Test files should now ignore has_wal_read_bug() so long as
wait_for_catchup() is their only known way of reaching the bug. That's
at least five files today, a number expected to grow over time. This
commit removes skip logic from three. By doing so, systems having the
bug regain the ability to catch other kinds of defects via those three
tests. The other two, 002_databases.pl and 031_recovery_conflict.pl,
have been unprotected. Back-patch to v15, where done_testing() first
became our standard.
Discussion: https://postgr.es/m/20221030031639.GA3082137@rfd.leadboat.com
When executing a utility statement, we must fetch everything
we need out of the PlannedStmt data structure before calling
standard_ProcessUtility. In certain cases (possibly only ROLLBACK
in extended query protocol), that data structure will get freed
during command execution. The situation is probably often harmless
in production builds, but in debug builds we intentionally overwrite
the freed memory with garbage, leading to picking up garbage values
of statement location and length, typically causing an assertion
failure later in pg_stat_statements. In non-debug builds, if
something did go wrong it would likely lead to storing garbage
for the query string.
Report and fix by zhaoqigui (with cosmetic adjustments by me).
It's an old problem, so back-patch to all supported versions.
Discussion: https://postgr.es/m/17663-a344fd0675f92128@postgresql.org
Discussion: https://postgr.es/m/1667307420050.56657@hundsun.com
This problem has been introduced by commit 272248a0c1 where we started
assigning the subtransactions to the top-level transaction when we mark
both the top-level transaction and its subtransactions as containing
catalog changes. After we assign subtransactions to the top-level
transaction, we were not allowed to execute any invalidations associated
with it when we decide to skip the transaction.
The reason to assign the subtransactions to the top-level transaction was
to avoid the assertion failure in AssertTXNLsnOrder() as they have the
same LSN when we sometimes start accumulating transaction changes for
partial transactions after the restart. Now that with commit 64ff0fe4e8,
we skip this assertion check until we reach the LSN at which we start
decoding the contents of the transaction, so, there is no reason for such
an assignment anymore.
The assignment change was introduced in 15 and prior versions but this bug
doesn't exist in branches prior to 14 since we don't add invalidation
messages to subtransactions. We decided to backpatch through 11 for
consistency but not for 10 since its final release is near.
Reported-by: Kuroda Hayato
Author: Masahiko Sawada
Reviewed-by: Amit Kapila
Backpatch-through: 11
Discussion: https://postgr.es/m/TYAPR01MB58660803BCAA7849C8584AA4F57E9%40TYAPR01MB5866.jpnprd01.prod.outlook.com
Discussion: https://postgr.es/m/a89b46b6-0239-2fd5-71a9-b19b1f7a7145%40enterprisedb.com
When the logical decoding restarts from NEW_CID, since there is no
association between the top transaction and its subtransaction, both are
created as top transactions and have the same LSN. This caused the
assertion failure in AssertTXNLsnOrder().
This patch skips the assertion check until we reach the LSN at which we
start decoding the contents of the transaction, specifically
start_decoding_at LSN in SnapBuild. This is okay because we don't
guarantee to make the association between top transaction and
subtransaction until we try to decode the actual contents of transaction.
The ordering of the records prior to the start_decoding_at LSN should have
been checked before the restart.
The other assertion failure is due to the reason that we forgot to track
that we have considered top-level transaction id in the list of catalog
changing transactions that were committed when one of its subtransactions
is marked as containing catalog change.
Reported-by: Tomas Vondra, Osumi Takamichi
Author: Masahiko Sawada, Kuroda Hayato
Reviewed-by: Amit Kapila, Dilip Kumar, Kuroda Hayato, Kyotaro Horiguchi, Masahiko Sawada
Backpatch-through: 10
Discussion: https://postgr.es/m/a89b46b6-0239-2fd5-71a9-b19b1f7a7145%40enterprisedb.com
Discussion: https://postgr.es/m/TYCPR01MB83733C6CEAE47D0280814D5AED7A9%40TYCPR01MB8373.jpnprd01.prod.outlook.com
Per discussion, the existing routine name able to initialize a SRF
function with materialize mode is unpopular, so rename it. Equally, the
flags of this function are renamed, as of:
- SRF_SINGLE_USE_EXPECTED -> MAT_SRF_USE_EXPECTED_DESC
- SRF_SINGLE_BLESS -> MAT_SRF_BLESS
The previous function and flags introduced in 9e98583 are kept around
for compatibility purposes, so as any extension code already compiled
with v15 continues to work as-is. The declarations introduced here for
compatibility will be removed from HEAD in a follow-up commit.
