Previously, the comments stated that there was no purpose to considering
startup cost for partial paths, but this is not the case: it's perfectly
reasonable to want a fast-start path for a plan that involves a LIMIT
(perhaps over an aggregate, so that there is enough data being processed
to justify parallel query but yet we don't want all the result rows).
Accordingly, rewrite add_partial_path and add_partial_path_precheck to
consider startup costs. This also fixes an independent bug in
add_partial_path_precheck: commit e222534679
failed to update it to do anything with the new disabled_nodes field.
That bug fix is formally separate from the rest of this patch and could
be committed separately, but I think it makes more sense to fix both
issues together, because then we can (as this commit does) just make
add_partial_path_precheck do the cost comparisons in the same way as
compare_path_costs_fuzzily, which hopefully reduces the chances of
ending up with something that's still incorrect.
This patch is based on earlier work on this topic by Tomas Vondra,
but I have rewritten a great deal of it.
Co-authored-by: Robert Haas <rhaas@postgresql.org>
Co-authored-by: Tomas Vondra <tomas@vondra.me>
Discussion: http://postgr.es/m/CA+TgmobRufbUSksBoxytGJS1P+mQY4rWctCk-d0iAUO6-k9Wrg@mail.gmail.com
Commit 4020b370f2 had the idea that it
would be a good idea to handle testing PGS_CONSIDER_NONPARTIAL within
cost_material to save callers the trouble, but that turns out not to be
a very good idea. One concern is that it makes cost_material() dependent
on the caller having initialized certain fields in the MaterialPath,
which is a bit awkward for materialize_finished_plan, which wants to use
a dummy path.
Another problem is that it can result in generated materialized nested
loops where the Materialize node is disabled, contrary to the intention
of joinpath.c's logic in match_unsorted_outer() and
consider_parallel_nestloop(), which aims to consider such paths only
when they would not need to be disabled. In the previous coding, it was
possible for the pgs_mask on the joinrel to have PGS_CONSIDER_NONPARTIAL
set, while the inner rel had the same bit clear. In that case, we'd
generate and then disable a Materialize path.
That seems wrong, so instead, pull up the logic to test the
PGS_CONSIDER_NONPARTIAL bit into joinpath.c, restoring the historical
behavior that either we don't generate a given materialized nested loop
in the first place, or we don't disable it.
Discussion: http://postgr.es/m/CA+TgmoawzvCoZAwFS85tE5+c8vBkqgcS8ZstQ_ohjXQ9wGT9sw@mail.gmail.com
Discussion: http://postgr.es/m/CA+TgmoYS4ZCVAF2jTce=bMP0Oq_db_srocR4cZyO0OBp9oUoGg@mail.gmail.com
Each RelOptInfo now has a pgs_mask member which is a mask of acceptable
strategies. For most rels, this is populated from PlannerGlobal's
default_pgs_mask, which is computed from the values of the enable_*
GUCs at the start of planning.
For baserels, get_relation_info_hook can be used to adjust pgs_mask for
each new RelOptInfo, at least for rels of type RTE_RELATION. Adjusting
pgs_mask is less useful for other types of rels, but if it proves to
be necessary, we can revisit the way this hook works or add a new one.
For joinrels, two new hooks are added. joinrel_setup_hook is called each
time a joinrel is created, and one thing that can be done from that hook
is to manipulate pgs_mask for the new joinrel. join_path_setup_hook is
called each time we're about to add paths to a joinrel by considering
some particular combination of an outer rel, an inner rel, and a join
type. It can modify the pgs_mask propagated into JoinPathExtraData to
restrict strategy choice for that particular combination of rels.
To make joinrel_setup_hook work as intended, the existing calls to
build_joinrel_partition_info are moved later in the calling functions;
this is because that function checks whether the rel's pgs_mask includes
PGS_CONSIDER_PARTITIONWISE, so we want it to only be called after
plugins have had a chance to alter pgs_mask.
Upper rels currently inherit pgs_mask from the input relation. It's
unclear that this is the most useful behavior, but at the moment there
are no hooks to allow the mask to be set in any other way.
Reviewed-by: Lukas Fittl <lukas@fittl.com>
Reviewed-by: Jakub Wartak <jakub.wartak@enterprisedb.com>
Reviewed-by: Greg Burd <greg@burd.me>
Reviewed-by: Jacob Champion <jacob.champion@enterprisedb.com>
Reviewed-by: Amit Langote <amitlangote09@gmail.com>
Reviewed-by: Haibo Yan <tristan.yim@gmail.com>
Discussion: http://postgr.es/m/CA+TgmoZ-Jh1T6QyWoCODMVQdhTUPYkaZjWztzP1En4=ZHoKPzw@mail.gmail.com
Liujinyang reported the one in binaryheap.c, I then found and analyzed
the rest.
For future patches, we require git archaelogical analysis before we
accept patches of this nature.
Co-authored-by: liujinyang <21043272@qq.com>
Co-authored-by: Álvaro Herrera <alvherre@kurilemu.de>
Discussion: https://postgr.es/m/tencent_6B302BFCAF6F010E00AB5C2C0ECB7AA3F205@qq.com
RIGHT_SEMI joins rely on the HEAP_TUPLE_HAS_MATCH flag to guarantee
that only the first match for each inner tuple is considered.
