XLogRecordAssemble() may be called multiple times before inserting a
record in XLogInsertRecord(), and the amount of FPIs generated inside
a record whose insertion is attempted multiple times may vary.
The logic added in f9a09aa295 touched directly pgWalUsage in
XLogRecordAssemble(), meaning that it could be possible for pgWalUsage
to be incremented multiple times for a single record. This commit
changes the code to use the same logic as the number of FPIs added to a
record, where XLogRecordAssemble() returns this information and feeds it
to XLogInsertRecord(), updating pgWalUsage only when a record is
inserted.
Reported-by: Shinya Kato <shinya11.kato@gmail.com>
Discussion: https://postgr.es/m/CAOzEurSiSr+rusd0GzVy8Bt30QwLTK=ugVMnF6=5WhsSrukvvw@mail.gmail.com
This type is just char * underneath, it provides no real value, no
type safety, and just makes the code one level more mysterious. It is
more idiomatic to refer to blobs of memory by a combination of void *
and size_t, so change it to that.
Also, since this type hides the pointerness, we can't apply qualifiers
to what is pointed to, which requires some unconstify nonsense. This
change allows fixing that.
Extension code that uses the Item type can change its code to use
void * to be backward compatible.
Reviewed-by: Nathan Bossart <nathandbossart@gmail.com>
Reviewed-by: Peter Geoghegan <pg@bowt.ie>
Discussion: https://www.postgresql.org/message-id/flat/c75cccf5-5709-407b-a36a-2ae6570be766@eisentraut.org
Instead of emitting a separate XLOG_HEAP2_VISIBLE WAL record for each
page that becomes all-visible in vacuum's third phase, specify the VM
changes in the already emitted XLOG_HEAP2_PRUNE_VACUUM_CLEANUP record.
Visibility checks are now performed before marking dead items unused.
This is safe because the heap page is held under exclusive lock for the
entire operation.
This reduces the number of WAL records generated by VACUUM phase III by
up to 50%.
Author: Melanie Plageman <melanieplageman@gmail.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Robert Haas <robertmhaas@gmail.com>
Reviewed-by: Kirill Reshke <reshkekirill@gmail.com>
Discussion: https://postgr.es/m/flat/CAAKRu_ZMw6Npd_qm2KM%2BFwQ3cMOMx1Dh3VMhp8-V7SOLxdK9-g%40mail.gmail.com
Stop including utils/relcache.h in access/genam.h, and stop including
htup_details.h in nodes/tidbitmap.h. Both these files (genam.h and
tidbitmap.h) are widely used in other header files, so it's in our best
interest that they remain as lean as reasonable.
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/202509291356.o5t6ny2hoa3q@alvherre.pgsql
This commit corrects several issues in function comments:
* The parameter "rel" was incorrectly referred to as "relation" in the comments
for table_tuple_delete(), table_tuple_update(), and table_tuple_lock().
* In table_tuple_delete(), "changingPart" was listed as an output parameter
in the comments but is actually input.
* In table_tuple_update(), "slot" was listed as an input parameter
in the comments but is actually output.
* The comment for "update_indexes" in table_tuple_update() was mis-indented.
* The comments for heap_lock_tuple() incorrectly referenced a non-existent
"tid" parameter.
Author: Chao Li <lic@highgo.com>
Reviewed-by: Fujii Masao <masao.fujii@gmail.com>
Discussion: https://postgr.es/m/CAEoWx2nB6Ay8g=KEn7L3qbYX_4+sLk9XOMkV0XZqHR4cTY8ZvQ@mail.gmail.com
The comment mistakenly had "the others" for "the other", but this
commit also reorders the comment so it matches the macros below. Now we
describe the levels in increasing strictness. In addition, it seems
easier to follow if we introduce one level at a time, rather than
describing two, followed by "the other" (and then jumping back to one of
the first two). Finally, reword the sentence about the purpose of the
macros, which was slightly off-point.
