Tom Lane [Wed, 14 Jun 2017 15:10:05 +0000 (11:10 -0400)]
Fix no-longer-valid shortcuts in expression_returns_set().
expression_returns_set() used to short-circuit its recursion upon
seeing certain node types, such as DistinctExpr, that it knew the
executor did not support set-valued arguments for. That was never
inherent, though, just a reflection of laziness in execQual.c.
With the new implementation of SRFs there is no reason to think
that any scalar-valued expression node could not have a set-valued
subexpression, except for AggRefs and WindowFuncs where we know there
is a parser check rejecting it. And indeed, the shortcut causes
unexpected failures for cases such as a SRF underneath DistinctExpr,
because the planner stops looking for SRFs too soon.
Tom Lane [Wed, 14 Jun 2017 14:26:46 +0000 (10:26 -0400)]
Fix violations of CatalogTupleInsert/Update/Delete abstraction.
In commits 2f5c9d9c9 and ab0289651 we invented an abstraction layer
to insulate catalog manipulations from direct heap update calls.
But evidently some patches that hadn't landed in-tree at that point
didn't get the memo completely. Fix a couple of direct calls to
simple_heap_delete to use CatalogTupleDelete instead; these appear
to have been added in commits 7c4f52409 and 7b504eb28. This change is
purely cosmetic ATM, but there's no point in having an abstraction layer
if we allow random code to break it.
Dean Rasheed [Wed, 14 Jun 2017 08:00:01 +0000 (09:00 +0100)]
Teach PL/pgSQL about partitioned tables.
Table partitioning, introduced in commit f0e44751d7, added a new
relkind - RELKIND_PARTITIONED_TABLE. Update a couple of places in
PL/pgSQL to handle it. Specifically plpgsql_parse_cwordtype() and
build_row_from_class() needed updating in order to make table%ROWTYPE
and table.col%TYPE work for partitioned tables.
Dean Rasheed [Wed, 14 Jun 2017 07:43:40 +0000 (08:43 +0100)]
Teach RemoveRoleFromObjectPolicy() about partitioned tables.
Table partitioning, introduced in commit f0e44751d7, added a new
relkind - RELKIND_PARTITIONED_TABLE. Update
RemoveRoleFromObjectPolicy() to handle it, otherwise DROP OWNED BY
will fail if the role has any RLS policies referring to partitioned
tables.
Tom Lane [Wed, 14 Jun 2017 03:46:39 +0000 (23:46 -0400)]
Disallow set-returning functions inside CASE or COALESCE.
When we reimplemented SRFs in commit 69f4b9c85, our initial choice was
to allow the behavior to vary from historical practice in cases where a
SRF call appeared within a conditional-execution construct (currently,
only CASE or COALESCE). But that was controversial to begin with, and
subsequent discussion has resulted in a consensus that it's better to
throw an error instead of executing the query differently from before,
so long as we can provide a reasonably clear error message and a way to
rewrite the query.
Hence, add a parser mechanism to allow detection of such cases during
parse analysis. The mechanism just requires storing, in the ParseState,
a pointer to the set-returning FuncExpr or OpExpr most recently emitted
by parse analysis. Then the parsing functions for CASE and COALESCE can
detect the presence of a SRF in their arguments by noting whether this
pointer changes while analyzing their arguments. Furthermore, if it does,
it provides a suitable error cursor location for the complaint. (This
means that if there's more than one SRF in the arguments, the error will
point at the last one to be analyzed not the first. While connoisseurs of
parsing behavior might find that odd, it's unlikely the average user would
ever notice.)
While at it, we can also provide more specific error messages than before
about some pre-existing restrictions, such as no-SRFs-within-aggregates.
Also, reject at parse time cases where a NULLIF or IS DISTINCT FROM
construct would need to return a set. We've never supported that, but the
restriction is depended on in more subtle ways now, so it seems wise to
detect it at the start.
Also, provide some documentation about how to rewrite a SRF-within-CASE
query using a custom wrapper SRF.
It turns out that the information_schema.user_mapping_options view
contained an instance of exactly the behavior we're now forbidding; but
rewriting it makes it more clear and safer too.
initdb forced because of user_mapping_options change.
Patch by me, with error message suggestions from Alvaro Herrera and
Andres Freund, pursuant to a complaint from Regina Obe.
Peter Eisentraut [Tue, 13 Jun 2017 20:10:11 +0000 (16:10 -0400)]
doc: Update example version numbers in pg_upgrade documentation
The exact numbers don't matter, since they are examples, but it was
looking quite dated.
For the target version, we now automatically substitute the current
major version. The updated example source version should be good for a
couple of years.
Tom Lane [Tue, 13 Jun 2017 17:05:59 +0000 (13:05 -0400)]
Re-run pgindent.