The new names have been suggested by Andres Freund and Melanie
Plageman.
Discussion: https://postgr.es/m/20221013194820.ciktb2sbbpw7cljm@awork3.anarazel.de
Backpatch-through: 15
The file name used for its temporary destination, before renaming it to
the real deal, has been using a microseconds in a timestamp aimed to be
originally in milli-seconds. This is harmless as this is aimed at being
a safeguard against name collisions (note MyProcPid in the name), but
let's be correct with the maths.
While on it, add a note in the module's makefile to document why
installcheck is not supported.
Author: Nathan Bossart
Reviewed-by: Bharath Rupireddy
Discussion: https://postgr.es/m/20221014044106.GA1673343@nathanxps13
Backpatch-through: 15
While directly targetting a foreign table with MERGE was already
expressly forbidden, we failed to catch the case of a partitioned table
that has a foreign table as a partition; and the result if you try is an
incomprehensible error. Fix that by adding a specific check.
Backpatch to 15.
Reported-by: Tatsuhiro Nakamori <bt22nakamorit@oss.nttdata.com>
Discussion: https://postgr.es/m/bt22nakamorit@oss.nttdata.com
This reverts commit db0d67db24 and
several follow-on fixes. The idea of making a cost-based choice
of the order of the sorting columns is not fundamentally unsound,
but it requires cost information and data statistics that we don't
really have. For example, relying on procost to distinguish the
relative costs of different sort comparators is pretty pointless
so long as most such comparator functions are labeled with cost 1.0.
Moreover, estimating the number of comparisons done by Quicksort
requires more than just an estimate of the number of distinct values
in the input: you also need some idea of the sizes of the larger
groups, if you want an estimate that's good to better than a factor of
three or so. That's data that's often unknown or not very reliable.
Worse, to arrive at estimates of the number of calls made to the
lower-order-column comparison functions, the code needs to make
estimates of the numbers of distinct values of multiple columns,
which are necessarily even less trustworthy than per-column stats.
Even if all the inputs are perfectly reliable, the cost algorithm
as-implemented cannot offer useful information about how to order
sorting columns beyond the point at which the average group size
is estimated to drop to 1.
Close inspection of the code added by db0d67db2 shows that there
are also multiple small bugs. These could have been fixed, but
there's not much point if we don't trust the estimates to be
accurate in-principle.
Finally, the changes in cost_sort's behavior made for very large
changes (often a factor of 2 or so) in the cost estimates for all
sorting operations, not only those for multi-column GROUP BY.
That naturally changes plan choices in many situations, and there's
precious little evidence to show that the changes are for the better.
Given the above doubts about whether the new estimates are really
trustworthy, it's hard to summon much confidence that these changes
are better on the average.
Since we're hard up against the release deadline for v15, let's
revert these changes for now. We can always try again later.
Note: in v15, I left T_PathKeyInfo in place in nodes.h even though
it's unreferenced. Removing it would be an ABI break, and it seems
a bit late in the release cycle for that.
Discussion: https://postgr.es/m/TYAPR01MB586665EB5FB2C3807E893941F5579@TYAPR01MB5866.jpnprd01.prod.outlook.com
We weren't jumbling the merge action list, so wildly different commands
would be considered to use the same query ID. Add that, mention it in
the docs, and some test lines.
Backpatch to 15.
Author: Tatsu <bt22nakamorit@oss.nttdata.com>
Reviewed-by: Julien Rouhaud <rjuju123@gmail.com>
Discussion: https://postgr.es/m/d87e391694db75a038abc3b2597828e8@oss.nttdata.com
The tlist of the EvalPlanQual outer plan for a ForeignScan node is
adjusted to produce a tuple whose descriptor matches the scan tuple slot
for the ForeignScan node. But in the case where the outer plan contains
an extra Sort node, if the new tlist contained columns required only for
evaluating PlaceHolderVars or columns required only for evaluating local
conditions, this would cause setrefs.c to fail with the error.
