In the planner, it was possible, given an extreme enough case containing a
large number of joins for the number of estimated rows to become infinite.
This could cause problems in initial_cost_mergejoin() where we perform
some calculations based on those row estimates.
A problem case, presented by Onder Kalaci showed an Assert failure from
an Assert checking outerstartsel <= outerendsel. In his test case this
was effectively NaN <= Inf, which is false. The NaN outerstartsel came
from multiplying the infinite outer_path_rows by 0.0.
In master, this problem was fixed by a90c950fc, however, that fix was too
invasive for the backbranches. Here we just relax the Asserts to allow
them to pass. The worst that appears to happen from this is that we show
NaN cost values and infinite row estimates in EXPLAIN. add_path() would
have had a hard time doing anything useful with such costs, but that does
not really matter as if the row estimates were even close to accurate,
such plan would not complete this side of the heat death of the universe.
Reported-by: Onder Kalaci
Backpatch: 9.5 to 13
Discussion: https://postgr.es/m/DM6PR21MB1211FF360183BCA901B27F04D80B0@DM6PR21MB1211.namprd21.prod.outlook.com
When allocating a EVP context, it would have been possible to leak some
memory allocated directly by OpenSSL, that PostgreSQL lost track of if
the initialization of the context allocated failed. The cleanup can be
done with EVP_MD_CTX_destroy().
Note that EVP APIs exist since OpenSSL 0.9.7 and we have in the tree
equivalent implementations for older versions since ce9b75d (code
removed with 9b7cd59a as of 10~). However, in 9.5 and 9.6, the existing
code makes use of EVP_MD_CTX_destroy() and EVP_MD_CTX_create() without
an equivalent implementation when building the tree with OpenSSL 0.9.6
or older, meaning that this code is in reality broken with such versions
since it got introduced in e2838c5. As we have heard no complains about
that, it does not seem worth bothering with in 9.5 and 9.6, so I have
left that out for simplicity.
Author: Michael Paquier
Discussion: https://postgr.es/m/20201015072212.GC2305@paquier.xyz
Backpatch-through: 9.5
Section 8.5.1.4, which defines these literals, made only a vague
reference to the fact that they might be evaluated too soon to be
safe in non-interactive contexts. Provide a more explicit caution
against misuse. Also, generalize the wording in the related tip in
section 9.9.4: while it clearly described this problem, it implied
(or really, stated outright) that the problem only applies to table
DEFAULT clauses.
Per gripe from Tijs van Dam. Back-patch to all supported branches.
Discussion: https://postgr.es/m/c2LuRv9BiRT3bqIo5mMQiVraEXey_25B4vUn0kDqVqilwOEu_iVF1tbtvLnyQK7yDG3PFaz_GxLLPil2SDkj1MCObNRVaac-7j1dVdFERk8=@thalex.com
This changes zic's default output format from "-b fat" to "-b slim".
We were already using "slim" in v13/HEAD, so those branches drop
the explicit -b switch in the Makefiles. Instead, add an explicit
"-b fat" in v12 and before, so that we don't change the output file
format in those branches. (This is perhaps excessively conservative,
but we decided not to do so in a12079109, and I'll stick with that.)
Other non-cosmetic changes are to drop support for zic's long-obsolete
"-y" switch, and to ensure that strftime() does not change errno
unless it fails.
As usual with tzcode changes, back-patch to all supported branches.
In theory, the second px_find_digest call in px_crypt_md5 could fail
even though the first one succeeded, since resource allocation is
required. Don't skip testing for a failure. (If one did happen,
the likely result would be a crash rather than clean recovery from
an OOM failure.)
The code's been like this all along, so back-patch to all supported
branches.
Daniel Gustafsson
Discussion: https://postgr.es/m/AA8D6FE9-4AB2-41B4-98CB-AE64BA668C03@yesql.se
Commit a97e85f2b caused "exceed the available area" warnings in PDF
builds. Fine-tune colwidth values to avoid that.
Back-patch to 9.6, like the prior patch. (This is of dubious value
before v13, since we were far from free of such warnings in older
branches. But we might as well keep the SGML looking the same in all
branches.)
Per buildfarm.
Unfortunately in LLVM 3.9 LLVMGetAttributeCountAtIndex(func, index)
crashes when called with an index that has 0 attributes. Since there's
no way to work around this in the C API, add a small C++ wrapper doing
so.
