Now that the upper planner uses paths, and now that we have proper hooks
to inject paths into the upper planning process, it's possible for
foreign data wrappers to arrange to push aggregates to the remote side
instead of fetching all of the rows and aggregating them locally. This
figures to be a massive win for performance, so teach postgres_fdw to
do it.
Jeevan Chalke and Ashutosh Bapat. Reviewed by Ashutosh Bapat with
additional testing by Prabhat Sahu. Various mostly cosmetic changes
by me.
You can use ALTER FOREIGN TABLE SET WITH OIDS on a foreign table, but the
oid column read out as zeros, because the postgres_fdw didn't know about
it. Teach postgres_fdw how to fetch it.
Etsuro Fujita, with an additional test case by me.
Discussion: <56E90A76.5000503@lab.ntt.co.jp>
Commits 4452000f3 et al established semantics for NullTest.argisrow that
are a bit different from its initial conception: rather than being merely
a cache of whether we've determined the input to have composite type,
the flag now has the further meaning that we should apply field-by-field
testing as per the standard's definition of IS [NOT] NULL. If argisrow
is false and yet the input has composite type, the construct instead has
the semantics of IS [NOT] DISTINCT FROM NULL. Update the comments in
primnodes.h to clarify this, and fix ruleutils.c and deparse.c to print
such cases correctly. In the case of ruleutils.c, this merely results in
cosmetic changes in EXPLAIN output, since the case can't currently arise
in stored rules. However, it represents a live bug for deparse.c, which
would formerly have sent a remote query that had semantics different
from the local behavior. (From the user's standpoint, this means that
testing a remote nested-composite column for null-ness could have had
unexpected recursive behavior much like that fixed in 4452000f3.)
In a related but somewhat independent fix, make plancat.c set argisrow
to false in all NullTest expressions constructed to represent "attnotnull"
constructs. Since attnotnull is actually enforced as a simple null-value
check, this is a more accurate representation of the semantics; we were
previously overpromising what it meant for composite columns, which might
possibly lead to incorrect planner optimizations. (It seems that what the
SQL spec expects a NOT NULL constraint to mean is an IS NOT NULL test, so
arguably we are violating the spec and should fix attnotnull to do the
other thing. If we ever do, this part should get reverted.)
Back-patch, same as the previous commit.
Discussion: <10682.1469566308@sss.pgh.pa.us>
As pointed out by Ashutosh Bapat, the header comments for this file
say that schema-qualification is needed for all and only those types
outside pg_catalog. pg_catalog.text is not outside pg_catalog.
something.* IS NOT NULL means that every attribute of the row is not
NULL, not that the row itself is non-NULL (e.g. because it's coming
from below an outer join. Use (somevar.*)::pg_catalog.text IS NOT
NULL instead.
Ashutosh Bapat, per a report by Rushabh Lathia. Reviewed by
Amit Langote and Etsuro Fujita. Schema-qualification added by me.
As discovered by Andreas Seltenreich via sqlsmith, it's possible for a
remote join to need to generate a target list which contains a
PlaceHolderVar which would need to be evaluated on the remote server.
This happens when we try to push down a join tree which contains outer
joins and the nullable side of the join contains a subquery which
evauates some expression which can go to NULL above the level of the
join. Since the deparsing logic can't build a remote query that
involves subqueries, it fails while trying to produce an SQL query
that can be sent to the remote side. Detect such cases and don't try
to push down the join at all.
It's actually fine to push down the join if the PlaceHolderVar needs
to be evaluated at the current join level. This patch makes a small
change to build_tlist_to_deparse so that this case will work.
Amit Langote, Ashutosh Bapat, and me.
Andreas Seltenreich reports that it is possible for a PlaceHolderVar
to creep into this tlist, and I fear that even after that's fixed we
might have other, similar bugs in this area either now or in the
future. There's a lot of action-at-a-distance here, because the
validity of this assertion depends on core planner behavior; so, let's
use elog() to make sure we catch this even in non-assert builds,
rather than just crashing.
