joins a bit better, ie, understand the differing cost functions for matched
and unmatched outer tuples. There is more that could be done in cost_hashjoin
but this already helps a great deal. Per discussions with Robert Haas.
Stefan Kaltenbrunner. The most reasonable behavior (at least for the near
term) seems to be to ignore the PlaceHolderVar and examine its argument
instead. In support of this, change the API of pull_var_clause() to allow
callers to request recursion into PlaceHolderVars. Currently
estimate_num_groups() is the only customer for that behavior, but where
there's one there may be others.
constants through full joins, as in
select * from tenk1 a full join tenk1 b using (unique1)
where unique1 = 42;
which should generate a fairly cheap plan where we apply the constraint
unique1 = 42 in each relation scan. This had been broken by my patch of
2008-06-27, which is now reverted in favor of a more invasive but hopefully
less incorrect approach. That patch was meant to prevent incorrect extraction
of OR'd indexclauses from OR conditions above an outer join. To do that
correctly we need more information than the outerjoin_delay flag can provide,
so add a nullable_relids field to RestrictInfo that records exactly which
relations are nulled by outer joins that are underneath a particular qual
clause. A side benefit is that we can make the test in create_or_index_quals
more specific: it is now smart enough to extract an OR'd indexclause into the
outer side of an outer join, even though it must not do so in the inner side.
The old coding couldn't distinguish these cases so it could not do either.
"physical tlist" optimization on the outer relation (ie, force a projection
step to occur in its scan). This avoids storing useless column values when
the outer relation's tuples are written to temporary batch files.
Modified version of a patch by Michael Henderson and Ramon Lawrence.
distribution, by creating a special fast path for the (first few) most common
values of the outer relation. Tuples having hashvalues matching the MCVs
are effectively forced to be in the first batch, so that we never write
them out to the batch temp files.
Bryce Cutt and Ramon Lawrence, with some editorialization by me.
exact-match pattern (no wildcard) can be index-optimized in some cases where a
prefix-match pattern cannot; specifically, since the required index clause is
simple equality, it works for regular text/varchar indexes even when the
locale is not C. I'm not sure how often this case really comes up, but since
it requires hardly any additional work to handle it, we might as well get it
right. Motivated by a discussion on the JDBC list.
by the planning process. This prevents the "failed to locate grouping columns"
error recently reported by Dickson Guedes. That happens because planning
replaces SubLinks by SubPlans in the subquery's targetlist, and exprTypmod()
is smarter about the former than the latter, causing the apparent type of
the subquery's output columns to change. This seems to be a deficiency we
should fix in exprTypmod(), but that will be a much more invasive patch
with possible side-effects elsewhere, so I'll do that only in HEAD.
Back-patch to 8.3. Arguably the lack of a copying step is broken/dangerous
all the way back, but in the absence of known problems I'll refrain from
making the older branches pay the extra cost. (The reason this particular
symptom didn't appear before is that exprTypmod() wasn't smart about SubLinks
either, until 8.3.)
amgettuple or only implement amgetbitmap, instead of the former assumption
that every AM supports both APIs. Extracted with minor editorialization
from Teodor's fast-GIN-insert patch; whatever becomes of that, this seems
like a simple and reasonable generalization of the index AM interface spec.
input lists before we grovel through the lists. This doesn't save much,
but testing shows that the case of both inputs NIL is common enough that
it saves something. And this is used enough to be a hotspot.
unique for a particular query, if the index predicate is satisfied. This
requires a bit of reordering of operations so that we check the predicates
before doing any selectivity estimates, but shouldn't really cause any
noticeable slowdown. Per a comment from Michal Politowski.
keys when considering a semi or anti join. This requires estimating the
selectivity of the merge qual as though it were a regular inner join condition.
To allow caching both that and the real outer-join-aware selectivity, split
RestrictInfo.this_selec into two fields.
This fixes one of the problems reported by Kevin Grittner.
the cheapest-total inner path as a new candidate while truncating the
sort key list, if it already matched the full sort key list. This is
too much of a corner case to be worth back-patching, since it's unusual
for the cheapest total path to be sorted, and anyway no real harm is
done (except in JOIN_SEMI/ANTI cases where cost_mergejoin is a bit
broken at the moment). But it wasn't behaving as intended, so fix it.