The only reason this didn't fail before 72559438f9 is that there
always are function attributes...
Author: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/20201016001254.w2nfj7gd74jmb5in@alap3.anarazel.de
Backpatch: 11-, like 72559438f9
This commit required support for inline variable definition, which is
not a requirement.
RELEASE NOTE AUTHOR: the author of commit 3c0471b5fd
(pg_upgrade/tablespaces) was Justin Pryzby, not me.
Reported-by: Andres Freund
Discussion: https://postgr.es/m/20201016001959.h24fkywfubkv2pc5@alap3.anarazel.de
Backpatch-through: 9.5
Previously, if pg_upgrade failed, and the user recreated the cluster but
did not remove the new cluster tablespace directory, a later pg_upgrade
would fail since the new tablespace directory would already exists.
This adds error reporting for this during check.
Reported-by: Justin Pryzby
Discussion: https://postgr.es/m/20200925005531.GJ23631@telsasoft.com
Backpatch-through: 9.5
Previously we only copied the function attributes. That caused problems at
least on s390x: Because we didn't copy the 'zeroext' attribute for
ExecAggTransReparent()'s *IsNull parameters, expressions invoking it didn't
ensure that the upper bytes of the registers were zeroed. In the - relatively
rare - cases where not, ExecAggTransReparent() wrongly ended up in the
newValueIsNull branch due to the register not being zero. Subsequently causing
a crash.
It's quite possible that this would cause problems on other platforms, and in
other places than just ExecAggTransReparent() on s390x.
Thanks to Christoph (and the Debian project) for providing me with access to a
s390x machine, allowing me to debug this.
Reported-By: Christoph Berg
Author: Andres Freund
Discussion: https://postgr.es/m/20201015083246.kie5726xerdt3ael@alap3.anarazel.de
Backpatch: 11-, where JIT was added
POSIX sigaction(2) can be told to block a set of signals while a
signal handler executes. Make use of that instead of manually
blocking and unblocking signals in the postmaster's signal handlers.
This should save a few cycles, but more importantly it prevents
recursive invocation of signal handlers when many signals arrive in
close succession. (Assuming that the platform's signal infrastructure
is designed to avoid consuming stack space in that case, but this is
demonstrably true at least on Linux.) The existing code has been seen
to recurse to the point of stack overflow, either in the postmaster
or in a forked-off child.
Back-patch of commit 9abb2bfc0. At the time, we'd only seen excess
postmaster stack consumption in the buildfarm; but we now have a
user report of it, and that commit has aged enough to have a fair
amount of confidence that it doesn't break anything.
This still doesn't change anything about the way that it works on
Windows. Perhaps someone else would like to fix that?
Per bug #16673 from David Geier. Back-patch to 9.6. Although
the problem exists in principle before that, we've only seen it
actually materialize in connection with heavy use of parallel
workers, so it doesn't seem necessary to do anything in 9.5;
and the relevant code is different there, too.
Discussion: https://postgr.es/m/16673-d278c604f8e34ec0@postgresql.org
Discussion: https://postgr.es/m/14878.1570820201@sss.pgh.pa.us
Previously it was documented that toast_tuple_target affected column
marked as only External or Extended. But this description is not correct
and toast_tuple_target affects also column marked as Main.
Back-patch to v11 where toast_tuple_target reloption was introduced.
Author: Shinya Okano
Reviewed-by: Tatsuhito Kasahara, Fujii Masao
Discussion: https://postgr.es/m/93f46e311a67422e89e770d236059817@oss.nttdata.com
The prohibitValueChange code paths in set_config_option(), which
are executed whenever we re-read a PGC_POSTMASTER variable from
postgresql.conf, neglected to free anything before exiting. Thus
we'd leak the proposed new value of a PGC_STRING variable, as noted
by BoChen in bug #16666. For all variable types, if the check hook
creates an "extra" chunk, we'd also leak that.
These are malloc not palloc chunks, so there is no mechanism for
recovering the leaks before process exit. Fortunately, the values
are typically not very large, meaning you'd have to go through an
awful lot of SIGHUP configuration-reload cycles to make the leakage
amount to anything. Still, for a long-lived postmaster process it
could potentially be a problem.
