Previously the changes to the "data" part of the sequence, i.e. the
one containing the current value, were not transactional, whereas the
definition, including minimum and maximum value were. That leads to
odd behaviour if a schema change is rolled back, with the potential
that out-of-bound sequence values can be returned.
To avoid the issue create a new relfilenode fork whenever ALTER
SEQUENCE is executed, similar to how TRUNCATE ... RESTART IDENTITY
already is already handled.
This commit also makes ALTER SEQUENCE RESTART transactional, as it
seems to be too confusing to have some forms of ALTER SEQUENCE behave
transactionally, some forms not. This way setval() and nextval() are
not transactional, but DDL is, which seems to make sense.
This commit also rolls back parts of the changes made in 3d092fe540
and f8dc1985f as they're now not needed anymore.
Author: Andres Freund
Discussion: https://postgr.es/m/20170522154227.nvafbsm62sjpbxvd@alap3.anarazel.de
Backpatch: Bug is in master/v10 only
In 1753b1b027, the pg_sequence system
catalog was introduced. This made sequence metadata changes
transactional, while the actual sequence values are still behaving
nontransactionally. This requires some refinement in how ALTER
SEQUENCE, which operates on both, locks the sequence and the catalog.
The main problems were:
- Concurrent ALTER SEQUENCE causes "tuple concurrently updated" error,
caused by updates to pg_sequence catalog.
- Sequence WAL writes and catalog updates are not protected by same
lock, which could lead to inconsistent recovery order.
- nextval() disregarding uncommitted ALTER SEQUENCE changes.
To fix, nextval() and friends now lock the sequence using
RowExclusiveLock instead of AccessShareLock. ALTER SEQUENCE locks the
sequence using ShareRowExclusiveLock. This means that nextval() and
ALTER SEQUENCE block each other, and ALTER SEQUENCE on the same sequence
blocks itself. (This was already the case previously for the OWNER TO,
RENAME, and SET SCHEMA variants.) Also, rearrange some code so that the
entire AlterSequence is protected by the lock on the sequence.
As an exception, use reduced locking for ALTER SEQUENCE ... RESTART.
Since that is basically a setval(), it does not require the full locking
of other ALTER SEQUENCE actions. So check whether we are only running a
RESTART and run with less locking if so.
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Reported-by: Jason Petersen <jason@citusdata.com>
Reported-by: Andres Freund <andres@anarazel.de>
This improves code coverage and lays a foundation for testing
similar issues in a distributed environment.
Author: Thomas Munro <thomas.munro@enterprisedb.com>
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Concurrent auto-analyze could be holding a snapshot, affecting the
removal of deleted row versions. Remove the deletion to avoid this
happening. Per buildfarm.
In passing, make the test independent of assumptions of physical row
order, just out of sheer paranoia.
presence of page pins, which leads to serious estimation errors in the
planner. This particularly affects small heavily-accessed tables,
especially where locking (e.g. from FK constraints) forces frequent
vacuums for mxid cleanup.
Fix by keeping separate track of pages whose live tuples were actually
counted vs. pages that were only scanned for freezing purposes. Thus,
reltuples can only be set to 0 if all pages of the relation were
actually counted.
Backpatch to all supported versions.
Per bug #14057 from Nicolas Baccelli, analyzed by me.
Discussion: https://postgr.es/m/20160331103739.8956.94469@wrigleys.postgresql.org
A transaction that conflicts against itself, for example
INSERT INTO t(pk) VALUES (1),(1) ON CONFLICT DO NOTHING;
should behave the same regardless of isolation level. It certainly
shouldn't throw a serialization error, as retrying will not help.
We got this wrong due to the ON CONFLICT logic not considering the case,
as reported by Jason Dusek.
Core of this patch is by Peter Geoghegan (based on an earlier patch by
Thomas Munro), though I didn't take his proposed code refactoring for fear
that it might have unexpected side-effects. Test cases by Thomas Munro
and myself.
