Formerly, we treated only portals created in the current subtransaction as
having failed during subtransaction abort. However, if the error occurred
while running a portal created in an outer subtransaction (ie, a cursor
declared before the last savepoint), that has to be considered broken too.
To allow reliable detection of which ones those are, add a bookkeeping
field to struct Portal that tracks the innermost subtransaction in which
each portal has actually been executed. (Without this, we'd end up
failing portals containing functions that had called the subtransaction,
thereby breaking plpgsql exception blocks completely.)
In addition, when we fail an outer-subtransaction Portal, transfer its
resources into the subtransaction's resource owner, so that they're
released early in cleanup of the subxact. This fixes a problem reported by
Jim Nasby in which a function executed in an outer-subtransaction cursor
could cause an Assert failure or crash by referencing a relation created
within the inner subtransaction.
The proximate cause of the Assert failure is that AtEOSubXact_RelationCache
assumed it could blow away a relcache entry without first checking that the
entry had zero refcount. That was a bad idea on its own terms, so add such
a check there, and to the similar coding in AtEOXact_RelationCache. This
provides an independent safety measure in case there are still ways to
provoke the situation despite the Portal-level changes.
This has been broken since subtransactions were invented, so back-patch
to all supported branches.
Tom Lane and Michael Paquier
Commit f3b5565dd4 was a couple of bricks shy
of a load; specifically, it missed putting pg_trigger_tgrelid_tgname_index
into the relcache init file, because that index is not used by any
syscache. However, we have historically nailed that index into cache for
performance reasons. The upshot was that load_relcache_init_file always
decided that the init file was busted and silently ignored it, resulting
in a significant hit to backend startup speed.
To fix, reinstantiate RelationIdIsInInitFile() as a wrapper around
RelationSupportsSysCache(), which can know about additional relations
that should be in the init file despite being unknown to syscache.c.
Also install some guards against future mistakes of this type: make
write_relcache_init_file Assert that all nailed relations get written to
the init file, and make load_relcache_init_file emit a WARNING if it takes
the "wrong number of nailed relations" exit path. Now that we remove the
init files during postmaster startup, that case should never occur in the
field, even if we are starting a minor-version update that added or removed
rels from the nailed set. So the warning shouldn't ever be seen by end
users, but it will show up in the regression tests if somebody breaks this
logic.
Back-patch to all supported branches, like the previous commit.
When we invalidate the relcache entry for a system catalog or index, we
must also delete the relcache "init file" if the init file contains a copy
of that rel's entry. The old way of doing this relied on a specially
maintained list of the OIDs of relations present in the init file: we made
the list either when reading the file in, or when writing the file out.
The problem is that when writing the file out, we included only rels
present in our local relcache, which might have already suffered some
deletions due to relcache inval events. In such cases we correctly decided
not to overwrite the real init file with incomplete data --- but we still
used the incomplete initFileRelationIds list for the rest of the current
session. This could result in wrong decisions about whether the session's
own actions require deletion of the init file, potentially allowing an init
file created by some other concurrent session to be left around even though
it's been made stale.
Since we don't support changing the schema of a system catalog at runtime,
the only likely scenario in which this would cause a problem in the field
involves a "vacuum full" on a catalog concurrently with other activity, and
even then it's far from easy to provoke. Remarkably, this has been broken
since 2002 (in commit 7863404417), but we had
never seen a reproducible test case until recently. If it did happen in
the field, the symptoms would probably involve unexpected "cache lookup
failed" errors to begin with, then "could not open file" failures after the
next checkpoint, as all accesses to the affected catalog stopped working.
Recovery would require manually removing the stale "pg_internal.init" file.
To fix, get rid of the initFileRelationIds list, and instead consult
syscache.c's list of relations used in catalog caches to decide whether a
relation is included in the init file. This should be a tad more efficient
anyway, since we're replacing linear search of a list with ~100 entries
with a binary search. It's a bit ugly that the init file contents are now
so directly tied to the catalog caches, but in practice that won't make
much difference.
Back-patch to all supported branches.
The corner case where a relcache invalidation tried to rebuild the
entry for a referenced relation but couldn't find it in the catalog
wasn't correct.
The code tried to RelationCacheDelete/RelationDestroyRelation the
entry. That didn't work when assertions are enabled because the latter
contains an assertion ensuring the refcount is zero. It's also more
generally a bad idea, because by virtue of being referenced somebody
might actually look at the entry, which is possible if the error is
trapped and handled via a subtransaction abort.