However, in a parallel hash join, the inner relation is stored in a
shared global hash table that can be probed by multiple workers
concurrently. This allows different workers to inspect and set the
match flags of the same inner tuples at the same time.
If two workers probe the same inner tuple concurrently, both may see
the match flag as unset and emit the same tuple, leading to duplicate
output rows and violating RIGHT_SEMI join semantics.
For now, we disable parallel plans for RIGHT_SEMI joins. In the long
term, it may be possible to support parallel execution by performing
atomic operations on the match flag, for example using a CAS or
similar mechanism.
Backpatch to v18, where RIGHT_SEMI join was introduced.
Bug: #19094
Reported-by: Lori Corbani <Lori.Corbani@jax.org>
Diagnosed-by: Tom Lane <tgl@sss.pgh.pa.us>
Author: Richard Guo <guofenglinux@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/19094-6ed410eb5b256abd@postgresql.org
Backpatch-through: 18
There are two implementation techniques for semijoins: one uses the
JOIN_SEMI jointype, where the executor emits at most one matching row
per left-hand side (LHS) row; the other unique-ifies the right-hand
side (RHS) and then performs a plain inner join.
The latter technique currently has some drawbacks related to the
unique-ification step.
* Only the cheapest-total path of the RHS is considered during
unique-ification. This may cause us to miss some optimization
opportunities; for example, a path with a better sort order might be
overlooked simply because it is not the cheapest in total cost. Such
a path could help avoid a sort at a higher level, potentially
resulting in a cheaper overall plan.
* We currently rely on heuristics to choose between hash-based and
sort-based unique-ification. A better approach would be to generate
paths for both methods and allow add_path() to decide which one is
preferable, consistent with how path selection is handled elsewhere in
the planner.
* In the sort-based implementation, we currently pay no attention to
the pathkeys of the input subpath or the resulting output. This can
result in redundant sort nodes being added to the final plan.
This patch improves semijoin planning by creating a new RelOptInfo for
the RHS rel to represent its unique-ified version. It then generates
multiple paths that represent elimination of distinct rows from the
RHS, considering both a hash-based implementation using the cheapest
total path of the original RHS rel, and sort-based implementations
that either exploit presorted input paths or explicitly sort the
cheapest total path. All resulting paths compete in add_path(), and
those deemed worthy of consideration are added to the new RelOptInfo.
Finally, the unique-ified rel is joined with the other side of the
semijoin using a plain inner join.
As a side effect, most of the code related to the JOIN_UNIQUE_OUTER
and JOIN_UNIQUE_INNER jointypes -- used to indicate that the LHS or
RHS path should be made unique -- has been removed. Besides, the
T_Unique path now has the same meaning for both semijoins and upper
DISTINCT clauses: it represents adjacent-duplicate removal on
presorted input. This patch unifies their handling by sharing the
same data structures and functions.
This patch also removes the UNIQUE_PATH_NOOP related code along the
way, as it is dead code -- if the RHS rel is provably unique, the
semijoin should have already been simplified to a plain inner join by
analyzejoins.c.
Author: Richard Guo <guofenglinux@gmail.com>
Reviewed-by: Alexandra Wang <alexandra.wang.oss@gmail.com>
Reviewed-by: wenhui qiu <qiuwenhuifx@gmail.com>
Discussion: https://postgr.es/m/CAMbWs4-EBnaRvEs7frTLbsXiweSTUXifsteF-d3rvv01FKO86w@mail.gmail.com
Currently, we do not support Memoize for SEMI and ANTI joins because
nested loop SEMI/ANTI joins do not scan the inner relation to
completion, which prevents Memoize from marking the cache entry as
complete. One might argue that we could mark the cache entry as
complete after fetching the first inner tuple, but that would not be
safe: if the first inner tuple and the current outer tuple do not
satisfy the join clauses, a second inner tuple matching the parameters
would find the cache entry already marked as complete.
However, if the inner side is provably unique, this issue doesn't
arise, since there would be no second matching tuple. That said, this
doesn't help in the case of SEMI joins, because a SEMI join with a
provably unique inner side would already have been reduced to an inner
join by reduce_unique_semijoins.
Therefore, in this patch, we check whether the inner relation is
provably unique for ANTI joins and enable the use of Memoize in such
cases.
Author: Richard Guo <guofenglinux@gmail.com>
Reviewed-by: wenhui qiu <qiuwenhuifx@gmail.com>
Reviewed-by: Andrei Lepikhov <lepihov@gmail.com>
Discussion: https://postgr.es/m/CAMbWs48FdLiMNrmJL-g6mDvoQVt0yNyJAqMkv4e2Pk-5GKCZLA@mail.gmail.com
Commit 85e5e222b, which added (a forerunner of) this logic,
argued that
Adding the necessary complexity to make this work doesn't seem like
it would be repaid in significantly better plans, because in cases
where such a PHV exists, there is probably a corresponding join order
constraint that would allow a good plan to be found without using the
star-schema exception.
The flaw in this claim is that there may be other join-order
restrictions that prevent us from finding a join order that doesn't
involve a "dangerous" PHV. In particular we now recognize that
small join_collapse_limit or from_collapse_limit could prevent it.
Therefore, let's bite the bullet and make the case work.