Noted while examining a test case from Kevin Grittner. This error doesn't
explain his issue, but it does explain why "set enable_seqscan = off"
seemed to reproduce it for me.
outer join clauses. Given, say,
... from a left join b on a.a1 = b.b1 where a.a1 = 42;
we'll deduce a clause b.b1 = 42 and then mark the original join clause
redundant (we can't remove it completely for reasons I don't feel like
squeezing into this log entry). However the original implementation of
that wasn't bulletproof, because clause_selectivity() wouldn't honor
this_selec if given nonzero varRelid --- which in practice meant that
it worked as desired *except* when considering index scan quals. Which
resulted in bogus underestimation of the size of the indexscan result for
an inner indexscan in an outer join, and consequently a possibly bad
choice of indexscan vs. bitmap scan. Fix by introducing an explicit test
into clause_selectivity(). Also, to make sure we don't trigger that test
in corner cases, change the convention to be that this_selec > 1, not
this_selec = 1, means it's been marked redundant. Per trouble report from
Scara Maccai.
Back-patch to 8.2, where the problem was introduced.
RHS that can't be unique-ified --- join_is_legal has to check that before
deciding to build a join, else we'll have an unimplementable joinrel.
Per report from Greg Stark.
though it is an inner rather than outer join type. This essentially means
that we don't bother to separate "pushed down" qual conditions from actual
join quals at a semijoin plan node; which is okay because the restrictions of
SQL syntax make it impossible to have a pushed-down qual that references the
inner side of a semijoin. This allows noticeably better optimization of
IN/EXISTS cases than we had before, since the equivalence-class machinery can
now use those quals. Also fix a couple of other mistakes that had essentially
disabled the ability to unique-ify the inner relation and then join it to just
a subset of the left-hand relations. An example case using the regression
database is
select * from tenk1 a, tenk1 b
where (a.unique1,b.unique2) in (select unique1,unique2 from tenk1 c);
which is planned reasonably well by 8.3 and earlier but had been forcing a
cartesian join of a/b in CVS HEAD.
return the tableoid as well as the ctid for any FOR UPDATE targets that
have child tables. All child tables are listed in the ExecRowMark list,
but the executor just skips the ones that didn't produce the current row.
Curiously, this longstanding restriction doesn't seem to have been documented
anywhere; so no doc changes.
we extended the appendrel mechanism to support UNION ALL optimization. The
reason nobody noticed was that we are not actually using attr_needed data for
appendrel children; hence it seems more reasonable to rip it out than fix it.
Back-patch to 8.2 because an Assert failure is possible in corner cases.
Per examination of an example from Jim Nasby.
In HEAD, also get rid of AppendRelInfo.col_mappings, which is quite inadequate
to represent UNION ALL situations; depend entirely on translated_vars instead.
that represent some expression that we desire to compute below the top level
of the plan, and then let that value "bubble up" as though it were a plain
Var (ie, a column value).
The immediate application is to allow sub-selects to be flattened even when
they are below an outer join and have non-nullable output expressions.
Formerly we couldn't flatten because such an expression wouldn't properly
go to NULL when evaluated above the outer join. Now, we wrap it in a
PlaceHolderVar and arrange for the actual evaluation to occur below the outer
join. When the resulting Var bubbles up through the join, it will be set to
NULL if necessary, yielding the correct results. This fixes a planner
limitation that's existed since 7.1.
In future we might want to use this mechanism to re-introduce some form of
Hellerstein's "expensive functions" optimization, ie place the evaluation of
an expensive function at the most suitable point in the plan tree.
set_rel_width(). The code had been catering for the possibility of different
varnos in the relation targetlist, but this is impossible for a base relation
(and if it were possible, putting all the widths in the same RelOptInfo would
be wrong anyway).
There are some unimplemented aspects: recursive queries must use UNION ALL
(should allow UNION too), and we don't have SEARCH or CYCLE clauses.
These might or might not get done for 8.4, but even without them it's a
pretty useful feature.
There are also a couple of small loose ends and definitional quibbles,
which I'll send a memo about to pgsql-hackers shortly. But let's land
the patch now so we can get on with other development.
Yoshiyuki Asaba, with lots of help from Tatsuo Ishii and Tom Lane
btree. We can't easily tell whether clauses generated from the equivalence
class could be used with such an index, so just assume that they might be.
