Ilmari Mannsåker
Instead of having a single knob (checkpoint_segments) that both triggers
checkpoints, and determines how many checkpoints to recycle, they are now
separate concerns. There is still an internal variable called
CheckpointSegments, which triggers checkpoints. But it no longer determines
how many segments to recycle at a checkpoint. That is now auto-tuned by
keeping a moving average of the distance between checkpoints (in bytes),
and trying to keep that many segments in reserve. The advantage of this is
that you can set max_wal_size very high, but the system won't actually
consume that much space if there isn't any need for it. The min_wal_size
sets a floor for that; you can effectively disable the auto-tuning behavior
by setting min_wal_size equal to max_wal_size.
The max_wal_size setting is now the actual target size of WAL at which a
new checkpoint is triggered, instead of the distance between checkpoints.
Previously, you could calculate the actual WAL usage with the formula
"(2 + checkpoint_completion_target) * checkpoint_segments + 1". With this
patch, you set the desired WAL usage with max_wal_size, and the system
calculates the appropriate CheckpointSegments with the reverse of that
formula. That's a lot more intuitive for administrators to set.
Reviewed by Amit Kapila and Venkata Balaji N.
Previously when the standby server failed to retrieve WAL files from any sources
(i.e., streaming replication, local pg_xlog directory or WAL archive), it always
waited for five seconds (hard-coded) before the next attempt. For example,
this is problematic in warm-standby because restore_command can fail
every five seconds even while new WAL file is expected to be unavailable for
a long time and flood the log files with its error messages.
This commit adds new parameter, wal_retrieve_retry_interval, to control that
wait time.
Alexey Vasiliev and Michael Paquier, reviewed by Andres Freund and me.
Andres Freund
Robert Haas
Peter Eisentraut
Greg Stark
Tom Lane
Kevin Grittner
Fujii Masao
Noah Misch
Alvaro Herrera
Jeff Davis
Joe Conway
Heikki Linnakangas