The cause of this is that when creating the outer plan by injecting the
Sort node into an alternative local join plan that could emit such extra
columns as well, we fail to arrange for the outer plan to propagate them
up through the Sort node, causing setrefs.c to fail to match up them in
the new tlist to what is available from the outer plan. Repair.
Per report from Alexander Pyhalov.
Richard Guo and Etsuro Fujita, reviewed by Alexander Pyhalov and Tom Lane.
Backpatch to all supported versions.
Discussion: http://postgr.es/m/cfb17bf6dfdf876467bd5ef533852d18%40postgrespro.ru
pg_walinspect uses datatype double (double precision floating point
number) for WAL stats percentile calculations and expose them via
float4 (single precision floating point number), which an unnecessary
loss of precision and confusing. Even though, it's harmless that way,
let's use float8 (double precision floating-point number) to be in
sync with what pg_walinspect does internally and what it exposes to
the users. This seems to be the pattern used elsewhere in the code.
Reported-by: Peter Eisentraut
Author: Bharath Rupireddy
Reviewed-by: Peter Eisentraut
Discussion: https://www.postgresql.org/message-id/36ee692b-232f-0484-ce94-dc39d82021ad%40enterprisedb.com
When using the BSD UUID functions, contrib/uuid-ossp expects
uuid_create() to produce a version-1 UUID. FreeBSD still does so,
but in recent NetBSD releases that function produces a version-4
(random) UUID instead. That's not acceptable for our purposes:
if the user wanted v4 she would have asked for v4, not v1.
Hence, check the version digit and complain if it's not '1'.
Also drop the documentation's claim that the NetBSD implementation
is usable. It might be, depending on which OS version you're using,
but we're not going to get into that kind of detail.
(Maybe someday we should ditch all these external libraries
and just write our own UUID code, but today is not that day.)
Nazir Bilal Yavuz, with cosmetic adjustments and docs by me.
Backpatch to all supported versions.
Discussion: https://postgr.es/m/3848059.1661038772@sss.pgh.pa.us
Discussion: https://postgr.es/m/17358-89806e7420797025@postgresql.org
Usage of ReadNextXLogRecord()'s first_record parameter for error
reporting isn't always correct. For instance, in GetWALRecordsInfo()
and GetWalStats(), we're reading multiple records, and first_record
is always passed as the LSN of the first record which is then used
for error reporting for later WAL record read failures. This isn't
correct.
The correct parameter to use for error reports in case of WAL
reading failures is xlogreader->EndRecPtr. This change fixes it.
While on it, removed an unnecessary Assert in pg_walinspect code.
Reported-by: Robert Haas
Author: Bharath Rupireddy
Reviewed-by: Robert Haas
Discussion: https://www.postgresql.org/message-id/CA%2BTgmoZAOGzPUifrcZRjFZ2vbtcw3mp-mN6UgEoEcQg6bY3OVg%40mail.gmail.com
Backpatch-through: 15
Previously, we relied on HEAP2_NEW_CID records and XACT_INVALIDATION
records to know if the transaction has modified the catalog, and that
information is not serialized to snapshot. Therefore, after the restart,
if the logical decoding decodes only the commit record of the transaction
that has actually modified a catalog, we will miss adding its XID to the
snapshot. Thus, we will end up looking at catalogs with the wrong
snapshot.
To fix this problem, this changes the snapshot builder so that it
remembers the last-running-xacts list of the decoded RUNNING_XACTS record
after restoring the previously serialized snapshot. Then, we mark the
transaction as containing catalog changes if it's in the list of initial
running transactions and its commit record has XACT_XINFO_HAS_INVALS. To
avoid ABI breakage, we store the array of the initial running transactions
in the static variables InitialRunningXacts and NInitialRunningXacts,
instead of storing those in SnapBuild or ReorderBuffer.
This approach has a false positive; we could end up adding the transaction
that didn't change catalog to the snapshot since we cannot distinguish
whether the transaction has catalog changes only by checking the COMMIT
record. It doesn't have the information on which (sub) transaction has
catalog changes, and XACT_XINFO_HAS_INVALS doesn't necessarily indicate
that the transaction has catalog change. But that won't be a problem since
we use snapshot built during decoding only to read system catalogs.
On the master branch, we took a more future-proof approach by writing
catalog modifying transactions to the serialized snapshot which avoids the
above false positive. But we cannot backpatch it because of a change in
the SnapBuild.