We don't have to extend the executor's support for nestloop parameters
as I thought at the time, because we can instead push the evaluation
of the placeholder's expression into the left-hand input of the
NestLoop node. So there's not really a lot of downside to this
solution, and giving the planner more join-order flexibility should
have value beyond just avoiding failure.
Having said that, there surely is a nonzero risk of introducing
new bugs. Since this failure mode escaped detection for ten years,
such cases don't seem common enough to justify a lot of risk.
Therefore, let's put this fix into master but leave the back branches
alone (for now anyway).
Bug: #18953
Reported-by: Alexander Lakhin <exclusion@gmail.com>
Diagnosed-by: Richard Guo <guofenglinux@gmail.com>
Author: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/18953-1c9883a9d4afeb30@postgresql.org
When creating an explicit Sort node for the outer path of a mergejoin,
we need to determine the number of presorted keys of the outer path to
decide whether explicit incremental sort can be applied. Currently,
this is done by repeatedly calling pathkeys_count_contained_in.
This patch caches the number of presorted outer pathkeys in MergePath,
allowing us to save several calls to pathkeys_count_contained_in. It
can be considered a complement to the changes in commit 828e94c9d.
Reported-by: David Rowley <dgrowleyml@gmail.com>
Author: Richard Guo <guofenglinux@gmail.com>
Reviewed-by: Tender Wang <tndrwang@gmail.com>
Discussion: https://postgr.es/m/CAApHDvqvBireB_w6x8BN5txdvBEHxVgZBt=rUnpf5ww5P_E_ww@mail.gmail.com
Memoize typically marks cache entries as complete after fully scanning
the inner side of a join. However, in the case of unique joins, we
skip to the next outer tuple as soon as the first matching inner tuple
is found, leaving no opportunity to scan the inner side to completion.
To work around that, we mark cache entries as complete after fetching
the first matching inner tuple in unique joins.
This approach is only safe when all of the join's restriction clauses
are parameterized; otherwise, there is no guarantee that reading just
one tuple from the inner side is sufficient.
Currently, we check for this by verifying that the number of clauses
in ppi_clauses is no less than the number of the join's restriction
clauses. However, this check isn't entirely reliable, as ppi_clauses
includes join clauses available from all outer rels, not just the
current outer rel. This means the check could pass even if a
restriction clause isn't parameterized, as long as another join
clause, which doesn't belong to the current join, is included in
ppi_clauses.
To fix this, we explicitly check whether each restriction clause of
the current join is present in ppi_clauses.
While we're here, remove the XXX comment from the modified code, as
it's not justified; in certain cases, it's not possible to move a join
clause to the inner side.
This is arguably a bugfix, but no backpatch given the lack of field
reports.
Author: Richard Guo <guofenglinux@gmail.com>
Reviewed-by: wenhui qiu <qiuwenhuifx@gmail.com>
Reviewed-by: Andrei Lepikhov <lepihov@gmail.com>
Discussion: https://postgr.es/m/CAMbWs4-8JPouj=wBDj4DhK-WO4+Xdx=A2jbjvvyyTBQneJ1=BQ@mail.gmail.com
Two near-identical copies of clause_sides_match_join() existed in
joinpath.c and analyzejoins.c. Deduplicate this by moving the function
into restrictinfo.h.
It isn't quite clear that keeping the inline property of this function
is worthwhile, but this commit is just an exercise in code
deduplication. More effort would be required to determine if the inline
property is worth keeping.
Author: James Hunter <james.hunter.pg@gmail.com>
Discussion: https://postgr.es/m/CAJVSvF7Nm_9kgMLOch4c-5fbh3MYg%3D9BdnDx3Dv7Fcb64zr64Q%40mail.gmail.com
When creating merge or hash join plans in createplan.c, the merge or
hash clauses may need to get commuted to ensure that the outer var is
on the left and the inner var is on the right if they are not already
in the expected form. This requires that their operators have
commutators. Failing to find a commutator at this stage would result
in 'ERROR: could not find commutator for operator xxx', with no
opportunity to select an alternative plan.
Typically, this is not an issue because mergejoinable or hashable
operators are expected to always have valid commutators. But in some
artificial cases this assumption may not hold true. Therefore, here
in this patch we check the validity of commutators for clauses in the
form "inner op outer" when selecting mergejoin/hash clauses, and
consider a clause unusable for the current pair of outer and inner
relations if it lacks a commutator.
There are not (and should not be) any such operators built into
Postgres that are mergejoinable or hashable but have no commutators;
so we leverage the alias type 'int8alias1' created in equivclass.sql
to build the test case. This is why the test case is included in
equivclass.sql rather than in join.sql.
Although this is arguably a bug fix, it cannot be reproduced without
installing an incomplete opclass, which is unlikely to happen in
practice, so no back-patch.
Reported-by: Alexander Pyhalov
Author: Richard Guo
Reviewed-by: Tom Lane
Discussion: https://postgr.es/m/c59ec04a2fef94d9ffc35a9b17dfc081@postgrespro.ru
Previously, when a path type was disabled by e.g. enable_seqscan=false,
we either avoided generating that path type in the first place, or
more commonly, we added a large constant, called disable_cost, to the
estimated startup cost of that path. This latter approach can distort
planning. For instance, an extremely expensive non-disabled path
could seem to be worse than a disabled path, especially if the full
cost of that path node need not be paid (e.g. due to a Limit).