Oversight in commit 2594cf0e8. Back-patch to all supported branches.
Discussion: https://postgr.es/m/16666-2c41a4eec61b03e1@postgresql.org
It appears that commit cf63c641c, which intended to prevent
misoptimization of the result-building step in makeOrderedSetArgs,
didn't go far enough: buildfarm member hornet's version of xlc
is now optimizing back to the old, broken behavior in which
list_length(directargs) is fetched only after list_concat() has
changed that value. I'm not entirely convinced whether that's
an undeniable compiler bug or whether it can be justified by a
sufficiently aggressive interpretation of C sequence points.
So let's just change the code to make it harder to misinterpret.
Back-patch to all supported versions, just in case.
Discussion: https://postgr.es/m/1830491.1601944935@sss.pgh.pa.us
Commit 3eb3d3e78 was a few bricks shy of a load: while it correctly
set the table's "interesting" flag when deciding to dump the data of
an extension config table, it was not correct to clear that flag
if we concluded we shouldn't dump the data. This led to the crash
reported in bug #16655, because in fact we'll traverse dumpTableSchema
anyway for all extension tables (to see if they have user-added
seclabels or RLS policies).
The right thing to do is to force "interesting" true in makeTableDataInfo,
and otherwise leave the flag alone. (Doing it there is more future-proof
in case additional calls are added, and it also avoids setting the flag
unnecessarily if that function decides the table is non-dumpable.)
This investigation also showed that while only the --inserts code path
had an obvious failure in the case considered by 3eb3d3e78, the COPY
code path also has a problem with not having loaded table subsidiary
data. That causes fmtCopyColumnList to silently return an empty string
instead of the correct column list. That accidentally mostly works,
which perhaps is why we didn't notice this before. It would only fail
if the restore column order is different from the dump column order,
which only happens in weird inheritance cases, so it's not surprising
nobody had hit the case with an extension config table. Nonetheless,
it's a bug, and it goes a long way back, not just to v12 where the
--inserts code path started to have a problem with this.
In hopes of catching such cases a bit sooner in future, add some
Asserts that "interesting" has been set in both dumpTableData and
dumpTableSchema. Adjust the test case added by 3eb3d3e78 so that it
checks the COPY rather than INSERT form of that bug, allowing it to
detect the longer-standing symptom.
Per bug #16655 from Cameron Daniel. Back-patch to all supported
branches.
Discussion: https://postgr.es/m/16655-5c92d6b3a9438137@postgresql.org
Discussion: https://postgr.es/m/18048b44-3414-b983-8c7c-9165b177900d@2ndQuadrant.com
get_eclass_for_sort_expr() computes expr_relids and nullable_relids
early on, even though they won't be needed unless we make a new
EquivalenceClass, which we often don't. Aside from the probably-minor
inefficiency, there's a memory management problem: these bitmapsets will
be built in the caller's context, leading to dangling pointers if that
is shorter-lived than root->planner_cxt. This would be a live bug if
get_eclass_for_sort_expr() could be called with create_it = true during
GEQO join planning. So far as I can find, the core code never does
that, but it's hard to be sure that no extensions do, especially since
the comments make it clear that that's supposed to be a supported case.
Fix by not computing these values until we've switched into planner_cxt
to build the new EquivalenceClass.
generate_join_implied_equalities() uses inner_rel->relids to look up
relevant eclasses, but it ought to be using nominal_inner_relids.
This is presently harmless because a child RelOptInfo will always have
exactly the same eclass_indexes as its topmost parent; but that might
not be true forever, and anyway it makes the code confusing.
The first of these is old (introduced by me in f3b3b8d5b), so back-patch
to all supported branches. The second only dates to v13, but we might
as well back-patch it to keep the code looking similar across branches.
Discussion: https://postgr.es/m/1508010.1601832581@sss.pgh.pa.us
I noticed while trying to run the regression tests under a low
geqo_threshold that one query on information_schema.columns had
unstable (as in, variable from one run to the next) output order.
This is pretty unsurprising given the complexity of the underlying
plan. Interestingly, of this test's three nigh-identical queries on
information_schema.columns, the other two already had ORDER BY clauses
guaranteeing stable output. Let's make this one look the same.
Back-patch to v10 where this test was added. We've not heard field
reports of the test failing, but this experience shows that it can
happen when testing under even slightly unusual conditions.