Author: Paul Jungwirth <pj@illuminatedcomputing.com>
Reviewed-by: Rustam ALLAKOV <rustamallakov@gmail.com>
Reviewed-by: Dean Rasheed <dean.a.rasheed@gmail.com>
Reviewed-by: Robert Haas <robertmhaas@gmail.com>
Discussion: https://postgr.es/m/CA+renyUp=xja80rBaB6NpY3RRdi750y046x28bo_xg29zKY72Q@mail.gmail.com
hash_xlog.h included descriptions for the blocks used in WAL records
that were was not completely consistent with how the records are
generated, with one block missing for SQUEEZE_PAGE, and inconsistent
descriptions used for block 0 in VACUUM_ONE_PAGE and MOVE_PAGE_CONTENTS.
This information was incorrect since c11453ce0a, cross-checking the
logic for the record generation.
Author: Kirill Reshke <reshkekirill@gmail.com>
Reviewed-by: Andrey Borodin <x4mmm@yandex-team.ru>
Discussion: https://postgr.es/m/CALdSSPj1j=a1d1hVA3oabRFz0hSU3KKrYtZPijw4UPUM7LY9zw@mail.gmail.com
Backpatch-through: 13
This commit fixes three issues:
1) When a disabled subscription is created with retain_dead_tuples set to true,
the launcher is not woken up immediately, which may lead to delays in creating
the conflict detection slot.
Creating the conflict detection slot is essential even when the subscription is
not enabled. This ensures that dead tuples are retained, which is necessary for
accurately identifying the type of conflict during replication.
2) Conflict-related data was unnecessarily retained when the subscription does
not have a table.
3) Conflict-relevant data could be prematurely removed before applying
prepared transactions on the publisher that are in the commit critical section.
This issue occurred because the backend executing COMMIT PREPARED was not
accounted for during the computation of oldestXid in the commit phase on
the publisher. As a result, the subscriber could advance the conflict
slot's xmin without waiting for such COMMIT PREPARED transactions to
complete.
We fixed this issue by identifying prepared transactions that are in the
commit critical section during computation of oldestXid in commit phase.
Author: Zhijie Hou <houzj.fnst@fujitsu.com>
Reviewed-by: shveta malik <shveta.malik@gmail.com>
Reviewed-by: Dilip Kumar <dilipbalaut@gmail.com>
Reviewed-by: Nisha Moond <nisha.moond412@gmail.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Discussion: https://postgr.es/m/OS9PR01MB16913DACB64E5721872AA5C02943BA@OS9PR01MB16913.jpnprd01.prod.outlook.com
Discussion: https://postgr.es/m/OS9PR01MB16913F67856B0DA2A909788129400A@OS9PR01MB16913.jpnprd01.prod.outlook.com
When executing a MERGE UPDATE action, if there is more than one
concurrent update of the target row, the lock-and-retry code would
sometimes incorrectly identify the latest version of the target tuple,
leading to incorrect results.
This was caused by using the ctid field from the TM_FailureData
returned by table_tuple_lock() in a case where the result was TM_Ok,
which is unsafe because the TM_FailureData struct is not guaranteed to
be fully populated in that case. Instead, it should use the tupleid
passed to (and updated by) table_tuple_lock().
To reduce the chances of similar errors in the future, improve the
commentary for table_tuple_lock() and TM_FailureData to make it
clearer that table_tuple_lock() updates the tid passed to it, and most
fields of TM_FailureData should not be relied on in non-failure cases.
An exception to this is the "traversed" field, which is set in both
success and failure cases.
Reported-by: Dmitry <dsy.075@yandex.ru>
Author: Yugo Nagata <nagata@sraoss.co.jp>
Reviewed-by: Dean Rasheed <dean.a.rasheed@gmail.com>
Reviewed-by: Chao Li <li.evan.chao@gmail.com>
Discussion: https://postgr.es/m/1570d30e-2b95-4239-b9c3-f7bf2f2f8556@yandex.ru
Backpatch-through: 15
SLRU bank locks are referred as "bank locks" or "SLRU bank locks" in the
code comments. The comments updated in this commit use the latter term.
Oversight in 53c2a97a92, that has replaced the single ControlLock by
the bank control locks.