This is just to have a clean base state for testing of Piotr Stefaniak's
latest version of FreeBSD indent. I fixed up a couple of places where
pgindent would have changed format not-nicely. perltidy not included.
This doesn't actually matter at present, because the current code
never consults null_index for range partitions. However, leaving
it uninitialized is still a bad idea, so let's not do that.
Dean Rasheed [Tue, 13 Jun 2017 16:30:36 +0000 (17:30 +0100)]
Teach relation_is_updatable() about partitioned tables.
Table partitioning, introduced in commit f0e44751d7, added a new
relkind - RELKIND_PARTITIONED_TABLE. Update relation_is_updatable() to
handle it. Specifically, partitioned tables and simple views built on
top of them are updatable.
This affects the SQL-callable functions pg_relation_is_updatable() and
pg_column_is_updatable(), and the views information_schema.views and
information_schema.columns.
Tom Lane [Tue, 13 Jun 2017 14:54:39 +0000 (10:54 -0400)]
In initdb, defend against assignment of NULL values to not-null columns.
Previously, you could write _null_ in a BKI DATA line for a column that's
supposed to be NOT NULL and initdb would let it pass, probably breaking
subsequent accesses to the row. No doubt the original coding overlooked
this simple sanity check because in the beginning we didn't have any way
to mark catalog columns NOT NULL at initdb time.
Tom Lane [Tue, 13 Jun 2017 03:29:44 +0000 (23:29 -0400)]
Fix confusion about number of subplans in partitioned INSERT setup.
ExecInitModifyTable() thought there was a plan per partition, but no,
there's only one. The problem had escaped detection so far because there
would only be visible misbehavior if there were a SubPlan (not an InitPlan)
in the quals being duplicated for each partition. However, valgrind
detected a bogus memory access in test cases added by commit 4f7a95be2,
and investigation of that led to discovery of the bug. The additional
test case added here crashes without the patch.
Tom Lane [Tue, 13 Jun 2017 00:04:32 +0000 (20:04 -0400)]
Assert that we don't invent relfilenodes or type OIDs in binary upgrade.
During pg_upgrade's restore run, all relfilenode choices should be
overridden by commands in the dump script. If we ever find ourselves
choosing a relfilenode in the ordinary way, someone blew it. Likewise for
pg_type OIDs. Since pg_upgrade might well succeed anyway, if there happens
not to be a conflict during the regression test run, we need assertions
here to keep us on the straight and narrow.
We might someday be able to remove the assertion in GetNewRelFileNode,
if pg_upgrade is rewritten to remove its assumption that old and new
relfilenodes always match. But it's hard to see how to get rid of the
pg_type OID constraint, since those OIDs are embedded in user tables
in some cases.
Back-patch as far as 9.5, because of the risk of back-patches breaking
something here even if it works in HEAD. I'd prefer to go back further,
but 9.4 fails both assertions due to get_rel_infos()'s use of a temporary
table. We can't use the later-branch solution of a CTE for compatibility
reasons (cf commit 5d16332e9), and it doesn't seem worth inventing some
other way to do the query. (I did check, by dint of changing the Asserts
to elog(WARNING), that there are no other cases of unwanted OID assignments
during 9.4's regression test run.)
Tom Lane [Mon, 12 Jun 2017 20:57:31 +0000 (16:57 -0400)]
Fix ALTER SEQUENCE OWNED BY to not rewrite the sequence relation.
It's not necessary for it to do that, since OWNED BY requires only ordinary
catalog updates and doesn't affect future sequence values. And pg_upgrade
needs to use OWNED BY without having it change the sequence's relfilenode.
Commit 3d79013b9 broke this by making all forms of ALTER SEQUENCE change
the relfilenode; that seems to be the explanation for the hard-to-reproduce
buildfarm failures we've been seeing since then.
Peter Eisentraut [Mon, 12 Jun 2017 19:57:22 +0000 (15:57 -0400)]
Add ICU_CFLAGS to global CPPFLAGS
The original code only added ICU_CFLAGS to the backend build. But it is
also needed for building external modules that include pg_locale.h. So
add it to the global CPPFLAGS. (This is only relevant if ICU is not in
a compiler default path, so it apparently hasn't bitten many.)
Stop table sync workers when subscription relation entry is removed
When a table sync worker is in waiting state and the subscription table
entry is removed because of a concurrent subscription refresh, the
worker could be left orphaned. To avoid that, explicitly stop the
worker when the pg_subscription_rel entry is removed.
