Previously, all heaps had FSMs. For very small tables, this means that the
FSM took up more space than the heap did. This is wasteful, so now we
refrain from creating the FSM for heaps with 4 pages or fewer. If the last
known target block has insufficient space, we still try to insert into some
other page before giving up and extending the relation, since doing
otherwise leads to table bloat. Testing showed that trying every page
penalized performance slightly, so we compromise and try every other page.
This way, we visit at most two pages. Any pages with wasted free space
become visible at next relation extension, so we still control table bloat.
As a bonus, directly attempting one or two pages can even be faster than
consulting the FSM would have been.
Once the FSM is created for a heap we don't remove it even if somebody
deletes all the rows from the corresponding relation. We don't think it is
a useful optimization as it is quite likely that relation will again grow
to the same size.
Author: John Naylor, Amit Kapila
Reviewed-by: Amit Kapila
Tested-by: Mithun C Y
Discussion: https://www.postgresql.org/message-id/CAJVSVGWvB13PzpbLEecFuGFc5V2fsO736BsdTakPiPAcdMM5tQ@mail.gmail.com
Previously we initialized pages when bulk extending in
RelationAddExtraBlocks(). That has a major disadvantage: It ties
RelationAddExtraBlocks() to heap, as other types of storage are likely
to need different amounts of special space, have different amount of
free space (previously determined by PageGetHeapFreeSpace()).
That we're relying on initializing pages, but not WAL logging the
initialization, also means the risk for getting
"WARNING: relation \"%s\" page %u is uninitialized --- fixing"
style warnings in vacuums after crashes/immediate shutdowns, is
considerably higher. The warning sounds much more serious than what
they are.
Fix those two issues together by not initializing pages in
RelationAddExtraPages() (but continue to do so in
RelationGetBufferForTuple(), which is linked much more closely to
heap), and accepting uninitialized pages as normal in
vacuumlazy.c. When vacuumlazy encounters an empty page it now adds it
to the FSM, but does nothing else. We chose to not issue a debug
message, much less a warning in that case - it seems rarely useful,
and quite likely to scare people unnecessarily.
For now empty pages aren't added to the VM, because standbys would not
re-discover such pages after a promotion. In contrast to other sources
for empty pages, there's no corresponding WAL records triggering FSM
updates during replay.
Previously when extending the relation, there was a moment between
extending the relation, and acquiring an exclusive lock on the new
page, in which another backend could lock the page. To avoid new
content being put on that new page, vacuumlazy needed to acquire the
extension lock for a brief moment when encountering a new page. A
second corner case, only working somewhat by accident, was that
RelationGetBufferForTuple() sometimes checks the last page in a
relation for free space, without consulting the FSM; that only worked
because PageGetHeapFreeSpace() interprets the zero page header in a
new page as no free space. The lack of handling this properly
required reverting the previous attempt in 684200543b.
This issue can be solved by using RBM_ZERO_AND_LOCK when extending the
relation, thereby avoiding this window. There's some added complexity
when RelationGetBufferForTuple() is called with another buffer (for
updates), to avoid deadlocks, but that's rarely hit at runtime.
Author: Andres Freund
Reviewed-By: Tom Lane
Discussion: https://postgr.es/m/20181219083945.6khtgm36mivonhva@alap3.anarazel.de
Add PG_CFLAGS, PG_CXXFLAGS, and PG_LDFLAGS variables to pgxs.mk which
will be appended or prepended to the corresponding make variables.
Notably, there was previously no way to pass custom CXXFLAGS to third
party extension module builds, COPT and PROFILE supporting only CFLAGS
and LDFLAGS.
Backpatch all the way down to ease integration with existing
extensions.
Author: Christoph Berg
Reviewed-by: Andres Freund, Tom Lane, Michael Paquier
Discussion: https://postgr.es/m/20181113104005.GA32154@msg.credativ.de
Backpatch-through: 9.4
To avoid deadlock, backend acquires a lock on heap pages in block
number order. In certain cases, lock on heap pages is dropped and
reacquired. In this case, the locks are dropped for reading in
corresponding VM page/s. The issue is we re-acquire locks in bufferId
order whereas the intention was to acquire in blockid order.
