In commit e2c2c2e8b1 I made use of nested
list structures to show which clauses went with which index columns, but
on reflection that's a data structure that only an old-line Lisp hacker
could love. Worse, it adds unnecessary complication to the many places
that don't much care which clauses go with which index columns. Revert
to the previous arrangement of flat lists of clauses, and instead add a
parallel integer list of column numbers. The places that care about the
pairing can chase both lists with forboth(), while the places that don't
care just examine one list the same as before.
The only real downside to this is that there are now two more lists that
need to be passed to amcostestimate functions in case they care about
column matching (which btcostestimate does, so not passing the info is not
an option). Rather than deal with 11-argument amcostestimate functions,
pass just the IndexPath and expect the functions to extract fields from it.
That gets us down to 7 arguments which is better than 11, and it seems
more future-proof against likely additions to the information we keep
about an index path.
It's potentially useful for an index to repeat the same indexable column
or expression in multiple index columns, if the columns have different
opclasses. (If they share opclasses too, the duplicate column is pretty
useless, but nonetheless we've allowed such cases since 9.0.) However,
the planner failed to cope with this, because createplan.c was relying on
simple equal() matching to figure out which index column each index qual
is intended for. We do have that information available upstream in
indxpath.c, though, so the fix is to not flatten the multi-level indexquals
list when putting it into an IndexPath. Then we can rely on the sublist
structure to identify target index columns in createplan.c. There's a
similar issue for index ORDER BYs (the KNNGIST feature), so introduce a
multi-level-list representation for that too. This adds a bit more
representational overhead, but we might more or less buy that back by not
having to search for matching index columns anymore in createplan.c;
likewise btcostestimate saves some cycles.
Per bug #6351 from Christian Rudolph. Likely symptoms include the "btree
index keys must be ordered by attribute" failure shown there, as well as
"operator MMMM is not a member of opfamily NNNN".
Although this is a pre-existing problem that can be demonstrated in 9.0 and
9.1, I'm not going to back-patch it, because the API changes in the planner
seem likely to break things such as index plugins. The corner cases where
this matters seem too narrow to justify possibly breaking things in a minor
release.
When a view is marked as a security barrier, it will not be pulled up
into the containing query, and no quals will be pushed down into it,
so that no function or operator chosen by the user can be applied to
rows not exposed by the view. Views not configured with this
option cannot provide robust row-level security, but will perform far
better.
Patch by KaiGai Kohei; original problem report by Heikki Linnakangas
(in October 2009!). Review (in earlier versions) by Noah Misch and
others. Design advice by Tom Lane and myself. Further review and
cleanup by me.
The need for this was debated when we put in the index-only-scan feature,
but at the time we had no near-term expectation of having AMs that could
support such scans for only some indexes; so we kept it simple. However,
the SP-GiST AM forces the issue, so let's fix it.
This patch only installs the new API; no behavior actually changes.
While logically correct, these two Asserts could fail depending on the
vagaries of floating-point arithmetic. In particular, on machines with
floating-point registers wider than standard "double" values, it was
possible for the compiler to compare a rounded-to-double value already
stored in memory with an unrounded long double value still in a register.
Given the preceding checks, these assertions aren't adding much, so let's
just get rid of them rather than try to find a compiler-proof fix.
Per report from Pavel Stehule.
Given the lack of previous complaints, and the fact that only developers
would be likely to trip over it, I'm only going to change this in HEAD,
even though the code has been like this for a long time.
If the right-hand side of a semijoin is unique, then we can treat it like a
normal join (or another way to say that is: we don't need to explicitly
unique-ify the data before doing it as a normal join). We were recognizing
such cases when the RHS was a sub-query with appropriate DISTINCT or GROUP
BY decoration, but there's another way: if the RHS is a plain relation with
unique indexes, we can check if any of the indexes prove the output is
unique. Most of the infrastructure for that was there already in the join
removal code, though I had to rearrange it a bit. Per reflection about a
recent example in pgsql-performance.
The uniqueness condition might fail to hold intra-transaction, and assuming
it does can give incorrect query results. Per report from Marti Raudsepp,
though this is not his proposed patch.