Author: Julien Rouhaud <julien.rouhaud@free.fr>
Discussion: https://postgr.es/m/aLUT2UO8RjJOzZNq@jrouhaud
Backpatch-through: 17
This provides a single entry point to access some information about the
state of MultiXacts, able to return some data about multixacts offsets
and counts. Originally this function was only able to return some
information about the number of multixacts and multixact members,
extended here to provide some data about the oldest multixact ID in use
and the oldest offset, if known.
This change has been proposed in a patch that aims at providing more
monitoring capabilities for multixacts, and it is useful on its own.
GetMultiXactInfo() is added to multixact.h, becoming available for
out-of-core code.
Extracted from a larger patch by the same author.
Author: Naga Appani <nagnrik@gmail.com>
Reviewed-by: Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/CA+QeY+AAsYK6WvBW4qYzHz4bahHycDAY_q5ECmHkEV_eB9ckzg@mail.gmail.com
Doing that seems rather random and unnecessary. This commit removes
those and fixes fallout, which is pretty minimal. We do need to add a
forward declaration of struct TM_IndexDeleteOp (whose full definition
appears in tableam.h) so that _bt_delitems_delete_check()'s declaration
can use it.
Author: Álvaro Herrera <alvherre@kurilemu.de>
Reviewed-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Discussion: https://postgr.es/m/202508051109.lzk3lcuzsaxo@alvherre.pgsql
Remove conditionally-compiled code for smaller Datum widths,
and simplify comments that describe cases no longer of interest.
I also fixed up a few more places that were not using
DatumGetIntXX where they should, and made some cosmetic
adjustments such as using sizeof(int64) not sizeof(Datum)
in places where that fit better with the surrounding code.
One thing I remembered while preparing this part is that SP-GiST
stores pass-by-value prefix keys as Datums, so that the on-disk
representation depends on sizeof(Datum). That's even more
unfortunate than the existing commentary makes it out to be,
because now there is a hazard that the change of sizeof(Datum)
will break SP-GiST indexes on 32-bit machines. It appears that
there are no existing SP-GiST opclasses that are actually
affected; and if there are some that I didn't find, the number
of installations that are using them on 32-bit machines is
doubtless tiny. So I'm proceeding on the assumption that we
can get away with this, but it's something to worry about.
(gininsert.c looks like it has a similar problem, but it's okay
because the "tuples" it's constructing are just transient data
within the tuplesort step. That's pretty poorly documented
though, so I added some comments.)
Author: Tom Lane <tgl@sss.pgh.pa.us>
Reviewed-by: Peter Eisentraut <peter@eisentraut.org>
Discussion: https://postgr.es/m/1749799.1752797397@sss.pgh.pa.us
Fix a couple more places where an explicit Datum conversion
is needed (not clear how we missed these in ff89e182d and
previous commits).
Replace the minority usage "(Datum) NULL" with "(Datum) 0".
The former depends on the assumption that Datum is the same
width as Pointer, the latter doesn't. Anyway consistency
is a good thing.
This is, I believe, the last of the notational mop-up needed
before we can consider changing Datum to uint64 everywhere.
It's also important cleanup for more aggressive ideas such
as making Datum a struct.
Discussion: https://postgr.es/m/1749799.1752797397@sss.pgh.pa.us
Discussion: https://postgr.es/m/8246d7ff-f4b7-4363-913e-827dadfeb145@eisentraut.org
The new name more accurately reflects the effects of this flag on a
requested checkpoint. Checkpoint-related log messages (i.e., those
controlled by the log_checkpoints configuration parameter) will now
say "fast" instead of "immediate", too. Likewise, references to
"immediate" checkpoints in the documentation have been updated to
say "fast". This is preparatory work for a follow-up commit that
will add a MODE option to the CHECKPOINT command.
Author: Christoph Berg <myon@debian.org>
Discussion: https://postgr.es/m/aDnaKTEf-0dLiEfz%40msg.df7cb.de
The new name more accurately relects the effects of this flag on a
requested checkpoint. Checkpoint-related log messages (i.e., those
controlled by the log_checkpoints configuration parameter) will now
say "flush-unlogged" instead of "flush-all", too. This is
preparatory work for a follow-up commit that will add a
FLUSH_UNLOGGED option to the CHECKPOINT command.