Reported-by: Mike Oh
Author: Masahiko Sawada
Reviewed-by: Amit Kapila, Shi yu, Takamichi Osumi, Kyotaro Horiguchi, Bertrand Drouvot, Ahsan Hadi
Backpatch-through: 10
Discussion: https://postgr.es/m/81D0D8B0-E7C4-4999-B616-1E5004DBDCD2%40amazon.com
When inserting a view referencing a foreign table that has WITH CHECK
OPTION constraints, in single-insert mode postgres_fdw retrieves the
data that was actually inserted on the remote side so that the WITH
CHECK OPTION constraints are enforced with the data locally, but in
batch-insert mode it cannot currently retrieve the data (except for the
row first inserted through the view), resulting in enforcing the WITH
CHECK OPTION constraints with the data passed from the core (except for
the first-inserted row), which led to incorrect results when inserting
into a view referencing a foreign table in which a remote BEFORE ROW
INSERT trigger changes the rows inserted through the view so that they
violate the view's WITH CHECK OPTION constraint. Also, the query
inserting into the view caused an assertion failure in assert-enabled
builds.
Fix these by disabling batch insertion when inserting into such a view.
Back-patch to v14 where batch insertion was added.
Discussion: https://postgr.es/m/CAPmGK17LpbTZs4m4a_6THP54UBeK9fHvX8aVVA%2BC6yEZDZwQcg%40mail.gmail.com
We've heard a couple of reports of people having trouble with
multi-gigabyte-sized query-texts files. It occurred to me that on
32-bit platforms, there could be an issue with integer overflow
of calculations associated with the total query text size.
Address that with several changes:
1. Limit pg_stat_statements.max to INT_MAX / 2 not INT_MAX.
The hashtable code will bound it to that anyway unless "long"
is 64 bits. We still need overflow guards on its use, but
this helps.
2. Add a check to prevent extending the query-texts file to
more than MaxAllocHugeSize. If it got that big, qtext_load_file
would certainly fail, so there's not much point in allowing it.
Without this, we'd need to consider whether extent, query_offset,
and related variables shouldn't be off_t not size_t.
3. Adjust the comparisons in need_gc_qtexts() to be done in 64-bit
arithmetic on all platforms. It appears possible that under duress
those multiplications could overflow 32 bits, yielding a false
conclusion that we need to garbage-collect the texts file, which
could lead to repeatedly garbage-collecting after every hash table
insertion.
Per report from Bruno da Silva. I'm not convinced that these
issues fully explain his problem; there may be some other bug that's
contributing to the query-texts file becoming so large in the first
place. But it did get that big, so #2 is a reasonable defense,
and #3 could explain the reported performance difficulties.
(See also commit 8bbe4cbd9, which addressed some related bugs.
The second Discussion: link is the thread that led up to that.)
This issue is old, and is primarily a problem for old platforms,
so back-patch.
Discussion: https://postgr.es/m/CAB+Nuk93fL1Q9eLOCotvLP07g7RAv4vbdrkm0cVQohDVMpAb9A@mail.gmail.com
Discussion: https://postgr.es/m/5601D354.5000703@BlueTreble.com
In commit 7c34555f8, I overlooked the need to configure SSPI
on Windows to allow login as the non-superuser role.
Fix that by adding auth_extra/--create-role incantation
(which, oddly enough, doesn't actually create the role).
Per buildfarm.
While here, upgrade the mechanism for temporarily setting
$ENV{PGUSER}, as per recommendation from ilmari.
Discussion: https://postgr.es/m/87edy7j1zz.fsf@wibble.ilmari.org
We weren't exercising the session_preload_libraries option in any
meaningful way. auto_explain is a good testbed for doing so, since
it's one of the primary use-cases for session_preload_libraries.
Hence, adjust its TAP test to load the library via
session_preload_libraries not shared_preload_libraries. While at it,
feed test-specific settings to the backend via PGOPTIONS rather than
tediously rewriting postgresql.conf.
Also, since auto_explain has some PGC_SUSET parameters, we can use it
to provide a test case for the permissions-checking bug just fixed
by commit b35617de3.
Back-patch to v15 so that we have coverage for the permissions issue
in that branch too. To do that, I back-patched the refactoring
recently done by commit 550bc0a6c.