This commit ensures that we will always acquire locks on heap pages in
blockid order.
Reported-by: Nishant Fnu
Author: Nishant Fnu
Reviewed-by: Amit Kapila and Robert Haas
Backpatch-through: 9.4
Discussion: https://postgr.es/m/5883C831-2ED1-47C8-BFAC-2D5BAE5A8CAE@amazon.com
The documentation includes sections to be able to initialize and start
Postgres via a couple of commands. Some of its recommendations involve
using directly "postgres", which is inconsistent with the recommendation
given by initdb. At the same time make some other command calls more
consistent with the rest, by using an absolute path when creating a
database.
Author: Andreas Scherbaum
Reviewed-by: Michael Banck, Ryan Lambert
An upcoming patch which changes how inheritance planning works requires
adding a new function that does a similar job to set_append_rel_size() but
for child target relations. To save it from having to duplicate the qual
building code, move that to a separate function first.
Here we also change things so that we never attempt to build security quals
after detecting some const false child quals. We needlessly used to do this
just before we marked the child relation as a dummy rel.
In passing, this also moves the partition pruned check to before the qual
building code. We don't need to build the child quals before we check if
the partition has been pruned.
Author: David Rowley
Discussion: https://postgr.es/m/CAKJS1f_i+jrrD+if8qC7KPuTAAWsd=dtepgY_7u=P86GDEwm7A@mail.gmail.com
A timeout of 5s is used when waiting for WAL to become available at
recovery so as the startup process is able to react promptly if a
trigger file shows up. However this missed the fact that the startup
process also relies on the timeout to check periodically the status of
any active WAL receiver.
Discussion: https://postgr.es/m/20190131070956.GE13429@paquier.xyz
When logging the replica identity of a deleted tuple, XLOG_HEAP_DELETE
records include references of the old tuple. Its data is stored in an
intermediate variable used to register this information for the WAL
record, but this variable gets away from the stack when the record gets
actually inserted.
Spotted by clang's AddressSanitizer.
Author: Stas Kelvish
Discussion: https://postgr.es/m/085C8825-AD86-4E93-AF80-E26CDF03D1EA@postgrespro.ru
Backpatch-through: 9.4
These two new options can be used to improve the selectivity of
relations to vacuum or analyze even further depending on the age of
respectively their transaction ID or multixact ID, so as it is possible
to prioritize tables to prevent wraparound of one or the other.
Combined with --table, it is possible to target a subset of tables to
choose as potential processing targets.
Author: Nathan Bossart
Reviewed-by: Michael Paquier, Masahiko Sawada
Discussion: https://postgr.es/m/FFE5373C-E26A-495B-B5C8-911EC4A41C5E@amazon.com
We can't allow these pseudo-types to be used as table column types,
because storing an anonymous record value in a table would result
in data that couldn't be understood by other sessions. However,
it seems like there's no harm in allowing the case in a column
definition list that's specifying what a function-returning-record
returns. The data involved is all local to the current session,
so we should be just as able to resolve its actual tuple type as
we are for the function-returning-record's top-level tuple output.
Elvis Pranskevichus, with cosmetic changes by me
Discussion: https://postgr.es/m/11038447.kQ5A9Uj5xi@hammer.magicstack.net
Logging the PostgreSQL version on startup is useful for two reasons:
There is a clear marker in the log file that a new postmaster is
beginning, and it's useful for tracking the server version across
startup while upgrading.
Author: Christoph Berg <christoph.berg@credativ.de>
Discussion: https://www.postgresql.org/message-id/flat/20181121144611.GJ15795@msg.credativ.de/
When the syslogger was originally
added (bdf8ef6925), nothing was normally
logged before the point where it was started. But since
f9dfa5c977, the creation of sockets
causes messages of level LOG to be written routinely, so those don't
go to the syslogger now.
To improve that, arrange the sequence in PostmasterMain() slightly so
that the syslogger is started early enough to capture those messages.