Author: Christoph Berg <myon@debian.org>
Discussion: https://postgr.es/m/aDnaKTEf-0dLiEfz%40msg.df7cb.de
This commit standardizes the output format for LSNs to ensure consistent
representation across various tools and messages. Previously, LSNs were
inconsistently printed as `%X/%X` in some contexts, while others used
zero-padding. This often led to confusion when comparing.
To address this, the LSN format is now uniformly set to `%X/%08X`,
ensuring the lower 32-bit part is always zero-padded to eight
hexadecimal digits.
Author: Japin Li <japinli@hotmail.com>
Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com>
Reviewed-by: Álvaro Herrera <alvherre@kurilemu.de>
Discussion: https://postgr.es/m/ME0P300MB0445CA53CA0E4B8C1879AF84B641A@ME0P300MB0445.AUSP300.PROD.OUTLOOK.COM
This refactoring is a follow-up of the work done in 5a1dfde833, that
has switched 2PC file names to use FullTransactionIds when written on
disk. This will help with the integration of a follow-up solution
related to the handling of two-phase files during recovery, to address
older defects while reading these from disk after a crash.
This change is useful in itself as it reduces the need to build the
file names from epoch numbers and TransactionIds, because we can use
directly FullTransactionIds from which the 2PC file names are guessed.
So this avoids a lot of back-and-forth between the FullTransactionIds
retrieved from the file names and how these are passed around in the
internal 2PC logic.
Note that the core of the change is the use of a FullTransactionId
instead of a TransactionId in GlobalTransactionData, that tracks 2PC
file information in shared memory. The change in TwoPhaseCallback makes
this commit unfit for stable branches.
Noah has contributed a good chunk of this patch. I have spent some time
on it as well while working on the issues with two-phase state files and
recovery.
Author: Noah Misch <noah@leadboat.com>
Co-Authored-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/Z5sd5O9JO7NYNK-C@paquier.xyz
Discussion: https://postgr.es/m/20250116205254.65.nmisch@google.com
Commit 08aa89b326 removed the COMMIT_TS_SETTS WAL record,
leaving xl_commit_ts_set and SizeOfCommitTsSet unused. However,
it missed removing these definitions. This commit cleans up
the leftover code.
Since this is a cleanup rather than a bug fix, it is applied only to
the master branch.
Author: Andy Fan <zhihuifan1213@163.com>
Reviewed-by: Fujii Masao <masao.fujii@gmail.com>
Discussion: https://postgr.es/m/87ecuzmkqf.fsf@163.com
This commit refactors the vacuum routines that rely on VacuumParams,
adding const markers where necessary to force a new policy in the code.
This structure should not use a pointer as it may be used across
multiple relations, and its contents should never be updated.
vacuum_rel() stands as an exception as it touches the "index_cleanup"
and "truncate" options.
VacuumParams has been introduced in 0d83138974, and 661643deda has
fixed a bug impacting VACUUM operating on multiple relations. The
changes done in tableam.h break ABI compatibility, so this commit can
only happen on HEAD.
Author: Shihao Zhong <zhong950419@gmail.com>
Co-authored-by: Michael Paquier <michael@paquier.xyz>
Reviewed-by: Nathan Bossart <nathandbossart@gmail.com>
Reviewed-by: Junwang Zhao <zhjwpku@gmail.com>
Discussion: https://postgr.es/m/CAGRkXqTo+aK=GTy5pSc-9cy8H2F2TJvcrZ-zXEiNJj93np1UUw@mail.gmail.com
Our maintenance of typedefs.list has been a little haphazard
(and apparently we can't alphabetize worth a darn). Replace
the file with the authoritative list from our buildfarm, and
run pgindent using that.
I also updated the additions/exclusions lists in pgindent where
necessary to keep pgindent from messing things up significantly.