Previously, querying the xmin column of a single postgres_fdw foreign
table fetched the tuple length, xmax the typmod, and cmin or cmax the
composite type OID of the tuple. However, when you queried several
such tables and the join got shipped to the remote side, these columns
ended up containing the remote values of the corresponding columns.
Both behaviors are rather unprincipled, the former for obvious reasons
and the latter because the remote values of these columns don't have
any local significance; our transaction IDs are in a different space
than those of the remote machine. Clean this up by setting all of
these fields to 0 in both cases. Also fix the handling of tableoid
to be sane.
Robert Haas and Ashutosh Bapat, reviewed by Etsuro Fujita.
The two get_tle_by_resno() calls introduced by this commit lacked any
check for a NULL return, unlike any other calls of that function anywhere
in our tree. Coverity quite properly complained about it. Also fix a
misindented line in process_query_params(), which Coverity also complained
about on the grounds that the bad indentation suggested possible programmer
misinterpretation.
postgres_fdw can now sent an UPDATE or DELETE statement directly to
the foreign server in simple cases, rather than sending a SELECT FOR
UPDATE statement and then updating or deleting rows one-by-one.
Etsuro Fujita, reviewed by Rushabh Lathia, Shigeru Hanada, Kyotaro
Horiguchi, Albe Laurenz, Thom Brown, and me.
In commit 19a541143a I did not make PathTarget a subtype of Node,
and embedded a RelOptInfo's reltarget directly into it rather than having
a separately-allocated Node. In hindsight that was misguided
micro-optimization, enabled by the fact that at that point we didn't have
any Paths with custom PathTargets. Now that PathTarget processing has
been fleshed out some more, it's easier to see that it's better to have
PathTarget as an indepedent Node type, even if it does cost us one more
palloc to create a RelOptInfo. So change it while we still can.
This commit just changes the representation, without doing anything more
interesting than that.
In commit 1d97c19a0f and later c1d9579dd8, we extended
pull_var_clause's API by adding enum-type arguments. That's sort of a pain
to maintain, though, because it means every time we add a new behavior we
must touch every last one of the call sites, even if there's a reasonable
default behavior that most of them could use. Let's switch over to using a
bitmask of flags, instead; that seems more maintainable and might save a
nanosecond or two as well. This commit changes no behavior in itself,
though I'm going to follow it up with one that does add a new behavior.
In passing, remove flatten_tlist(), which has not been used since 9.1
and would otherwise need the same API changes.
Removing these enums means that optimizer/tlist.h no longer needs to
depend on optimizer/var.h. Changing that caused a number of C files to
need addition of #include "optimizer/var.h" (probably we can thank old
runs of pgrminclude for that); but on balance it seems like a good change
anyway.
Previously, we included NULLS FIRST when appropriate but relied on the
default behavior to be NULLS LAST. This is, however, not true for a
sort in descending order and seems like a fragile assumption anyway.
Report by Rajkumar Raghuwanshi. Patch by Ashutosh Bapat. Review
comments from Michael Paquier and Tom Lane.
Up to now, there's been an assumption that all Paths for a given relation
compute the same output column set (targetlist). However, there are good
reasons to remove that assumption. For example, an indexscan on an
expression index might be able to return the value of an expensive function
"for free". While we have the ability to generate such a plan today in
simple cases, we don't have a way to model that it's cheaper than a plan
that computes the function from scratch, nor a way to create such a plan
in join cases (where the function computation would normally happen at
the topmost join node). Also, we need this so that we can have Paths
representing post-scan/join steps, where the targetlist may well change
from one step to the next. Therefore, invent a "struct PathTarget"
representing the columns we expect a plan step to emit. It's convenient
to include the output tuple width and tlist evaluation cost in this struct,
and there will likely be additional fields in future.
