Previously, md.c and checkpointer.c were tightly integrated so that
fsync calls could be handed off and processed in the background.
Introduce a system of callbacks and file tags, so that other modules
can hand off fsync work in the same way.
For now only md.c uses the new interface, but other users are being
proposed. Since there may be use cases that are not strictly SMGR
implementations, use a new function table for sync handlers rather
than extending the traditional SMGR one.
Instead of using a bitmapset of segment numbers for each RelFileNode
in the checkpointer's hash table, make the segment number part of the
key. This requires sending explicit "forget" requests for every
segment individually when relations are dropped, but suits the file
layout schemes of proposed future users better (ie sparse or high
segment numbers).
Author: Shawn Debnath and Thomas Munro
Reviewed-by: Thomas Munro, Andres Freund
Discussion: https://postgr.es/m/CAEepm=2gTANm=e3ARnJT=n0h8hf88wqmaZxk0JYkxw+b21fNrw@mail.gmail.com
This commit makes existing GIN operator classes jsonb_ops and json_path_ops
support "jsonb @@ jsonpath" and "jsonb @? jsonpath" operators. Basic idea is
to extract statements of following form out of jsonpath.
key1.key2. ... .keyN = const
The rest of jsonpath is rechecked from heap.
Catversion is bumped.
Discussion: https://postgr.es/m/fcc6fc6a-b497-f39a-923d-aa34d0c588e8%402ndQuadrant.com
Author: Nikita Glukhov, Alexander Korotkov
Reviewed-by: Jonathan Katz, Pavel Stehule
Provide GetTopFullTransactionId() and GetCurrentFullTransactionId().
The intended users of these interfaces are access methods that use
xids for visibility checks but don't want to have to go back and
"freeze" existing references some time later before the 32 bit xid
counter wraps around.
Use a new struct to serialize the transaction state for parallel
query, because FullTransactionId doesn't fit into the previous
serialization scheme very well.
Author: Thomas Munro
Reviewed-by: Heikki Linnakangas
Discussion: https://postgr.es/m/CAA4eK1%2BMv%2Bmb0HFfWM9Srtc6MVe160WFurXV68iAFMcagRZ0dQ%40mail.gmail.com
Instead of inferring epoch progress from xids and checkpoints,
introduce a 64 bit FullTransactionId type and use it to track xid
generation. This fixes an unlikely bug where the epoch is reported
incorrectly if the range of active xids wraps around more than once
between checkpoints.
The only user-visible effect of this commit is to correct the epoch
used by txid_current() and txid_status(), also visible with
pg_controldata, in those rare circumstances. It also creates some
basic infrastructure so that later patches can use 64 bit
transaction IDs in more places.
The new type is a struct that we pass by value, as a form of strong
typedef. This prevents the sort of accidental confusion between
TransactionId and FullTransactionId that would be possible if we
were to use a plain old uint64.
Author: Thomas Munro
Reported-by: Amit Kapila
Reviewed-by: Andres Freund, Tom Lane, Heikki Linnakangas
Discussion: https://postgr.es/m/CAA4eK1%2BMv%2Bmb0HFfWM9Srtc6MVe160WFurXV68iAFMcagRZ0dQ%40mail.gmail.com
To support building indexes over tables of different AMs, the scans to
do so need to be routed through the table AM. While moving a fair
amount of code, nearly all the changes are just moving code to below a
callback.
Currently the range based interface wouldn't make much sense for non
block based table AMs. But that seems aceptable for now.
Author: Andres Freund
Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
Previously the xid horizon was only computed during WAL replay. That
had two major problems:
1) It relied on knowing what the table pointed to looks like. That was
easy enough before the introducing of tableam (we knew it had to be
heap, although some trickery around logging the heap relfilenodes
was required). But to properly handle table AMs we need
per-database catalog access to look up the AM handler, which
recovery doesn't allow.
2) Not knowing the xid horizon also makes it hard to support logical
decoding on standbys. When on a catalog table, we need to be able
to conflict with slots that have an xid horizon that's too old. But
computing the horizon by visiting the heap only works once
consistency is reached, but we always need to be able to detect
conflicts.
There's also a secondary problem, in that the current method performs
redundant work on every standby. But that's counterbalanced by
potentially computing the value when not necessary (either because
there's no standby, or because there's no connected backends).
Solve 1) and 2) by moving computation of the xid horizon to the
primary and by involving tableam in the computation of the horizon.