Notably, now that regex_t and some related names are macros not real
typedefs, we have to whitelist them explicitly. The exclusions list
has also drifted noticeably, presumably due to changes of system
headers on the buildfarm animals that contribute to the list.
Unlike in prior years, I've not manually added typedef names that
are missing from the buildfarm's list because they are not used to
declare any variables or fields. So there are a few places where
the typedef declaration itself is formatted worse than before,
e.g. typedef enum IoMethod. I could preserve the names that were
manually added to the list previously, but I'd really prefer to find
a less manual way of dealing with these cases. A quick grep finds
about 75 such symbols, most of which have never gotten any special
treatment.
Per discussion among pgsql-release, doing this now seems appropriate
even though we're still a week or two away from making the v18 branch.
Delay calling BufferGetLSNAtomic() until we finish reading a page that
actually contains items that btgettuple will return to the executor.
This reduces the number of calls during plain index scans (we'll only
call BufferGetLSNAtomic() when _bt_readpage returns true), and totally
eliminates calls during index-only scans, bitmap index scans, and plain
index scans of an unlogged relation.
Currently, when checksums (or wal_log_hints) are enabled, acquiring a
page's LSN in BufferGetLSNAtomic() involves locking the buffer header
(which involves the use of spinlocks). Testing has shown that enabling
page-level checksums causes large regressions with certain workloads,
especially on larger multi-socket systems.
The regression isn't tied to any Postgres 18 commit. However, Postgres
18 commit 04bec894 made initdb use checksums by default, so it seems
prudent to address the problem now.
Author: Peter Geoghegan <pg@bowt.ie>
Reviewed-By: Tomas Vondra <tomas@vondra.me>
Discussion: https://postgr.es/m/941f0190-e3c6-4622-9ac7-c04e936e5fdb@vondra.me
Discussion: https://postgr.es/m/CAH2-Wzk-Dg5XWs_jDuiHt4_7ryrSY+n=vxmHY51EVqPDFsKXmg@mail.gmail.com
Commit 7406ab623f added a gist support function that we internally
refer to by the symbol GIST_STRATNUM_PROC. This translated from
"well-known" strategy numbers to opfamily-specific strategy numbers.
However, we later (commit 630f9a43ce) changed this to fit into
index-AM-level compare type mapping, so this function actually now
maps from compare type to opfamily-specific strategy numbers. So this
name is no longer fitting.
Moreover, the index AM level also supports the opposite, a function to
map from strategy number to compare type. This is currently not
supported in gist, but one might wonder what this function is supposed
to be called when it is added.
This patch changes the naming of the gist-level functionality to be
more in line with the index-AM-level functionality. This makes sense
because these are essentially the same thing on different levels.
This also changes the names of the externally visible functions that
are provided for use as such a support function.
Reviewed-by: Paul A Jungwirth <pj@illuminatedcomputing.com>
Discussion: https://www.postgresql.org/message-id/37ebb1d9-9036-485f-a215-e55435689917%40eisentraut.org
Previously, postgres_fdw always 1) opened a remote transaction in READ
WRITE mode even when the local transaction was READ ONLY, causing a READ
ONLY transaction using it that references a foreign table mapped to a
remote view executing a volatile function to write in the remote side,
and 2) opened the remote transaction in NOT DEFERRABLE mode even when
the local transaction was DEFERRABLE, causing a SERIALIZABLE READ ONLY
DEFERRABLE transaction using it to abort due to a serialization failure
in the remote side.
To avoid these, modify postgres_fdw to open a remote transaction in the
same access/deferrable modes as the local transaction. This commit also
modifies it to open a remote subtransaction in the same access mode as
the local subtransaction.
Although these issues exist since the introduction of postgres_fdw,
there have been no reports from the field. So it seems fine to just fix
them in master only.
Author: Etsuro Fujita <etsuro.fujita@gmail.com>
Reviewed-by: Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/CAPmGK16n_hcUUWuOdmeUS%2Bw4Q6dZvTEDHb%3DOP%3D5JBzo-M3QmpQ%40mail.gmail.com
This allows them to be added without scanning the table, and validating
them afterwards without holding access exclusive lock on the table after
any violating rows have been deleted or fixed.