While Path nodes that actually do have custom outputs will need their own
PathTargets, it will still be true that most Paths for a given relation
will compute the same tlist. To reduce the overhead added by this patch,
keep a "default PathTarget" in RelOptInfo, and allow Paths that compute
that column set to just point to their parent RelOptInfo's reltarget.
(In the patch as committed, actually every Path is like that, since we
do not yet have any cases of custom PathTargets.)
I took this opportunity to provide some more-honest costing of
PlaceHolderVar evaluation. Up to now, the assumption that "scan/join
reltargetlists have cost zero" was applied not only to Vars, where it's
reasonable, but also PlaceHolderVars where it isn't. Now, we add the eval
cost of a PlaceHolderVar's expression to the first plan level where it can
be computed, by including it in the PathTarget cost field and adding that
to the cost estimates for Paths. This isn't perfect yet but it's much
better than before, and there is a way forward to improve it more. This
costing change affects the join order chosen for a couple of the regression
tests, changing expected row ordering.
If we've got a relatively straightforward join between two tables,
this pushes that join down to the remote server instead of fetching
the rows for each table and performing the join locally. Some cases
are not handled yet, such as SEMI and ANTI joins. Also, we don't
yet attempt to create presorted join paths or parameterized join
paths even though these options do get tried for a base relation
scan. Nevertheless, this seems likely to be a very significant win
in many practical cases.
Shigeru Hanada and Ashutosh Bapat, reviewed by Robert Haas, with
additional review at various points by Tom Lane, Etsuro Fujita,
KaiGai Kohei, and Jeevan Chalke.
deparseReturningList ended up adding up RETURNING NULL to the code, but
code elsewhere saw an empty list of attributes and concluded that it
should not expect tuples from the remote side.
Etsuro Fujita and Robert Haas, reviewed by Thom Brown
The code that generates a complete SQL query for a given foreign relation
was repeated in two places, and they didn't quite agree: the EXPLAIN case
left out the locking clause. Centralize the code so we get the same
behavior everywhere, and adjust calling conventions and which functions
are static vs. extern accordingly . Centralize the code so we get the same
behavior everywhere, and adjust calling conventions and which functions
are static vs. extern accordingly.
Ashutosh Bapat, reviewed and slightly adjusted by me.
The upcoming patch to allow join pushdown in postgres_fdw needs to use
this code multiple times, which requires moving it to deparse.c. That
seems like a good idea anyway, so do that now both on general principle
and to simplify the future patch.
Inspired by a patch by Shigeru Hanada and Ashutosh Bapat, but I did
it a little differently than what that patch did.
The user can whitelist specified extension(s) in the foreign server's
options, whereupon we will treat immutable functions and operators of those
extensions as candidates to be sent for remote execution.
Whitelisting an extension in this way basically promises that the extension
exists on the remote server and behaves compatibly with the local instance.
We have no way to prove that formally, so we have to rely on the user to
get it right. But this seems like something that people can usually get
right in practice.
We might in future allow functions and operators to be whitelisted
individually, but extension granularity is a very convenient special case,
so it got done first.
The patch as-committed lacks any regression tests, which is unfortunate,
but introducing dependencies on other extensions for testing purposes
would break "make installcheck" scenarios, which is worse. I have some
ideas about klugy ways around that, but it seems like material for a
separate patch. For the moment, leave the problem open.
Paul Ramsey, hacked up a bit more by me
If the join problem's entire ORDER BY clause can be pushed to the
remote server, consider a path that adds this ORDER BY clause. If
use_remote_estimate is on, we cost this path using an additional
remote EXPLAIN. If not, we just estimate that the path costs 20%
more, which is intended to be large enough that we won't request a
remote sort when it's not helpful, but small enough that we'll have
the remote side do the sort when in doubt. In some cases, the remote
sort might actually be free, because the remote query plan might
happen to produce output that is ordered the way we need, but without
remote estimates we have no way of knowing that.
It might also be useful to request sorted output from the remote side
if it enables an efficient merge join, but this patch doesn't attempt
to handle that case.