Back-patch to 9.0, where both these features were introduced. In the
released branches, add the new IndexOptInfo field to the end of the struct,
to try to minimize ABI breakage for third-party code that may be examining
that struct.
Add a column pg_class.relallvisible to remember the number of pages that
were all-visible according to the visibility map as of the last VACUUM
(or ANALYZE, or some other operations that update pg_class.relpages).
Use relallvisible/relpages, instead of an arbitrary constant, to estimate
how many heap page fetches can be avoided during an index-only scan.
This is pretty primitive and will no doubt see refinements once we've
acquired more field experience with the index-only scan mechanism, but
it's way better than using a constant.
Note: I had to adjust an underspecified query in the window.sql regression
test, because it was changing answers when the plan changed to use an
index-only scan. Some of the adjacent tests perhaps should be adjusted
as well, but I didn't do that here.
This commit changes index-only scans so that data is read directly from the
index tuple without first generating a faux heap tuple. The only immediate
benefit is that indexes on system columns (such as OID) can be used in
index-only scans, but this is necessary infrastructure if we are ever to
support index-only scans on expression indexes. The executor is now ready
for that, though the planner still needs substantial work to recognize
the possibility.
To do this, Vars in index-only plan nodes have to refer to index columns
not heap columns. I introduced a new special varno, INDEX_VAR, to mark
such Vars to avoid confusion. (In passing, this commit renames the two
existing special varnos to OUTER_VAR and INNER_VAR.) This allows
ruleutils.c to handle them with logic similar to what we use for subplan
reference Vars.
Since index-only scans are now fundamentally different from regular
indexscans so far as their expression subtrees are concerned, I also chose
to change them to have their own plan node type (and hence, their own
executor source file).
When a btree index contains all columns required by the query, and the
visibility map shows that all tuples on a target heap page are
visible-to-all, we don't need to fetch that heap page. This patch depends
on the previous patches that made the visibility map reliable.
There's a fair amount left to do here, notably trying to figure out a less
chintzy way of estimating the cost of an index-only scan, but the core
functionality seems ready to commit.
Robert Haas and Ibrar Ahmed, with some previous work by Heikki Linnakangas.
If an indexable operator for a non-collatable indexed datatype has a
collatable right-hand input type, any OpExpr for it will be marked with a
nonzero inputcollid (since having one collatable input is sufficient to
make that happen). However, an index on a non-collatable column certainly
doesn't have any collation. This caused us to fail to match such operators
to their indexes, because indxpath.c required an exact match of index
collation and clause collation. It seems correct to allow a match when the
index is collation-less regardless of the clause's inputcollid: an operator
with both noncollatable and collatable inputs could perhaps depend on the
collation of the collatable input, but it could hardly expect the index for
the noncollatable input to have that same collation.
Per bug #6232 from Pierre Ducroquet. His example is specifically about
"hstore ? text" but the problem seems quite generic.
The constraint exclusion feature checks for contradictions among scan
restriction clauses, as well as contradictions between those clauses and a
table's CHECK constraints. The first aspect of this testing can be useful
for non-table relations (such as subqueries or functions-in-FROM), but the
feature was coded with only the CHECK case in mind so we were applying it
only to plain-table RTEs. Move the relation_excluded_by_constraints call
so that it is applied to all RTEs not just plain tables. With the default
setting of constraint_exclusion this results in no extra work, but with
constraint_exclusion = ON we will detect optimizations that we missed
before (at the cost of more planner cycles than we expended before).
Per a gripe from Gunnlaugur Þór Briem. Experimentation with
his example also showed we were not being very bright about the case where
constraint exclusion is proven within a subquery within UNION ALL, so tweak
the code to allow set_append_rel_pathlist to recognize such cases.
Formerly, set_subquery_pathlist and other creators of plans for subqueries
saved only the rangetable and rowMarks lists from the lower-level
PlannerInfo. But there's no reason not to remember the whole PlannerInfo,
and indeed this turns out to simplify matters in a number of places.
The immediate reason for doing this was so that the subroot will still be
accessible when we're trying to extract column statistics out of an
already-planned subquery. But now that I've done it, it seems like a good
code-beautification effort in its own right.