Dagfinn Ilmari Mannsåker and Tom Lane
Discussion: https://postgr.es/m/CABwTF4VEpwTHhRQ+q5MiC5ucngN-whN-PdcKeufX7eLSoAfbZA@mail.gmail.com
When postgres_fdw begins an asynchronous data fetch, it submits FETCH query
by using PQsendQuery(). If PQsendQuery() fails and returns 0, postgres_fdw
should report an error. But, previously, postgres_fdw reported an error
only when the return value is less than 0, though PQsendQuery() never return
the values other than 0 and 1. Therefore postgres_fdw could not handle
the failure to send FETCH query in an asynchronous data fetch.
This commit fixes postgres_fdw so that it reports an error
when PQsendQuery() returns 0.
Back-patch to v14 where asynchronous execution was supported in postgres_fdw.
Author: Fujii Masao
Reviewed-by: Japin Li, Tom Lane
Discussion: https://postgr.es/m/b187a7cf-d4e3-5a32-4d01-8383677797f3@oss.nttdata.com
Detail and hint messages should be full sentences and should end with a
period, but some of the messages newly-introduced in v15 did not follow
that.
Author: Justin Pryzby
Reviewed-by: Álvaro Herrera
Discussion: https://postgr.es/m/20220719120948.GF12702@telsasoft.com
Backpatch-through: 15
As $gzip is embedded in postgresql.conf \ needs to be escaped, otherwise guc.c
will take it as a string escape. Similarly, if "$gzip" contains spaces, the
prior incantation will fail. Both of these are common on windows.
Reviewed-by: Peter Eisentraut <peter.eisentraut@enterprisedb.com>
Discussion: https://postgr.es/m/ce1b6eb3-5736-6f38-9775-b7020128b8d8@enterprisedb.com
Backpatch: 15-, where the test was added in 027fa0fd72
The motivation for this is to ensure successful transmission of the
values of constants of regconfig and other reg* types. The remote
will be reading them with search_path = 'pg_catalog', so schema
qualification is necessary when referencing objects in other schemas.
Per bug #17483 from Emmanuel Quincerot. Back-patch to all supported
versions. (There's some other stuff to do here, but it's less
back-patchable.)
Discussion: https://postgr.es/m/1423433.1652722406@sss.pgh.pa.us
Commit a117cebd63 used the original userid
for ACL checks located directly in DefineIndex(), but it still adopted
the table owner userid for more ACL checks than intended. That broke
dump/reload of indexes that refer to an operator class, collation, or
exclusion operator in a schema other than "public" or "pg_catalog".
Back-patch to v10 (all supported versions), like the earlier commit.
Nathan Bossart and Noah Misch
Discussion: https://postgr.es/m/f8a4105f076544c180a87ef0c4822352@stmuk.bayern.de
The macro is being applied to a TOAST pointer, not a varlena header.
Therefore the use of VARATT_IS_COMPRESSED() is wrong. We can check
VARATT_EXTERNAL_IS_COMPRESSED(), but then we don't need the length
check that follows.
Report and fix by Kyotaro Horiguchi.
Discussion: http://postgr.es/m/20220517.162719.1671558681467343711.horikyota.ntt@gmail.com
Since a117cebd6, some older gcc versions issue "variable may be used
uninitialized in this function" complaints for brin_summarize_range.
Silence that using the same coding pattern as in bt_index_check_internal;
arguably, a117cebd6 had too narrow a view of which compilers might give
trouble.
Nathan Bossart and Tom Lane. Back-patch as the previous commit was.
Discussion: https://postgr.es/m/20220601163537.GA2331988@nathanxps13
Currently, preloaded libraries are expected to request additional
shared memory and LWLocks in _PG_init(). However, it is not unusal
for such requests to depend on MaxBackends, which won't be
initialized at that time. Such requests could also depend on GUCs
that other modules might change. This introduces a new hook where
modules can safely use MaxBackends and GUCs to request additional
shared memory and LWLocks.
Furthermore, this change restricts requests for shared memory and
LWLocks to this hook. Previously, libraries could make requests
until the size of the main shared memory segment was calculated.
Unlike before, we no longer silently ignore requests received at
invalid times. Instead, we FATAL if someone tries to request
additional shared memory or LWLocks outside of the hook.
