When reading a WAL record fails to find continuation record(s) of the
proper length, report what it expects, for clarity.
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/20200903212152.GA15319@alvherre.pgsql
When multiple relations are reindexed, a scan of pg_class is done first
to build the list of relations to work on. However the REINDEX logic
has never checked if a relation listed still exists when beginning the
work on it, causing for example sudden cache lookup failures.
This commit adds safeguards against dropped relations for REINDEX,
similarly to VACUUM or CLUSTER where we try to open the relation,
ignoring it if it is missing. A new option is added to the REINDEX
routines to control if a missed relation is OK to ignore or not.
An isolation test, based on REINDEX SCHEMA, is added for the concurrent
and non-concurrent cases.
Author: Michael Paquier
Reviewed-by: Anastasia Lubennikova
Discussion: https://postgr.es/m/20200813043805.GE11663@paquier.xyz
Since other sessions aren't allowed to look into a temporary table
of our own session, we do not need to worry about the global xmin
horizon when setting the vacuum XID cutoff. Indeed, if we're not
inside a transaction block, we may set oldestXmin to be the next
XID, because there cannot be any in-doubt tuples in a temp table,
nor any tuples that are dead but still visible to some snapshot of
our transaction. (VACUUM, of course, is never inside a transaction
block; but we need to test that because CLUSTER shares the same code.)
This approach allows us to always clean out a temp table completely
during VACUUM, independently of concurrent activity. Aside from
being useful in its own right, that simplifies building reproducible
test cases.
Discussion: https://postgr.es/m/3490536.1598629609@sss.pgh.pa.us
Historically, we've considered the state with relpages and reltuples
both zero as indicating that we do not know the table's tuple density.
This is problematic because it's impossible to distinguish "never yet
vacuumed" from "vacuumed and seen to be empty". In particular, a user
cannot use VACUUM or ANALYZE to override the planner's normal heuristic
that an empty table should not be believed to be empty because it is
probably about to get populated. That heuristic is a good safety
measure, so I don't care to abandon it, but there should be a way to
override it if the table is indeed intended to stay empty.
Hence, represent the initial state of ignorance by setting reltuples
to -1 (relpages is still set to zero), and apply the minimum-ten-pages
heuristic only when reltuples is still -1. If the table is empty,
VACUUM or ANALYZE (but not CREATE INDEX) will override that to
reltuples = relpages = 0, and then we'll plan on that basis.
This requires a bunch of fiddly little changes, but we can get rid of
some ugly kluges that were formerly needed to maintain the old definition.
One notable point is that FDWs' GetForeignRelSize methods will see
baserel->tuples = -1 when no ANALYZE has been done on the foreign table.
That seems like a net improvement, since those methods were formerly
also in the dark about what baserel->tuples = 0 really meant. Still,
it is an API change.
I bumped catversion because code predating this change would get confused
by seeing reltuples = -1.
Discussion: https://postgr.es/m/F02298E0-6EF4-49A1-BCB6-C484794D9ACC@thebuild.com
collectMatchBitmap() needs to re-find the index tuple it was previously
looking at, after transiently dropping lock on the index page it's on.
The tuple should still exist and be at its prior position or somewhere
to the right of that, since ginvacuum never removes tuples but
concurrent insertions could add one. However, there was a thinko in
that logic, to the effect of expecting any inserted tuples to have the
same index "attnum" as what we'd been scanning. Since there's no
physical separation of tuples with different attnums, it's not terribly
hard to devise scenarios where this fails, leading to transient "lost
saved point in index" errors. (While I've duplicated this with manual
testing, it seems impossible to make a reproducible test case with our
available testing technology.)
Fix by just continuing the scan when the attnum doesn't match.
While here, improve the error message used if we do fail, so that it
matches the wording used in btree for a similar case.
collectMatchBitmap()'s posting-tree code path was previously not
exercised at all by our regression tests. While I can't make
a regression test that exhibits the bug, I can at least improve
the code coverage here, so do that. The test case I made for this
is an extension of one added by 4b754d6c1, so it only works in
HEAD and v13; didn't seem worth trying hard to back-patch it.
Per bug #16595 from Jesse Kinkead. This has been broken since
multicolumn capability was added to GIN (commit 27cb66fdf),
so back-patch to all supported branches.