Report: <CAO3NbwOycQjt2Oqy2VW-eLTq2M5uGMyHnGm=RNga4mjqcYD7gQ@mail.gmail.com>
Related-Discussion: <57EE93C8.8080504@postgrespro.ru>
ExecInitCteScan supposed that it didn't have to do anything to the extra
tuplestore read pointer it gets from tuplestore_alloc_read_pointer.
However, it needs this read pointer to be positioned at the start of the
tuplestore, while tuplestore_alloc_read_pointer is actually defined as
cloning the current position of read pointer 0. In normal situations
that accidentally works because we initialize the whole plan tree at once,
before anything gets read. But it fails in an EvalPlanQual recheck, as
illustrated in bug #14328 from Dima Pavlov. To fix, just forcibly rewind
the pointer after tuplestore_alloc_read_pointer. The cost of doing so is
negligible unless the tuplestore is already in TSS_READFILE state, which
wouldn't happen in normal cases. We could consider altering tuplestore's
API to make that case cheaper, but that would make for a more invasive
back-patch and it doesn't seem worth it.
This has been broken probably for as long as we've had CTEs, so back-patch
to all supported branches.
Discussion: <32468.1474548308@sss.pgh.pa.us>
INSERT .. ON CONFLICT runs a pre-check of the possible conflicting
constraints before performing the actual speculative insertion. In case
the inserted tuple included TOASTed columns the ON CONFLICT condition
would be handled correctly in case the conflict was caught by the
pre-check, but if two transactions entered the speculative insertion
phase at the same time, one would have to re-try, and the code for
aborting a speculative insertion did not handle deleting the
speculatively inserted TOAST datums correctly.
TOAST deletion would fail with "ERROR: attempted to delete invisible
tuple" as we attempted to remove the TOAST tuples using
simple_heap_delete which reasoned that the given tuples should not be
visible to the command that wrote them.
This commit updates the heap_abort_speculative() function which aborts
the conflicting tuple to use itself, via toast_delete, for deleting
associated TOAST datums. Like before, the inserted toast rows are not
marked as being speculative.
This commit also adds a isolationtester spec test, exercising the
relevant code path. Unfortunately 9.5 cannot handle two waiting
sessions, and thus cannot execute this test.
Reported-By: Viren Negi, Oskari Saarenmaa
Author: Oskari Saarenmaa, edited a bit by me
Bug: #14150
Discussion: <20160519123338.12513.20271@wrigleys.postgresql.org>
Backpatch: 9.5, where ON CONFLICT was introduced
When key-share locking a tuple that has been not-key-updated, and the
update is a committed transaction, in some cases we raised
serializability errors:
ERROR: could not serialize access due to concurrent update
Because the key-share doesn't conflict with the update, the error is
unnecessary and inconsistent with the case that the update hasn't
committed yet. This causes problems for some usage patterns, even if it
can be claimed that it's sufficient to retry the aborted transaction:
given a steady stream of updating transactions and a long locking
transaction, the long transaction can be starved indefinitely despite
multiple retries.
To fix, we recognize that HeapTupleSatisfiesUpdate can return
HeapTupleUpdated when an updating transaction has committed, and that we
need to deal with that case exactly as if it were a non-committed
update: verify whether the two operations conflict, and if not, carry on
normally. If they do conflict, however, there is a difference: in the
HeapTupleBeingUpdated case we can just sleep until the concurrent
transaction is gone, while in the HeapTupleUpdated case this is not
possible and we must raise an error instead.
Per trouble report from Olivier Dony.
In addition to a couple of test cases that verify the changed behavior,
I added a test case to verify the behavior that remains unchanged,
namely that errors are raised when a update that modifies the key is
used. That must still generate serializability errors. One
pre-existing test case changes behavior; per discussion, the new
behavior is actually the desired one.
Discussion: https://www.postgresql.org/message-id/560AA479.4080807@odoo.comhttps://www.postgresql.org/message-id/20151014164844.3019.25750@wrigleys.postgresql.org
Backpatch to 9.3, where the problem appeared.