To address the potentially increased overhead, increase the efficiency
of the xid horizon computation for heap by sorting the tids, and
eliminating redundant buffer accesses. When prefetching is available,
additionally perform prefetching of buffers. As this is more of a
maintenance task, rather than something routinely done in every read
only query, we add an arbitrary 10 to the effective concurrency -
thereby using IO concurrency, when not globally enabled. That's
possibly not the perfect formula, but seems good enough for now.
Bumps WAL format, as latestRemovedXid is now part of the records, and
the heap's relfilenode isn't anymore.
Author: Andres Freund, Amit Khandekar, Robert Haas
Reviewed-By: Robert Haas
Discussion:
https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.dehttps://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.dehttps://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
This adds new, required, table AM callbacks for insert/delete/update
and lock_tuple. To be able to reasonably use those, the EvalPlanQual
mechanism had to be adapted, moving more logic into the AM.
Previously both delete/update/lock call-sites and the EPQ mechanism had
to have awareness of the specific tuple format to be able to fetch the
latest version of a tuple. Obviously that needs to be abstracted
away. To do so, move the logic that find the latest row version into
the AM. lock_tuple has a new flag argument,
TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last
version, rather than the current one. It'd have been possible to do
so via a separate callback as well, but finding the last version
usually also necessitates locking the newest version, making it
sensible to combine the two. This replaces the previous use of
EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously
signaled either a concurrent update or delete, is now split into two,
to avoid callers needing AM specific knowledge to differentiate.
The move of finding the latest row version into tuple_lock means that
encountering a row concurrently moved into another partition will now
raise an error about "tuple to be locked" rather than "tuple to be
updated/deleted" - which is accurate, as that always happens when
locking rows. While possible slightly less helpful for users, it seems
like an acceptable trade-off.
As part of this commit HTSU_Result has been renamed to TM_Result, and
its members been expanded to differentiated between updating and
deleting. HeapUpdateFailureData has been renamed to TM_FailureData.
The interface to speculative insertion is changed so nodeModifyTable.c
does not have to set the speculative token itself anymore. Instead
there's a version of tuple_insert, tuple_insert_speculative, that
performs the speculative insertion (without requiring a flag to signal
that fact), and the speculative insertion is either made permanent
with table_complete_speculative(succeeded = true) or aborted with
succeeded = false).
Note that multi_insert is not yet routed through tableam, nor is
COPY. Changing multi_insert requires changes to copy.c that are large
enough to better be done separately.
Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED
VIEW are also only going to be adjusted in a later commit.
Author: Andres Freund and Haribabu Kommi
Discussion:
https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.dehttps://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.dehttps://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
An offline cluster can now work with more modes in pg_checksums:
- --enable enables checksums in a cluster, updating all blocks with a
correct checksum, and updating the control file at the end.
- --disable disables checksums in a cluster, updating only the control
file.
- --check is an extra option able to verify checksums for a cluster, and
the default used if no mode is specified.
When running --enable or --disable, the data folder gets fsync'd for
durability, and then it is followed by a control file update and flush
to keep the operation consistent should the tool be interrupted, killed
or the host unplugged. If no mode is specified in the options, then
--check is used for compatibility with older versions of pg_checksums
(named pg_verify_checksums in v11 where it was introduced).
Author: Michael Banck, Michael Paquier
Reviewed-by: Fabien Coelho, Magnus Hagander, Sergei Kornilov
Discussion: https://postgr.es/m/20181221201616.GD4974@nighthawk.caipicrew.dd-dns.de
SQL 2016 standards among other things contains set of SQL/JSON features for
JSON processing inside of relational database. The core of SQL/JSON is JSON
path language, allowing access parts of JSON documents and make computations
over them. This commit implements partial support JSON path language as
separate datatype called "jsonpath". The implementation is partial because
it's lacking datetime support and suppression of numeric errors. Missing
features will be added later by separate commits.
Support of SQL/JSON features requires implementation of separate nodes, and it
will be considered in subsequent patches. This commit includes following
set of plain functions, allowing to execute jsonpath over jsonb values:
* jsonb_path_exists(jsonb, jsonpath[, jsonb, bool]),
* jsonb_path_match(jsonb, jsonpath[, jsonb, bool]),
* jsonb_path_query(jsonb, jsonpath[, jsonb, bool]),
* jsonb_path_query_array(jsonb, jsonpath[, jsonb, bool]).