Or, as in the regression test whose expected output changes with this
commit, the addition of disable_cost can make two paths that would
normally be distinguishible in cost seem to have fuzzily the same cost.
To fix that, we now count the number of disabled path nodes and
consider that a high-order component of both the startup cost and the
total cost. Hence, the path list is now sorted by disabled_nodes and
then by total_cost, instead of just by the latter, and likewise for
the partial path list. It is important that this number is a count
and not simply a Boolean; else, as soon as we're unable to respect
disabled path types in all portions of the path, we stop trying to
avoid them where we can.
Because the path list is now sorted by the number of disabled nodes,
the join prechecks must compute the count of disabled nodes during
the initial cost phase instead of postponing it to final cost time.
Counts of disabled nodes do not cross subquery levels; at present,
there is no reason for them to do so, since the we do not postpone
path selection across subquery boundaries (see make_subplan).
Reviewed by Andres Freund, Heikki Linnakangas, and David Rowley.
Discussion: http://postgr.es/m/CA+TgmoZ_+MS+o6NeGK2xyBv-xM+w1AfFVuHE4f_aq6ekHv7YSQ@mail.gmail.com
Parameterized partial paths are not supported, and we have several
checks in try_partial_xxx_path functions to enforce this. For a
partial nestloop join path, we need to ensure that if the inner path
is parameterized, the parameterization is fully satisfied by the
proposed outer path. For a partial merge/hashjoin join path, we need
to ensure that the inner path is not parameterized. In all cases, we
need to ensure that the outer path is not parameterized.
However, the comment in try_partial_hashjoin_path does not describe
this correctly. This patch fixes that.
In addtion, this patch simplifies the checks peformed in
try_partial_hashjoin_path and try_partial_mergejoin_path with the help
of macro PATH_REQ_OUTER, and also adds asserts that the outer path is
not parameterized in try_partial_xxx_path functions.
Author: Richard Guo
Discussion: https://postgr.es/m/CAMbWs48mKJ6g_GnYNa7dnw04MHaMK-jnAEBrMVhTp2uUg3Ut4A@mail.gmail.com
In sort_inner_and_outer, we create mergejoin join paths by explicitly
sorting both relations on each possible ordering of the available
mergejoin clauses. However, if there are no available mergejoin
clauses, we can skip this process entirely.
This patch introduces a check for mergeclause_list at the beginning of
sort_inner_and_outer and exits the function if it is found to be
empty. This might help skip all the statements that come before the
call to select_outer_pathkeys_for_merge, including the build of
UniquePaths in the case of JOIN_UNIQUE_OUTER or JOIN_UNIQUE_INNER.
I doubt there's any measurable performance improvement, but throughout
the run of the regression tests, sort_inner_and_outer is called a
total of 44,424 times. Among these calls, there are 11,064 instances
where mergeclause_list is found to be empty, which accounts for
approximately one-fourth. I think this suggests that implementing
this shortcut is worthwhile.
Author: Richard Guo
Reviewed-by: Ashutosh Bapat
Discussion: https://postgr.es/m/CAMbWs48RKiZGFEd5A0JtztRY5ZdvVvNiHh0AKeuoz21F+0dVjQ@mail.gmail.com
If we intend to generate a Memoize node on top of a path, we need
cache keys of some sort. Currently we search for the cache keys in
the parameterized clauses of the path as well as the lateral_vars of
its parent. However, it turns out that this is not sufficient because
there might be lateral references derived from PlaceHolderVars, which
we fail to take into consideration.
This oversight can cause us to miss opportunities to utilize the
Memoize node. Moreover, in some plans, failing to recognize all the
cache keys could result in performance regressions. This is because
without identifying all the cache keys, we would need to purge the
entire cache every time we get a new outer tuple during execution.
This patch fixes this issue by extracting lateral Vars from within
PlaceHolderVars and subsequently including them in the cache keys.
In passing, this patch also includes a comment clarifying that Memoize
nodes are currently not added on top of join relation paths. This
explains why this patch only considers PlaceHolderVars that are due to
be evaluated at baserels.
Author: Richard Guo
Reviewed-by: Tom Lane, David Rowley, Andrei Lepikhov
Discussion: https://postgr.es/m/CAMbWs48jLxn0pAPZpJ50EThZ569Xrw+=4Ac3QvkpQvNszbeoNg@mail.gmail.com
When generating non-parallel nestloop paths for each available outer
path, we always consider materializing the cheapest inner path if
feasible. Similarly, in this patch, we also consider materializing
the cheapest inner path when building partial nestloop paths. This
approach potentially reduces the need to rescan the inner side of a
partial nestloop path for each outer tuple.
Author: Tender Wang
Reviewed-by: Richard Guo, Robert Haas, David Rowley, Alena Rybakina
Reviewed-by: Tomasz Rybak, Paul Jungwirth, Yuki Fujii
Discussion: https://postgr.es/m/CAHewXNkPmtEXNfVQMou_7NqQmFABca9f4etjBtdbbm0ZKDmWvw@mail.gmail.com
Hash joins can support semijoin with the LHS input on the right, using
the existing logic for inner join, combined with the assurance that only
the first match for each inner tuple is considered, which can be
achieved by leveraging the HEAP_TUPLE_HAS_MATCH flag. This can be very
useful in some cases since we may now have the option to hash the
smaller table instead of the larger.