This bit of over-optimization prevented use of non-btree indexes for nestloop
inner indexscans, in any case where the join uses an equality operator that
is also a btree operator --- which in particular is typically true for hash
indexes. Noted while trying to test the current hash index patch.
inserting a materialize node above an inner-side sort node, when the sort is
expected to spill to disk. (The materialize protects the sort from having
to support mark/restore, allowing it to do its final merge pass on-the-fly.)
We neglected to teach cost_mergejoin about that hack, so it was failing to
include the materialize's costs in the estimated cost of the mergejoin.
The materialize's costs are generally going to be pretty negligible in
comparison to the sort's, so this is only a small error and probably not
worth back-patching; but it's still wrong.
In the similar case where a materialize is inserted to protect an inner-side
node that can't do mark/restore at all, it's still true that the materialize
should not spill to disk, and so we should cost it cheaply rather than
expensively.
Noted while thinking about a question from Tom Raney.
into nodes/nodeFuncs, so as to reduce wanton cross-subsystem #includes inside
the backend. There's probably more that should be done along this line,
but this is a start anyway.
subqueries into the same thing you'd have gotten from IN (except always with
unknownEqFalse = true, so as to get the proper semantics for an EXISTS).
I believe this fixes the last case within CVS HEAD in which an EXISTS could
give worse performance than an equivalent IN subquery.
The tricky part of this is that if the upper query probes the EXISTS for only
a few rows, the hashing implementation can actually be worse than the default,
and therefore we need to make a cost-based decision about which way to use.
But at the time when the planner generates plans for subqueries, it doesn't
really know how many times the subquery will be executed. The least invasive
solution seems to be to generate both plans and postpone the choice until
execution. Therefore, in a query that has been optimized this way, EXPLAIN
will show two subplans for the EXISTS, of which only one will actually get
executed.
There is a lot more that could be done based on this infrastructure: in
particular it's interesting to consider switching to the hash plan if we start
out using the non-hashed plan but find a lot more upper rows going by than we
expected. I have therefore left some minor inefficiencies in place, such as
initializing both subplans even though we will currently only use one.
eval_const_expressions will generally throw away anything that's ANDed with
constant FALSE, what we're left with given an example like
select * from tenk1 a where (unique1,0) in (select unique2,1 from tenk1 b);
is a cartesian product computation, which is really not acceptable.
This is a regression in CVS HEAD compared to previous releases, which were
able to notice the impossible join condition in this case --- though not in
some related cases that are also improved by this patch, such as
select * from tenk1 a left join tenk1 b on (a.unique1=b.unique2 and 0=1);
Fix by skipping evaluation of the appropriate side of the outer join in
cases where it's demonstrably unnecessary.
and anti joins. To do this, pass the SpecialJoinInfo struct for the current
join as an additional optional argument to operator join selectivity
estimation functions. This allows the estimator to tell not only what kind
of join is being formed, but which variable is on which side of the join;
a requirement long recognized but not dealt with till now. This also leaves
the door open for future improvements in the estimators, such as accounting
for the null-insertion effects of lower outer joins. I didn't do anything
about that in the current patch but the information is in principle deducible
from what's passed.
The patch also clarifies the definition of join selectivity for semi/anti
joins: it's the fraction of the left input that has (at least one) match
in the right input. This allows getting rid of some very fuzzy thinking
that I had committed in the original 7.4-era IN-optimization patch.
There's probably room to estimate this better than the present patch does,
but at least we know what to estimate.
Since I had to touch CREATE OPERATOR anyway to allow a variant signature
for join estimator functions, I took the opportunity to add a couple of
additional checks that were missing, per my recent message to -hackers:
* Check that estimator functions return float8;
* Require execute permission at the time of CREATE OPERATOR on the
operator's function as well as the estimator functions;
* Require ownership of any pre-existing operator that's modified by
the command.