Commit 151c0c5f7 neglected the possibility that a TEMP_CONFIG file
would explicitly set max_wal_senders=0; as indeed buildfarm member
thorntail does, so that it can test wal_level=minimal in other test
suites. Hence, rather than assuming that max_wal_senders=10 will
prevail if we say nothing, set it explicitly.
Set max_replication_slots=10 explicitly too, just to be safe.
Back-patch to v10, like the previous patch.
Discussion: https://postgr.es/m/723911.1601417626@sss.pgh.pa.us
The error message about columns in the primary key not including all of
the partition key was unclear; reword it.
Backpatch all the way to pg11, where it appeared.
Reported-by: Nagaraj Raj <nagaraj.sf@yahoo.com>
Discussion: https://postgr.es/m/64062533.78364.1601415362244@mail.yahoo.com
Previously, a conversion such as
to_date('-44-02-01','YYYY-MM-DD')
would result in '0045-02-01 BC', as the code attempted to interpret
the negative year as BC, but failed to apply the correction needed
for our internal handling of BC years. Fix the off-by-one problem.
Also, arrange for the combination of a negative year and an
explicit "BC" marker to cancel out and produce AD. This is how
the negative-century case works, so it seems sane to do likewise.
Continue to read "year 0000" as 1 BC. Oracle would throw an error,
but we've accepted that case for a long time so I'm hesitant to
change it in a back-patch.
Per bug #16419 from Saeed Hubaishan. Back-patch to all supported
branches.
Dar Alathar-Yemen and Tom Lane
Discussion: https://postgr.es/m/16419-d8d9db0a7553f01b@postgresql.org
Explicitly mention that primary key constraints are also included in the
limitation that the constraint columns must be a superset of the partition key
columns.
Wording suggestion from Tom Lane.
Discussion: https://postgr.es/m/64062533.78364.1601415362244@mail.yahoo.com
Backpatch-through: 11, where unique constraints on partitioned tables were added
PostgresNode.pm set "max_wal_senders = 5" for replication testing,
but this seems to be slightly too low for our current test suite.
Slower buildfarm members frequently report "number of requested standby
connections exceeds max_wal_senders" failures, due to old walsenders
not exiting instantaneously. Usually, the test does not fail overall
because of automatic walreceiver restart, but sometimes the failure
becomes visible; and in any case such retries slow down the test.
That value came in with commit 89ac7004d, but was soon obsoleted by
f6d6d2920, which raised the built-in default from zero to 10; so that
PostgresNode.pm is actually setting it to less than the conservative
built-in default. That seems pretty pointless, so let's remove the
special setting and let the default prevail, in hopes of making
the TAP tests more robust.
Likewise, the setting "max_replication_slots = 5" is obsolete and
can be removed.
While here, reverse-engineer a comment about why we're choosing
less-than-default values for some other settings.
(Note: before v12, max_wal_senders counted against max_connections
so that the latter setting also needs some fiddling with.)
Back-patch to v10 where the subscription tests were added.
It's likely that the older branches aren't pushing the boundaries
of max_wal_senders, but I'm disinclined to spend time trying to
figure out exactly when it started to be a problem.
Discussion: https://postgr.es/m/723911.1601417626@sss.pgh.pa.us
When executing a CALL or DO in a non-atomic context (i.e., not inside
a function or query), plpgsql creates a new plan each time through,
as a rather hacky solution to some resource management issues. But
it failed to free this plan until exit of the current procedure or DO
block, resulting in serious memory bloat in procedures that called
other procedures many times. Fix by remembering to free the plan,
and by being more honest about restoring the previous state (otherwise,
recursive procedure calls have a problem).
There was also a smaller leak associated with recalculation of the
"target" list of output variables. Fix that by using the statement-
lifespan context to hold non-permanent values.
Back-patch to v11 where procedures were introduced.
Pavel Stehule and Tom Lane
Discussion: https://postgr.es/m/CAFj8pRDiiU1dqym+_P4_GuTWm76knJu7z9opWayBJTC0nQGUUA@mail.gmail.com
Previously the standby server didn't archive timeline history files
streamed from the primary even when archive_mode is set to "always",
while it archives the streamed WAL files. This could cause the PITR to
fail because there was no required timeline history file in the archive.