Discussion: https://www.postgresql.org/message-id/d5d50936-20b9-85f1-06bc-94a01c5040c1%402ndquadrant.com
Reviewed-by: Christoph Berg <christoph.berg@credativ.de>
The function called can result in an out of memory error that subsequently was
disregarded. Instead it should set the appropriate SQL error variables and be
checked by whatever whenever statement is defined.
The bug was that determining which columns are part of the replica
identity index using RelationGetIndexAttrBitmap() would run
eval_const_expressions() on index expressions and predicates across
all indexes of the table, which in turn might require a snapshot, but
there wasn't one set, so it crashes. There were actually two separate
bugs, one on the publisher and one on the subscriber.
To trigger the bug, a table that is part of a publication or
subscription needs to have an index with a predicate or expression
that lends itself to constant expressions simplification.
The fix is to avoid the constant expressions simplification in
RelationGetIndexAttrBitmap(), so that it becomes safe to call in these
contexts. The constant expressions simplification comes from the
calls to RelationGetIndexExpressions()/RelationGetIndexPredicate() via
BuildIndexInfo(). But RelationGetIndexAttrBitmap() calling
BuildIndexInfo() is overkill. The latter just takes pg_index catalog
information, packs it into the IndexInfo structure, which former then
just unpacks again and throws away. We can just do this directly with
less overhead and skip the troublesome calls to
eval_const_expressions(). This also removes the awkward
cross-dependency between relcache.c and index.c.
Bug: #15114
Reported-by: Петър Славов <pet.slavov@gmail.com>
Reviewed-by: Noah Misch <noah@leadboat.com>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://www.postgresql.org/message-id/flat/152110589574.1223.17983600132321618383@wrigleys.postgresql.org/
The old name of this file was never a very good indication of what it
was for. Now that there's also access/relation.h, we have a potential
confusion hazard as well, so let's rename it to something more apropos.
Per discussion, "pathnodes.h" is reasonable, since a good fraction of
the file is Path node definitions.
While at it, tweak a couple of other headers that were gratuitously
importing relation.h into modules that don't need it.
Discussion: https://postgr.es/m/7719.1548688728@sss.pgh.pa.us
Create a new header optimizer/optimizer.h, which exposes just the
planner functions that can be used "at arm's length", without need
to access Paths or the other planner-internal data structures defined
in nodes/relation.h. This is intended to provide the whole planner
API seen by most of the rest of the system; although FDWs still need
to use additional stuff, and more thought is also needed about just
what selfuncs.c should rely on.
The main point of doing this now is to limit the amount of new
#include baggage that will be needed by "planner support functions",
which I expect to introduce later, and which will be in relevant
datatype modules rather than anywhere near the planner.
This commit just moves relevant declarations into optimizer.h from
other header files (a couple of which go away because everything
got moved), and adjusts #include lists to match. There's further
cleanup that could be done if we want to decide that some stuff
being exposed by optimizer.h doesn't belong in the planner at all,
but I'll leave that for another day.
Discussion: https://postgr.es/m/11460.1548706639@sss.pgh.pa.us
Move a few very simple node-creation and node-type-testing functions
from the planner's clauses.c to nodes/makefuncs and nodes/nodeFuncs.
There's nothing planner-specific about them, as evidenced by the
number of other places that were using them.
While at it, rename and_clause() etc to is_andclause() etc, to clarify
that they are node-type-testing functions not node-creation functions.
And use "static inline" implementations for the shortest ones.
Also, modify flatten_join_alias_vars() and some subsidiary functions
to take a Query not a PlannerInfo to define the join structure that
Vars should be translated according to. They were only using the
"parse" field of the PlannerInfo anyway, so this just requires removing
one level of indirection. The advantage is that now parse_agg.c can
use flatten_join_alias_vars() without the horrid kluge of creating an
incomplete PlannerInfo, which will allow that file to be decoupled from
relation.h in a subsequent patch.
Discussion: https://postgr.es/m/11460.1548706639@sss.pgh.pa.us
Someone running these test could have key or certificate files in
their ~/.postgresql/, which would interfere with the tests. The way
to override that is to specify sslcert=invalid and/or
sslrootcert=invalid if no actual certificate is used for a particular
test. Document that and fix up one test that had a risk of failing in
these circumstances.