I also moved the lookup of the functions out of OperatorCreate() and
into operatorcmds.c, since that seemed more consistent with most of
the other catalog object creation processes, eg CREATE TYPE.
the old JOIN_IN code, but antijoins are new functionality.) Teach the planner
to convert appropriate EXISTS and NOT EXISTS subqueries into semi and anti
joins respectively. Also, LEFT JOINs with suitable upper-level IS NULL
filters are recognized as being anti joins. Unify the InClauseInfo and
OuterJoinInfo infrastructure into "SpecialJoinInfo". With that change,
it becomes possible to associate a SpecialJoinInfo with every join attempt,
which permits some cleanup of join selectivity estimation. That needs to be
taken much further than this patch does, but the next step is to change the
API for oprjoin selectivity functions, which seems like material for a
separate patch. So for the moment the output size estimates for semi and
especially anti joins are quite bogus.
as per my recent proposal:
1. Fold SortClause and GroupClause into a single node type SortGroupClause.
We were already relying on them to be struct-equivalent, so using two node
tags wasn't accomplishing much except to get in the way of comparing items
with equal().
2. Add an "eqop" field to SortGroupClause to carry the associated equality
operator. This is cheap for the parser to get at the same time it's looking
up the sort operator, and storing it eliminates the need for repeated
not-so-cheap lookups during planning. In future this will also let us
represent GROUP/DISTINCT operations on datatypes that have hash opclasses
but no btree opclasses (ie, they have equality but no natural sort order).
The previous representation simply didn't work for that, since its only
indicator of comparison semantics was a sort operator.
3. Add a hasDistinctOn boolean to struct Query to explicitly record whether
the distinctClause came from DISTINCT or DISTINCT ON. This allows removing
some complicated and not 100% bulletproof code that attempted to figure
that out from the distinctClause alone.
This patch doesn't in itself create any new capability, but it's necessary
infrastructure for future attempts to use hash-based grouping for DISTINCT
and UNION/INTERSECT/EXCEPT.
taking the maximum of any child rel's width, we should weight the widths
proportionally to the number of rows expected from each child. In hindsight
this is obviously correct because row width is really a proxy for the total
physical size of the relation. Per discussion with Scott Carey (bug #4264).
the associated datatype as their equality member. This means that these
opclasses can now support plain equality comparisons along with LIKE tests,
thus avoiding the need for an extra index in some applications. This
optimization was not possible when the pattern opclasses were first introduced,
because we didn't insist that text equality meant bitwise equality; but we
do now, so there is no semantic difference between regular and pattern
equality operators.
I removed the name_pattern_ops opclass altogether, since it's really useless:
name's regular comparisons are just strcmp() and are unlikely to become
something different. Instead teach indxpath.c that btree name_ops can be
used for LIKE whether or not the locale is C. This might lead to a useful
speedup in LIKE queries on the system catalogs in non-C locales.
The ~=~ and ~<>~ operators are gone altogether. (It would have been nice to
keep them for backward compatibility's sake, but since the pg_amop structure
doesn't allow multiple equality operators per opclass, there's no way.)
A not-immediately-obvious incompatibility is that the sort order within
bpchar_pattern_ops indexes changes --- it had been identical to plain
strcmp, but is now trailing-blank-insensitive. This will impact
in-place upgrades, if those ever happen.
Per discussions a couple months ago.
unnecessary #include lines in it. Also, move some tuple routine prototypes and
macros to htup.h, which allows removal of heapam.h inclusion from some .c
files.
For this to work, a new header file access/sysattr.h needed to be created,
initially containing attribute numbers of system columns, for pg_dump usage.
While at it, make contrib ltree, intarray and hstore header files more
consistent with our header style.
no particular need to do get_op_opfamily_properties() while building an
indexscan plan. Postpone that lookup until executor start. This simplifies
createplan.c a lot more than it complicates nodeIndexscan.c, and makes things
more uniform since we already had to do it that way for RowCompare
expressions. Should be a bit faster too, at least for plans that aren't
re-used many times, since we avoid palloc'ing and perhaps copying the
intermediate list data structure.
eval_const_expressions needs to be passed the PlannerInfo ("root") structure,
because in some cases we want it to substitute values for Param nodes.
(So "constant" is not so constant as all that ...) This mistake partially
disabled optimization of unnamed extended-Query statements in 8.3: in
particular the LIKE-to-indexscan optimization would never be applied if the
LIKE pattern was passed as a parameter, and constraint exclusion depending
on a parameter value didn't work either.
strings. This patch introduces four support functions cstring_to_text,
cstring_to_text_with_len, text_to_cstring, and text_to_cstring_buffer, and
two macros CStringGetTextDatum and TextDatumGetCString. A number of
existing macros that provided variants on these themes were removed.