Merge join could likely support "Right Semi Join" too. However, the
benefit of swapping inputs tends to be small here, so we do not address
that in this patch.
Note that this patch also modifies a test query in join.sql to ensure it
continues testing as intended. With this patch the original query would
result in a right-semi-join rather than semi-join, compromising its
original purpose of testing the fix for neqjoinsel's behavior for
semi-joins.
Author: Richard Guo
Reviewed-by: wenhui qiu, Alena Rybakina, Japin Li
Discussion: https://postgr.es/m/CAMbWs4_X1mN=ic+SxcyymUqFx9bB8pqSLTGJ-F=MHy4PW3eRXw@mail.gmail.com
When considering nestloop paths for individual partitions within
a partitionwise join, if the inner path is parameterized, it is
parameterized by the topmost parent of the outer rel, not the
corresponding outer rel itself. Therefore, we need to translate the
parameterization so that the inner path is parameterized by the
corresponding outer rel.
Up to now, we did this while generating join paths. However, that's
problematic because we must also translate some expressions that are
shared across all paths for a relation, such as restriction clauses
(kept in the RelOptInfo and/or IndexOptInfo) and TableSampleClauses
(kept in the RangeTblEntry). The existing code fails to translate
these at all, leading to wrong answers, odd failures such as
"variable not found in subplan target list", or executor crashes.
But we can't modify them during path generation, because that would
break things if we end up choosing some non-partitioned-join path.
So this patch postpones reparameterization of the inner path until
createplan.c, where it is safe to modify the referenced RangeTblEntry,
RelOptInfo or IndexOptInfo, because we have made a final choice of which
Path to use. We do still have to check during path generation that
the reparameterization will be possible. So we introduce a new
function path_is_reparameterizable_by_child() to detect that.
The duplication between path_is_reparameterizable_by_child() and
reparameterize_path_by_child() is a bit annoying, but there seems
no other good answer. A small benefit is that we can avoid building
useless reparameterized trees in cases where a non-partitioned join
is ultimately chosen. Also, reparameterize_path_by_child() can now
be allowed to scribble on the input paths, saving a few cycles.
This fix repairs the same problems previously addressed in the
back branches by commits 62f120203 et al.
Richard Guo, reviewed at various times by Ashutosh Bapat, Andrei
Lepikhov, Alena Rybakina, Robert Haas, and myself
Discussion: https://postgr.es/m/CAMbWs496+N=UAjOc=rcD3P7B6oJe4rZw08e_TZRUsWbPxZW3Tw@mail.gmail.com
It was possible when determining the cache keys for a Memoize path that
if the same expr appeared twice in the parameterized path's ppi_clauses
and/or in the Nested Loop's inner relation's lateral_vars. If this
happened the Memoize node's cache keys would contain duplicates. This
isn't a problem for correctness, all it means is that the cache lookups
will be suboptimal due to having redundant work to do on every hash table
lookup and insert.
Here we adjust paraminfo_get_equal_hashops() to look for duplicates and
ignore them when we find them.
Author: David Rowley
Reviewed-by: Richard Guo
Discussion: https://postgr.es/m/422277.1706207562%40sss.pgh.pa.us
This was previously fixed in 9e215378d but got broken again as a result
of 2489d76c4. It seems that commit causes ppi_clauses to contain
duplicate clauses and it's no longer safe to check the list_length of
that list to determine if there are join conditions other than what's
mentioned in ppi_clauses.
Here we adjust the check to count the distinct rinfo_serial mentioned in
ppi_clauses. We expect that extra->restrictlist won't have duplicate
rinfo_serials.
Reported-by: Amadeo Gallardo
Author: Richard Guo
Discussion: https://postgr.es/m/CADFREbW-BLJd7-a5J%2B5wjVumeFG1ByXiSOFzMtkmY_SDWckTxw%40mail.gmail.com
Backpatch-through: 16, where 2489d76c4 was introduced.
These were not testing the same thing as the comparable Assert
in calc_nestloop_required_outer(), because we neglected to map
the given Paths' relids to top-level relids. When considering
a partition child join the latter is the correct thing to do.
This oversight is old, but since it's only an overly-weak Assert
check there doesn't seem to be much value in back-patching.
Richard Guo (with cosmetic changes and comment updates by me)
Discussion: https://postgr.es/m/CAMbWs49sqbe9GBZ8sy8dSfKRNURgicR85HX8vgzcgQsPF0XY1w@mail.gmail.com
The comment introduced by commit e7cb7ee14 was a bit too terse, which
could lead to extensions doing different things within the hook function
than we intend to allow. Extend the comment to explain what they can do
within the hook function.
Back-patch to all supported branches.
In passing, I rephrased a nearby comment that I recently added to the
back branches.
Reviewed by David Rowley and Andrei Lepikhov.