Discussion: https://www.postgresql.org/message-id/flat/398754d8-6bb5-c5cf-e7b8-22e5f0983caf@2ndquadrant.com/
estimate_path_cost_size() failed to re-use cached costs when the cached
startup/total cost was 0, so it calculated the costs redundantly.
This is an oversight in commit aa09cd242f; but apply the patch to HEAD
only because there are no reports of actual trouble from that.
Author: Etsuro Fujita
Discussion: https://postgr.es/m/5C4AF3F3.4060409%40lab.ntt.co.jp
vacuumdb would use a catalog query only when the command caller does not
define a list of tables. Switching to a catalog table represents two
advantages:
- Relation existence check can happen before running any VACUUM or
ANALYZE query. Before this change, if multiple relations are defined
using --table, the utility would fail only after processing the
firstly-defined ones, which may be a long some depending on the size of
the relation. This adds checks for the relation names, and does
nothing, at least yet, for the attribute names.
- More filtering options can become available for the utility user.
These options, which may be introduced later on, are based on the
relation size or the relation age, and need to be made available even if
the user does not list any specific table with --table.
Author: Nathan Bossart
Reviewed-by: Michael Paquier, Masahiko Sawada
Discussion: https://postgr.es/m/FFE5373C-E26A-495B-B5C8-911EC4A41C5E@amazon.com
Previously llvmjit.h #error'ed when USE_LLVM was not defined, to
prevent it from being included from code not having #ifdef USE_LLVM
guards - but that's not actually that useful after, during the
development of JIT support, LLVM related code was moved into a
separately compiled .so. Having that #error means cpluspluscheck
doesn't work when llvm support isn't enabled, which isn't great.
Similarly add USE_LLVM guards to llvmjit_emit.h, and additionally make
sure it compiles standalone.
Per complaint from Tom Lane.
Author: Andres Freund
Discussion: https://postgr.es/m/19808.1548692361@sss.pgh.pa.us
Backpatch: 11, where JIT support was added
This reverts commit fc02e6724f and
e6799d5a53.
Parts of the buildfarm error out with
ERROR: page %u of relation "%s" should be empty but is not
errors, and so far I/we do not know why. fc02e672 didn't fix the
issue. As I cannot reproduce the issue locally, it seems best to get
the buildfarm green again, and reproduce the issue without time
pressure.
In e6799d5a53 I removed vacuumlazy.c trickery around re-checking
whether a page is actually empty after acquiring an extension lock on
the relation, because the page is not PageInit()ed anymore, and
entries in the FSM ought not to lead to user-visible errors.
As reported by various buildfarm animals that is not correct, given
the way to code currently stands: If vacuum processes a page that's
just been newly added by either RelationGetBufferForTuple() or
RelationAddExtraBlocks(), it could add that page to the FSM and it
could be reused by other backends, before those two functions check
whether the newly added page is actually new. That's a relatively
narrow race, but several buildfarm machines appear to be able to hit
it.
While it seems wrong that the FSM, given it's lack of durability and
approximative nature, can trigger errors like this, that seems better
fixed in a separate commit. Especially given that a good portion of
the buildfarm is red, and this is just re-introducing logic that
existed a few hours ago.
Author: Andres Freund
Discussion: https://postgr.es/m/20190128222259.zhi7ovzgtkft6em6@alap3.anarazel.de
In batching mode, COPY was using the same (per-tuple) memory context for
allocations with longer lifetime. This was confusing but harmless, until
commit 31f3817402 added COPY FROM ... WHERE feature, introducing a risk
of memory leak.
The "per-tuple" memory context was reset only when starting new batch,
but as the rows may be filtered out by the WHERE clauses, that may not
happen at all. The WHERE clause however has to be evaluated for all
rows, before filtering them out.
This commit separates the per-tuple and per-batch contexts, removing the
ambiguity. Expressions (both defaults and WHERE clause) are evaluated
in the per-tuple context, while tuples are formed in the batch context.
This allows resetting the contexts at appropriate times.
The main complexity is related to partitioning, in which case we need to
reset the batch context after forming the tuple (which happens before
routing to leaf partition). Instead of switching between two contexts
as before, we simply copy the last tuple aside, reset the context and
then copy the tuple back. The performance impact is negligible, and
juggling with two contexts is not free either.