* jsonb_path_query_first(jsonb, jsonpath[, jsonb, bool]).
This commit also implements "jsonb @? jsonpath" and "jsonb @@ jsonpath", which
are wrappers over jsonpath_exists(jsonb, jsonpath) and jsonpath_predicate(jsonb,
jsonpath) correspondingly. These operators will have an index support
(implemented in subsequent patches).
Catversion bumped, to add new functions and operators.
Code was written by Nikita Glukhov and Teodor Sigaev, revised by me.
Documentation was written by Oleg Bartunov and Liudmila Mantrova. The work
was inspired by Oleg Bartunov.
Discussion: https://postgr.es/m/fcc6fc6a-b497-f39a-923d-aa34d0c588e8%402ndQuadrant.com
Author: Nikita Glukhov, Teodor Sigaev, Alexander Korotkov, Oleg Bartunov, Liudmila Mantrova
Reviewed-by: Tomas Vondra, Andrew Dunstan, Pavel Stehule, Alexander Korotkov
Too allow table accesses to be not directly dependent on heap, several
new abstractions are needed. Specifically:
1) Heap scans need to be generalized into table scans. Do this by
introducing TableScanDesc, which will be the "base class" for
individual AMs. This contains the AM independent fields from
HeapScanDesc.
The previous heap_{beginscan,rescan,endscan} et al. have been
replaced with a table_ version.
There's no direct replacement for heap_getnext(), as that returned
a HeapTuple, which is undesirable for a other AMs. Instead there's
table_scan_getnextslot(). But note that heap_getnext() lives on,
it's still used widely to access catalog tables.
This is achieved by new scan_begin, scan_end, scan_rescan,
scan_getnextslot callbacks.
2) The portion of parallel scans that's shared between backends need
to be able to do so without the user doing per-AM work. To achieve
that new parallelscan_{estimate, initialize, reinitialize}
callbacks are introduced, which operate on a new
ParallelTableScanDesc, which again can be subclassed by AMs.
As it is likely that several AMs are going to be block oriented,
block oriented callbacks that can be shared between such AMs are
provided and used by heap. table_block_parallelscan_{estimate,
intiialize, reinitialize} as callbacks, and
table_block_parallelscan_{nextpage, init} for use in AMs. These
operate on a ParallelBlockTableScanDesc.
3) Index scans need to be able to access tables to return a tuple, and
there needs to be state across individual accesses to the heap to
store state like buffers. That's now handled by introducing a
sort-of-scan IndexFetchTable, which again is intended to be
subclassed by individual AMs (for heap IndexFetchHeap).
The relevant callbacks for an AM are index_fetch_{end, begin,
reset} to create the necessary state, and index_fetch_tuple to
retrieve an indexed tuple. Note that index_fetch_tuple
implementations need to be smarter than just blindly fetching the
tuples for AMs that have optimizations similar to heap's HOT - the
currently alive tuple in the update chain needs to be fetched if
appropriate.
Similar to table_scan_getnextslot(), it's undesirable to continue
to return HeapTuples. Thus index_fetch_heap (might want to rename
that later) now accepts a slot as an argument. Core code doesn't
have a lot of call sites performing index scans without going
through the systable_* API (in contrast to loads of heap_getnext
calls and working directly with HeapTuples).
Index scans now store the result of a search in
IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the
target is not generally a HeapTuple anymore that seems cleaner.
To be able to sensible adapt code to use the above, two further
callbacks have been introduced:
a) slot_callbacks returns a TupleTableSlotOps* suitable for creating
slots capable of holding a tuple of the AMs
type. table_slot_callbacks() and table_slot_create() are based
upon that, but have additional logic to deal with views, foreign
tables, etc.
While this change could have been done separately, nearly all the
call sites that needed to be adapted for the rest of this commit
also would have been needed to be adapted for
table_slot_callbacks(), making separation not worthwhile.
b) tuple_satisfies_snapshot checks whether the tuple in a slot is
currently visible according to a snapshot. That's required as a few
places now don't have a buffer + HeapTuple around, but a
slot (which in heap's case internally has that information).
Additionally a few infrastructure changes were needed:
I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now
internally uses a slot to keep track of tuples. While
systable_getnext() still returns HeapTuples, and will so for the
foreseeable future, the index API (see 1) above) now only deals with
slots.