While prior to this patch the user-visible effect on the database
of any set of successfully committed serializable transactions was
always consistent with some one-at-a-time order of execution of
those transactions, the presence of declarative constraints could
allow errors to occur which were not possible in any such ordering,
and developers had no good workarounds to prevent user-facing
errors where they were not necessary or desired. This patch adds
a check for serialization failure ahead of duplicate key checking
so that if a developer explicitly (redundantly) checks for the
pre-existing value they will get the desired serialization failure
where the problem is caused by a concurrent serializable
transaction; otherwise they will get a duplicate key error.
While it would be better if the reads performed by the constraints
could count as part of the work of the transaction for
serialization failure checking, and we will hopefully get there
some day, this patch allows a clean and reliable way for developers
to work around the issue. In many cases existing code will already
be doing the right thing for this to "just work".
Author: Thomas Munro, with minor editing of docs by me
Reviewed-by: Marko Tiikkaja, Kevin Grittner
In commit afb9249d06, we (probably I) made ExecLockRows assign
null test tuples to all relations of the query while setting up to do an
EvalPlanQual recheck for a newly-updated locked row. This was sheerest
brain fade: we should only set test tuples for relations that are lockable
by the LockRows node, and in particular empty test tuples are only sensible
for inheritance child relations that weren't the source of the current
tuple from their inheritance tree. Setting a null test tuple for an
unrelated table causes it to return NULLs when it should not, as exhibited
in bug #14034 from Bronislav Houdek. To add insult to injury, doing it the
wrong way required two loops where one would suffice; so the corrected code
is even a bit shorter and faster.
Add a regression test case based on his example, and back-patch to 9.5
where the bug was introduced.
I noticed that the async-notify test results in log messages like these:
LOG: could not send data to client: Broken pipe
FATAL: connection to client lost
This is because it unceremoniously disconnects a client session that is
about to have some NOTIFY messages delivered to it. Such log messages
during a regression test might well cause people to go looking for a
problem that doesn't really exist (it did cause me to waste some time that
way). We can shut it up by adding an UNLISTEN command to session teardown.
Patch HEAD only; this doesn't seem significant enough to back-patch.
The original coding of the test was relying too much on the ordering in
which backends are awakened once an advisory lock which they wait for is
released. Change the code so that each backend uses its own advisory
lock instead, so that the output becomes stable. Also add a few seconds
of sleep between lock releases, so that the test isn't broken in
overloaded buildfarm animals, as suggested by Tom Lane.
Per buildfarm members spoonbill and guaibasaurus.
Discussion: https://www.postgresql.org/message-id/19294.1456551587%40sss.pgh.pa.us
In 0e5680f473, I fixed a bug in heapam that caused spurious deadlocks
when multiple updates concurrently attempted to modify the old version
of an updated tuple whose new version was key-share locked. I proposed
an isolationtester spec file that reproduced the bug, but back then
isolationtester wasn't mature enough to be able to run it. Now that
38f8bdcac4 is in the tree, we can have this spec file too.
Discussion: https://www.postgresql.org/message-id/20141212205254.GC1768%40alvh.no-ip.org
The previous value of 5s is inadequate for the buildfarm's
CLOBBER_CACHE_ALWAYS animals: they take long enough to do the is-it-waiting
queries that the timeout expires, allowing the database state to change,
before isolationtester is done looking. Perhaps 10s will be enough.
(If it isn't, I'm inclined to reduce the number of sessions involved.)
It turns out that there is a second race condition in the new deadlock-hard
test: once the deadlock detector fires, it's uncertain whether step s7a8 or
step s8a1 will report first, because killing s8's transaction unblocks s7.
So far, s7 has only been seen to report first in CLOBBER_CACHE_ALWAYS
builds, but it's pretty reproducible there, and in theory it should
sometimes occur in normal builds too. If s7 were a bit slower than usual,
that could also break the test, since the existing expected-file assumes
that we'll see s7a8 report the first time we check it after s8a1 completes.
To fix, add a post-lock delay to s7a8.