Discussion: https://postgr.es/m/CAPmGK15SBPA1nr3Aqsdm%2BYyS-ay0Ayo2BRYQ8_A2To9eLqwopQ%40mail.gmail.com
This was disabled in commit 6f80a8d9c due to the lack of support for
handling of pseudoconstant quals assigned to replaced joins in
createplan.c. To re-allow it, this patch adds the support by 1)
modifying the ForeignPath and CustomPath structs so that if they
represent foreign and custom scans replacing a join with a scan, they
store the list of RestrictInfo nodes to apply to the join, as in
JoinPaths, and by 2) modifying create_scan_plan() in createplan.c so
that it uses that list in that case, instead of the baserestrictinfo
list, to get pseudoconstant quals assigned to the join, as mentioned in
the commit message for that commit.
Important item for the release notes: this is non-backwards-compatible
since it modifies the ForeignPath and CustomPath structs, as mentioned
above, and changes the argument lists for FDW helper functions
create_foreignscan_path(), create_foreign_join_path(), and
create_foreign_upper_path().
Richard Guo, with some additional changes by me, reviewed by Nishant
Sharma, Suraj Kharage, and Richard Guo.
Discussion: https://postgr.es/m/CADrsxdbcN1vejBaf8a%2BQhrZY5PXL-04mCd4GDu6qm6FigDZd6Q%40mail.gmail.com
The use of Memoize was already disabled in normal joins when the join
conditions had volatile functions per the code in
match_opclause_to_indexcol(). Ordinarily, the parameterization for the
inner side of a nested loop will be an Index Scan or at least eventually
lead to an index scan (perhaps nested several joins deep). However, for
lateral joins, that's not the case and seq scans can be parameterized
too, so we can't rely on match_opclause_to_indexcol().
Here we explicitly check the parameterization for volatile functions and
don't consider the generation of a Memoize path when such functions
are present.
Author: Richard Guo
Discussion: https://postgr.es/m/CAMbWs49nHFnHbpepLsv_yF3qkpCS4BdB-v8HoJVv8_=Oat0u_w@mail.gmail.com
Backpatch-through: 14, where Memoize was introduced
Commit e7cb7ee14, which introduced the infrastructure for FDWs and
custom scan providers to replace joins with scans, failed to add support
handling of pseudoconstant quals assigned to replaced joins in
createplan.c, leading to an incorrect plan without a gating Result node
when postgres_fdw replaced a join with such a qual.
To fix, we could add the support by 1) modifying the ForeignPath and
CustomPath structs to store the list of RestrictInfo nodes to apply to
the join, as in JoinPaths, if they represent foreign and custom scans
replacing a join with a scan, and by 2) modifying create_scan_plan() in
createplan.c to use that list in that case, instead of the
baserestrictinfo list, to get pseudoconstant quals assigned to the join;
but #1 would cause an ABI break. So fix by modifying the infrastructure
to just disallow replacing joins with such quals.
Back-patch to all supported branches.
Reported by Nishant Sharma. Patch by me, reviewed by Nishant Sharma and
Richard Guo.
Discussion: https://postgr.es/m/CADrsxdbcN1vejBaf8a%2BQhrZY5PXL-04mCd4GDu6qm6FigDZd6Q%40mail.gmail.com
An outer join cannot be formed using an input path that is parameterized
by a value that is supposed to be nulled by the outer join. This is
obviously nonsensical, and it could lead to a bad plan being selected;
although currently it seems that we'll hit various sanity-check
assertions first.
I think that such cases were formerly prevented by the delay_upper_joins
mechanism, but now that that's gone we need an explicit check.
(Perhaps we should avoid generating baserel paths that could
lead to this situation in the first place; but it seems like
having a defense at the join level would be a good idea anyway.)
Richard Guo and Tom Lane, per report from Jaime Casanova
Discussion: https://postgr.es/m/CAJKUy5g2uZRrUDZJ8p-=giwcSHVUn0c9nmdxPSY0jF0Ov8VoEA@mail.gmail.com
It turns out that the fixes we applied in commits bfd332b3f
and 63e4f13d2 were not nearly enough to solve the problem.
We'd focused narrowly on subquery RTEs with lateral references,
but lateral references can occur in several other RTE kinds
such as function RTEs. Putting the same hack into half a dozen
code paths seems quite unattractive. Hence, revert the code changes
(but not the test cases) from those commits and instead solve it
centrally in identify_current_nestloop_params(), as Richard proposed
originally. This is a bit annoying because it could mask erroneous
nullingrels in nestloop params that are generated from non-LATERAL
parameterized paths; but on balance I don't see a better way.
Maybe at some future time we'll be motivated to find a more rigorous
approach to nestloop params, but that's not happening for beta2.
Richard Guo and Tom Lane
Discussion: https://postgr.es/m/CAMbWs48Jcw-NvnxT23WiHP324wG44DvzcH1j4hc0Zn+3sR9cfg@mail.gmail.com
The issue fixed in commit bfd332b3f can also bite Memoize plans,
because of the separate copies of lateral reference Vars made
by paraminfo_get_equal_hashops. Apply the same hacky fix there.
(In passing, clean up shaky grammar in the existing comments
for this function.)
Richard Guo
Discussion: https://postgr.es/m/CAMbWs4-krwk0Wbd6WdufMAupuou_Ua73ijQ4XQCr1Mb5BaVtKQ@mail.gmail.com
Merge and hash joins can support antijoin with the non-nullable input
on the right, using very simple combinations of their existing logic
for right join and anti join. This gives the planner more freedom
about how to order the join. It's particularly useful for hash join,
since we may now have the option to hash the smaller table instead
of the larger.