The remainder, and largest part, of this commit is then adjusting all
scans in postgres to use the new APIs.
Author: Andres Freund, Haribabu Kommi, Alvaro Herrera
Discussion:
https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.dehttps://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
This introduces the concept of table access methods, i.e. CREATE
ACCESS METHOD ... TYPE TABLE and
CREATE TABLE ... USING (storage-engine).
No table access functionality is delegated to table AMs as of this
commit, that'll be done in following commits.
Subsequent commits will incrementally abstract table access
functionality to be routed through table access methods. That change
is too large to be reviewed & committed at once, so it'll be done
incrementally.
Docs will be updated at the end, as adding them incrementally would
likely make them less coherent, and definitely is a lot more work,
without a lot of benefit.
Table access methods are specified similar to index access methods,
i.e. pg_am.amhandler returns, as INTERNAL, a pointer to a struct with
callbacks. In contrast to index AMs that struct needs to live as long
as a backend, typically that's achieved by just returning a pointer to
a constant struct.
Psql's \d+ now displays a table's access method. That can be disabled
with HIDE_TABLEAM=true, which is mainly useful so regression tests can
be run against different AMs. It's quite possible that this behaviour
still needs to be fine tuned.
For now it's not allowed to set a table AM for a partitioned table, as
we've not resolved how partitions would inherit that. Disallowing
allows us to introduce, if we decide that's the way forward, such a
behaviour without a compatibility break.
Catversion bumped, to add the heap table AM and references to it.
Author: Haribabu Kommi, Andres Freund, Alvaro Herrera, Dimitri Golgov and others
Discussion:
https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.dehttps://postgr.es/m/20160812231527.GA690404@alvherre.pgsqlhttps://postgr.es/m/20190107235616.6lur25ph22u5u5av@alap3.anarazel.dehttps://postgr.es/m/20190304234700.w5tmhducs5wxgzls@alap3.anarazel.de
Before this change FunctionCallInfoData, the struct arguments etc for
V1 function calls are stored in, always had space for
FUNC_MAX_ARGS/100 arguments, storing datums and their nullness in two
arrays. For nearly every function call 100 arguments is far more than
needed, therefore wasting memory. Arg and argnull being two separate
arrays also guarantees that to access a single argument, two
cachelines have to be touched.
Change the layout so there's a single variable-length array with pairs
of value / isnull. That drastically reduces memory consumption for
most function calls (on x86-64 a two argument function now uses
64bytes, previously 936 bytes), and makes it very likely that argument
value and its nullness are on the same cacheline.
Arguments are stored in a new NullableDatum struct, which, due to
padding, needs more memory per argument than before. But as usually
far fewer arguments are stored, and individual arguments are cheaper
to access, that's still a clear win. It's likely that there's other
places where conversion to NullableDatum arrays would make sense,
e.g. TupleTableSlots, but that's for another commit.
Because the function call information is now variable-length
allocations have to take the number of arguments into account. For
heap allocations that can be done with SizeForFunctionCallInfoData(),
for on-stack allocations there's a new LOCAL_FCINFO(name, nargs) macro
that helps to allocate an appropriately sized and aligned variable.
Some places with stack allocation function call information don't know
the number of arguments at compile time, and currently variably sized
stack allocations aren't allowed in postgres. Therefore allow for
FUNC_MAX_ARGS space in these cases. They're not that common, so for
now that seems acceptable.
Because of the need to allocate FunctionCallInfo of the appropriate
size, older extensions may need to update their code. To avoid subtle
breakages, the FunctionCallInfoData struct has been renamed to
FunctionCallInfoBaseData. Most code only references FunctionCallInfo,
so that shouldn't cause much collateral damage.
This change is also a prerequisite for more efficient expression JIT
compilation (by allocating the function call information on the stack,
allowing LLVM to optimize it away); previously the size of the call
information caused problems inside LLVM's optimizer.
Author: Andres Freund
Reviewed-By: Tom Lane
Discussion: https://postgr.es/m/20180605172952.x34m5uz6ju6enaem@alap3.anarazel.de
This is in preparation for allowing the same snapshot be used for
different table AMs. With the current callback based approach we would
need one callback for each supported AM, which clearly would not be
extensible. Thus add a new Snapshot->snapshot_type field, and move
the dispatch into HeapTupleSatisfiesVisibility() (which is now a
function). Later work will then dispatch calls to
HeapTupleSatisfiesVisibility() and other AMs visibility functions
depending on the type of the table. The central SnapshotType enum
also seems like a good location to centralize documentation about the
intended behaviour of various types of snapshots.