Discussion: https://postgr.es/m/16595-633118be8eef9ce2@postgresql.org
The additional information added will be an offset number for heap
operations. This information will help us in finding the exact tuple due
to which the error has occurred.
Author: Mahendra Singh Thalor and Amit Kapila
Reviewed-by: Sawada Masahiko, Justin Pryzby and Amit Kapila
Discussion: https://postgr.es/m/CAKYtNApK488TDF4bMbw+1QH8HJf9cxdNDXquhU50TK5iv_FtCQ@mail.gmail.com
We were displaying the wrong phase information for 'info' message in the
index clean up phase because we were switching to the previous phase a bit
early. We were also not displaying context information for heap phase
unless the block number is valid which is fine for error cases but for
messages at 'info' or lower error level it appears to be inconsistent with
index phase information.
Reported-by: Sawada Masahiko
Author: Sawada Masahiko
Reviewed-by: Amit Kapila
Backpatch-through: 13, where it was introduced
Discussion: https://postgr.es/m/CA+fd4k4HcbhPnCs7paRTw1K-AHin8y4xKomB9Ru0ATw0UeTy2w@mail.gmail.com
If a commit or abort record includes "dropped relfilenodes", then replaying
the record will remove data files. That is surely a "special rel update",
but the records were not marked as such. Fix that, teach pg_rewind to
expect and ignore them, and add a test case to cover it.
It's always been like this, but no backporting for fear of breaking
existing applications. If an application parsed the WAL but was not
handling commit/abort records, it would stop working. That might be a good
thing if it really needed to handle the dropped rels, but it will be caught
when the application is updated to work with PostgreSQL v14 anyway.
Discussion: https://www.postgresql.org/message-id/07b33e2c-46a6-86a1-5f9e-a7da73fddb95%40iki.fi
Reviewed-by: Amit Kapila, Michael Paquier
Reuse cautionary language from src/test/ssl/README in
src/test/kerberos/README. SLRUs have had access to six-character
segments names since commit 73c986adde,
and recovery stopped calling HeapTupleHeaderAdvanceLatestRemovedXid() in
commit 558a9165e0. The other corrections
are more self-evident.
The SimpleLruTruncate() header comment states the new coding rule. To
achieve this, add locktype "frozenid" and two LWLocks. This closes a
rare opportunity for data loss, which manifested as "apparent
wraparound" or "could not access status of transaction" errors. Data
loss is more likely in pg_multixact, due to released branches' thin
margin between multiStopLimit and multiWrapLimit. If a user's physical
replication primary logged ": apparent wraparound" messages, the user
should rebuild standbys of that primary regardless of symptoms. At less
risk is a cluster having emitted "not accepting commands" errors or
"must be vacuumed" warnings at some point. One can test a cluster for
this data loss by running VACUUM FREEZE in every database. Back-patch
to 9.5 (all supported versions).
Discussion: https://postgr.es/m/20190218073103.GA1434723@rfd.leadboat.com
Similar to the previous changes this increases the chance that data
frequently needed by GetSnapshotData() stays in l2 cache. In many
workloads subtransactions are very rare, and this makes the check for
that considerably cheaper.
As this removes the last member of PGXACT, there is no need to keep it
around anymore.
On a larger 2 socket machine this and the two preceding commits result
in a ~1.07x performance increase in read-only pgbench. For read-heavy
mixed r/w workloads without row level contention, I see about 1.1x.
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
Similar to the previous commit this increases the chance that data
frequently needed by GetSnapshotData() stays in l2 cache. As we now
take care to not unnecessarily write to ProcGlobal->vacuumFlags, there
should be very few modifications to the ProcGlobal->vacuumFlags array.
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
The new array contains the xids for all connected backends / in-use
PGPROC entries in a dense manner (in contrast to the PGPROC/PGXACT
arrays which can have unused entries interspersed).
This improves performance because GetSnapshotData() always needs to
scan the xids of all live procarray entries and now there's no need to
go through the procArray->pgprocnos indirection anymore.
As the set of running top-level xids changes rarely, compared to the
number of snapshots taken, this substantially increases the likelihood
of most data required for a snapshot being in l2 cache. In
read-mostly workloads scanning the xids[] array will sufficient to
build a snapshot, as most backends will not have an xid assigned.
To keep the xid array dense ProcArrayRemove() needs to move entries
behind the to-be-removed proc's one further up in the array. Obviously
moving array entries cannot happen while a backend sets it
xid. I.e. locking needs to prevent that array entries are moved while
a backend modifies its xid.