Instead just error out, without deleting the entry. As the entry is
marked invalid, the worst that can happen is that the invalid (and at
some point unused) entry lingers in the relcache.
Discussion: 22459.1418656530@sss.pgh.pa.us
There should be no way to hit this case < 9.4 where logical decoding
introduced a bug that can hit this. But since the code for handling
the corner case is there it should do something halfway sane, so
backpatch all the the way back. The logical decoding bug will be
handled in a separate commit.
This only happens if a client issues a Parse message with an empty query
string, which is a bit odd; but since it is explicitly called out as legal
by our FE/BE protocol spec, we'd probably better continue to allow it.
Fix by adding tests everywhere that the raw_parse_tree field is passed to
functions that don't or shouldn't accept NULL. Also make it clear in the
relevant comments that NULL is an expected case.
This reverts commits a73c9dbab0 and
2e9650cbcf, which fixed specific crash
symptoms by hacking things at what now seems to be the wrong end, ie the
callee functions. Making the callees allow NULL is superficially more
robust, but it's not always true that there is a defensible thing for the
callee to do in such cases. The caller has more context and is better
able to decide what the empty-query case ought to do.
Per followup discussion of bug #11335. Back-patch to 9.2. The code
before that is sufficiently different that it would require development
of a separate patch, which doesn't seem worthwhile for what is believed
to be an essentially cosmetic change.
This was not changed in HEAD, but will be done later as part of a
pgindent run. Future pgindent runs will also do this.
Report by Tom Lane
Backpatch through all supported branches, but not HEAD
The previous coding in plancache.c essentially used 10% of the estimated
runtime as its cost estimate for planning. This can be pretty bogus,
especially when the estimated runtime is very small, such as in a simple
expression plan created by plpgsql, or a simple INSERT ... VALUES.
While we don't have a really good handle on how planning time compares
to runtime, it seems reasonable to use an estimate based on the number of
relations referenced in the query, with a rather large multiplier. This
patch uses 1000 * cpu_operator_cost * (nrelations + 1), so that even a
trivial query will be charged 1000 * cpu_operator_cost for planning.
This should address the problem reported by Marc Cousin and others that
9.2 and up prefer custom plans in cases where the planning time greatly
exceeds what can be saved.
We need to increment the refcount on the composite type's cached tuple
descriptor while we do lookups of its column types. Otherwise a cache
flush could occur and release the tuple descriptor before we're done with
it. This fails reliably with -DCLOBBER_CACHE_ALWAYS, but the odds of a
failure in a production build seem rather low (since the pfree'd descriptor
typically wouldn't get scribbled on immediately). That may explain the
lack of any previous reports. Buildfarm issue noted by Christian Ullrich.
Back-patch to 9.1 where the bogus code was added.
When creating or manipulating a cached plan for a transaction control
command (particularly ROLLBACK), we must not perform any catalog accesses,
since we might be in an aborted transaction. However, plancache.c busily
saved or examined the search_path for every cached plan. If we were
unlucky enough to do this at a moment where the path's expansion into
schema OIDs wasn't already cached, we'd do some catalog accesses; and with
some more bad luck such as an ill-timed signal arrival, that could lead to
crashes or Assert failures, as exhibited in bug #8095 from Nachiket Vaidya.
Fortunately, there's no real need to consider the search path for such
commands, so we can just skip the relevant steps when the subject statement
is a TransactionStmt. This is somewhat related to bug #5269, though the
failure happens during initial cached-plan creation rather than
revalidation.
This bug has been there since the plan cache was invented, so back-patch
to all supported branches.
SPI_execute() and related functions create a CachedPlan, execute it once,
and immediately discard it, so that the functionality offered by
plancache.c is of no value in this code path. And performance measurements
show that the extra data copying and invalidation checking done by
plancache.c slows down simple queries by 10% or more compared to 9.1.
However, enough of the SPI code is shared with functions that do need plan
caching that it seems impractical to bypass plancache.c altogether.
Instead, let's invent a variant version of cached plans that preserves
99% of the API but doesn't offer any of the actual functionality, nor the
overhead. This puts SPI_execute() performance back on par, or maybe even
slightly better, than it was before. This change should resolve recent
complaints of performance degradation from Dong Ye, Pavel Stehule, and
others.