Tom Lane [Sun, 11 Jun 2017 23:00:01 +0000 (19:00 -0400)]
Handle unqualified SEQUENCE NAME options properly in parse_utilcmd.c.
generateSerialExtraStmts() was sloppy about handling the case where
SEQUENCE NAME is given with a not-schema-qualified name. It was generating
a CreateSeqStmt with an unqualified sequence name, and an AlterSeqStmt
whose "owned_by" DefElem contained a T_String Value with a null string
pointer in the schema-name position. The generated nextval() argument was
also underqualified. This accidentally failed to fail at runtime, but only
so long as the current default creation namespace at runtime is the right
namespace. That's bogus; the parse-time transformation is supposed to be
inserting the right schema name in all cases, so as to avoid any possible
skew in that selection. I'm not sure this could fail in pg_dump's usage,
but it's still wrong; we have had real bugs in this area before adopting
the policy that parse_utilcmd.c should generate only fully-qualified
auxiliary commands. A slightly lesser problem, which is what led me to
notice this in the first place, is that pprint() dumped core on the
AlterSeqStmt because of the bogus T_String.
Noted while poking into the open problem with ALTER SEQUENCE breaking
pg_upgrade.
Joe Conway [Sun, 11 Jun 2017 15:51:18 +0000 (08:51 -0700)]
Apply RLS policies to partitioned tables.
The new partitioned table capability added a new relkind, namely
RELKIND_PARTITIONED_TABLE. Update fireRIRrules() to apply RLS
policies on RELKIND_PARTITIONED_TABLE as it does RELKIND_RELATION.
In addition, add RLS regression test coverage for partitioned tables.
Issue raised by Fakhroutdinov Evgenievich and patch by Mike Palmiotto.
Regression test editorializing by me.
Andrew Dunstan [Sat, 10 Jun 2017 14:19:06 +0000 (10:19 -0400)]
Take PROVE_FLAGS from the command line but not the environment
This reverts commit 56b6ef893fee9e9bf47d927a02f4d1ea911f4d9c and instead
makes vcregress.pl parse out PROVE_FLAGS from a command line argument
when doing a TAP test, thus making it consistent with the makefile
treatment.
Silence warning about uninitialized 'ret' variable on some compilers.
If the compiler doesn't notice that the switch-statement handles all
possible values of the enum, it might complain that 'ret' is being used
without initialization. Jeff Janes reported that on gcc 4.4.7.
Improve tablesync behavior with concurrent changes
When a table is removed from a subscription before the tablesync worker
could start, this would previously result in an error when reading
pg_subscription_rel. Now we just ignore this.
Give a better error message on invalid hostaddr option.
If you accidentally pass a host name in the hostaddr option, e.g.
hostaddr=localhost, you get an error like:
psql: could not translate host name "localhost" to address: Name or service not known
That's a bit confusing, because it implies that we tried to look up
"localhost" in DNS, but it failed. To make it more clear that we tried to
parse "localhost" as a numeric network address, change the message to:
psql: could not parse network address "localhost": Name or service not known
Andres Freund [Thu, 8 Jun 2017 22:00:53 +0000 (15:00 -0700)]
Use standard interrupt handling in logical replication launcher.
Previously the exit handling was only able to exit from within the
main loop, and not from within the backend code it calls. Fix that by
using the standard die() SIGTERM handler, and adding the necessary
CHECK_FOR_INTERRUPTS() call.
This requires adding yet another process-type-specific branch to
ProcessInterrupts(), which hints that we probably should generalize
that handling. But that's work for another day.
Author: Petr Jelinek Reviewed-By: Andres Freund
Discussion: https://postgr.es/m/fe072153-babd-3b5d-8052-73527a6eb657@2ndquadrant.com
Andres Freund [Thu, 8 Jun 2017 21:42:18 +0000 (14:42 -0700)]
Again report a useful error message when walreceiver's connection closes.
Since 7c4f52409a8c (merged in v10), a shutdown master is reported as
FATAL: unexpected result after CommandComplete: server closed the connection unexpectedly
by walsender. It used to be
LOG: replication terminated by primary server
FATAL: could not send end-of-streaming message to primary: no COPY in progress
while the old message clearly is not perfect, it's definitely better
than what's reported now.
The change comes from the attempt to handle finished COPYs without
erroring out, needed for the new logical replication, which wasn't
needed before.
There's probably better ways to handle this, but for now just
explicitly check for a closed connection.
Author: Petr Jelinek Reviewed-By: Andres Freund
Discussion: https://postgr.es/m/f7c7dd08-855c-e4ed-41f4-d064a6c0665a@2ndquadrant.com
Backpatch: -
Tom Lane [Thu, 8 Jun 2017 17:48:27 +0000 (13:48 -0400)]
Fix bit-rot in pg_upgrade's test.sh, and improve documentation.