The new deadlock-soft-2 test has a timing dependency too: it supposes
that isolationtester will detect step s1b as waiting before the deadlock
detector runs and grants it the lock. Adjust deadlock_timeout to ensure
that that's true even in CLOBBER_CACHE_ALWAYS builds, where the wait
detection query is quite slow. Per buildfarm member jaguarundi.
The original formulation of 4c9864b9b4
was extremely timing-sensitive, because it arranged for the deadlock
detector to be running (and possibly unblocking the current query)
at almost exactly the same time as isolationtester would be probing
to see if the query is blocked. The committed expected-file assumed
that the deadlock detection would finish first, but we see the opposite
on both fast and slow buildfarm animals. Adjust the deadlock timeout
settings to make it predictable that isolationtester *will* see the
query as waiting before deadlock detection unblocks it.
I used a 5s timeout for the same reasons mentioned in
a7921f71a3.
This is a necessary prerequisite for forthcoming changes to allow deadlock
scenarios to be tested by the isolation tester. It is also a good idea on
general principle, since these scenarios add no useful test coverage not
provided by other scenarios, but do to take time to execute.
Commit 2834855cb added a not-very-carefully-thought-out isolation test
to check a BRIN index bug fix. The test depended on the availability
of the pageinspect contrib module, which meant it did not work in
several common testing scenarios such as "make check-world". It's not
clear whether we want a core test depending on a contrib module like
that, but in any case, failing to deal with the possibility that the
module isn't present in the installation-under-test is not acceptable.
Remove that test pending some better solution.
For correctness of summarization results, it is critical that the
snapshot used during the summarization scan is able to see all tuples
that are live to all transactions -- including tuples inserted or
deleted by in-progress transactions. Otherwise, it would be possible
for a transaction to insert a tuple, then idle for a long time while a
concurrent transaction executes summarization of the range: this would
result in the inserted value not being considered in the summary.
Previously we were trying to use a MVCC snapshot in conjunction with
adding a "placeholder" tuple in the index: the snapshot would see all
committed tuples, and the placeholder tuple would catch insertions by
any new inserters. The hole is that prior insertions by transactions
that are still in progress by the time the MVCC snapshot was taken were
ignored.
Kevin Grittner reported this as a bogus error message during vacuum with
default transaction isolation mode set to repeatable read (because the
error report mentioned a function name not being invoked during), but
the problem is larger than that.
To fix, tweak IndexBuildHeapRangeScan to have a new mode that behaves
the way we need using SnapshotAny visibility rules. This change
simplifies the BRIN code a bit, mainly by removing large comments that
were mistaken. Instead, rely on the SnapshotAny semantics to provide
what it needs. (The business about a placeholder tuple needs to remain:
that covers the case that a transaction inserts a a tuple in a page that
summarization already scanned.)
Discussion: https://www.postgresql.org/message-id/20150731175700.GX2441@postgresql.org
In passing, remove a couple of unused declarations from brin.h and
reword a comment to be proper English. This part submitted by Kevin
Grittner.
Backpatch to 9.5, where BRIN was introduced.
A Salesforce colleague of mine griped that the regression tests don't
exercise EvalPlanQualFetchRowMarks() and allied routines. Which is
a fair complaint. Add test cases that go through the REFERENCE and COPY
code paths. Unfortunately we don't have sufficient infrastructure right
now to exercise the FDW code path in the isolation tests, but this is
surely better than before.
This tells you what fraction of NOTIFY's queue is currently filled.
Brendan Jurd, reviewed by Merlin Moncure and Gurjeet Singh. A few
further tweaks by me.
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.
Up to now, the "child" executor state trees generated for EvalPlanQual
rechecks have simply shared the ResultRelInfo arrays used for the original
execution tree. However, this leads to dangling-pointer problems, because
ExecInitModifyTable() is all too willing to scribble on some fields of the
ResultRelInfo(s) even when it's being run in one of those child trees.