Ashutosh Bapat with revisions by me. Also reviewed by Fabrízio de Royes
Mello and Jeevan Chalke.
If we have a local Var of say varchar type with default collation, and
we apply a RelabelType to convert that to text with default collation, we
don't want to consider that as creating an FDW_COLLATE_UNSAFE situation.
It should be okay to compare that to a remote Var, so long as the remote
Var determines the comparison collation. (When we actually ship such an
expression to the remote side, the local Var would become a Param with
default collation, meaning the remote Var would in fact control the
comparison collation, because non-default implicit collation overrides
default implicit collation in parse_collate.c.) To fix, be more precise
about what FDW_COLLATE_NONE means: it applies either to a noncollatable
data type or to a collatable type with default collation, if that collation
can't be traced to a remote Var. (When it can, FDW_COLLATE_SAFE is
appropriate.) We were essentially using that interpretation already at
the Var/Const/Param level, but we weren't bubbling it up properly.
An alternative fix would be to introduce a separate FDW_COLLATE_DEFAULT
value to describe the second situation, but that would add more code
without changing the actual behavior, so it didn't seem worthwhile.
Also, since we're clarifying the rule to be that we care about whether
operator/function input collations match, there seems no need to fail
immediately upon seeing a Const/Param/non-foreign-Var with nondefault
collation. We only have to reject if it appears in a collation-sensitive
context (for example, "var IS NOT NULL" is perfectly safe from a collation
standpoint, whatever collation the var has). So just set the state to
UNSAFE rather than failing immediately.
Per report from Jeevan Chalke. This essentially corrects some sloppy
thinking in commit ed3ddf918b, so back-patch
to 9.3 where that logic appeared.
The newly added ON CONFLICT clause allows to specify an alternative to
raising a unique or exclusion constraint violation error when inserting.
ON CONFLICT refers to constraints that can either be specified using a
inference clause (by specifying the columns of a unique constraint) or
by naming a unique or exclusion constraint. DO NOTHING avoids the
constraint violation, without touching the pre-existing row. DO UPDATE
SET ... [WHERE ...] updates the pre-existing tuple, and has access to
both the tuple proposed for insertion and the existing tuple; the
optional WHERE clause can be used to prevent an update from being
executed. The UPDATE SET and WHERE clauses have access to the tuple
proposed for insertion using the "magic" EXCLUDED alias, and to the
pre-existing tuple using the table name or its alias.
This feature is often referred to as upsert.
This is implemented using a new infrastructure called "speculative
insertion". It is an optimistic variant of regular insertion that first
does a pre-check for existing tuples and then attempts an insert. If a
violating tuple was inserted concurrently, the speculatively inserted
tuple is deleted and a new attempt is made. If the pre-check finds a
matching tuple the alternative DO NOTHING or DO UPDATE action is taken.
If the insertion succeeds without detecting a conflict, the tuple is
deemed inserted.
To handle the possible ambiguity between the excluded alias and a table
named excluded, and for convenience with long relation names, INSERT
INTO now can alias its target table.
Bumps catversion as stored rules change.
Author: Peter Geoghegan, with significant contributions from Heikki
Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes.
Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs,
Dean Rasheed, Stephen Frost and many others.
postgres_fdw would send query conditions involving system columns to the
remote server, even though it makes no effort to ensure that system
columns other than CTID match what the remote side thinks. tableoid,
in particular, probably won't match and might have some use in queries.
Hence, prevent sending conditions that include non-CTID system columns.
Also, create_foreignscan_plan neglected to check local restriction
conditions while determining whether to set fsSystemCol for a foreign
scan plan node. This again would bollix the results for queries that
test a foreign table's tableoid.
Back-patch the first fix to 9.3 where postgres_fdw was introduced.
Back-patch the second to 9.2. The code is probably broken in 9.1 as
well, but the patch doesn't apply cleanly there; given the weak state
of support for FDWs in 9.1, it doesn't seem worth fixing.