The cause of this issue was that walreceiver didn't mark those files as
ready for archiving.
This commit makes walreceiver mark those streamed timeline history
files as ready for archiving if archive_mode=always. Then the archiver
process archives the marked timeline history files.
Back-patch to all supported versions.
Reported-by: Grigory Smolkin
Author: Grigory Smolkin, Fujii Masao
Reviewed-by: David Zhang, Anastasia Lubennikova
Discussion: https://postgr.es/m/54b059d4-2b48-13a4-6f43-95a087c92367@postgrespro.ru
This function leaked some memory while loading qual clauses for
an RLS policy. While ordinarily negligible, that could build up
in some repeated-reload cases, as reported by Konstantin Knizhnik.
We can improve matters by borrowing the coding long used in
RelationBuildRuleLock: build stringToNode's result directly in
the target context, and remember to explicitly pfree the
input string.
This patch by no means completely guarantees zero leaks within
this function, since we have no real guarantee that the catalog-
reading subroutines it calls don't leak anything. However,
practical tests suggest that this is enough to resolve the issue.
In any case, any remaining leaks are similar to those risked by
RelationBuildRuleLock and other relcache-loading subroutines.
If we need to fix them, we should adopt a more global approach
such as that used by the RECOVER_RELATION_BUILD_MEMORY hack.
While here, let's remove the need for an expensive PG_TRY block by
using MemoryContextSetParent to reparent an initially-short-lived
context for the RLS data.
Back-patch to all supported branches.
Discussion: https://postgr.es/m/21356c12-8917-8249-b35f-1c447231922b@postgrespro.ru
Parallel pg_dump failed if its -d parameter was a connection string
containing any essential information other than host, port, or username.
The same was true for pg_restore with --create.
The reason is that these scenarios failed to preserve the connection
string from the command line; the code felt free to replace that with
just the database name when reconnecting from a pg_dump parallel worker
or after creating the target database. By chance, parallel pg_restore
did not suffer this defect, as long as you didn't say --create.
In practice it seems that the error would be obvious only if the
connstring included essential, non-default SSL or GSS parameters.
This may explain why it took us so long to notice. (It also makes
it very difficult to craft a regression test case illustrating the
problem, since the test would fail in builds without those options.)
Fix by refactoring so that ConnectDatabase always receives all the
relevant options directly from the command line, rather than
reconstructed values. Inject a different database name, when necessary,
by relying on libpq's rules for handling multiple "dbname" parameters.
While here, let's get rid of the essentially duplicate _connectDB
function, as well as some obsolete nearby cruft.
Per bug #16604 from Zsolt Ero. Back-patch to all supported branches.
Discussion: https://postgr.es/m/16604-933f4b8791227b15@postgresql.org
Harmonize behavior by moving reponsibility for fsyncing directories down
into slru.c. In 10 and later, only the multixact directories were
missed (see commit 1b02be21), and in older branches all SLRUs were
missed.
Back-patch to all supported releases.
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/CA%2BhUKGLtsTUOScnNoSMZ-2ZLv%2BwGh01J6kAo_DM8mTRq1sKdSQ%40mail.gmail.com
tsearch_readline() saves the string pointer it returns to the caller
for possible use in the associated error context callback. However,
the caller will usually pfree that string sometime before it next
calls tsearch_readline(), so that there is a window where an ereport
will try to print an already-freed string.
The built-in users of tsearch_readline() happen to all do that pfree
at the bottoms of their loops, so that the window is effectively
empty for them. However, this is not documented as a requirement,
and contrib/dict_xsyn doesn't do it like that, so it seems likely
that third-party dictionaries might have live bugs here.
The practical consequences of this seem pretty limited in any case,
since production builds wouldn't clobber the freed string immediately,
besides which you'd not expect syntax errors in dictionary files
being used in production. Still, it's clearly a bug waiting to bite
somebody.
Fix by pstrdup'ing the string to be saved for the error callback,
and then pfree'ing it next time through. It's been like this for
a long time, so back-patch to all supported branches.
Discussion: https://postgr.es/m/48A4FA71-524E-41B9-953A-FD04EF36E2E7@yesql.se
These two SQL functions are aliases for the same C function, so this
change has no semantic effect. However, because we dropped the
numeric_fac alias in HEAD (commit 76f412ab3), operator definitions
based on that one don't port forward, causing problems for cross-version
upgrade tests based on the regression database.