To avoid locking ProcArrayLock in GetNewTransactionId() - a fairly hot
spot already - ProcArrayAdd() / ProcArrayRemove() now needs to hold
XidGenLock in addition to ProcArrayLock. Adding / Removing a procarray
entry is not a very frequent operation, even taking 2PC into account.
Due to the above, the dense array entries can only be read or modified
while holding ProcArrayLock and/or XidGenLock. This prevents a
concurrent ProcArrayRemove() from shifting the dense array while it is
accessed concurrently.
While the new dense array is very good when needing to look at all
xids it is less suitable when accessing a single backend's xid. In
particular it would be problematic to have to acquire a lock to access
a backend's own xid. Therefore a backend's xid is not just stored in
the dense array, but also in PGPROC. This also allows a backend to
only access the shared xid value when the backend had acquired an
xid.
The infrastructure added in this commit will be used for the remaining
PGXACT fields in subsequent commits. They are kept separate to make
review easier.
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
Now that xmin isn't needed for GetSnapshotData() anymore, it leads to
unnecessary cacheline ping-pong to have it in PGXACT, as it is updated
considerably more frequently than the other PGXACT members.
After the changes in dc7420c2c9, this is a very straight-forward change.
For highly concurrent, snapshot acquisition heavy, workloads this change alone
can significantly increase scalability. E.g. plain pgbench on a smaller 2
socket machine gains 1.07x for read-only pgbench, 1.22x for read-only pgbench
when submitting queries in batches of 100, and 2.85x for batches of 100
'SELECT';. The latter numbers are obviously not to be expected in the
real-world, but micro-benchmark the snapshot computation
scalability (previously spending ~80% of the time in GetSnapshotData()).
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
When a table is scanned by heapam_index_build_range_scan (née
IndexBuildHeapScan) and the table lock being held allows concurrent data
changes, it is possible for new HOT chains to sprout in a page that were
unknown when the scan of a page happened. This leads to an error such
as
ERROR: failed to find parent tuple for heap-only tuple at (X,Y) in table "tbl"
because the root tuple was not present when we first obtained the list
of the page's root tuples. This can be fixed by re-obtaining the list
of root tuples, if we see that a heap-only tuple appears to point to a
non-existing root.
This was reported by Anastasia as occurring for BRIN summarization
(which exists since 9.5), but I think it could theoretically also happen
with CREATE INDEX CONCURRENTLY (much older) or REINDEX CONCURRENTLY
(very recent). It seems a happy coincidence that BRIN forces us to
backpatch this all the way to 9.5.
Reported-by: Anastasia Lubennikova <a.lubennikova@postgrespro.ru>
Diagnosed-by: Anastasia Lubennikova <a.lubennikova@postgrespro.ru>
Co-authored-by: Anastasia Lubennikova <a.lubennikova@postgrespro.ru>
Co-authored-by: Álvaro Herrera <alvherre@alvh.no-ip.org>
Discussion: https://postgr.es/m/602d8487-f0b2-5486-0088-0f372b2549fa@postgrespro.ru
Backpatch: 9.5 - master
To make GetSnapshotData() more scalable, it cannot not look at at each proc's
xmin: While snapshot contents do not need to change whenever a read-only
transaction commits or a snapshot is released, a proc's xmin is modified in
those cases. The frequency of xmin modifications leads to, particularly on
higher core count systems, many cache misses inside GetSnapshotData(), despite
the data underlying a snapshot not changing. That is the most
significant source of GetSnapshotData() scaling poorly on larger systems.
Without accessing xmins, GetSnapshotData() cannot calculate accurate horizons /
thresholds as it has so far. But we don't really have to: The horizons don't
actually change that much between GetSnapshotData() calls. Nor are the horizons
actually used every time a snapshot is built.
The trick this commit introduces is to delay computation of accurate horizons
until there use and using horizon boundaries to determine whether accurate
horizons need to be computed.
The use of RecentGlobal[Data]Xmin to decide whether a row version could be
removed has been replaces with new GlobalVisTest* functions. These use two
thresholds to determine whether a row can be pruned:
1) definitely_needed, indicating that rows deleted by XIDs >= definitely_needed
are definitely still visible.
2) maybe_needed, indicating that rows deleted by XIDs < maybe_needed can
definitely be removed
GetSnapshotData() updates definitely_needed to be the xmin of the computed
snapshot.