I also chose to get rid of the transient subrtable and subrowmark fields in
SubqueryScan nodes, in favor of having setrefs.c look up the subquery's
RelOptInfo. That required changing all the APIs in setrefs.c to pass
PlannerInfo not PlannerGlobal, which was a large but quite mechanical
transformation.
One side-effect not foreseen at the beginning is that this finally broke
inheritance_planner's assumption that replanning the same subquery RTE N
times would necessarily give interchangeable results each time. That
assumption was always pretty risky, but now we really have to make a
separate RTE for each instance so that there's a place to carry the
separate subroots.
set_append_rel_pathlist supposed that, while computing per-column width
estimates for the appendrel, it could ignore child rels for which the
translated reltargetlist entry wasn't a Var. This gave rise to completely
silly estimates in some common cases, such as constant outputs from some or
all of the arms of a UNION ALL. Instead, fall back on get_typavgwidth to
estimate from the value's datatype; which might be a poor estimate but at
least it's not completely wacko.
That problem was exposed by an Assert in set_subquery_size_estimates, which
unfortunately was still overoptimistic even with that fix, since we don't
compute attr_widths estimates for appendrels that are entirely excluded by
constraints. So remove the Assert; we'll just fall back on get_typavgwidth
in such cases.
Also, since set_subquery_size_estimates calls set_baserel_size_estimates
which calls set_rel_width, there's no need for set_subquery_size_estimates
to call get_typavgwidth; set_rel_width will handle it for us if we just
leave the estimate set to zero. Remove the unnecessary code.
Per report from Erik Rijkers and subsequent investigation.
A PlaceHolderVar's expression might contain another, lower-level
PlaceHolderVar. If the outer PlaceHolderVar is used, the inner one
certainly will be also, and so we have to make sure that both of them get
into the placeholder_list with correct ph_may_need values during the
initial pre-scan of the query (before deconstruct_jointree starts).
We did this correctly for PlaceHolderVars appearing in the query quals,
but overlooked the issue for those appearing in the top-level targetlist;
with the result that nested placeholders referenced only in the targetlist
did not work correctly, as illustrated in bug #6154.
While at it, add some error checking to find_placeholder_info to ensure
that we don't try to create new placeholders after it's too late to do so;
they have to all be created before deconstruct_jointree starts.
Back-patch to 8.4 where the PlaceHolderVar mechanism was introduced.
Regular aggregate functions in combination with, or within the arguments
of, window functions are OK per spec; they have the semantics that the
aggregate output rows are computed and then we run the window functions
over that row set. (Thus, this combination is not really useful unless
there's a GROUP BY so that more than one aggregate output row is possible.)
The case without GROUP BY could fail, as recently reported by Jeff Davis,
because sloppy construction of the Agg node's targetlist resulted in extra
references to possibly-ungrouped Vars appearing outside the aggregate
function calls themselves. See the added regression test case for an
example.
Fixing this requires modifying the API of flatten_tlist and its underlying
function pull_var_clause. I chose to make pull_var_clause's API for
aggregates identical to what it was already doing for placeholders, since
the useful behaviors turn out to be the same (error, report node as-is, or
recurse into it). I also tightened the error checking in this area a bit:
if it was ever valid to see an uplevel Var, Aggref, or PlaceHolderVar here,
that was a long time ago, so complain instead of ignoring them.
Backpatch into 9.1. The failure exists in 8.4 and 9.0 as well, but seeing
that it only occurs in a basically-useless corner case, it doesn't seem
worth the risks of changing a function API in a minor release. There might
be third-party code using pull_var_clause.
The previous coding failed to account properly for the costs of evaluating
the input expressions of aggregates and window functions, as seen in a
recent gripe from Claudio Freire. (I said at the time that it wasn't
counting these costs at all; but on closer inspection, it was effectively
charging these costs once per output tuple. That is completely wrong for
aggregates, and not exactly right for window functions either.)
There was also a hard-wired assumption that aggregates and window functions
had procost 1.0, which is now fixed to respect the actual cataloged costs.
The costing of WindowAgg is still pretty bogus, since it doesn't try to
estimate the effects of spilling data to disk, but that seems like a
separate issue.