Etsuro Fujita, reviewed by Ashutosh Bapat, and somewhat modified by me
This command provides an automated way to create foreign table definitions
that match remote tables, thereby reducing tedium and chances for error.
In this patch, we provide the necessary core-server infrastructure and
implement the feature fully in the postgres_fdw foreign-data wrapper.
Other wrappers will throw a "feature not supported" error until/unless
they are updated.
Ronan Dunklau and Michael Paquier, additional work by me
Since most of the system thinks AND and OR are N-argument expressions
anyway, let's have the grammar generate a representation of that form when
dealing with input like "x AND y AND z AND ...", rather than generating
a deeply-nested binary tree that just has to be flattened later by the
planner. This avoids stack overflow in parse analysis when dealing with
queries having more than a few thousand such clauses; and in any case it
removes some rather unsightly inconsistencies, since some parts of parse
analysis were generating N-argument ANDs/ORs already.
It's still possible to get a stack overflow with weirdly parenthesized
input, such as "x AND (y AND (z AND ( ... )))", but such cases are not
mainstream usage. The maximum depth of parenthesization is already
limited by Bison's stack in such cases, anyway, so that the limit is
probably fairly platform-independent.
Patch originally by Gurjeet Singh, heavily revised by me
We were emitting "(SELECT null::typename)", which is usually interpreted
as a scalar subselect, but not so much in the context "x = ANY(...)".
This led to remote-side parsing failures when remote_estimate is enabled.
A quick and ugly fix is to stick in an extra cast step,
"((SELECT null::typename)::typename)". The cast will be thrown away as
redundant by parse analysis, but not before it's done its job of making
sure the grammar sees the ANY argument as an a_expr rather than a
select_with_parens. Per an example from Hannu Krosing.
For variadic functions (other than VARIADIC ANY), the syntaxes foo(x,y,...)
and foo(VARIADIC ARRAY[x,y,...]) should be considered equivalent, since the
former is converted to the latter at parse time. They have indeed been
equivalent, in all releases before 9.3. However, commit 75b39e790 made an
ill-considered decision to record which syntax had been used in FuncExpr
nodes, and then to make equal() test that in checking node equality ---
which caused the syntaxes to not be seen as equivalent by the planner.
This is the underlying cause of bug #9817 from Dmitry Ryabov.
It might seem that a quick fix would be to make equal() disregard
FuncExpr.funcvariadic, but the same commit made that untenable, because
the field actually *is* semantically significant for some VARIADIC ANY
functions. This patch instead adopts the approach of redefining
funcvariadic (and aggvariadic, in HEAD) as meaning that the last argument
is a variadic array, whether it got that way by parser intervention or was
supplied explicitly by the user. Therefore the value will always be true
for non-ANY variadic functions, restoring the principle of equivalence.
(However, the planner will continue to consider use of VARIADIC as a
meaningful difference for VARIADIC ANY functions, even though some such
functions might disregard it.)
In HEAD, this change lets us simplify the decompilation logic in
ruleutils.c, since the funcvariadic/aggvariadic flag tells directly whether
to print VARIADIC. However, in 9.3 we have to continue to cope with
existing stored rules/views that might contain the previous definition.
Fortunately, this just means no change in ruleutils.c, since its existing
behavior effectively ignores funcvariadic for all cases other than VARIADIC
ANY functions.
In HEAD, bump catversion to reflect the fact that FuncExpr.funcvariadic
changed meanings; this is sort of pro forma, since I don't believe any
built-in views are affected.
Unfortunately, this patch doesn't magically fix everything for affected
9.3 users. After installing 9.3.5, they might need to recreate their
rules/views/indexes containing variadic function calls in order to get
everything consistent with the new definition. As in the cited bug,
the symptom of a problem would be failure to use a nominally matching
index that has a variadic function call in its definition. We'll need
to mention this in the 9.3.5 release notes.