Doing a cross-version upgrade test with test.sh evidently hasn't been
tested since circa 9.2, because the script lacked case branches for
old-version servers newer than 9.1. Future-proof that a bit, and
clean up breakage induced by our recent drop of V0 function call
protocol (namely that oldstyle_length() isn't in the regression
suite anymore).
(This isn't enough to make the test work perfectly cleanly across
versions, but at least it finishes and provides dump files that
you can diff manually. One issue I didn't touch is that we might
want to execute the "reindex_hash.sql" file in the new DB before
dumping it, so that the hash indexes don't vanish from the dump.)
Improve the TESTING doc file: put the tl;dr version at the top not
the bottom, and bring its explanation of how to run a cross-version
test up to speed, since the installcheck target isn't there and won't
be resurrected. Improve the comment in the Makefile about why not.
In passing, teach .gitignore and "make clean" about a couple more
junk output files.
Most of the improvements were in the new SCRAM code:
* In SCRAM protocol violation messages, use errdetail to provide the
details.
* If pg_backend_random() fails, throw an ERROR rather than just LOG. We
shouldn't continue authentication if we can't generate a random nonce.
* Use ereport() rather than elog() for the "invalid SCRAM verifier"
messages. They shouldn't happen, if everything works, but it's not
inconceivable that someone would have invalid scram verifiers in
pg_authid, e.g. if a broken client application was used to generate the
verifier.
But this change applied to old code:
* Use ERROR rather than COMMERROR for protocol violation errors. There's
no reason to not tell the client what they did wrong. The client might be
confused already, so that it cannot read and display the error correctly,
but let's at least try. In the "invalid password packet size" case, we
used to actually continue with authentication anyway, but that is now a
hard error.
Patch by Michael Paquier and me. Thanks to Daniel Varrazzo for spotting
the typo in one of the messages that spurred the discussion and these
larger changes.
Joe Conway [Thu, 8 Jun 2017 00:54:33 +0000 (17:54 -0700)]
Fix contrib/sepgsql regr tests for tup-routing constraint check change.
Commit 15ce775 changed tuple-routing constraint checking logic.
This affects the expected output for contrib/sepgsql, because
there's no longer LOG entries reporting allowance of int4eq()
execution. Per buildfarm.
Tom Lane [Wed, 7 Jun 2017 21:23:38 +0000 (17:23 -0400)]
Docs: improve CREATE TABLE ref page's discussion of partition bounds.
Clarify in the syntax synopsis that partition bound values must be
exactly numeric literals or string literals; previously it
said "bound_literal" which was defined nowhere.
Replace confusing --- and, I think, incorrect in detail --- definition
of how range bounds work with a reference to row-wise comparison plus
a concrete example (which I stole from Robert Haas).
Robert Haas [Wed, 7 Jun 2017 19:14:55 +0000 (15:14 -0400)]
postgres_fdw: Allow cancellation of transaction control commands.
Commit f039eaac7131ef2a4cf63a10cf98486f8bcd09d2, later back-patched
with commit 1b812afb0eafe125b820cc3b95e7ca03821aa675, allowed many of
the queries issued by postgres_fdw to fetch remote data to respond to
cancel interrupts in a timely fashion. However, it didn't do anything
about the transaction control commands, which remained
noninterruptible.
Improve the situation by changing do_sql_command() to retrieve query
results using pgfdw_get_result(), which uses the asynchronous
interface to libpq so that it can check for interrupts every time
libpq returns control. Since this might result in a situation
where we can no longer be sure that the remote transaction state
matches the local transaction state, add a facility to force all
levels of the local transaction to abort if we've lost track of
the remote state; without this, an apparently-successful commit of
the local transaction might fail to commit changes made on the
remote side. Also, add a 60-second timeout for queries issue during
transaction abort; if that expires, give up and mark the state of
the connection as unknown. Drop all such connections when we exit
the local transaction. Together, these changes mean that if we're
aborting the local toplevel transaction anyway, we can just drop the
remote connection in lieu of waiting (possibly for a very long time)
for it to complete an abort.
This still leaves quite a bit of room for improvement. PQcancel()
has no asynchronous interface, so if we get stuck sending the cancel
request we'll still hang. Also, PQsetnonblocking() is not used, which
means we could block uninterruptibly when sending a query. There
might be some other optimizations possible as well. Nonetheless,
this allows us to escape a wait for an unresponsive remote server
quickly in many more cases than previously.
Report by Suraj Kharage. Patch by me and Rafia Sabih. Review
and testing by Amit Kapila and Tushar Ahuja.
A logical replication worker should not insert new rows into
pg_subscription_rel, only update existing rows, so that there are no
races if a concurrent refresh removes rows. Adjust the API to be able
to choose that behavior.