Richard Guo, reviewed by Ronan Dunklau and myself
Discussion: https://postgr.es/m/CAMbWs48xh9hMzXzSy3VaPzGAz+fkxXXTUbCLohX1_L8THFRm2Q@mail.gmail.com
Full and right outer joins were not supported in the initial
implementation of Parallel Hash Join because of deadlock hazards (see
discussion). Therefore FULL JOIN inhibited parallelism, as the other
join strategies can't do that in parallel either.
Add a new PHJ phase PHJ_BATCH_SCAN that scans for unmatched tuples on
the inner side of one batch's hash table. For now, sidestep the
deadlock problem by terminating parallelism there. The last process to
arrive at that phase emits the unmatched tuples, while others detach and
are free to go and work on other batches, if there are any, but
otherwise they finish the join early.
That unfairness is considered acceptable for now, because it's better
than no parallelism at all. The build and probe phases are run in
parallel, and the new scan-for-unmatched phase, while serial, is usually
applied to the smaller of the two relations and is either limited by
some multiple of work_mem, or it's too big and is partitioned into
batches and then the situation is improved by batch-level parallelism.
Author: Melanie Plageman <melanieplageman@gmail.com>
Author: Thomas Munro <thomas.munro@gmail.com>
Reviewed-by: Thomas Munro <thomas.munro@gmail.com>
Discussion: https://postgr.es/m/CA%2BhUKG%2BA6ftXPz4oe92%2Bx8Er%2BxpGZqto70-Q_ERwRaSyA%3DafNg%40mail.gmail.com
Late in the development of commit 2489d76c4, I (tgl) incorrectly
concluded that the new function have_unsafe_outer_join_ref couldn't
ever reach its inner loop. That should be the case if the inner
rel's parameterization is based on just one Var, but it could be
based on Vars from several relations, and then not only is the
inner loop reachable but it's wrongly coded.
Despite those errors, it still appears that the whole thing is
redundant given previous join_is_legal checks, so let's arrange
to only run it in assert-enabled builds.
Diagnosis and patch by Richard Guo, per fuzz testing by Justin Pryzby.
Discussion: https://postgr.es/m/20230212235823.GW1653@telsasoft.com
EquivalenceClasses are now understood as applying within a "join
domain", which is a set of inner-joined relations (possibly underneath
an outer join). We no longer need to treat an EC from below an outer
join as a second-class citizen.
I have hopes of eventually being able to treat outer-join clauses via
EquivalenceClasses, by means of only applying deductions within the
EC's join domain. There are still problems in the way of that, though,
so for now the reconsider_outer_join_clause logic is still here.
I haven't been able to get rid of RestrictInfo.is_pushed_down either,
but I wonder if that could be recast using JoinDomains.
I had to hack one test case in postgres_fdw.sql to make it still test
what it was meant to, because postgres_fdw is inconsistent about
how it deals with quals containing non-shippable expressions; see
https://postgr.es/m/1691374.1671659838@sss.pgh.pa.us. That should
be improved, but I don't think it's within the scope of this patch
series.
Patch by me; thanks to Richard Guo for review.
Discussion: https://postgr.es/m/830269.1656693747@sss.pgh.pa.us
Traditionally we used the same Var struct to represent the value
of a table column everywhere in parse and plan trees. This choice
predates our support for SQL outer joins, and it's really a pretty
bad idea with outer joins, because the Var's value can depend on
where it is in the tree: it might go to NULL above an outer join.
So expression nodes that are equal() per equalfuncs.c might not
represent the same value, which is a huge correctness hazard for
the planner.
To improve this, decorate Var nodes with a bitmapset showing
which outer joins (identified by RTE indexes) may have nulled
them at the point in the parse tree where the Var appears.
This allows us to trust that equal() Vars represent the same value.
A certain amount of klugery is still needed to cope with cases
where we re-order two outer joins, but it's possible to make it
work without sacrificing that core principle. PlaceHolderVars
receive similar decoration for the same reason.
In the planner, we include these outer join bitmapsets into the relids
that an expression is considered to depend on, and in consequence also
add outer-join relids to the relids of join RelOptInfos. This allows
us to correctly perceive whether an expression can be calculated above
or below a particular outer join.
This change affects FDWs that want to plan foreign joins. They *must*
follow suit when labeling foreign joins in order to match with the
core planner, but for many purposes (if postgres_fdw is any guide)
they'd prefer to consider only base relations within the join.
To support both requirements, redefine ForeignScan.fs_relids as
base+OJ relids, and add a new field fs_base_relids that's set up by
the core planner.
Large though it is, this commit just does the minimum necessary to
install the new mechanisms and get check-world passing again.
Follow-up patches will perform some cleanup. (The README additions
and comments mention some stuff that will appear in the follow-up.)
Patch by me; thanks to Richard Guo for review.
Discussion: https://postgr.es/m/830269.1656693747@sss.pgh.pa.us
A couple of places weren't up to speed for this. By sheer good
luck, we didn't fail but just selected a non-memoized join plan,
at least in the test case we have. Nonetheless, it's a bug,
and I'm not quite sure that it couldn't have worse consequences
in other examples. So back-patch to v14 where Memoize came in.