Discussion: https://www.postgresql.org/message-id/flat/CALAY4q_DdpWDuB5-Zyi-oTtO2uSk8pmy+dupiRe3AvAc++1imA@mail.gmail.com
The fact that "SELECT expression" has no base relations has long been a
thorn in the side of the planner. It makes it hard to flatten a sub-query
that looks like that, or is a trivial VALUES() item, because the planner
generally uses relid sets to identify sub-relations, and such a sub-query
would have an empty relid set if we flattened it. prepjointree.c contains
some baroque logic that works around this in certain special cases --- but
there is a much better answer. We can replace an empty FROM clause with a
dummy RTE that acts like a table of one row and no columns, and then there
are no such corner cases to worry about. Instead we need some logic to
get rid of useless dummy RTEs, but that's simpler and covers more cases
than what was there before.
For really trivial cases, where the query is just "SELECT expression" and
nothing else, there's a hazard that adding the extra RTE makes for a
noticeable slowdown; even though it's not much processing, there's not
that much for the planner to do overall. However testing says that the
penalty is very small, close to the noise level. In more complex queries,
this is able to find optimizations that we could not find before.
The new RTE type is called RTE_RESULT, since the "scan" plan type it
gives rise to is a Result node (the same plan we produced for a "SELECT
expression" query before). To avoid confusion, rename the old ResultPath
path type to GroupResultPath, reflecting that it's only used in degenerate
grouping cases where we know the query produces just one grouped row.
(It wouldn't work to unify the two cases, because there are different
rules about where the associated quals live during query_planner.)
Note: although this touches readfuncs.c, I don't think a catversion
bump is required, because the added case can't occur in stored rules,
only plans.
Patch by me, reviewed by David Rowley and Mark Dilger
Discussion: https://postgr.es/m/15944.1521127664@sss.pgh.pa.us
There's no reason not to install these, and jit.h can be useful for
users of e.g. planner hooks.
Author: Donald Dong
Reviewed-By: Andres Freund
Discussion: https://postgr.es/m/296D405F-7F95-49F1-B565-389D6AA78505@csumb.edu
Backpatch: 11-, where JIT compilation was introduced
Previously we initialized pages when bulk extending in
RelationAddExtraBlocks(). That has a major disadvantage: It ties
RelationAddExtraBlocks() to heap, as other types of storage are likely
to need different amounts of special space, have different amount of
free space (previously determined by PageGetHeapFreeSpace()).
That we're relying on initializing pages, but not WAL logging the
initialization, also means the risk for getting
"WARNING: relation \"%s\" page %u is uninitialized --- fixing"
style warnings in vacuums after crashes/immediate shutdowns, is
considerably higher. The warning sounds much more serious than what
they are.
Fix those two issues together by not initializing pages in
RelationAddExtraPages() (but continue to do so in
RelationGetBufferForTuple(), which is linked much more closely to
heap), and accepting uninitialized pages as normal in
vacuumlazy.c. When vacuumlazy encounters an empty page it now adds it
to the FSM, but does nothing else. We chose to not issue a debug
message, much less a warning in that case - it seems rarely useful,
and quite likely to scare people unnecessarily.
For now empty pages aren't added to the VM, because standbys would not
re-discover such pages after a promotion. In contrast to other sources
for empty pages, there's no corresponding WAL records triggering FSM
updates during replay.
Author: Andres Freund
Reviewed-By: Tom Lane
Discussion: https://postgr.es/m/20181219083945.6khtgm36mivonhva@alap3.anarazel.de
Previously, all heaps had FSMs. For very small tables, this means that the
FSM took up more space than the heap did. This is wasteful, so now we
refrain from creating the FSM for heaps with 4 pages or fewer. If the last
known target block has insufficient space, we still try to insert into some
other page before giving up and extending the relation, since doing
otherwise leads to table bloat. Testing showed that trying every page
penalized performance slightly, so we compromise and try every other page.
This way, we visit at most two pages. Any pages with wasted free space
become visible at next relation extension, so we still control table bloat.