As tqual.h isn't included by bufmgr.h any more (as HeapTupleSatisfies*
isn't referenced by TestForOldSnapshot() anymore) a few files now need
to include it directly.
Author: Andres Freund, loosely based on earlier work by Haribabu Kommi
Discussion:
https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.dehttps://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
This extends cluster_rel() in such a way that more options can be added
in the future, which will reduce the amount of chunk code for an
upcoming SKIP_LOCKED aimed for VACUUM. As VACUUM FULL is a different
flavor of CLUSTER, we want to make that extensible to ease integration.
This only reworks the API and its callers, without providing anything
user-facing. Two options are present now: verbose mode and relation
recheck when doing the cluster command work across multiple
transactions. This could be used as well as a base to extend the
grammar of CLUSTER later on.
Author: Michael Paquier
Reviewed-by: Nathan Bossart
Discussion: https://postgr.es/m/20180723031058.GE2854@paquier.xyz
Since the old logic was completely unaware of subtransactions, a
change made in a subsequently-aborted subtransaction would still cause
workers to be stopped at toplevel transaction commit. Fix that by
managing a stack of worker lists rather than just one.
Amit Khandekar and Robert Haas
Discussion: http://postgr.es/m/CAJ3gD9eaG_mWqiOTA2LfAug-VRNn1hrhf50Xi1YroxL37QkZNg@mail.gmail.com
Update typedefs.list from current buildfarm results. Adjust pgindent's
typedef blacklist to block some more unfortunate typedef names that have
snuck in since last time. Manually tweak a few places where I didn't
like the initial results of pgindent'ing.
This reverts commits d204ef6377,
83454e3c2b and a few more commits thereafter
(complete list at the end) related to MERGE feature.
While the feature was fully functional, with sufficient test coverage and
necessary documentation, it was felt that some parts of the executor and
parse-analyzer can use a different design and it wasn't possible to do that in
the available time. So it was decided to revert the patch for PG11 and retry
again in the future.
Thanks again to all reviewers and bug reporters.
List of commits reverted, in reverse chronological order:
f1464c5380 Improve parse representation for MERGE
ddb4158579 MERGE syntax diagram correction
530e69e59b Allow cpluspluscheck to pass by renaming variable
01b88b4df5 MERGE minor errata
3af7b2b0d4 MERGE fix variable warning in non-assert builds
a5d86181ec MERGE INSERT allows only one VALUES clause
4b2d44031f MERGE post-commit review
4923550c20 Tab completion for MERGE
aa3faa3c7a WITH support in MERGE
83454e3c2b New files for MERGE
d204ef6377 MERGE SQL Command following SQL:2016
Author: Pavan Deolasee
Reviewed-by: Michael Paquier
MERGE performs actions that modify rows in the target table
using a source table or query. MERGE provides a single SQL
statement that can conditionally INSERT/UPDATE/DELETE rows
a task that would other require multiple PL statements.
e.g.
MERGE INTO target AS t
USING source AS s
ON t.tid = s.sid
WHEN MATCHED AND t.balance > s.delta THEN
UPDATE SET balance = t.balance - s.delta
WHEN MATCHED THEN
DELETE
WHEN NOT MATCHED AND s.delta > 0 THEN
INSERT VALUES (s.sid, s.delta)
WHEN NOT MATCHED THEN
DO NOTHING;
MERGE works with regular and partitioned tables, including
column and row security enforcement, as well as support for
row, statement and transition triggers.
MERGE is optimized for OLTP and is parameterizable, though
also useful for large scale ETL/ELT. MERGE is not intended
to be used in preference to existing single SQL commands
for INSERT, UPDATE or DELETE since there is some overhead.
MERGE can be used statically from PL/pgSQL.
MERGE does not yet support inheritance, write rules,
RETURNING clauses, updatable views or foreign tables.
MERGE follows SQL Standard per the most recent SQL:2016.
Includes full tests and documentation, including full
isolation tests to demonstrate the concurrent behavior.
This version written from scratch in 2017 by Simon Riggs,
using docs and tests originally written in 2009. Later work
from Pavan Deolasee has been both complex and deep, leaving
the lead author credit now in his hands.