By avoiding data copying, this change also reduces the amount of memory
needed to execute many-statement SPI_execute() strings, as for instance in
a recent complaint from Tomas Vondra.
An additional benefit of this change is that multi-statement SPI_execute()
query strings are now processed fully serially, that is we complete
execution of earlier statements before running parse analysis and planning
on following ones. This eliminates a long-standing POLA violation, in that
DDL that affects the behavior of a later statement will now behave as
expected.
Back-patch to 9.2, since this was a performance regression compared to 9.1.
(In 9.2, place the added struct fields so as to avoid changing the offsets
of existing fields.)
Heikki Linnakangas and Tom Lane
During crash recovery, we remove disk files belonging to temporary tables,
but the system catalog entries for such tables are intentionally not
cleaned up right away. Instead, the first backend that uses a temp schema
is expected to clean out any leftover objects therein. This approach
requires that we be careful to ignore leftover temp tables (since any
actual access attempt would fail), *even if their BackendId matches our
session*, if we have not yet established use of the session's corresponding
temp schema. That worked fine in the past, but was broken by commit
debcec7dc3 which incorrectly removed the
rd_islocaltemp relcache flag. Put it back, and undo various changes
that substituted tests like "rel->rd_backend == MyBackendId" for use
of a state-aware flag. Per trouble report from Heikki Linnakangas.
Back-patch to 9.1 where the erroneous change was made. In the back
branches, be careful to add rd_islocaltemp in a spot in the struct that
was alignment padding before, so as not to break existing add-on code.
Commit 8cb53654db, which introduced DROP
INDEX CONCURRENTLY, managed to break CREATE INDEX CONCURRENTLY via a poor
choice of catalog state representation. The pg_index state for an index
that's reached the final pre-drop stage was the same as the state for an
index just created by CREATE INDEX CONCURRENTLY. This meant that the
(necessary) change to make RelationGetIndexList ignore about-to-die indexes
also made it ignore freshly-created indexes; which is catastrophic because
the latter do need to be considered in HOT-safety decisions. Failure to
do so leads to incorrect index entries and subsequently wrong results from
queries depending on the concurrently-created index.
To fix, make the final state be indisvalid = true and indisready = false,
which is otherwise nonsensical. This is pretty ugly but we can't add
another column without forcing initdb, and it's too late for that in 9.2.
(There's a cleaner fix in HEAD.)
In addition, change CREATE/DROP INDEX CONCURRENTLY so that the pg_index
flag changes they make without exclusive lock on the index are made via
heap_inplace_update() rather than a normal transactional update. The
latter is not very safe because moving the pg_index tuple could result in
concurrent SnapshotNow scans finding it twice or not at all, thus possibly
resulting in index corruption. This is a pre-existing bug in CREATE INDEX
CONCURRENTLY, which was copied into the DROP code.
In addition, fix various places in the code that ought to check to make
sure that the indexes they are manipulating are valid and/or ready as
appropriate. These represent bugs that have existed since 8.2, since
a failed CREATE INDEX CONCURRENTLY could leave a corrupt or invalid
index behind, and we ought not try to do anything that might fail with
such an index.
Also fix RelationReloadIndexInfo to ensure it copies all the pg_index
columns that are allowed to change after initial creation. Previously we
could have been left with stale values of some fields in an index relcache
entry. It's not clear whether this actually had any user-visible
consequences, but it's at least a bug waiting to happen.
In addition, do some code and docs review for DROP INDEX CONCURRENTLY;
some cosmetic code cleanup but mostly addition and revision of comments.
Portions of this need to be back-patched even further, but I'll work
on that separately.
Problem reported by Amit Kapila, diagnosis by Pavan Deolasee,
fix by Tom Lane and Andres Freund.
This change ensures that the planner will see implicit and explicit casts
as equivalent for all purposes, except in the minority of cases where
there's actually a semantic difference (as reflected by having a 3-argument
cast function). In particular, this fixes cases where the EquivalenceClass
machinery failed to consider two references to a varchar column as
equivalent if one was implicitly cast to text but the other was explicitly
cast to text, as seen in bug #7598 from Vaclav Juza. We have had similar
bugs before in other parts of the planner, so I think it's time to fix this
problem at the core instead of continuing to band-aid around it.