Robert Haas [Wed, 7 Jun 2017 16:45:32 +0000 (12:45 -0400)]
Prevent BEFORE triggers from violating partitioning constraints.
Since tuple-routing implicitly checks the partitioning constraints
at least for the levels of the partitioning hierarchy it traverses,
there's normally no need to revalidate the partitioning constraint
after performing tuple routing. However, if there's a BEFORE trigger
on the target partition, it could modify the tuple, causing the
partitioning constraint to be violated. Catch that case.
Also, instead of checking the root table's partition constraint after
tuple-routing, check it beforehand. Otherwise, the rules for when
the partitioning constraint gets checked get too complicated, because
you sometimes have to check part of the constraint but not all of it.
This effectively reverts commit 39162b2030fb0a35a6bb28dc636b5a71b8df8d1c
in favor of a different approach altogether.
Report by me. Initial debugging by Jeevan Ladhe. Patch by Amit
Langote, reviewed by me.
Clear auth context correctly when re-connecting after failed auth attempt.
If authentication over an SSL connection fails, with sslmode=prefer,
libpq will reconnect without SSL and retry. However, we did not clear
the variables related to GSS, SSPI, and SASL authentication state, when
reconnecting. Because of that, the second authentication attempt would
always fail with a "duplicate GSS/SASL authentication request" error.
pg_SSPI_startup did not check for duplicate authentication requests like
the corresponding GSS and SASL functions, so with SSPI, you would leak
some memory instead.
Another way this could manifest itself, on version 10, is if you list
multiple hostnames in the "host" parameter. If the first server requests
Kerberos or SCRAM authentication, but it fails, the attempts to connect to
the other servers will also fail with "duplicate authentication request"
errors.
To fix, move the clearing of authentication state from closePGconn to
pgDropConnection, so that it is cleared also when re-connecting.
Patch by Michael Paquier, with some kibitzing by me.
Backpatch down to 9.3. 9.2 has the same bug, but the code around closing
the connection is somewhat different, so that this patch doesn't apply.
To fix this in 9.2, I think we would need to back-port commit 210eb9b743
first, and then apply this patch. However, given that we only bumped into
this in our own testing, we haven't heard any reports from users about
this, and that 9.2 will be end-of-lifed in a couple of months anyway, it
doesn't seem worth the risk and trouble.
The logic to free the buffer after the gss_init_sec_context() call was
always a bit wonky. Because gss_init_sec_context() sets the GSS context
variable, conn->gctx, we would in fact always attempt to free the buffer.
That only works, because previously conn->ginbuf.value was initialized to
NULL, and free(NULL) is a no-op. Commit 61bf96cab0 refactored things so
that the GSS input token buffer is allocated locally in pg_GSS_continue,
and not held in the PGconn object. After that, the now-local ginbuf.value
variable isn't initialized when it's not used, so we pass a bogus pointer
to free().
To fix, only try to free the input buffer if we allocated it. That was the
intention, certainly after the refactoring, and probably even before that.
But because there's no live bug before the refactoring, I refrained from
backpatching this.
The bug was also independently reported by Graham Dutton, as bug #14690.
Patch reviewed by Michael Paquier.
Consistently use subscription name as application name
The logical replication apply worker uses the subscription name as
application name, except for table sync. This was incorrectly set to
use the replication slot name, which might be different, in one case.
Also add a comment why the other case is different.
Andres Freund [Tue, 6 Jun 2017 23:13:00 +0000 (16:13 -0700)]
Clean up latch related code.
The larger part of this patch replaces usages of MyProc->procLatch
with MyLatch. The latter works even early during backend startup,
where MyProc->procLatch doesn't yet. While the affected code
shouldn't run in cases where it's not initialized, it might get copied
into places where it might. Using MyLatch is simpler and a bit faster
to boot, so there's little point to stick with the previous coding.
While doing so I noticed some weaknesses around newly introduced uses
of latches that could lead to missed events, and an omitted
CHECK_FOR_INTERRUPTS() call in worker_spi.
As all the actual bugs are in v10 code, there doesn't seem to be
sufficient reason to backpatch this.
Author: Andres Freund
Discussion:
https://postgr.es/m/20170606195321.sjmenrfgl2nu6j63@alap3.anarazel.de
https://postgr.es/m/20170606210405.sim3yl6vpudhmufo@alap3.anarazel.de
Backpatch: -
Robert Haas [Tue, 6 Jun 2017 16:24:44 +0000 (12:24 -0400)]
Fix some cases of "the the" split across two lines.
Kevin Grittner observed that 2186b608b3cb859fe0ec04015a5c4e4cbf69caed
introduced a new occurence of this by copying existing text, and I
found a few more cases using grep.