This trashes those fields from the perspective of the parent tree, because
even if the generated subtree is logically identical to what was in use in
the parent, it's in a memory context that will go away when we're done
with the child state tree.
We do however want to share information in the direction from the parent
down to the children; in particular, fields such as es_instrument *must*
be shared or we'll lose the stats arising from execution of the children.
So the simplest fix is to make a copy of the parent's ResultRelInfo array,
but not copy any fields back at end of child execution.
Per report from Manuel Kniep. The added isolation test is based on his
example. In an unpatched memory-clobber-enabled build it will reliably
fail with "ctid is NULL" errors in all branches back to 9.1, as a
consequence of junkfilter->jf_junkAttNo being overwritten with $7f7f.
This test cannot be run as-is before that for lack of WITH syntax; but
I have no doubt that some variant of this problem can arise in older
branches, so apply the code change all the way back.
Commit 0e5680f473 contained a thinko
mixing LOCKMODE with LockTupleMode. This caused misbehavior in the case
where a tuple is marked with a multixact with at most a FOR SHARE lock,
and another transaction tries to acquire a FOR NO KEY EXCLUSIVE lock;
this case should block but doesn't.
Include a new isolation tester spec file to explicitely try all the
tuple lock combinations; without the fix it shows the problem:
starting permutation: s1_begin s1_lcksvpt s1_tuplock2 s2_tuplock3 s1_commit
step s1_begin: BEGIN;
step s1_lcksvpt: SELECT * FROM multixact_conflict FOR KEY SHARE; SAVEPOINT foo;
a
1
step s1_tuplock2: SELECT * FROM multixact_conflict FOR SHARE;
a
1
step s2_tuplock3: SELECT * FROM multixact_conflict FOR NO KEY UPDATE;
a
1
step s1_commit: COMMIT;
With the fixed code, step s2_tuplock3 blocks until session 1 commits,
which is the correct behavior.
All other cases behave correctly.
Backpatch to 9.3, like the commit that introduced the problem.
In READ COMMITTED mode, if a SELECT FOR UPDATE discovers it has to redo
WHERE-clause checking on rows that have been updated since the SELECT's
snapshot, it invokes EvalPlanQual processing to do that. If this first
occurs within a non-first child table of an inheritance tree, the previous
coding could accidentally re-return a matching row from an earlier,
already-scanned child table. (And, to add insult to injury, I think this
could make it miss returning a row that should have been returned, if the
updated row that this happens on should still have passed the WHERE qual.)
Per report from Kyotaro Horiguchi; the added isolation test is based on his
test case.
This has been broken for quite awhile, so back-patch to all supported
branches.
This clause changes the behavior of SELECT locking clauses in the
presence of locked rows: instead of causing a process to block waiting
for the locks held by other processes (or raise an error, with NOWAIT),
SKIP LOCKED makes the new reader skip over such rows. While this is not
appropriate behavior for general purposes, there are some cases in which
it is useful, such as queue-like tables.
Catalog version bumped because this patch changes the representation of
stored rules.
Reviewed by Craig Ringer (based on a previous attempt at an
implementation by Simon Riggs, who also provided input on the syntax
used in the current patch), David Rowley, and Álvaro Herrera.
Author: Thomas Munro
If SELECT FOR UPDATE NOWAIT tries to lock a tuple that is concurrently
being updated, it might fail to honor its NOWAIT specification and block
instead of raising an error.
Fix by adding a no-wait flag to EvalPlanQualFetch which it can pass down
to heap_lock_tuple; also use it in EvalPlanQualFetch itself to avoid
blocking while waiting for a concurrent transaction.
Authors: Craig Ringer and Thomas Munro, tweaked by Álvaro
http://www.postgresql.org/message-id/51FB6703.9090801@2ndquadrant.com
Per Thomas Munro in the course of his SKIP LOCKED feature submission,
who also provided one of the isolation test specs.
Backpatch to 9.4, because that's as far back as it applies without
conflicts (although the bug goes all the way back). To that branch also
backpatch Thomas Munro's new NOWAIT test cases, committed in master by
Heikki as commit 9ee16b49f0 .