Doing ALTER TABLE ... SET NOT NULL for a column that has an invalid
not-null constraint validates that constraint. ALTER TABLE .. VALIDATE
CONSTRAINT is also supported. There are various checks on whether an
invalid constraint is allowed in a child table when the parent table has
a valid constraint; this should match what we do for enforced/not
enforced constraints.
pg_attribute.attnotnull is now only an indicator for whether a not-null
constraint exists for the column; whether it's valid or invalid must be
queried in pg_constraint. Applications can continue to query
pg_attribute.attnotnull as before, but now it's possible that NULL rows
are present in the column even when that's set to true.
For backend internal purposes, we cache the nullability status in
CompactAttribute->attnullability that each tuple descriptor carries
(replacing CompactAttribute.attnotnull, which was a mirror of
Form_pg_attribute.attnotnull). During the initial tuple descriptor
creation, based on the pg_attribute scan, we set this to UNRESTRICTED if
pg_attribute.attnotnull is false, or to UNKNOWN if it's true; then we
update the latter to VALID or INVALID depending on the pg_constraint
scan. This flag is also copied when tupledescs are copied.
Comparing tuple descs for equality must also compare the
CompactAttribute.attnullability flag and return false in case of a
mismatch.
pg_dump deals with these constraints by storing the OIDs of invalid
not-null constraints in a separate array, and running a query to obtain
their properties. The regular table creation SQL omits them entirely.
They are then dealt with in the same way as "separate" CHECK
constraints, and dumped after the data has been loaded. Because no
additional pg_dump infrastructure was required, we don't bump its
version number.
I decided not to bump catversion either, because the old catalog state
works perfectly in the new world. (Trying to run with new catalog state
and the old server version would likely run into issues, however.)
System catalogs do not support invalid not-null constraints (because
commit 14e87ffa5c didn't allow them to have pg_constraint rows
anyway.)
Author: Rushabh Lathia <rushabh.lathia@gmail.com>
Author: Jian He <jian.universality@gmail.com>
Reviewed-by: Álvaro Herrera <alvherre@alvh.no-ip.org>
Tested-by: Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>
Discussion: https://postgr.es/m/CAGPqQf0KitkNack4F5CFkFi-9Dqvp29Ro=EpcWt=4_hs-Rt+bQ@mail.gmail.com
Don't allow nbtree scans with skip arrays to end any primitive scan on
its first leaf page without giving some consideration to how many times
the scan's arrays advanced while changing at least one skip array
(though continue not caring about the number of array advancements that
only affected SAOP arrays, even during skip scans with SAOP arrays).
Now when a scan performs more than 3 such array advancements in the
course of reading a single leaf page, it is taken as a signal that the
next page is unlikely to be skippable. We'll therefore continue the
ongoing primitive index scan, at least until we can perform a recheck
against the next page's finaltup.
Testing has shown that this new heuristic occasionally makes all the
difference with skip scans that were expected to rely on the "passed
first page" heuristic added by commit 9a2e2a28. Without it, there is a
remaining risk that certain kinds of skip scans will never quite manage
to clear the initial hurdle of performing a primitive scan that lasts
beyond its first leaf page (or that such a skip scan will only clear
that initial hurdle when it has already wasted noticeably-many cycles
due to inefficient primitive scan scheduling).
Follow-up to commits 92fe23d9 and 9a2e2a28.
Author: Peter Geoghegan <pg@bowt.ie>
Reviewed-By: Matthias van de Meent <boekewurm+postgres@gmail.com>
Discussion: https://postgr.es/m/CAH2-Wz=RVdG3zWytFWBsyW7fWH7zveFvTHed5JKEsuTT0RCO_A@mail.gmail.com
Postgres 17 commit e0b1ee17 added two complementary optimizations to
nbtree: the "prechecked" and "firstmatch" optimizations. _bt_readpage
was made to avoid needlessly evaluating keys that are guaranteed to be
satisfied by applying page-level context. "prechecked" did this for
keys required in the current scan direction, while "firstmatch" did it
for keys required in the opposite-to-scan direction only.