Extensive discussion of concurrency from Peter Geoghegan,
with thanks for the time and effort contributed.
Various issues reported via sqlsmith by Andreas Seltenreich
Authors: Pavan Deolasee, Simon Riggs
Reviewer: Peter Geoghegan, Amit Langote, Tomas Vondra, Simon Riggs
Discussion:
https://postgr.es/m/CANP8+jKitBSrB7oTgT9CY2i1ObfOt36z0XMraQc+Xrz8QB0nXA@mail.gmail.comhttps://postgr.es/m/CAH2-WzkJdBuxj9PO=2QaO9-3h3xGbQPZ34kJH=HukRekwM-GZg@mail.gmail.com
A Bloom filter is a space-efficient, probabilistic data structure that
can be used to test set membership. Callers will sometimes incur false
positives, but never false negatives. The rate of false positives is a
function of the total number of elements and the amount of memory
available for the Bloom filter.
Two classic applications of Bloom filters are cache filtering, and data
synchronization testing. Any user of Bloom filters must accept the
possibility of false positives as a cost worth paying for the benefit in
space efficiency.
This commit adds a test harness extension module, test_bloomfilter. It
can be used to get a sense of how the Bloom filter implementation
performs under varying conditions.
This is infrastructure for the upcoming "heapallindexed" amcheck patch,
which verifies the consistency of a heap relation against one of its
indexes.
Author: Peter Geoghegan
Reviewed-By: Andrey Borodin, Michael Paquier, Thomas Munro, Andres Freund
Discussion: https://postgr.es/m/CAH2-Wzm5VmG7cu1N-H=nnS57wZThoSDQU+F5dewx3o84M+jY=g@mail.gmail.com
In addition to the interpretation of expressions (which back
evaluation of WHERE clauses, target list projection, aggregates
transition values etc) support compiling expressions to native code,
using the infrastructure added in earlier commits.
To avoid duplicating a lot of code, only support emitting code for
cases that are likely to be performance critical. For expression steps
that aren't deemed that, use the existing interpreter.
The generated code isn't great - some architectural changes are
required to address that. But this already yields a significant
speedup for some analytics queries, particularly with WHERE clauses
filtering a lot, or computing multiple aggregates.
Author: Andres Freund
Tested-By: Thomas Munro
Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
Disable JITing for VALUES() nodes.
VALUES() nodes are only ever executed once. This is primarily helpful
for debugging, when forcing JITing even for cheap queries.
Author: Andres Freund
Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
This commit introduces the ability to actually generate code using
LLVM. In particular, this adds:
- Ability to emit code both in heavily optimized and largely
unoptimized fashion
- Batching facility to allow functions to be defined in small
increments, but optimized and emitted in executable form in larger
batches (for performance and memory efficiency)
- Type and function declaration synchronization between runtime
generated code and normal postgres code. This is critical to be able
to access struct fields etc.
- Developer oriented jit_dump_bitcode GUC, for inspecting / debugging
the generated code.
- per JitContext statistics of number of functions, time spent
generating code, optimizing, and emitting it. This will later be
employed for EXPLAIN support.
This commit doesn't yet contain any code actually generating
functions. That'll follow in later commits.
Documentation for GUCs added, and for JIT in general, will be added in
later commits.
Author: Andres Freund, with contributions by Pierre Ducroquet
Testing-By: Thomas Munro, Peter Eisentraut
Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
If the partition keys of input relation are part of the GROUP BY
clause, all the rows belonging to a given group come from a single
partition. This allows aggregation/grouping over a partitioned
relation to be broken down * into aggregation/grouping on each
partition. This should be no worse, and often better, than the normal
approach.
If the GROUP BY clause does not contain all the partition keys, we can
still perform partial aggregation for each partition and then finalize
aggregation after appending the partial results. This is less certain
to be a win, but it's still useful.
Jeevan Chalke, Ashutosh Bapat, Robert Haas. The larger patch series
of which this patch is a part was also reviewed and tested by Antonin
Houska, Rajkumar Raghuwanshi, David Rowley, Dilip Kumar, Konstantin
Knizhnik, Pascal Legrand, and Rafia Sabih.