Robert Haas [Tue, 6 Jun 2017 15:07:20 +0000 (11:07 -0400)]
Clean up partcollation handling for OID 0.
Consistent with what we do for indexes, we shouldn't try to record
dependencies on collation OID 0 or the default collation OID (which
is pinned). Also, the fact that indcollation and partcollation can
contain zero OIDs when the data type is not collatable should be
documented.
Andres Freund [Tue, 6 Jun 2017 01:53:41 +0000 (18:53 -0700)]
Wire up query cancel interrupt for walsender backends.
This allows to cancel commands run over replication connections. While
it might have some use before v10, it has become important now that
normal SQL commands are allowed in database connected walsender
connections.
Author: Petr Jelinek Reviewed-By: Andres Freund, Michael Paquier
Discussion: https://postgr.es/m/7966f454-7cd7-2b0c-8b70-cdca9d5a8c97@2ndquadrant.com
Andres Freund [Tue, 6 Jun 2017 01:53:41 +0000 (18:53 -0700)]
Unify SIGHUP handling between normal and walsender backends.
Because walsender and normal backends share the same main loop it's
problematic to have two different flag variables, set in signal
handlers, indicating a pending configuration reload. Only certain
walsender commands reach code paths checking for the
variable (START_[LOGICAL_]REPLICATION, CREATE_REPLICATION_SLOT
... LOGICAL, notably not base backups).
This is a bug present since the introduction of walsender, but has
gotten worse in releases since then which allow walsender to do more.
A later patch, not slated for v10, will similarly unify SIGHUP
handling in other types of processes as well.
Author: Petr Jelinek, Andres Freund Reviewed-By: Michael Paquier
Discussion: https://postgr.es/m/20170423235941.qosiuoyqprq4nu7v@alap3.anarazel.de
Backpatch: 9.2-, bug is present since 9.0
Andres Freund [Tue, 6 Jun 2017 01:53:41 +0000 (18:53 -0700)]
Prevent possibility of panics during shutdown checkpoint.
When the checkpointer writes the shutdown checkpoint, it checks
afterwards whether any WAL has been written since it started and
throws a PANIC if so. At that point, only walsenders are still
active, so one might think this could not happen, but walsenders can
also generate WAL, for instance in BASE_BACKUP and logical decoding
related commands (e.g. via hint bits). So they can trigger this panic
if such a command is run while the shutdown checkpoint is being
written.
To fix this, divide the walsender shutdown into two phases. First,
checkpointer, itself triggered by postmaster, sends a
PROCSIG_WALSND_INIT_STOPPING signal to all walsenders. If the backend
is idle or runs an SQL query this causes the backend to shutdown, if
logical replication is in progress all existing WAL records are
processed followed by a shutdown. Otherwise this causes the walsender
to switch to the "stopping" state. In this state, the walsender will
reject any further replication commands. The checkpointer begins the
shutdown checkpoint once all walsenders are confirmed as
stopping. When the shutdown checkpoint finishes, the postmaster sends
us SIGUSR2. This instructs walsender to send any outstanding WAL,
including the shutdown checkpoint record, wait for it to be replicated
to the standby, and then exit.
Author: Andres Freund, based on an earlier patch by Michael Paquier Reported-By: Fujii Masao, Andres Freund Reviewed-By: Michael Paquier
Discussion: https://postgr.es/m/20170602002912.tqlwn4gymzlxpvs2@alap3.anarazel.de
Backpatch: 9.4, where logical decoding was introduced
Andres Freund [Tue, 6 Jun 2017 01:53:41 +0000 (18:53 -0700)]
Have walsenders participate in procsignal infrastructure.
The non-participation in procsignal was a problem for both changes in
master, e.g. parallelism not working for normal statements run in
walsender backends, and older branches, e.g. recovery conflicts and
catchup interrupts not working for logical decoding walsenders.
This commit thus replaces the previous WalSndXLogSendHandler with
procsignal_sigusr1_handler. In branches since db0f6cad48 that can
lead to additional SetLatch calls, but that only rarely seems to make
a difference.
Author: Andres Freund Reviewed-By: Michael Paquier
Discussion: https://postgr.es/m/20170421014030.fdzvvvbrz4nckrow@alap3.anarazel.de
Backpatch: 9.4, earlier commits don't seem to benefit sufficiently
The approach implemented in the above commit has some issues. While
those could easily be fixed incrementally, doing so would make
backpatching considerably harder, so instead first revert this patch.
There was a grammar ambiguity between SET PUBLICATION name REFRESH and
SET PUBLICATION SKIP REFRESH, because SKIP is not a reserved word. To
resolve that, fold the refresh choice into the WITH options. Refreshing
is the default now.