Remove set_coercionform_dontcare(), which represents the band-aid
previously in use for allowing matching of index and constraint expressions
with inconsistent cast labeling. (We can probably get rid of
COERCE_DONTCARE altogether, but I don't think removing that enum value in
back branches would be wise; it's possible there's third party code
referring to it.)
Back-patch to 9.2. We could go back further, and might want to once this
has been tested more; but for the moment I won't risk destabilizing plan
choices in long-since-stable branches.
When (re) loading the typcache comparison cache for an enum type's values,
use an up-to-date MVCC snapshot, not the transaction's existing snapshot.
This avoids problems if we encounter an enum OID that was created since our
transaction started. Per report from Andres Freund and diagnosis by Robert
Haas.
To ensure this is safe even if enum comparison manages to get invoked
before we've set a transaction snapshot, tweak GetLatestSnapshot to
redirect to GetTransactionSnapshot instead of throwing error when
FirstSnapshotSet is false. The existing uses of GetLatestSnapshot (in
ri_triggers.c) don't care since they couldn't be invoked except in a
transaction that's already done some work --- but it seems just conceivable
that this might not be true of enums, especially if we ever choose to use
enums in system catalogs.
Note that the comparable coding in enum_endpoint and enum_range_internal
remains GetTransactionSnapshot; this is perhaps debatable, but if we
changed it those functions would have to be marked volatile, which doesn't
seem attractive.
Back-patch to 9.1 where ALTER TYPE ADD VALUE was added.
The original syntax wasn't universally loved, and it didn't allow its
usage in CREATE TABLE, only ALTER TABLE. It now works everywhere, and
it also allows using ALTER TABLE ONLY to add an uninherited CHECK
constraint, per discussion.
The pg_constraint column has accordingly been renamed connoinherit.
This commit partly reverts some of the changes in
61d81bd28d, particularly some pg_dump and
psql bits, because now pg_get_constraintdef includes the necessary NO
INHERIT within the constraint definition.
Author: Nikhil Sontakke
Some tweaks by me
This restores the pre-9.0 situation that it's possible to add new indexes
on pg_class and other mapped-but-not-shared catalogs, so long as you broke
the glass and flipped the big red Dont-Touch-Me switch. As before, there
are a lot of gotchas, and you'd have to be pretty desperate to try this
on a production database; but there doesn't seem to be a reason for
relmapper.c to be preventing such things all by itself. Per
experimentation with a case suggested by Cody Cutrer.
Making this operation look like a utility statement seems generally a good
idea, and particularly so in light of the desire to provide command
triggers for utility statements. The original choice of representing it as
SELECT with an IntoClause appendage had metastasized into rather a lot of
places, unfortunately, so that this patch is a great deal more complicated
than one might at first expect.
In particular, keeping EXPLAIN working for SELECT INTO and CREATE TABLE AS
subcommands required restructuring some EXPLAIN-related APIs. Add-on code
that calls ExplainOnePlan or ExplainOneUtility, or uses
ExplainOneQuery_hook, will need adjustment.
Also, the cases PREPARE ... SELECT INTO and CREATE RULE ... SELECT INTO,
which formerly were accepted though undocumented, are no longer accepted.
The PREPARE case can be replaced with use of CREATE TABLE AS EXECUTE.
The CREATE RULE case doesn't seem to have much real-world use (since the
rule would work only once before failing with "table already exists"),
so we'll not bother with that one.
Both SELECT INTO and CREATE TABLE AS still return a command tag of
"SELECT nnnn". There was some discussion of returning "CREATE TABLE nnnn",
but for the moment backwards compatibility wins the day.
Andres Freund and Tom Lane
Now that cache invalidation callbacks get only a hash value, and not a
tuple TID (per commits 632ae6829f and
b5282aa893), the only way they can restrict
what they invalidate is to know what the hash values mean. setrefs.c was
doing this via a hard-wired assumption but that seems pretty grotty, and
it'll only get worse as more cases come up. So let's expose a calculation
function that takes the same parameters as SearchSysCache. Per complaint
from Marko Kreen.
We don't normally allow quals to be pushed down into a view created
with the security_barrier option, but functions without side effects
are an exception: they're OK. This allows much better performance in
common cases, such as when using an equality operator (that might
even be indexable).