Although rowcount estimates really ought not be NaN, a bug elsewhere
could perhaps result in that, and that would cause Assert failure in
cost_mergejoin, which I believe to be the explanation for bug #5977 from
Anton Kuznetsov. Seems like a good idea to expend a couple more cycles
to prevent that, even though the real bug is elsewhere. Not back-patching,
though, because we don't encourage running production systems with
Asserts on.
When we are doing GEQO join planning, the current memory context is a
short-lived context that will be reset at the end of geqo_eval(). However,
the RelOptInfos for base relations are set up before that and then re-used
across many GEQO cycles. Hence, any code that modifies a baserel during
join planning has to be careful not to put pointers to the short-lived
context into the baserel struct. mark_dummy_rel got this wrong, leading to
easy-to-reproduce-once-you-know-how crashes in 8.4, as reported off-list by
Leo Carson of SDSC. Some improvements made in 9.0 make it difficult to
demonstrate the crash in 9.0 or HEAD; but there's no doubt that there's
still a risk factor here, so patch all branches that have the function.
(Note: 8.3 has a similar function, but it's only applied to joinrels and
thus is not a hazard.)
Since collation is effectively an argument, not a property of the function,
FmgrInfo is really the wrong place for it; and this becomes critical in
cases where a cached FmgrInfo is used for varying purposes that might need
different collation settings. Fix by passing it in FunctionCallInfoData
instead. In particular this allows a clean fix for bug #5970 (record_cmp
not working). This requires touching a bit more code than the original
method, but nobody ever thought that collations would not be an invasive
patch...
This is necessary, not optional, now that ILIKE and regexes are collation
aware --- else we might derive a wrong comparison constant for index
optimized pattern matches.
Get rid of bogus collation test in match_special_index_operator (even for
ILIKE, the pattern match operator's collation doesn't matter here, and even
if it did the test was testing the wrong thing).
Fix broken looping logic in expand_indexqual_rowcompare.
Add collation check in match_clause_to_ordering_op.
Make naming and argument ordering more consistent; improve comments.
I'm not sure these have any non-cosmetic implications, but I'm not sure
they don't, either. In particular, ensure the CaseTestExpr generated
by transformAssignmentIndirection to represent the base target column
carries the correct collation, because parse_collate.c won't fix that.
Tweak lsyscache.c API so that we can get the appropriate collation
without an extra syscache lookup.
In nearly all cases, the caller already knows the correct collation, and
in a number of places, the value the caller has handy is more correct than
the default for the type would be. (In particular, this patch makes it
significantly less likely that eval_const_expressions will result in
changing the exposed collation of an expression.) So an internal lookup
is both expensive and wrong.
Instead of playing cute games with pathkeys, just build a direct
representation of the intended sub-select, and feed it through
query_planner to get a Path for the index access. This is a bit slower
than 9.1's previous method, since we'll duplicate most of the overhead of
query_planner; but since the whole optimization only applies to rather
simple single-table queries, that probably won't be much of a problem in
practice. The advantage is that we get to do the right thing when there's
a partial index that needs the implicit IS NOT NULL clause to be usable.
Also, although this makes planagg.c be a bit more closely tied to the
ordering of operations in grouping_planner, we can get rid of some coupling
to lower-level parts of the planner. Per complaint from Marti Raudsepp.
All expression nodes now have an explicit output-collation field, unless
they are known to only return a noncollatable data type (such as boolean
or record). Also, nodes that can invoke collation-aware functions store
a separate field that is the collation value to pass to the function.
This avoids confusion that arises when a function has collatable inputs
and noncollatable output type, or vice versa.
Also, replace the parser's on-the-fly collation assignment method with
a post-pass over the completed expression tree. This allows us to use
a more complex (and hopefully more nearly spec-compliant) assignment
rule without paying for it in extra storage in every expression node.
Fix assorted bugs in the planner's handling of collations by making
collation one of the defining properties of an EquivalenceClass and
by converting CollateExprs into discardable RelabelType nodes during
expression preprocessing.
While this will give wrong answers when estimating selectivity for a
comparison operator that's using a non-default collation, the estimation
error probably won't be large; and anyway the former approach created
estimation errors of its own by trying to use a histogram that might have
been computed with some other collation. So we'll adopt this simplified
approach for now and perhaps improve it sometime in the future.