Patch all active back branches to dodge the problem.
Discussion: https://postgr.es/m/449144.1600439950@sss.pgh.pa.us
For parallel btree scan to work for array of scan keys, it should reach
BTPARALLEL_DONE state once for every distinct combination of array keys.
This is required to ensure that the parallel workers don't try to seize
blocks at the same time for different scan keys. We missed to update this
state when we discovered that the scan keys can't be satisfied.
Author: James Hunter
Reviewed-by: Amit Kapila
Tested-by: Justin Pryzby
Backpatch-through: 10, where it was introduced
Discussion: https://postgr.es/m/4248CABC-25E3-4809-B4D0-128E1BAABC3C@amazon.com
The code recorded cache invalidation events by zeroing the "localreloid"
field of affected cache entries. However, it's possible for an inval
event to occur even while we have the entry open and locked. So an
ill-timed inval could result in "cache lookup failed for relation 0"
errors, if the worker's code tried to use the cleared field. We can
fix that by creating a separate bool field to record whether the entry
needs to be revalidated. (In the back branches, cram the bool into
what had been padding space, to avoid an ABI break in the somewhat
unlikely event that any extension is looking at this struct.)
Also, rearrange the logic in logicalrep_rel_open so that it
does the right thing in cases where table_open would fail.
We should retry the lookup by name in that case, but we didn't.
The real-world impact of this is probably small. In the first place,
the error conditions are very low probability, and in the second place,
the worker would just exit and get restarted. We only noticed because
in a CLOBBER_CACHE_ALWAYS build, the failure can occur repeatedly,
preventing the worker from making progress. Nonetheless, it's clearly
a bug, and it impedes a useful type of testing; so back-patch to v10
where this code was introduced.
Discussion: https://postgr.es/m/1032727.1600096803@sss.pgh.pa.us
transformCreateStmt() adjusts the transformed statement's RangeVar
to specify the target schema explicitly, for the express reason
of making sure that auxiliary statements derived by parse
transformation operate on the right table. But the refactoring
I did in commit 502898192 got this wrong and passed the untransformed
RangeVar to expandTableLikeClause(). This could lead to assertion
failures or weird misbehavior if the wrong table was accessed.
Per report from Alexander Lakhin. Like the previous patch, back-patch
to all supported branches.
Discussion: https://postgr.es/m/05051f9d-b32b-cb35-6735-0e9f2ab86b5f@gmail.com
Bring the signal handling for startup-packet collection into line
with the policy established in commits bedadc732 and 8e19a8264,
namely don't risk running atexit callbacks when handling SIGQUIT.
Ideally, we'd not do so for SIGTERM or timeout interrupts either,
but that change seems a bit too risky for the back branches.
For now, just improve the comments in this area to describe the risk.
Also relocate where BackendInitialize re-disables these interrupts,
to minimize the code span where they're active. This doesn't buy
a whole lot of safety, but it can't hurt.
In passing, rename startup_die() to remove confusion about whether
it is for the startup process.
Like the previous commits, back-patch to all supported branches.
Discussion: https://postgr.es/m/1850884.1599601164@sss.pgh.pa.us
We have had multiple reports that point to the
'@colReorder=latn-digit' collation customization being buggy. We have
reported this to ICU and are waiting for a fix. In the meantime,
remove references to this from the documentation and replace it by
another reordering example. Apparently, many users have been picking
up this example specifically from the documentation.
Author: Jehan-Guillaume de Rorthais <jgdr@dalibo.com>
Discussion: https://www.postgresql.org/message-id/flat/153201618542.1404.3611626898935613264%40wrigleys.postgresql.org
Commit 8e19a8264 changed the SIGQUIT handlers of almost all server
processes not to run atexit callbacks. The archiver process was
skipped, perhaps because it's not connected to shared memory; but
it's just as true here that running atexit callbacks in a signal
handler is unsafe. So let's make it work like the rest.
In HEAD and v13, we can use the common SignalHandlerForCrashExit
handler. Before that, just tweak pgarch_exit to use _exit(2)
explicitly.
Like the previous commit, back-patch to all supported branches.
Kyotaro Horiguchi, back-patching by me
Discussion: https://postgr.es/m/1850884.1599601164@sss.pgh.pa.us