If a tuple was locked by transaction A, and transaction B updated it,
the new version of the tuple created by B would be locked by A, yet
visible only to B; due to an oversight in HeapTupleSatisfiesUpdate, the
lock held by A wouldn't get checked if transaction B later deleted (or
key-updated) the new version of the tuple. This might cause referential
integrity checks to give false positives (that is, allow deletes that
should have been rejected).
This is an easy oversight to have made, because prior to improved tuple
locks in commit 0ac5ad5134 it wasn't possible to have tuples created by
our own transaction that were also locked by remote transactions, and so
locks weren't even considered in that code path.
It is recommended that foreign keys be rechecked manually in bulk after
installing this update, in case some referenced rows are missing with
some referencing row remaining.
Per bug reported by Daniel Wood in
CAPweHKe5QQ1747X2c0tA=5zf4YnS2xcvGf13Opd-1Mq24rF1cQ@mail.gmail.com
In 247c76a989, I added some code to do fine-grained checking of
MultiXact status of locking/updating transactions when traversing an
update chain. There was a thinko in that patch which would have the
traversing abort, that is return HeapTupleUpdated, when the other
transaction is a committed lock-only. In this case we should ignore it
and return success instead. Of course, in the case where there is a
committed update, HeapTupleUpdated is the correct return value.
A user-visible symptom of this bug is that in REPEATABLE READ and
SERIALIZABLE transaction isolation modes spurious serializability errors
can occur:
ERROR: could not serialize access due to concurrent update
In order for this to happen, there needs to be a tuple that's key-share-
locked and also updated, and the update must abort; a subsequent
transaction trying to acquire a new lock on that tuple would abort with
the above error. The reason is that the initial FOR KEY SHARE is seen
as committed by the new locking transaction, which triggers this bug.
(If the UPDATE commits, then the serialization error is correctly
reported.)
When running a query in READ COMMITTED mode, what happens is that the
locking is aborted by the HeapTupleUpdated return value, then
EvalPlanQual fetches the newest version of the tuple, which is then the
only version that gets locked. (The second time the tuple is checked
there is no misbehavior on the committed lock-only, because it's not
checked by the code that traverses update chains; so no bug.) Only the
newest version of the tuple is locked, not older ones, but this is
harmless.
The isolation test added by this commit illustrates the desired
behavior, including the proper serialization errors that get thrown.
Backpatch to 9.3.
The buildfarm members using -DCLOBBER_CACHE_ALWAYS still don't like this
test. Some experimentation shows that on my machine, isolationtester's
query to check for "waiting" state takes 2 to 2.5 seconds to bind+execute
under -DCLOBBER_CACHE_ALWAYS. Set the timeouts to 5 seconds to leave some
headroom for possibly-slower buildfarm critters.
Really we ought to fix the "waiting" query, which is not only horridly
slow but outright wrong in detail; and then maybe we can back off these
timeouts. But right now I'm just trying to get the buildfarm green again.
Buildfarm member friarbird doesn't like this test as-committed, evidently
because it's so slow that the test framework doesn't reliably notice that
the backend is waiting before the timeout goes off. (This is not totally
surprising, since friarbird builds with -DCLOBBER_CACHE_ALWAYS.) Increase
the timeout delay from 1 second to 2 in hopes of resolving that problem.
This GUC allows limiting the time spent waiting to acquire any one
heavyweight lock.
In support of this, improve the recently-added timeout infrastructure
to permit efficiently enabling or disabling multiple timeouts at once.
That reduces the performance hit from turning on lock_timeout, though
it's still not zero.
Zoltán Böszörményi, reviewed by Tom Lane,
Stephen Frost, and Hari Babu
This patch introduces two additional lock modes for tuples: "SELECT FOR
KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each
other, in contrast with already existing "SELECT FOR SHARE" and "SELECT
FOR UPDATE". UPDATE commands that do not modify the values stored in
the columns that are part of the key of the tuple now grab a SELECT FOR
NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently
with tuple locks of the FOR KEY SHARE variety.
Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this
means the concurrency improvement applies to them, which is the whole
point of this patch.
The added tuple lock semantics require some rejiggering of the multixact
module, so that the locking level that each transaction is holding can
be stored alongside its Xid. Also, multixacts now need to persist
across server restarts and crashes, because they can now represent not
only tuple locks, but also tuple updates. This means we need more
careful tracking of lifetime of pg_multixact SLRU files; since they now
persist longer, we require more infrastructure to figure out when they
can be removed. pg_upgrade also needs to be careful to copy
pg_multixact files over from the old server to the new, or at least part
of multixact.c state, depending on the versions of the old and new
servers.
Tuple time qualification rules (HeapTupleSatisfies routines) need to be
careful not to consider tuples with the "is multi" infomask bit set as
being only locked; they might need to look up MultiXact values (i.e.
possibly do pg_multixact I/O) to find out the Xid that updated a tuple,
whereas they previously were assured to only use information readily
available from the tuple header. This is considered acceptable, because
the extra I/O would involve cases that would previously cause some
commands to block waiting for concurrent transactions to finish.
Another important change is the fact that locking tuples that have
previously been updated causes the future versions to be marked as
locked, too; this is essential for correctness of foreign key checks.
This causes additional WAL-logging, also (there was previously a single
WAL record for a locked tuple; now there are as many as updated copies
of the tuple there exist.)
With all this in place, contention related to tuples being checked by
foreign key rules should be much reduced.
As a bonus, the old behavior that a subtransaction grabbing a stronger
tuple lock than the parent (sub)transaction held on a given tuple and
later aborting caused the weaker lock to be lost, has been fixed.
Many new spec files were added for isolation tester framework, to ensure
overall behavior is sane. There's probably room for several more tests.
There were several reviewers of this patch; in particular, Noah Misch
and Andres Freund spent considerable time in it. Original idea for the
patch came from Simon Riggs, after a problem report by Joel Jacobson.
Most code is from me, with contributions from Marti Raudsepp, Alexander
Shulgin, Noah Misch and Andres Freund.
This patch was discussed in several pgsql-hackers threads; the most
important start at the following message-ids:
AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com1290721684-sup-3951@alvh.no-ip.org1294953201-sup-2099@alvh.no-ip.org1320343602-sup-2290@alvh.no-ip.org1339690386-sup-8927@alvh.no-ip.org4FE5FF020200002500048A3D@gw.wicourts.gov4FEAB90A0200002500048B7D@gw.wicourts.gov
Serializable Snapshot Isolation used for serializable transactions
depends on acquiring SIRead locks on all heap relation tuples which
are used to generate the query result, so that a later delete or
update of any of the tuples can flag a read-write conflict between
transactions. This is normally handled in heapam.c, with tuple level
locking. Since an index-only scan avoids heap access in many cases,
building the result from the index tuple, the necessary predicate
locks were not being acquired for all tuples in an index-only scan.
To prevent problems with tuple IDs which are vacuumed and re-used
while the transaction still matters, the xmin of the tuple is part of
the tag for the tuple lock. Since xmin is not available to the
index-only scan for result rows generated from the index tuples, it
is not possible to acquire a tuple-level predicate lock in such
cases, in spite of having the tid. If we went to the heap to get the
xmin value, it would no longer be an index-only scan. Rather than
prohibit index-only scans under serializable transaction isolation,
we acquire an SIRead lock on the page containing the tuple, when it
was not necessary to visit the heap for other reasons.
Backpatch to 9.2.
Kevin Grittner and Tom Lane
Much more could be done here, but at least now we have *some* automated
test coverage of that mechanism. In particular this tests the writable-CTE
case reported by Phil Sorber.
In passing, remove isolationtester's arbitrary restriction on the number of
steps in a permutation list. I used this so that a single spec file could
be used to run several related test scenarios, but there are other possible
reasons to want a step series that's not exactly a permutation. Improve
documentation and fix a couple other nits as well.