When testing whether a row can be removed (with GlobalVisTestIsRemovableXid())
and the tested XID falls in between the two (i.e. XID >= maybe_needed && XID <
definitely_needed) the boundaries can be recomputed to be more accurate. As it
is not cheap to compute accurate boundaries, we limit the number of times that
happens in short succession. As the boundaries used by
GlobalVisTestIsRemovableXid() are never reset (with maybe_needed updated by
GetSnapshotData()), it is likely that further test can benefit from an earlier
computation of accurate horizons.
To avoid regressing performance when old_snapshot_threshold is set (as that
requires an accurate horizon to be computed), heap_page_prune_opt() doesn't
unconditionally call TransactionIdLimitedForOldSnapshots() anymore. Both the
computation of the limited horizon, and the triggering of errors (with
SetOldSnapshotThresholdTimestamp()) is now only done when necessary to remove
tuples.
This commit just removes the accesses to PGXACT->xmin from
GetSnapshotData(), but other members of PGXACT residing in the same
cache line are accessed. Therefore this in itself does not result in a
significant improvement. Subsequent commits will take advantage of the
fact that GetSnapshotData() now does not need to access xmins anymore.
Note: This contains a workaround in heap_page_prune_opt() to keep the
snapshot_too_old tests working. While that workaround is ugly, the tests
currently are not meaningful, and it seems best to address them separately.
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
If a page range is desummarized at just the right time concurrently with
an index walk, BRIN would raise an error indicating index corruption.
This is scary and unhelpful; silently returning that the page range is
not summarized is sufficient reaction.
This bug was introduced by commit 975ad4e602 as additional protection
against a bug whose actual fix was elsewhere. Backpatch equally.
Reported-By: Anastasia Lubennikova <a.lubennikova@postgrespro.ru>
Diagnosed-By: Alexander Lakhin <exclusion@gmail.com>
Discussion: https://postgr.es/m/2588667e-d07d-7e10-74e2-7e1e46194491@postgrespro.ru
Backpatch: 9.5 - master
The reason for doing so is that a subsequent commit will need that to
avoid wraparound issues. As the subsequent change is large this was
split out for easier review.
The reason this is not a perfect straight-forward change is that we do
not want track 64bit xids in the procarray or the WAL. Therefore we
need to advance lastestCompletedXid in relation to 32 bit xids. The
code for that is now centralized in MaintainLatestCompletedXid*.
Author: Andres Freund
Reviewed-By: Thomas Munro, Robert Haas, David Rowley
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
Including Full in variable names duplicates the type information and
leads to overly long names. As FullTransactionId cannot accidentally
be casted to TransactionId that does not seem necessary.
Author: Andres Freund
Discussion: https://postgr.es/m/20200724011143.jccsyvsvymuiqfxu@alap3.anarazel.de
They are equivalent, except that StrNCpy() zero-fills the entire
destination buffer instead of providing just one trailing zero. For
all but a tiny number of callers, that's just overhead rather than
being desirable.
Remove StrNCpy() as it is now unused.
In some cases, namestrcpy() is the more appropriate function to use.
While we're here, simplify the API of namestrcpy(): Remove the return
value, don't check for NULL input. Nothing was using that anyway.
Also, remove a few unused name-related functions.
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://www.postgresql.org/message-id/flat/44f5e198-36f6-6cdb-7fa9-60e34784daae%402ndquadrant.com
Instead of serializing the transaction to disk after reaching the
logical_decoding_work_mem limit in memory, we consume the changes we have
in memory and invoke stream API methods added by commit 45fdc9738b.
However, sometimes if we have incomplete toast or speculative insert we
spill to the disk because we can't generate the complete tuple and stream.
And, as soon as we get the complete tuple we stream the transaction
including the serialized changes.
We can do this incremental processing thanks to having assignments
(associating subxact with toplevel xacts) in WAL right away, and
thanks to logging the invalidation messages at each command end. These
features are added by commits 0bead9af48 and c55040ccd0 respectively.
Now that we can stream in-progress transactions, the concurrent aborts
may cause failures when the output plugin consults catalogs (both system
and user-defined).
We handle such failures by returning ERRCODE_TRANSACTION_ROLLBACK
sqlerrcode from system table scan APIs to the backend or WALSender
decoding a specific uncommitted transaction. The decoding logic on the
receipt of such a sqlerrcode aborts the decoding of the current
transaction and continue with the decoding of other transactions.