There is an outstanding issue here with the CREATE FUNCTION / ALTER
FUNCTION syntax: there's no way to use ALTER FUNCTION to unset the
leakproof flag. But I'm committing this as-is so that it doesn't
have to be rebased again; we can fix up the grammar in a future
commit.
KaiGai Kohei, with some wordsmithing by me.
When default_text_search_config, default_tablespace, or temp_tablespaces
setting is set per-user or per-database, with an "ALTER USER/DATABASE SET
..." statement, don't throw an error if the text search configuration or
tablespace does not exist. In case of text search configuration, even if
it doesn't exist in the current database, it might exist in another
database, where the setting is intended to have its effect. This behavior
is now the same as search_path's.
Tablespaces are cluster-wide, so the same argument doesn't hold for
tablespaces, but there's a problem with pg_dumpall: it dumps "ALTER USER
SET ..." statements before the "CREATE TABLESPACE" statements. Arguably
that's pg_dumpall's fault - it should dump the statements in such an order
that the tablespace is created first and then the "ALTER USER SET
default_tablespace ..." statements after that - but it seems better to be
consistent with search_path and default_text_search_config anyway. Besides,
you could still create a dump that throws an error, by creating the
tablespace, running "ALTER USER SET default_tablespace", then dropping the
tablespace and running pg_dumpall on that.
Backpatch to all supported versions.
Normally, accessing variable-length members of catalog structures past
the first one doesn't work at all. Here, it happened to work because
indnatts was checked to be 1, and so the defined FormData_pg_index
layout, using int2vector[1] and oidvector[1] for variable-length
arrays, happened to match the actual memory layout. But it's a very
fragile assumption, and it's not in a performance-critical path, so
code it properly using heap_getattr() instead.
bug analysis by Tom Lane
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.
This makes them enforceable only on the parent table, not on children
tables. This is useful in various situations, per discussion involving
people bitten by the restrictive behavior introduced in 8.4.
Message-Id:
8762mp93iw.fsf@comcast.netCAFaPBrSMMpubkGf4zcRL_YL-AERUbYF_-ZNNYfb3CVwwEqc9TQ@mail.gmail.com
Authors: Nikhil Sontakke, Alex Hunsaker
Reviewed by Robert Haas and myself
This patch creates an API whereby a btree index opclass can optionally
provide non-SQL-callable support functions for sorting. In the initial
patch, we only use this to provide a directly-callable comparator function,
which can be invoked with a bit less overhead than the traditional
SQL-callable comparator. While that should be of value in itself, the real
reason for doing this is to provide a datatype-extensible framework for
more aggressive optimizations, as in Peter Geoghegan's recent work.
Robert Haas and Tom Lane
Since record[] uses array_in, it needs to have its element type passed
as typioparam. In HEAD and 9.1, this fix essentially reverts commit
9bc933b212, which was a hack that is no
longer needed since domains don't set their typelem anymore. Before
that, adjust the logic so that only domains are excluded from being
treated like arrays, rather than assuming that only base types should
be included. Add a regression test to demonstrate the need for this.
Per report from Maxim Boguk.
Back-patch to 8.4, where type record[] was added.
Move the responsibility for caching specialized information about range
types into the type cache, so that the catalog lookups only have to occur
once per session. Rearrange APIs a bit so that fn_extra caching is
actually effective in the GiST support code. (Use of OidFunctionCallN is
bad enough for performance in itself, but it also prevents the function
from exploiting fn_extra caching.)
The range I/O functions are still not very bright about caching repeated
lookups, but that seems like material for a separate patch.
Also, avoid unnecessary use of memcpy to fetch/store the range type OID and
flags, and don't use the full range_deserialize machinery when all we need
to see is the flags value.
Also fix API error in range_gist_penalty --- it was failing to set *penalty
for any case involving an empty range.
If a tuple in a syscache contains an out-of-line toasted field, and we
try to fetch that field shortly after some other transaction has committed
an update or deletion of the tuple, there is a race condition: vacuum
could come along and remove the toast tuples before we can fetch them.
This leads to transient failures like "missing chunk number 0 for toast
value NNNNN in pg_toast_2619", as seen in recent reports from Andrew
Hammond and Tim Uckun.
The design idea of syscache is that access to stale syscache entries
should be prevented by relation-level locks, but that fails for at least
two cases where toasted fields are possible: ANALYZE updates pg_statistic
rows without locking out sessions that might want to plan queries on the
same table, and CREATE OR REPLACE FUNCTION updates pg_proc rows without
any meaningful lock at all.