The "prechecked" design had a number of notable issues. It didn't
account for the fact that an = array scan key's sk_argument field might
need to advance at the point of the page precheck (it didn't check the
precheck tuple against the key's array, only the key's sk_argument,
which needlessly made it ineffective in cases involving stepping to a
page having advanced the scan's arrays using a truncated high key).
"prechecked" was also completely ineffective when only one scan key
wasn't guaranteed to be satisfied by every tuple (it didn't recognize
that it was still safe to avoid evaluating other, earlier keys).
The "firstmatch" optimization had similar limitations. It could only be
applied after _bt_readpage found its first matching tuple, regardless of
why any earlier tuples failed to satisfy the scan's index quals. This
allowed unsatisfied non-required scan keys to impede the optimization.
Replace both optimizations with a new optimization, without any of these
limitations: the "startikey" optimization. Affected _bt_readpage calls
generate a page-level key offset ("startikey"), that their _bt_checkkeys
calls can then start at. This is an offset to the first key that isn't
known to be satisfied by every tuple on the page.
Although this is independently useful work, its main goal is to avoid
performance regressions with index scans that use skip arrays, but still
never manage to skip over irrelevant leaf pages. We must avoid wasting
CPU cycles on overly granular skip array maintenance in these cases.
The new "startikey" optimization helps with this by selectively
disabling array maintenance for the duration of a _bt_readpage call.
This has no lasting consequences for the scan's array keys (they'll
still reliably track the scan's progress through the index's key space
whenever the scan is "between pages").
Skip scan adds skip arrays during preprocessing using simple, static
rules, and decides how best to navigate/apply the scan's skip arrays
dynamically, at runtime. The "startikey" optimization enables this
approach. As a result of all this, the planner doesn't need to generate
distinct, competing index paths (one path for skip scan, another for an
equivalent traditional full index scan). The overall effect is to make
scan runtime close to optimal, even when the planner works off an
incorrect cardinality estimate. Scans will also perform well given a
skipped column with data skew: individual groups of pages with many
distinct values (in respect of a skipped column) can be read about as
efficiently as before -- without the scan being forced to give up on
skipping over other groups of pages that are provably irrelevant.
Many scans that cannot possibly skip will still benefit from the use of
skip arrays, since they'll allow the "startikey" optimization to be as
effective as possible (by allowing preprocessing to mark all the scan's
keys as required). A scan that uses a skip array on "a" for a qual
"WHERE a BETWEEN 0 AND 1_000_000 AND b = 42" is often much faster now,
even when every tuple read by the scan has its own distinct "a" value.
However, there are still some remaining regressions, affecting certain
trickier cases.
Scans whose index quals have several range skip arrays, each on some
high cardinality column, can still be slower than they were before the
introduction of skip scan -- even with the new "startikey" optimization.
There are also known regressions affecting very selective index scans
that use a skip array. The underlying issue with such selective scans
is that they never get as far as reading a second leaf page, and so will
never get a chance to consider applying the "startikey" optimization.
In principle, all regressions could be avoided by teaching preprocessing
to not add skip arrays whenever they aren't expected to help, but it
seems best to err on the side of robust performance.
Follow-up to commit 92fe23d9, which added nbtree skip scan.
Author: Peter Geoghegan <pg@bowt.ie>
Reviewed-By: Heikki Linnakangas <heikki.linnakangas@iki.fi>
Reviewed-By: Masahiro Ikeda <ikedamsh@oss.nttdata.com>
Reviewed-By: Matthias van de Meent <boekewurm+postgres@gmail.com>
Discussion: https://postgr.es/m/CAH2-Wz=Y93jf5WjoOsN=xvqpMjRy-bxCE037bVFi-EasrpeUJA@mail.gmail.com
Discussion: https://postgr.es/m/CAH2-WznWDK45JfNPNvDxh6RQy-TaCwULaM5u5ALMXbjLBMcugQ@mail.gmail.com