We have ReorderBufferTXN pointer in each ReorderBufferChange by which we
know which xact it belongs to. The output plugin can use this to decide
which changes to discard in case of stream_abort_cb (e.g. when a subxact
gets discarded).
We also provide a new option via SQL APIs to fetch the changes being
streamed.
Author: Dilip Kumar, Tomas Vondra, Amit Kapila, Nikhil Sontakke
Reviewed-by: Amit Kapila, Kuntal Ghosh, Ajin Cherian
Tested-by: Neha Sharma, Mahendra Singh Thalor and Ajin Cherian
Discussion: https://postgr.es/m/688b0b7f-2f6c-d827-c27b-216a8e3ea700@2ndquadrant.com
Make the nbtree page split REDO routine consistent with original
execution in its approach to acquiring and releasing buffer locks (at
least for pages on the tree level of the page being split). This brings
btree_xlog_split() in line with btree_xlog_unlink_page(), which was
taught to couple buffer locks by commit 9a9db08a.
Note that the precise order in which we both acquire and release sibling
buffer locks in btree_xlog_split() now matches original execution
exactly (the precise order in which the locks are released probably
doesn't matter much, but we might as well be consistent about it).
The rule for nbtree REDO routines from here on is that same-level locks
should be acquired in an order that's consistent with original
execution. It's not practical to have a similar rule for cross-level
page locks, since for the most part original execution holds those locks
for a period that spans multiple atomic actions/WAL records. It's also
not necessary, because clearly the cross-level lock coupling is only
truly needed during original execution because of the presence of
concurrent inserters.
This is not a bug fix (unlike the similar aforementioned commit, commit
9a9db08a). The immediate reason to tighten things up in this area is to
enable an upcoming enhancement to contrib/amcheck that allows it to
verify that sibling links are in agreement with only an AccessShareLock
(this check produced false positives when run on a replica server on
account of the inconsistency fixed by this commit). But that's not the
only reason to be stricter here.
It is generally useful to make locking on replicas be as close to what
happens during original execution as practically possible. It makes it
less likely that hard to catch bugs will slip in in the future. The
previous state of affairs seems to be a holdover from before the
introduction of Hot Standby, when buffer lock acquisitions during
recovery were totally unnecessary. See also: commit 3bbf668d, which
tightened things up in this area a few years after the introduction of
Hot Standby.
Discussion: https://postgr.es/m/CAH2-Wz=465cJj11YXD9RKH8z=nhQa2dofOZ_23h67EXUGOJ00Q@mail.gmail.com
Make the nbtree page split REDO routine variable names consistent with
_bt_split() (which handles the original execution of page splits).
These names make the code easier to follow by making the distinction
between the original page and the left half of the split clear. (The
left half of the split page is a temp page that REDO creates to replace
the origpage contents.)
Also reduce the elevel used when adding a new high key to the temp page
from PANIC to ERROR to be consistent. We already only raise an ERROR
when data item PageAddItem() temp page calls fail.
Currently, page unlink leaves remaining items "as is", but replay of
corresponding WAL-record re-initializes page leaving it with no items.
For the sake of consistency, this commit makes primary delete all the items
during page unlink as well.
Thanks to this change, we now don't mask contents of deleted btree page for
WAL consistency checking.
Discussion: https://postgr.es/m/CAPpHfdt_OTyQpXaPJcWzV2N-LNeNJseNB-K_A66qG%3DL518VTFw%40mail.gmail.com
Author: Alexander Korotkov
Reviewed-by: Peter Geoghegan
It was possible for the logic used by backward scans (which must reason
about concurrent page splits/deletions in its own peculiar way) to
become confused when running on a replica. Concurrent replay of a WAL
record that describes the second phase of page deletion could cause
_bt_walk_left() to get confused. btree_xlog_unlink_page() simply failed
to adhere to the same locking protocol that we use on the primary, which
is obviously wrong once you consider these two disparate functions
together. This bug is present in all stable branches.
More concretely, the problem was that nothing stopped _bt_walk_left()
from observing inconsistencies between the deletion's target page and
its original sibling pages when running on a replica. This is true even
though the second phase of page deletion is supposed to work as a single
atomic action. Queries running on replicas raised "could not find left
sibling of block %u in index %s" can't-happen errors when they went back
to their scan's "original" page and observed that the page has not been
marked deleted (even though it really was concurrently deleted).