Andres Freund [Mon, 5 Jun 2017 22:56:58 +0000 (15:56 -0700)]
Fix record length computation in pg_waldump/xlogdump.
The current method of computing the record length (excluding the
lenght of full-page images) has been wrong since the WAL format has
been revamped in 2c03216d831160bedd72d45f712601b6f7d03f1c. Only the
main record's length was counted, but that can be significantly too
little if there's data associated with further blocks.
Fix by computing the record length as total_lenght - fpi_length.
Tom Lane [Mon, 5 Jun 2017 18:50:52 +0000 (14:50 -0400)]
Code review for shm_toc.h/.c.
Declare the toc_nentry field as uint32 not Size. Since shm_toc_lookup()
reads the field without any lock, it has to be atomically readable, and
we do not assume that for fields wider than 32 bits. Performance would
be impossibly bad for entry counts approaching 2^32 anyway, so there is
no need to try to preserve maximum width here.
This is probably an academic issue, because even if reading int64 isn't
atomic, the high order half would never change in practice. Still, it's
a coding rule violation, so let's fix it.
Adjust some other not-terribly-well-chosen data types too, and copy-edit
some comments. Make shm_toc_attach's Asserts consistent with
shm_toc_create's.
None of this looks to be a live bug, so no need for back-patch.
Tom Lane [Mon, 5 Jun 2017 16:05:42 +0000 (12:05 -0400)]
Don't be so trusting that shm_toc_lookup() will always succeed.
Given the possibility of race conditions and so on, it seems entirely
unsafe to just assume that shm_toc_lookup() always finds the key it's
looking for --- but that was exactly what all but one call site were
doing. To fix, add a "bool noError" argument, similarly to what we
have in many other functions, and throw an error on an unexpected
lookup failure. Remove now-redundant Asserts that a rather random
subset of call sites had.
I doubt this will throw any light on buildfarm member lorikeet's
recent failures, because if an unnoticed lookup failure were involved,
you'd kind of expect a null-pointer-dereference crash rather than the
observed symptom. But you never know ... and this is better coding
practice even if it never catches anything.
Tom Lane [Sun, 4 Jun 2017 20:20:03 +0000 (16:20 -0400)]
Replace over-optimistic Assert in partitioning code with a runtime test.
get_partition_parent felt that it could simply Assert that systable_getnext
found a tuple. This is unlike any other caller of that function, and it's
unsafe IMO --- in fact, the reason I noticed it was that the Assert failed.
(OK, I was working with known-inconsistent catalog contents, but I wasn't
expecting the DB to fall over quite that violently. The behavior in a
non-assert-enabled build wouldn't be very nice, either.) Fix it to do what
other callers do, namely an actual runtime-test-and-elog.
Also, standardize the wording of elog messages that are complaining about
unexpected failure of systable_getnext. 90% of them say "could not find
tuple for <object>", so make the remainder do likewise. Many of the
holdouts were using the phrasing "cache lookup failed", which is outright
misleading since no catcache search is involved.
Tom Lane [Sun, 4 Jun 2017 17:34:05 +0000 (13:34 -0400)]
#ifdef out assorted unused GEQO code.
I'd always assumed that backend/optimizer/geqo/'s remarkably poor
showing on code coverage metrics was because we weren't exercising
it much in the regression tests. But it turns out that a good chunk
of the problem is that there's a bunch of code that is physically
unreachable (because the calls to it are #ifdef'd out in geqo_main.c)
but is being built anyway. Making the called code have #if guards
similar to the calling code saves a couple of kilobytes of executable
size and should make the coverage numbers more reflective of reality.
It's arguable that we should just delete all the unused recombination
mechanisms altogether, but I didn't feel a need to go that far today.
Tom Lane [Sun, 4 Jun 2017 16:02:31 +0000 (12:02 -0400)]
Disallow CREATE INDEX if table is already in use in current session.
If we allow this, whatever outer command has the table open will not know
about the new index and may fail to update it as needed, as shown in a
report from Laurenz Albe. We already had such a prohibition in place for
ALTER TABLE, but the CREATE INDEX syntax missed the check.
Fixing it requires an API change for DefineIndex(), which conceivably
would break third-party extensions if we were to back-patch it. Given
how long this problem has existed without being noticed, fixing it in
the back branches doesn't seem worth that risk.
Tom Lane [Sat, 3 Jun 2017 21:10:25 +0000 (17:10 -0400)]
Add some missing backslash commands to psql's tab-completion knowledge.
\if and related commands were overlooked here, as were \dRp and \dRs
from the logical-replication patch, as was \?.
While here, reformat the list to put each new first command letter on
a separate line; perhaps that will limit the need to reflow the whole
list when we add more commands in future.