This covers all the SQL-standard trigger types supported for regular
tables; it does not cover constraint triggers. The approach for
acquiring the old row mirrors that for view INSTEAD OF triggers. For
AFTER ROW triggers, we spool the foreign tuples to a tuplestore.
This changes the FDW API contract; when deciding which columns to
populate in the slot returned from data modification callbacks, writable
FDWs will need to check for AFTER ROW triggers in addition to checking
for a RETURNING clause.
In support of the feature addition, refactor the TriggerFlags bits and
the assembly of old tuples in ModifyTable.
Ronan Dunklau, reviewed by KaiGai Kohei; some additional hacking by me.
The previous coding supposed that it could consider just a single join
condition in any one parameterized path for the foreign table. But in
reality, the parameterized-path machinery forces all join clauses that are
"movable to" the foreign table to be evaluated at that node; including
clauses that we might not consider safe to send across. Such cases would
result in an Assert failure in an assert-enabled build, and otherwise in
sending an unsafe clause to the foreign server, which might result in
errors or silently-wrong answers. A lesser problem was that the
cost/rowcount estimates generated for the parameterized path failed to
account for any additional join quals that get assigned to the scan.
To fix, rewrite postgresGetForeignPaths so that it correctly collects all
the movable quals for any one outer relation when generating parameterized
paths; we'll now generate just one path per outer relation not one per join
qual. Also fix bogus assumptions in postgresGetForeignPlan and
estimate_path_cost_size that only safe-to-send join quals will be
presented.
Based on complaint from Etsuro Fujita that the path costs were being
miscalculated, though this is significantly different from his proposed
patch.
This should provide some marginal overall savings, since it surely takes
many more cycles for the remote server to deal with the NULL columns than
it takes for postgres_fdw not to emit them. But really the reason is to
keep the emitted queries from looking quite so silly ...
I wasn't going to ship this without having at least some example of how
to do that. This version isn't terribly bright; in particular it won't
consider any combinations of multiple join clauses. Given the cost of
executing a remote EXPLAIN, I'm not sure we want to be very aggressive
about doing that, anyway.
In support of this, refactor generate_implied_equalities_for_indexcol
so that it can be used to extract equivalence clauses that aren't
necessarily tied to an index.
Treat expressions as being remotely executable only if all collations used
in them are determined by Vars of the foreign table. This means that, if
the foreign server gets different answers than we do, it's the user's fault
for not having marked the foreign table columns with collations equivalent
to the remote table's. This rule allows most simple expressions such as
"var < 'constant'" to be sent to the remote side, because the constant
isn't determining the collation (the Var's collation would win). There's
still room for improvement, but it's hard to see how to do it without a
lot more knowledge and/or assumptions about what the remote side will do.
Adopt the position that only locally-defined defaults matter. Any defaults
defined in the remote database do not affect insertions performed through
a foreign table (unless they are for columns not known to the foreign
table). While it'd arguably be more useful to permit remote defaults to be
used, making that work in a consistent fashion requires far more work than
seems possible for 9.3.
For datatypes whose output formatting depends on one or more GUC settings,
we have to worry about whether the other server will interpret the value
the same way it was meant. pg_dump has been aware of this hazard for a
long time, but postgres_fdw needs to deal with it too. To fix data
retrieval from the remote server, set the necessary remote GUC settings at
connection startup. (We were already assuming that settings made then
would persist throughout the remote session.) To fix data transmission to
the remote server, temporarily force the relevant GUCs to the right values
when we're about to convert any data values to text for transmission.
This is all pretty grotty, and not very cheap either. It's tempting to
think of defining one uber-GUC that would override any settings that might
render printed data values unportable. But of course, older remote servers
wouldn't know any such thing and would still need this logic.
While at it, revert commit f7951eef89, since
this provides a real fix. (The timestamptz given in the error message
returned from the "remote" server will now reliably be shown in UTC.)
Robert Haas