There is no evidence that this actually happened in the real world. The
issue came to light during unrelated feature development work. Note
that _bt_walk_left() is the only code that cares about the difference
between a half-dead page and a fully deleted page that isn't also
exclusively used by nbtree VACUUM (unless you include contrib/amcheck
code). It seems very likely that backward scans are the only thing that
could become confused by the inconsistency. Even amcheck's complex
bt_right_page_check_scankey() dance was unaffected.
To fix, teach btree_xlog_unlink_page() to lock the left sibling, target,
and right sibling pages in that order before releasing any locks (just
like _bt_unlink_halfdead_page()). This is the simplest possible
approach. There doesn't seem to be any opportunity to be more clever
about lock acquisition in the REDO routine, and it hardly seems worth
the trouble in any case.
This fix might enable contrib/amcheck verification of leaf page sibling
links with only an AccessShareLock on the relation. An amcheck patch
from Andrey Borodin was rejected back in January because it clashed with
btree_xlog_unlink_page()'s lax approach to locking pages. It now seems
likely that the real problem was with btree_xlog_unlink_page(), not the
patch.
This is a low severity, low likelihood bug, so no backpatch.
Author: Michail Nikolaev
Diagnosed-By: Michail Nikolaev
Discussion: https://postgr.es/m/CANtu0ohkR-evAWbpzJu54V8eCOtqjJyYp3PQ_SGoBTRGXWhWRw@mail.gmail.com
Add a documenting assertion that's similar to the nearby assertion added
by commit cd8c73a3. This conveys that the entire call to _bt_pagedel()
does no work if it isn't possible to get a descent stack for the initial
scanblkno page.
We have edge-case bugs when assigning values in the last few dozen pages
before the wrap limit. We may introduce similar bugs in the future. At
default BLCKSZ, this makes such bugs unreachable outside of single-user
mode. Also, when VACUUM began to consume mxacts, multiStopLimit did not
change to compensate.
pg_upgrade may fail on a cluster that was already printing "must be
vacuumed" warnings. Follow the warning's instructions to clear the
warning, then run pg_upgrade again. One can still, peacefully consume
98% of XIDs or mxacts, so DBAs need not change routine VACUUM settings.
Discussion: https://postgr.es/m/20200621083513.GA3074645@rfd.leadboat.com
This allows AM-specific knowledge to be applied during creation of
pg_amop and pg_amproc entries. Specifically, the AM knows better than
core code which entries to consider as required or optional. Giving
the latter entries the appropriate sort of dependency allows them to
be dropped without taking out the whole opclass or opfamily; which
is something we'd like to have to correct obsolescent entries in
extensions.
This callback also opens the door to performing AM-specific validity
checks during opclass creation, rather than hoping than an opclass
developer will remember to test with "amvalidate". For the most part
I've not actually added any such checks yet; that can happen in a
follow-on patch. (Note that we shouldn't remove any tests from
"amvalidate", as those are still needed to cross-check manually
constructed entries in the initdb data. So adding tests to
"amadjustmembers" will be somewhat duplicative, but it seems like
a good idea anyway.)
Patch by me, reviewed by Alexander Korotkov, Hamid Akhtar, and
Anastasia Lubennikova.
Discussion: https://postgr.es/m/4578.1565195302@sss.pgh.pa.us
Avoid repeatedly calling lseek(SEEK_END) during recovery by caching
the size of each fork. For now, we can't use the same technique in
other processes, because we lack a shared invalidation mechanism.
Do this by generalizing the pre-existing caching used by FSM and VM
to support all forks.
Discussion: https://postgr.es/m/CAEepm%3D3SSw-Ty1DFcK%3D1rU-K6GSzYzfdD4d%2BZwapdN7dTa6%3DnQ%40mail.gmail.com
For pg_attribute, this allows to insert at once a full set of attributes
for a relation (roughly 15% of WAL reduction in extreme cases). For
pg_shdepend, this reduces the work done when creating new shared
dependencies from a database template. The number of slots used for the
insertion is capped at 64kB of data inserted for both, depending on the
number of items to insert and the length of the rows involved.
More can be done for other catalogs, like pg_depend. This part requires
a different approach as the number of slots to use depends also on the
number of entries discarded as pinned dependencies. This is also
related to the rework or dependency handling for ALTER TABLE and CREATE
TABLE, mainly.