Nathan Bossart and Julien Rouhaud
Discussion: https://postgr.es/m/20220412210112.GA2065815%40nathanxps13
Discussion: https://postgr.es/m/Yn2jE/lmDhKtkUdr@paquier.xyz
Expand the comment about the parallel_commit option to mention that the
default is false.
Also, since the comment about alteration of the keep_connections option,
which was located above the expanded comment, holds true for the
parallel_commit option, rewrite it to reflect this, and move it to after
the expanded comment.
Follow-up for commit 04e706d42.
Discussion: https://postgr.es/m/CAPmGK16Kg2Bf90sqzcZ4YM5cN_G-4h7wFUS01qQpqNB%2B2BG5_w%40mail.gmail.com
The code for unloading a library has been commented-out for over 12
years, ever since commit 602a9ef5a7, and we're
no closer to supporting it now than we were back then.
Nathan Bossart, reviewed by Michael Paquier and by me.
Discussion: http://postgr.es/m/Ynsc9bRL1caUSBSE@paquier.xyz
When a feature enumerates relations and runs functions associated with
all found relations, the feature's user shall not need to trust every
user having permission to create objects. BRIN-specific functionality
in autovacuum neglected to account for this, as did pg_amcheck and
CLUSTER. An attacker having permission to create non-temp objects in at
least one schema could execute arbitrary SQL functions under the
identity of the bootstrap superuser. CREATE INDEX (not a
relation-enumerating operation) and REINDEX protected themselves too
late. This change extends to the non-enumerating amcheck interface.
Back-patch to v10 (all supported versions).
Sergey Shinderuk, reviewed (in earlier versions) by Alexander Lakhin.
Reported by Alexander Lakhin.
Security: CVE-2022-1552
The PXE_CIPHER_INIT error is used to report initialization errors, so
appending a questionmark to the error isn't entirely accurate (using a
space before the questionmark doubly so).
Discussion: https://postgr.es/m/C89D932C-501E-4473-9750-638CFCD9095E@yesql.se
Report OpenSSL errors during initialization as PXE_CIPHER_INIT since
that's just what they were, and not generic unknown errors. This also
removes the last users of the generic error, and thus it can be removed.
Discussion: http://postgr.es/m/C89D932C-501E-4473-9750-638CFCD9095E@yesql.se
Instability in the test for pg_walinspect revealed that
pg_get_wal_records_info_till_end_of_wal(x) would try to decode all the
records with a start LSN earlier than the flush LSN, even though that
might include a partial record at the end of the range. In that case,
read_local_xlog_page_no_wait() would return NULL when it tried to read
past the flush LSN, which would be interpreted as an error by the
caller. That caused a test failure only on a BF animal that had been
restarted recently, but could be expected to happen in the wild quite
easily depending on the alignment of various parameters.
Fix by using private data in read_local_xlog_page_no_wait() to signal
end-of-wal to the caller, so that it can be properly distinguished
from a real error.
Discussion: https://postgr.es/m/Ymd/e5eeZMNAkrXo%40paquier.xyz
Discussion: https://postgr.es/m/111657.1650910309@sss.pgh.pa.us
Authors: Thomas Munro, Bharath Rupireddy.
mark_async_capable_plan(), which is called from create_append_plan() to
determine whether subplans are async-capable, failed to take into
account that the given subplan created from a given subpath might
include a gating Result node if the subpath is a SubqueryScanPath or
ForeignPath, causing a segmentation fault there when the subplan created
from a SubqueryScanPath includes the Result node, or causing
ExecAsyncRequest() to throw an error about an unrecognized node type
when the subplan created from a ForeignPath includes the Result node,
because in the latter case the Result node was unintentionally
considered as async-capable, but we don't currently support executing
Result nodes asynchronously. Fix by modifying mark_async_capable_plan()
to disable asynchronous execution in such cases. Also, adjust code in
the ProjectionPath case in mark_async_capable_plan(), for consistency
with other cases, and adjust/improve comments there.
is_async_capable_path() added in commit 27e1f1456, which was rewritten
to mark_async_capable_plan() in a later commit, has the same issue,
causing the error at execution mentioned above, so back-patch to v14
where the aforesaid commit went in.
Per report from Justin Pryzby.
Etsuro Fujita, reviewed by Zhihong Yu and Justin Pryzby.