This patch incorporates changes from Andres Freund to make sure that
selfuncs.c passes a valid collation OID to any datatype-specific function
it calls, in case that function wants collation information. Said OID will
now always be DEFAULT_COLLATION_OID, but at least we won't get errors.
The recent additions for FDW support required checking foreign-table-ness
in several places in the parse/plan chain. While it's not clear whether
that would really result in a noticeable slowdown, it seems best to avoid
any performance risk by keeping a copy of the relation's relkind in
RangeTblEntry. That might have some other uses later, anyway.
Per discussion.
This commit provides the core code and documentation needed. A contrib
module test case will follow shortly.
Shigeru Hanada, Jan Urbanski, Heikki Linnakangas
This adds collation support for columns and domains, a COLLATE clause
to override it per expression, and B-tree index support.
Peter Eisentraut
reviewed by Pavel Stehule, Itagaki Takahiro, Robert Haas, Noah Misch
Per my note of a couple days ago, create_index_paths would refuse to
consider any path at all for GIN indexes if the selectivity estimate came
out as 1.0; not even if you tried to force it with enable_seqscan. While
this isn't really a bad outcome in practice, it could be annoying for
testing purposes. Adjust the test for "is this path only useful for
sorting" so that it doesn't fire on paths with nil pathkeys, which will
include all GIN paths.
This is advantageous first because it allows us to hash the smaller table
regardless of the outer-join type, and second because hash join can be more
flexible than merge join in dealing with arbitrary join quals in a FULL
join. For merge join all the join quals have to be mergejoinable, but hash
join will work so long as there's at least one hashjoinable qual --- the
others can be any condition. (This is true essentially because we don't
keep per-inner-tuple match flags in merge join, while hash join can do so.)
To do this, we need a has-it-been-matched flag for each tuple in the
hashtable, not just one for the current outer tuple. The key idea that
makes this practical is that we can store the match flag in the tuple's
infomask, since there are lots of bits there that are of no interest for a
MinimalTuple. So we aren't increasing the size of the hashtable at all for
the feature.
To write this without turning the hash code into even more of a pile of
spaghetti than it already was, I rewrote ExecHashJoin in a state-machine
style, similar to ExecMergeJoin. Other than that decision, it was pretty
straightforward.
This is a heavily revised version of builtin_knngist_core-0.9. The
ordering operators are no longer mixed in with actual quals, which would
have confused not only humans but significant parts of the planner.
Instead, ordering operators are carried separately throughout planning and
execution.
Since the API for ambeginscan and amrescan functions had to be changed
anyway, this commit takes the opportunity to rationalize that a bit.
RelationGetIndexScan no longer forces a premature index_rescan call;
instead, callers of index_beginscan must call index_rescan too. Aside from
making the AM-side initialization logic a bit less peculiar, this has the
advantage that we do not make a useless extra am_rescan call when there are
runtime key values. AMs formerly could not assume that the key values
passed to amrescan were actually valid; now they can.
Teodor Sigaev and Tom Lane
Formerly we looked up the operators associated with each index (caching
them in relcache) and then the planner looked up the btree opfamily
containing such operators in order to build the btree-centric pathkey
representation that describes the index's sort order. This is quite
pointless for btree indexes: we might as well just use the index's opfamily
information directly. That saves syscache lookup cycles during planning,
and furthermore allows us to eliminate the relcache's caching of operators
altogether, which may help in reducing backend startup time.
I added code to plancat.c to perform the same type of double lookup
on-the-fly if it's ever faced with a non-btree amcanorder index AM.
If such a thing actually becomes interesting for production, we should
replace that logic with some more-direct method for identifying the
corresponding btree opfamily; but it's not worth spending effort on now.
There is considerably more to do pursuant to my recent proposal to get rid
of sort-operator-based representations of sort orderings, but this patch
grabs some of the low-hanging fruit. I'll look at the remainder of that
work after the current commitfest.
We no longer need the terminating zero entry in opfamily[], so get rid of
it. Also replace assorted ad-hoc looping logic with simple for and foreach
constructs. This code is now noticeably more readable than it was an hour
ago; credit to Robert for seeing that it could be simplified.