Author: Daniel Gustafsson
Reviewed-by: Andres Freund, Michael Paquier
Discussion: https://postgr.es/m/20190213182737.mxn6hkdxwrzgxk35@alap3.anarazel.de
When fast promotion was supported in 9.3, non-fast promotion became
undocumented feature and it's basically not available for ordinary users.
However we decided not to remove non-fast promotion at that moment,
to leave it for a release or two for debugging purpose or as an emergency
method because fast promotion might have some issues, and then to
remove it later. Now, several versions were released since that decision
and there is no longer reason to keep supporting non-fast promotion.
Therefore this commit removes non-fast promotion.
Author: Fujii Masao
Reviewed-by: Hamid Akhtar, Kyotaro Horiguchi
Discussion: https://postgr.es/m/76066434-648f-f567-437b-54853b43398f@oss.nttdata.com
Previously we would allocate blocks to parallel workers during a parallel
sequential scan 1 block at a time. Since other workers were likely to
request a block before a worker returns for another block number to work
on, this could lead to non-sequential I/O patterns in each worker which
could cause the operating system's readahead to perform poorly or not at
all.
Here we change things so that we allocate consecutive "chunks" of blocks
to workers and have them work on those until they're done, at which time
we allocate another chunk for the worker. The size of these chunks is
based on the size of the relation.
Initial patch here was by Thomas Munro which showed some good improvements
just having a fixed chunk size of 64 blocks with a simple ramp-down near
the end of the scan. The revisions of the patch to make the chunk size
based on the relation size and the adjusted ramp-down in powers of two was
done by me, along with quite extensive benchmarking to determine the
optimal chunk sizes.
For the most part, benchmarks have shown significant performance
improvements for large parallel sequential scans on Linux, FreeBSD and
Windows using SSDs. It's less clear how this affects the performance of
cloud providers. Tests done so far are unable to obtain stable enough
performance to provide meaningful benchmark results. It is possible that
this could cause some performance regressions on more obscure filesystems,
so we may need to later provide users with some ability to get something
closer to the old behavior. For now, let's leave that until we see that
it's really required.
Author: Thomas Munro, David Rowley
Reviewed-by: Ranier Vilela, Soumyadeep Chakraborty, Robert Haas
Reviewed-by: Amit Kapila, Kirk Jamison
Discussion: https://postgr.es/m/CA+hUKGJ_EErDv41YycXcbMbCBkztA34+z1ts9VQH+ACRuvpxig@mail.gmail.com
When wal_level=logical, write invalidations at command end into WAL so
that decoding can use this information.
This patch is required to allow the streaming of in-progress transactions
in logical decoding. The actual work to allow streaming will be committed
as a separate patch.
We still add the invalidations to the cache and write them to WAL at
commit time in RecordTransactionCommit(). This uses the existing
XLOG_INVALIDATIONS xlog record type, from the RM_STANDBY_ID resource
manager (see LogStandbyInvalidations for details).
So existing code relying on those invalidations (e.g. redo) does not need
to be changed.
The invalidations written at command end uses a new xlog record type
XLOG_XACT_INVALIDATIONS, from RM_XACT_ID resource manager. See
LogLogicalInvalidations for details.
These new xlog records are ignored by existing redo procedures, which
still rely on the invalidations written to commit records.
The invalidations are decoded and accumulated in top-transaction, and then
executed during replay. This obviates the need to decode the
invalidations as part of a commit record.
Bump XLOG_PAGE_MAGIC, since this introduces XLOG_XACT_INVALIDATIONS.
Author: Dilip Kumar, Tomas Vondra, Amit Kapila
Reviewed-by: Amit Kapila
Tested-by: Neha Sharma and Mahendra Singh Thalor
Discussion: https://postgr.es/m/688b0b7f-2f6c-d827-c27b-216a8e3ea700@2ndquadrant.com
Holding just a buffer pin (with no buffer lock) on an nbtree buffer/page
provides very weak guarantees, especially compared to heapam, where it's
often safe to read a page while only holding a buffer pin. This commit
has Valgrind enforce the following rule: it is never okay to access an
nbtree buffer without holding both a pin and a lock on the buffer.