Discussion: https://postgr.es/m/20220408124338.GK24419%40telsasoft.com
This reverts commits 2c902bb and ccfbd92. Per buildfarm members
kestrel, rorqual and calliphoridae, the assertions checking that a TLI
history file should not exist when created by a WAL receiver have been
failing, and switching to durable_rename() over durable_rename_excl()
would cause the newest TLI history file to overwrite the existing one.
We need to think harder about such cases, so revert the new logic for
now.
Note that all the failures have been reported in the test
025_stuck_on_old_timeline.
Discussion: https://postgr.es/m/511362.1651116498@sss.pgh.pa.us
durable_rename_excl() attempts to avoid overwriting any existing files
by using link() and unlink(), falling back to rename() on some platforms
(e.g., Windows where link() followed by unlink() is not concurrent-safe,
see 909b449). Most callers of durable_rename_excl() use it just in case
there is an existing file, but it happens that for all of them we never
expect a target file to exist (WAL segment recycling, creation of
timeline history file and basic_archive).
basic_archive used durable_rename_excl() to avoid overwriting an archive
concurrently created by another server. Now, there is a stat() call to
avoid overwriting an existing archive a couple of lines above, so note
that this change opens a small TOCTOU window in this module between the
stat() call and durable_rename().
Furthermore, as mentioned in the top comment of durable_rename_excl(),
this routine can result in multiple hard links to the same file and data
corruption, with two or more links to the same file in pg_wal/ if a
crash happens before the unlink() call during WAL recycling.
Specifically, this would produce links to the same file for the current
WAL file and the next one because the half-recycled WAL file was
re-recycled during crash recovery of a follow-up cluster restart.
This change replaces all calls to durable_rename_excl() with
durable_rename(). This removes the protection against accidentally
overwriting an existing file, but some platforms are already living
without it, and all those code paths never expect an existing file (a
couple of assertions are added to check after that, in case).
This is a bug fix, but knowing the unlikeliness of the problem involving
one of more crashes at an exceptionally bad moment, no backpatch is
done. This could be revisited in the future.
Author: Nathan Bossart
Reviewed-by: Robert Haas, Kyotaro Horiguchi, Michael Paquier
Discussion: https://postgr.es/m/20220407182954.GA1231544@nathanxps13
Previously, we allowed this, but such triggers might query the table to
insert into and act differently if the tuples that have already been
processed and prepared for insertion are not there, so disable it in
such cases.
Back-patch to v14 where batch insert was added.
Discussion: https://postgr.es/m/CAPmGK16_uPqsmgK0-LpLSUk54_BoK13bPrhxhfjSoSTVz414hA%40mail.gmail.com
We have some streaming tests that rely on the size of changes which can
fail if there are additional changes like invalidation messages by
background activity like auto analyze. Avoid such failures by increasing
autovacuum_naptime to a reasonably high value (1d).
Author: Dilip Kumar
Backpatch-through: 14
Discussion: https://postgr.es/m/1958043.1650129119@sss.pgh.pa.us
Getting from get_raw_page() an all-zero page is considered as a valid
case by the buffer manager and it can happen for example when finding a
corrupted page with zero_damaged_pages enabled (using zero_damaged_pages
to look at corrupted pages happens), or after a crash when a relation
file is extended before any WAL for its new data is generated (before a
vacuum or autovacuum job comes in to do some cleanup).
However, all the functions of pageinspect, as of the index AMs (except
hash that has its own idea of new pages), heap, the FSM or the page
header have never worked with all-zero pages, causing various crashes
when going through the page internals.
This commit changes all the pageinspect functions to be compliant with
all-zero pages, where the choice is made to return NULL or no rows for
SRFs when finding a new page. get_raw_page() still works the same way,
returning a batch of zeros in the bytea of the page retrieved. A hard
error could be used but NULL, while more invasive, is useful when
scanning relation files in full to get a batch of results for a single
relation in one query. Tests are added for all the code paths
impacted.
Reported-by: Daria Lepikhova
Author: Michael Paquier
Discussion: https://postgr.es/m/561e187b-3549-c8d5-03f5-525c14e65bd0@postgrespro.ru
Backpatch-through: 10
Noah Misch
Tom Lane
Amit Kapila
Michael Paquier
Alvaro Herrera
Etsuro Fujita
Peter Eisentraut
Jeff Davis
Robert Haas
Fujii Masao
Andres Freund
Daniel Gustafsson