As per the ancient comment for set_rel_width, it really wasn't much good
for relations that aren't plain tables: it would never find any stats and
would always fall back on datatype-based estimates, which are often pretty
silly. Fix that by copying up width estimates from the subquery planning
process.
At some point we might want to do this for CTEs too, but that would be a
significantly more invasive patch because the sub-PlannerInfo is no longer
accessible by the time it's needed. I refrained from doing anything about
that, partly for fear of breaking the unmerged CTE-related patches.
In passing, also generate less bogus width estimates for whole-row Vars.
Per a gripe from Jon Nelson.
Per my recent proposal, get rid of all the direct inspection of indexes
and manual generation of paths in planagg.c. Instead, set up
EquivalenceClasses for the aggregate argument expressions, and let the
regular path generation logic deal with creating paths that can satisfy
those sort orders. This makes planagg.c a bit more visible to the rest
of the planner than it was originally, but the approach is basically a lot
cleaner than before. A major advantage of doing it this way is that we get
MIN/MAX optimization on inheritance trees (using MergeAppend of indexscans)
practically for free, whereas in the old way we'd have had to add a whole
lot more duplicative logic.
One small disadvantage of this approach is that MIN/MAX aggregates can no
longer exploit partial indexes having an "x IS NOT NULL" predicate, unless
that restriction or something that implies it is specified in the query.
The previous implementation was able to use the added "x IS NOT NULL"
condition as an extra predicate proof condition, but in this version we
rely entirely on indexes that are considered usable by the main planning
process. That seems a fair tradeoff for the simplicity and functionality
gained.
It was reporting that these were fully indexed (hence cheap), when of
course they're the exact opposite of that. I'm not certain if the case
would arise in practice, since a clauseless semijoin is hard to produce
in SQL, but if it did happen we'd make some dumb decisions.
The core of this patch is hash_array() and associated typcache
infrastructure, which works just about exactly like the existing support
for array comparison.
In addition I did some work to ensure that the planner won't think that an
array type is hashable unless its element type is hashable, and similarly
for sorting. This includes adding a datatype parameter to op_hashjoinable
and op_mergejoinable, and adding an explicit "hashable" flag to
SortGroupClause. The lack of a cross-check on the element type was a
pre-existing bug in mergejoin support --- but it didn't matter so much
before, because if you couldn't sort the element type there wasn't any good
alternative to failing anyhow. Now that we have the alternative of hashing
the array type, there are cases where we can avoid a failure by being picky
at the planner stage, so it's time to be picky.
The issue of exactly how to combine the per-element hash values to produce
an array hash is still open for discussion, but the rest of this is pretty
solid, so I'll commit it as-is.
Zoltan Boszormenyi exhibited a test case in which planning time was
dominated by construction of EquivalenceClasses and PathKeys that had no
actual relevance to the query (and in fact got discarded immediately).
This happened because we generated PathKeys describing the sort ordering of
every index on every table in the query, and only after that checked to see
if the sort ordering was relevant. The EC/PK construction code is O(N^2)
in the number of ECs, which is all right for the intended number of such
objects, but it gets out of hand if there are ECs for lots of irrelevant
indexes.
To fix, twiddle the handling of mergeclauses a little bit to ensure that
every interesting EC is created before we begin path generation. (This
doesn't cost anything --- in fact I think it's a bit cheaper than before
--- since we always eventually created those ECs anyway.) Then, if an
index column can't be found in any pre-existing EC, we know that that sort
ordering is irrelevant for the query. Instead of creating a useless EC,
we can just not build a pathkey for the index column in the first place.
The index will still be considered if it's useful for non-order-related
reasons, but we will think of its output as unsorted.
This patch eliminates the former need to sort the output of an Append scan
when an ordered scan of an inheritance tree is wanted. This should be
particularly useful for fast-start cases such as queries with LIMIT.
Original patch by Greg Stark, with further hacking by Hans-Jurgen Schonig,
Robert Haas, and Tom Lane.
Tom Lane
Robert Haas
Bruce Momjian
Peter Eisentraut