As a bonus, directly attempting one or two pages can even be faster than
consulting the FSM would have been.
Once the FSM is created for a heap we don't remove it even if somebody
deletes all the rows from the corresponding relation. We don't think it is
a useful optimization as it is quite likely that relation will again grow
to the same size.
Author: John Naylor with design inputs and some code contribution by Amit Kapila
Reviewed-by: Amit Kapila
Tested-by: Mithun C Y
Discussion: https://www.postgresql.org/message-id/CAJVSVGWvB13PzpbLEecFuGFc5V2fsO736BsdTakPiPAcdMM5tQ@mail.gmail.com
vacuumdb generates by itself SQL queries to run ANALYZE or VACUUM on the
backend, but we never actually checked for query patterns with column
lists defined.
Author: Michael Paquier
Reviewed-by: Nathan Bossart
Discussion: https://postgr.es/m/FFE5373C-E26A-495B-B5C8-911EC4A41C5E@amazon.com
Given the right timing, psql could emit "connection to server was lost"
rather than one of the other messages that this test script checked for.
It looks like commit 4247db625 may have made this more likely, but
I don't really believe it was impossible before then. Rather than
stress about it, just add that spelling as one of the crash-successfully-
detected cases.
Discussion: https://postgr.es/m/19344.1548554028@sss.pgh.pa.us
Before this change FunctionCallInfoData, the struct arguments etc for
V1 function calls are stored in, always had space for
FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two
arrays. For nearly every function call 100 arguments is far more than
needed, therefore wasting memory. Arg and argnull being two separate
arrays also guarantees that to access a single argument, two
cachelines have to be touched.
Change the layout so there's a single variable-length array with pairs
of value / isnull. That drastically reduces memory consumption for
most function calls (on x86-64 a two argument function now uses
64bytes, previously 936 bytes), and makes it very likely that argument
value and its nullness are on the same cacheline.
Arguments are stored in a new NullableDatum struct, which, due to
padding, needs more memory per argument than before. But as usually
far fewer arguments are stored, and individual arguments are cheaper
to access, that's still a clear win. It's likely that there's other
places where conversion to NullableDatum arrays would make sense,
e.g. TupleTableSlots, but that's for another commit.
Because the function call information is now variable-length
allocations have to take the number of arguments into account. For
heap allocations that can be done with SizeForFunctionCallInfoData(),
for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro
that helps to allocate an appropriately sized and aligned variable.
Some places with stack allocation function call information don't know
the number of arguments at compile time, and currently variably sized
stack allocations aren't allowed in postgres. Therefore allow for
FUNC_MAX_ARGS space in these cases. They're not that common, so for
now that seems acceptable.
Because of the need to allocate FunctionCallInfo of the appropriate
size, older extensions may need to update their code. To avoid subtle
breakages, the FunctionCallInfoData struct has been renamed to
FunctionCallInfoBaseData. Most code only references FunctionCallInfo,
so that shouldn't cause much collateral damage.
This change is also a prerequisite for more efficient expression JIT
compilation (by allocating the function call information on the stack,
allowing LLVM to optimize it away); previously the size of the call
information caused problems inside LLVM's optimizer.
Author: Andres Freund
Reviewed-By: Tom Lane
Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
Previously, \g would successfully execute the COPY command, but
the target specification if any was ignored, so that the data was
always dumped to the regular query output target. This seems like
a clear bug, so let's not just fix it but back-patch it.
While at it, adjust the documentation for \copy to recommend
"COPY ... TO STDOUT \g foo" as a plausible alternative.
Back-patch to 9.5. The problem exists much further back, but the
code associated with \g was refactored enough in 9.5 that we'd
need a significantly different patch for 9.4, and it doesn't
seem worth the trouble.
Daniel Vérité, reviewed by Fabien Coelho
Discussion: https://postgr.es/m/15dadc39-e050-4d46-956b-dcc4ed098753@manitou-mail.org
Amit Kapila
Michael Paquier
Thomas Munro
Andres Freund
Alvaro Herrera
Peter Eisentraut
Tom Lane
Michael Meskes
Magnus Hagander
Etsuro Fujita
Tomas Vondra