Richard Guo
Discussion: https://postgr.es/m/CAMbWs48GkNom272sfp0-WeD6_0HSR19BJ4H1c9ZKSfbVnJsvRg@mail.gmail.com
In order to estimate the cache hit ratio of a Memoize node, one of the
inputs we require is the estimated number of times the Memoize node will
be rescanned. The higher this number, the large the cache hit ratio is
likely to become. Unfortunately, the value being passed as the number of
"calls" to the Memoize was incorrectly using the Nested Loop's
outer_path->parent->rows instead of outer_path->rows. This failed to
account for the fact that the outer_path might be parameterized by some
upper-level Nested Loop.
This problem could lead to Memoize plans appearing more favorable than
they might actually be. It could also lead to extended executor startup
times when work_mem values were large due to the planner setting overly
large MemoizePath->est_entries resulting in the Memoize hash table being
initially made much larger than might be required.
Fix this simply by passing outer_path->rows rather than
outer_path->parent->rows. Also, adjust the expected regression test
output for a plan change.
Reported-by: Pavel Stehule
Author: David Rowley
Discussion: https://postgr.es/m/CAFj8pRAMp%3DQsMi6sPQJ4W3hczoFJRvyXHJV3AZAZaMyTVM312Q%40mail.gmail.com
Backpatch-through: 14, where Memoize was introduced
In sort_inner_and_outer we iterate a list of PathKey elements, but the
variable is declared as (List *). This mistake is benign, because we
only pass the pointer to lcons() and never dereference it.
This exists since ~2004, but it's confusing. So fix and backpatch to all
supported branches.
Backpatch-through: 10
Discussion: https://postgr.es/m/bf3a6ea1-a7d8-7211-0669-189d5c169374%40enterprisedb.com
Memoize would always use the hash equality operator for the cache key
types to determine if the current set of parameters were the same as some
previously cached set. Certain types such as floating points where -0.0
and +0.0 differ in their binary representation but are classed as equal by
the hash equality operator may cause problems as unless the join uses the
same operator it's possible that whichever join operator is being used
would be able to distinguish the two values. In which case we may
accidentally return in the incorrect rows out of the cache.
To fix this here we add a binary mode to Memoize to allow it to the
current set of parameters to previously cached values by comparing
bit-by-bit rather than logically using the hash equality operator. This
binary mode is always used for LATERAL joins and it's used for normal
joins when any of the join operators are not hashable.
Reported-by: Tom Lane
Author: David Rowley
Discussion: https://postgr.es/m/3004308.1632952496@sss.pgh.pa.us
Backpatch-through: 14, where Memoize was added
In v14, because we don't have a field in RestrictInfo to cache both the
left and right type's hash equality operator, we just restrict the scope
of Memoize to only when the left and right types of a RestrictInfo are the
same.
In master we add another field to RestrictInfo and cache both hash
equality operators.
Reported-by: Jaime Casanova
Author: David Rowley
Discussion: https://postgr.es/m/20210929185544.GB24346%40ahch-to
Backpatch-through: 14
"Result Cache" was never a great name for this node, but nobody managed
to come up with another name that anyone liked enough. That was until
David Johnston mentioned "Node Memoization", which Tom Lane revised to
just "Memoize". People seem to like "Memoize", so let's do the rename.
Reviewed-by: Justin Pryzby
Discussion: https://postgr.es/m/20210708165145.GG1176@momjian.us
Backpatch-through: 14, where Result Cache was introduced
Code added in 9e215378d to disable building of Result Cache paths when
not all join conditions are part of the parameterization of a unique
join failed to first check if the inner path's param_info was set before
checking the param_info's ppi_clauses.
Add a check for NULL values here and just bail on trying to build the
path if param_info is NULL. lateral_vars are not considered when
deciding if the join is unique, so we're not missing out on doing the
optimization when there are lateral_vars and no param_info.
Reported-by: Coverity, via Tom Lane
Discussion: https://postgr.es/m/457998.1621779290@sss.pgh.pa.us
When the planner considered using a Result Cache node to cache results
from the inner side of a Nested Loop Join, it failed to consider that the
inner path's parameterization may not be the entire join condition. If
the join was marked as inner_unique then we may accidentally put the cache
in singlerow mode. This meant that entries would be marked as complete
after caching the first row. That was wrong as if only part of the join
condition was parameterized then the uniqueness of the unique join was not
guaranteed at the Result Cache's level. The uniqueness is only guaranteed
after Nested Loop applies the join filter. If subsequent rows were found,
this would lead to:
ERROR: cache entry already complete
This could have been fixed by only putting the cache in singlerow mode if
the entire join condition was parameterized. However, Nested Loop will
only read its inner side so far as the first matching row when the join is
unique, so that might mean we never get an opportunity to mark cache
entries as complete. Since non-complete cache entries are useless for
subsequent lookups, we just don't bother considering a Result Cache path
in this case.
In passing, remove the XXX comment that claimed the above ERROR might be
better suited to be an Assert. After there being an actual case which
triggered it, it seems better to keep it an ERROR.
Reported-by: David Christensen
Discussion: https://postgr.es/m/CAOxo6X+dy-V58iEPFgst8ahPKEU+38NZzUuc+a7wDBZd4TrHMQ@mail.gmail.com