Most of the places that need to make such conversions now require just one
function or macro call, in place of the multiple notational layers that used
to be needed. There are no longer any direct calls of textout or textin,
and we got most of the places that were using handmade conversions via
memcpy (there may be a few still lurking, though).
This commit doesn't make any serious effort to eliminate transient memory
leaks caused by detoasting toasted text objects before they reach
text_to_cstring. We changed PG_GETARG_TEXT_P to PG_GETARG_TEXT_PP in a few
places where it was easy, but much more could be done.
Brendan Jurd and Tom Lane
knowledge up through any joins it participates in. We were doing that already
in some special cases but not in the general case. Also, defend against zero
row estimates for the input relations in cost_mergejoin --- this fix may have
eliminated the only scenario in which that can happen, but be safe. Per
report from Alex Solovey.
for unhandled clause types ought to be 0.5, not 1.0. I fear I introduced
this silliness due to misreading the intent of the very-poorly-structured
code that was there when we inherited the file from Berkeley. The lack
of sanity in this behavior was exposed by an example from Sim Zacks.
(Arguably this is a bug fix and should be back-patched, but I'm a bit
hesitant to introduce a possible planner behavior change in the back
branches; it might detune queries that worked acceptably in the past.)
While at it, make estimation for DistinctExpr do something marginally
realistic, rather than just defaulting.
clauseless joins of relations that have unexploited join clauses. Rather
than looking at every other base relation in the query, the correct thing is
to examine the other relations in the "initial_rels" list of the current
make_rel_from_joinlist() invocation, because those are what we actually have
the ability to join against. This might be a subset of the whole query in
cases where join_collapse_limit or from_collapse_limit or full joins have
prevented merging the whole query into a single join problem. This is a bit
untidy because we have to pass those rels down through a new PlannerInfo
field, but it's necessary. Per bug #3865 from Oleg Kharin.
of poorer planning in 8.3 than 8.2:
1. After pushing a constant across an outer join --- ie, given
"a LEFT JOIN b ON (a.x = b.y) WHERE a.x = 42", we can deduce that b.y is
sort of equal to 42, in the sense that we needn't fetch any b rows where
it isn't 42 --- loop to see if any additional deductions can be made.
Previous releases did that by recursing, but I had mistakenly thought that
this was no longer necessary given the EquivalenceClass machinery.
2. Allow pushing constants across outer join conditions even if the
condition is outerjoin_delayed due to a lower outer join. This is safe
as long as the condition is strict and we re-test it at the upper join.
3. Keep the outer-join clause even if we successfully push a constant
across it. This is *necessary* in the outerjoin_delayed case, but
even in the simple case, it seems better to do this to ensure that the
join search order heuristics will consider the join as reasonable to
make. Mark such a clause as having selectivity 1.0, though, since it's
not going to eliminate very many rows after application of the constant
condition.
4. Tweak have_relevant_eclass_joinclause to report that two relations
are joinable when they have vars that are equated to the same constant.
We won't actually generate any joinclause from such an EquivalenceClass,
but again it seems that in such a case it's a good idea to consider
the join as worth costing out.
5. Fix a bug in select_mergejoin_clauses that was exposed by these
changes: we have to reject candidate mergejoin clauses if either side was
equated to a constant, because we can't construct a canonical pathkey list
for such a clause. This is an implementation restriction that might be
worth fixing someday, but it doesn't seem critical to get it done for 8.3.
the two join variables at both ends: not only trailing rows that need not be
scanned because there cannot be a match on the other side, but initial rows
that will be scanned without possibly having a match. This allows a more
realistic estimate of startup cost to be made, per recent pgsql-performance
discussion. In passing, fix a couple of bugs that had crept into
mergejoinscansel: it was not quite up to speed for the task of estimating
descending-order scans, which is a new requirement in 8.3.
clauselist_selectivity skip some analysis that's useless when there's only
one clause in the given list. Actually this can win even for not-so-simple
queries, because we also apply clauselist_selectivity to sublists such as the
quals matching an index; which are likely to have only a single entry even
when the total query is quite complicated.