Tom Lane [Sat, 3 Jun 2017 18:36:25 +0000 (14:36 -0400)]
Fix <> and pattern-NOT-match estimators to handle nulls correctly.
These estimators returned 1 minus the corresponding equality/match
estimate, which is incorrect: we need to subtract off the fraction
of nulls in the column, since those are neither equal nor not equal
to the comparison value. The error only becomes obvious if the
nullfrac is large, but it could be very bad in a mostly-nulls
column, as reported in bug #14676 from Marko Tiikkaja.
To fix the <> case, refactor eqsel() and neqsel() to call a common
support routine, which can be made to account for nullfrac correctly.
The pattern-match cases were already factored that way, and it was
simply an oversight that patternsel() wasn't subtracting off nullfrac.
neqjoinsel() has a similar problem, but since we're elsewhere discussing
changing its behavior entirely, I left it alone for now.
This is a very longstanding bug, but I'm hesitant to back-patch a fix for
it. Given the lack of prior complaints, such cases must not come up often,
so it's probably not worth the risk of destabilizing plans in stable
branches.
Tom Lane [Sat, 3 Jun 2017 17:48:15 +0000 (13:48 -0400)]
Fix old corner-case logic error in final_cost_nestloop().
When costing a nestloop with stop-at-first-inner-match semantics, and a
non-indexscan inner path, final_cost_nestloop() wants to charge the full
scan cost of the inner rel at least once, with additional scans charged
at inner_rescan_run_cost which might be less. However the logic for
doing this effectively assumed that outer_matched_rows is at least 1.
If it's zero, which is not unlikely for a small outer rel, we ended up
charging inner_run_cost plus N times inner_rescan_run_cost, as much as
double the correct charge for an outer rel with only one row that
we're betting won't be matched. (Unless the inner rel is materialized,
in which case it has very small inner_rescan_run_cost and the cost
is not so far off what it should have been.)
The upshot of this was that the planner had a tendency to select plans
that failed to make effective use of the stop-at-first-inner-match
semantics, and that might have Materialize nodes in them even when the
predicted number of executions of the Materialize subplan was only 1.
This was not so obvious before commit 9c7f5229a, because the case only
arose in connection with semi/anti joins where there's not freedom to
reverse the join order. But with the addition of unique-inner joins,
it could result in some fairly bad planning choices, as reported by
Teodor Sigaev. Indeed, some of the test cases added by that commit
have plans that look dubious on closer inspection, and are changed
by this patch.
Fix the logic to ensure that we don't charge for too many inner scans.
I chose to adjust it so that the full-freight scan cost is associated
with an unmatched outer row if possible, not a matched one, since that
seems like a better model of what would happen at runtime.
This is a longstanding bug, but given the lesser impact in back branches,
and the lack of field complaints, I won't risk a back-patch.
Receive invalidation messages correctly in tablesync worker
We didn't accept any invalidation messages until the whole sync process
had finished (because it flattens all the remote transactions in the
single one). So the sync worker didn't learn about subscription
changes/drop until it has finished. This could lead to "orphaned" sync
workers.
Author: Petr Jelinek <petr.jelinek@2ndquadrant.com> Reported-by: Masahiko Sawada <sawada.mshk@gmail.com>
Andres Freund [Sat, 3 Jun 2017 02:05:57 +0000 (19:05 -0700)]
Allow parallelism in COPY (query) TO ...;
Previously this was not allowed, as copy.c didn't set the
CURSOR_OPT_PARALLEL_OK flag when planning the query. Set it.
While the lack of parallel query for COPY isn't strictly speaking a
bug, it does prevent parallelism from being used in a facility
commonly used to run long running queries. Thus backpatch to 9.6.
Author: Andres Freund
Discussion: https://postgr.es/m/20170531231958.ihanapplorptykzm@alap3.anarazel.de
Backpatch: 9.6, where parallelism was introduced.
Peter Eisentraut [Tue, 30 May 2017 18:57:01 +0000 (14:57 -0400)]
Remove replication slot name check from ReplicationSlotAcquire()
When trying to access a replication slot that is supposed to already
exist, we don't need to check the naming rules again. If the slot
does not exist, we will then get a "does not exist" error message, which
is generally more useful from the perspective of an end user.
Fix signal handling in logical replication workers
The logical replication worker processes now use the normal die()
handler for SIGTERM and CHECK_FOR_INTERRUPTS() instead of custom code.
One problem before was that the apply worker would not exit promptly
when a subscription was dropped, which could lead to deadlocks.
Author: Petr Jelinek <petr.jelinek@2ndquadrant.com> Reported-by: Masahiko Sawada <sawada.mshk@gmail.com>
projects / postgresql / log