The least risky fix seems to be an idea that Heikki suggested when we
were dealing with a related problem back in August: forcibly detoast any
out-of-line fields before putting a tuple into syscache in the first place.
This avoids the problem because at the time we fetch the parent tuple from
the catalog, we should be holding an MVCC snapshot that will prevent
removal of the toast tuples, even if the parent tuple is outdated
immediately after we fetch it. (Note: I'm not convinced that this
statement holds true at every instant where we could be fetching a syscache
entry at all, but it does appear to hold true at the times where we could
fetch an entry that could have a toasted field. We will need to be a bit
wary of adding toast tables to low-level catalogs that don't have them
already.) An additional benefit is that subsequent uses of the syscache
entry should be faster, since they won't have to detoast the field.
Back-patch to all supported versions. The problem is significantly harder
to reproduce in pre-9.0 releases, because of their willingness to flush
every entry in a syscache whenever the underlying catalog is vacuumed
(cf CatalogCacheFlushRelation); but there is still a window for trouble.
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.
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.
The code path that tried a generic plan, didn't like it, and then made a
custom plan was mistakenly passing the same copy of the query_list to the
planner both times. This doesn't work too well for nontrivial queries,
since the planner tends to scribble on its input. Diagnosis and fix by
Yamamoto Takashi.
I had copied-and-pasted a claim that we couldn't reach this point when
dealing with utility statements, but that was a leftover from when the
caller was required to supply a plan to start with. We now will go
through here at least once when handling a utility statement, so it
seems worth a check to see whether a snapshot is actually needed.
(Note that analyze_requires_snapshot is quite a cheap test.)
Per suggestion from Yamamoto Takashi. I don't think I believe that this
resolves his reported assertion failure; but it's worth changing anyway,
just to save a cycle or two.
The regression tests were failing with CLOBBER_CACHE_ALWAYS enabled,
as reported by buildfarm member jaguar. There was an Assert in
BuildCachedPlan that asserted that the CachedPlanSource hadn't been
invalidated since we called RevalidateCachedQuery, which in theory can't
happen because we are holding locks on all the relevant database objects.
However, CLOBBER_CACHE_ALWAYS generates a false positive by making an
invalidation happen anyway; and on reflection, that could also occur as a
result of a badly-timed sinval reset due to queue overflow. We could just
remove the Assert and forge ahead with the not-really-stale querytree, but
it seems safer to do another RevalidateCachedQuery call just to make real
sure everything's OK.
Rewrite plancache.c so that a "cached plan" (which is rather a misnomer
at this point) can support generation of custom, parameter-value-dependent
plans, and can make an intelligent choice between using custom plans and
the traditional generic-plan approach. The specific choice algorithm
implemented here can probably be improved in future, but this commit is
all about getting the mechanism in place, not the policy.
In addition, restructure the API to greatly reduce the amount of extraneous
data copying needed. The main compromise needed to make that possible was
to split the initial creation of a CachedPlanSource into two steps. It's
worth noting in particular that SPI_saveplan is now deprecated in favor of
SPI_keepplan, which accomplishes the same end result with zero data
copying, and no need to then spend even more cycles throwing away the
original SPIPlan. The risk of long-term memory leaks while manipulating
SPIPlans has also been greatly reduced. Most of this improvement is based
on use of the recently-added MemoryContextSetParent primitive.
walsender.h should depend on xlog.h, not vice versa. (Actually, the
inclusion was circular until a couple hours ago, which was even sillier;
but Bruce broke it in the expedient rather than logically correct
direction.) Because of that poor decision, plus blind application of
pgrminclude, we had a situation where half the system was depending on
xlog.h to include such unrelated stuff as array.h and guc.h. Clean up
the header inclusion, and manually revert a lot of what pgrminclude had
done so things build again.
This episode reinforces my feeling that pgrminclude should not be run
without adult supervision. Inclusion changes in header files in particular
need to be reviewed with great care. More generally, it'd be good if we
had a clearer notion of module layering to dictate which headers can sanely
include which others ... but that's a big task for another day.
It is possible for VACUUM to scan no pages at all, if the visibility map
shows that all pages are all-visible. In this situation VACUUM has no new
information to report about the relation's tuple density, so it wasn't
changing pg_class.reltuples ... but it updated pg_class.relpages anyway.