A draft version of this patch detected questionable code that was
cleaned up by commits fa7ff642 and 7154aa16. The code in question used
to access an nbtree buffer page's special/opaque area with no buffer
lock (only a buffer pin). This practice (which isn't obviously unsafe)
is hereby formally disallowed in nbtree. There doesn't seem to be any
reason to allow it, and banning it keeps things simple for Valgrind.
The new checks are implemented by adding custom nbtree client requests
(located in LockBuffer() wrapper functions); these requests are
"superimposed" on top of the generic bufmgr.c Valgrind client requests
added by commit 1e0dfd16. No custom resource management cleanup code is
needed to undo the effects of marking buffers as non-accessible under
this scheme.
Author: Peter Geoghegan
Reviewed-By: Anastasia Lubennikova, Georgios Kokolatos
Discussion: https://postgr.es/m/CAH2-WzkLgyN3zBvRZ1pkNJThC=xi_0gpWRUb_45eexLH1+k2_Q@mail.gmail.com
max_slot_wal_keep_size that was added in v13 and wal_keep_segments are
the GUC parameters to specify how much WAL files to retain for
the standby servers. While max_slot_wal_keep_size accepts the number of
bytes of WAL files, wal_keep_segments accepts the number of WAL files.
This difference of setting units between those similar parameters could
be confusing to users.
To alleviate this situation, this commit renames wal_keep_segments to
wal_keep_size, and make users specify the WAL size in it instead of
the number of WAL files.
There was also the idea to rename max_slot_wal_keep_size to
max_slot_wal_keep_segments, in the discussion. But we have been moving
away from measuring in segments, for example, checkpoint_segments was
replaced by max_wal_size. So we concluded to rename wal_keep_segments
to wal_keep_size.
Back-patch to v13 where max_slot_wal_keep_size was added.
Author: Fujii Masao
Reviewed-by: Álvaro Herrera, Kyotaro Horiguchi, David Steele
Discussion: https://postgr.es/m/574b4ea3-e0f9-b175-ead2-ebea7faea855@oss.nttdata.com
The logical decoding infrastructure needs to know which top-level
transaction the subxact belongs to, in order to decode all the
changes. Until now that might be delayed until commit, due to the
caching (GPROC_MAX_CACHED_SUBXIDS), preventing features requiring
incremental decoding.
So we also write the assignment info into WAL immediately, as part
of the next WAL record (to minimize overhead) only when wal_level=logical.
We can not remove the existing XLOG_XACT_ASSIGNMENT WAL as that is
required for avoiding overflow in the hot standby snapshot.
Bump XLOG_PAGE_MAGIC, since this introduces XLR_BLOCK_ID_TOPLEVEL_XID.
Author: Tomas Vondra, Dilip Kumar, Amit Kapila
Reviewed-by: Amit Kapila
Tested-by: Neha Sharma and Mahendra Singh Thalor
Discussion: https://postgr.es/m/688b0b7f-2f6c-d827-c27b-216a8e3ea700@2ndquadrant.com
Valgrind builds with assertions enabled sometimes perform a
theoretically unsafe page access inside an assertion in
heapam_tuple_lock(). This happened when the eval-plan-qual isolation
test ran one of the permutations added by commit a2418f9e23.
Avoid complaints from Valgrind by moving the assertion ever so slightly.
This is minor cleanup for commit 1e0dfd16, which added Valgrind buffer
access instrumentation.
No backpatch, since this only happens within an assertion, and seems
very unlikely to cause any real problems even with assert-enabled
builds.
There is no advantage to attempting deduplication for a unique index
during CREATE INDEX, since there cannot possibly be any duplicates.
Doing so wastes cycles due to unnecessary copying. Make sure that we
avoid it consistently.
We already avoided unique index deduplication in the case where there
were some spool2 tuples to merge. That didn't account for the fact that
spool2 is removed early/unset in the common case where it has no tuples
that need to be merged (i.e. it failed to account for the "spool2 turns
out to be unnecessary" optimization in _bt_spools_heapscan()).
Oversight in commit 0d861bbb, which added nbtree deduplication
Backpatch: 13-, where nbtree deduplication was introduced.
Peter Eisentraut
Alvaro Herrera
Tom Lane
Bruce Momjian
Michael Paquier
Amit Kapila
Andres Freund
Heikki Linnakangas
Noah Misch
Peter Geoghegan
Alexander Korotkov
Thomas Munro
Fujii Masao
David Rowley