That's wrong in general, since there is no evidence to justify changing the
density ratio reltuples/relpages, but it's particularly bad if the previous
state was relpages=reltuples=0, which means "unknown tuple density".
We just replaced "unknown" with "zero". ANALYZE would eventually recover
from this, but it could take a lot of repetitions of ANALYZE to do so if
the relation size is much larger than the maximum number of pages ANALYZE
will scan, because of the moving-average behavior introduced by commit
b4b6923e03.
The only known situation where we could have relpages=reltuples=0 and yet
the visibility map asserts everything's visible is immediately following
a pg_upgrade. It might be advisable for pg_upgrade to try to preserve the
relpages/reltuples statistics; but in any case this code is wrong on its
own terms, so fix it. Per report from Sergey Koposov.
Back-patch to 8.4, where the visibility map was introduced, same as the
previous change.
This requires adjusting the API for syscache callback functions: they now
get a hash value, not a TID, to identify the target tuple. Most of them
weren't paying any attention to that argument anyway, but plancache did
require a small amount of fixing.
Also, improve performance a trifle by avoiding sending duplicate inval
messages when a heap_update isn't changing the catcache lookup columns.
The TID isn't stable enough: we might queue an sinval event before a VACUUM
FULL, and then process it afterwards, when the target tuple no longer has
the same TID. So we must invalidate entries on the basis of hash value
only. The old coding can be shown to result in various bizarre,
hard-to-reproduce errors in the presence of concurrent VACUUM FULLs on
system catalogs, and could easily result in permanent catalog corruption,
up to and including complete loss of tables.
This commit is just a minimal fix that removes the unsafe comparison.
We should remove transmission of the tuple TID from sinval messages
altogether, and then arrange to suppress the extra message in the common
case of a heap_update that doesn't change the key hashvalue. But that's
going to be much more invasive, and will only produce a probably-marginal
performance gain, so it doesn't seem like material for a back-patch.
Back-patch to 9.0. Before that, VACUUM FULL refused to do any tuple moving
if it found any INSERT_IN_PROGRESS or DELETE_IN_PROGRESS tuples (and
CLUSTER would give up altogether), so there was no risk of moving a tuple
that might be the subject of an unsent sinval message.
We have to be sure that we have revalidated each nailed-in-cache relcache
entry before we try to use it to load data for some other relcache entry.
The introduction of "mapped relations" in 9.0 broke this, because although
we updated the state kept in relmapper.c early enough, we failed to
propagate that information into relcache entries soon enough; in
particular, we could try to fetch pg_class rows out of pg_class before
we'd updated its relcache entry's rd_node.relNode value from the map.
This bug accounts for Dave Gould's report of failures after "vacuum full
pg_class", and I believe that there is risk for other system catalogs
as well.
The core part of the fix is to copy relmapper data into the relcache
entries during "phase 1" in RelationCacheInvalidate(), before they'll be
used in "phase 2". To try to future-proof the code against other similar
bugs, I also rearranged the order in which nailed relations are visited
during phase 2: now it's pg_class first, then pg_class_oid_index, then
other nailed relations. This should ensure that RelationClearRelation can
apply RelationReloadIndexInfo to all nailed indexes without risking use
of not-yet-revalidated relcache entries.
Back-patch to 9.0 where the relation mapper was introduced.
The previous code tried to synchronize by unlinking the init file twice,
but that doesn't actually work: it leaves a window wherein a third process
could read the already-stale init file but miss the SI messages that would
tell it the data is stale. The result would be bizarre failures in catalog
accesses, typically "could not read block 0 in file ..." later during
startup.
Instead, hold RelCacheInitLock across both the unlink and the sending of
the SI messages. This is more straightforward, and might even be a bit
faster since only one unlink call is needed.
This has been wrong since it was put in (in 2002!), so back-patch to all
supported releases.
get_op_btree_interpretation assumed this in order to save some duplication
of code, but it's not true in general anymore because we added <> support
to btree_gist. (We still assume it for btree opclasses, though.)
Also, essentially the same logic was baked into predtest.c. Get rid of
that duplication by generalizing get_op_btree_interpretation so that it
can be used by predtest.c.
Per bug report from Denis de Bernardy and investigation by Jeff Davis,
though I didn't use Jeff's patch exactly as-is.
Back-patch to 9.1; we do not support this usage before that.