Discussion: http://postgr.es/m/CAM2+6=V64_xhstVHie0Rz=KPEQnLJMZt_e314P0jaT_oJ9MR8A@mail.gmail.com
This commit introduces:
1) JIT provider abstraction, which allows JIT functionality to be
implemented in separate shared libraries. That's desirable because
it allows to install JIT support as a separate package, and because
it allows experimentation with different forms of JITing.
2) JITContexts which can be, using functions introduced in follow up
commits, used to emit JITed functions, and have them be cleaned up
on error.
3) The outline of a LLVM JIT provider, which will be fleshed out in
subsequent commits.
Documentation for GUCs added, and for JIT in general, will be added in
later commits.
Author: Andres Freund, with architectural input from Jeff Davis
Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
To make this work, tuplesort.c and logtape.c must also support
parallelism, so this patch adds that infrastructure and then applies
it to the particular case of parallel btree index builds. Testing
to date shows that this can often be 2-3x faster than a serial
index build.
The model for deciding how many workers to use is fairly primitive
at present, but it's better than not having the feature. We can
refine it as we get more experience.
Peter Geoghegan with some help from Rushabh Lathia. While Heikki
Linnakangas is not an author of this patch, he wrote other patches
without which this feature would not have been possible, and
therefore the release notes should possibly credit him as an author
of this feature. Reviewed by Claudio Freire, Heikki Linnakangas,
Thomas Munro, Tels, Amit Kapila, me.
Discussion: http://postgr.es/m/CAM3SWZQKM=Pzc=CAHzRixKjp2eO5Q0Jg1SoFQqeXFQ647JiwqQ@mail.gmail.com
Discussion: http://postgr.es/m/CAH2-Wz=AxWqDoVvGU7dq856S4r6sJAj6DBn7VMtigkB33N5eyg@mail.gmail.com
When an UPDATE causes a row to no longer match the partition
constraint, try to move it to a different partition where it does
match the partition constraint. In essence, the UPDATE is split into
a DELETE from the old partition and an INSERT into the new one. This
can lead to surprising behavior in concurrency scenarios because
EvalPlanQual rechecks won't work as they normally did; the known
problems are documented. (There is a pending patch to improve the
situation further, but it needs more review.)
Amit Khandekar, reviewed and tested by Amit Langote, David Rowley,
Rajkumar Raghuwanshi, Dilip Kumar, Amul Sul, Thomas Munro, Álvaro
Herrera, Amit Kapila, and me. A few final revisions by me.
Discussion: http://postgr.es/m/CAJ3gD9do9o2ccQ7j7+tSgiE1REY65XRiMb=yJO3u3QhyP8EEPQ@mail.gmail.com
Previously aggregate transition and combination functions were invoked
by special case code in nodeAgg.c, evaluating input and filters
separately using the expression evaluation machinery. That turns out
to not be great for performance for several reasons:
- repeated expression evaluations have some cost
- the transition functions invocations are poorly predicted, as
commonly there are multiple aggregates in a query, resulting in the
same call-stack invoking different functions.
- filter and input computation had to be done separately
- the special case code made it hard to implement JITing of the whole
transition function invocation
Address this by building one large expression that computes input,
evaluates filters, and invokes transition functions.
This leads to moderate speedups in queries bottlenecked by aggregate
computations, and enables large speedups for similar cases once JITing
is done.
There's potential for further improvement:
- It'd be nice if we could simplify the somewhat expensive
aggstate->all_pergroups lookups.
- right now there's still an advance_transition_function invocation in
nodeAgg.c, leading to some code duplication.
Author: Andres Freund
Discussion: https://postgr.es/m/20170901064131.tazjxwus3k2w3ybh@alap3.anarazel.de
Introduce parallel-aware hash joins that appear in EXPLAIN plans as Parallel
Hash Join with Parallel Hash. While hash joins could already appear in
parallel queries, they were previously always parallel-oblivious and had a
partial subplan only on the outer side, meaning that the work of the inner
subplan was duplicated in every worker.
After this commit, the planner will consider using a partial subplan on the
inner side too, using the Parallel Hash node to divide the work over the
available CPU cores and combine its results in shared memory. If the join
needs to be split into multiple batches in order to respect work_mem, then
workers process different batches as much as possible and then work together
on the remaining batches.
The advantages of a parallel-aware hash join over a parallel-oblivious hash
join used in a parallel query are that it:
* avoids wasting memory on duplicated hash tables
* avoids wasting disk space on duplicated batch files
* divides the work of building the hash table over the CPUs
One disadvantage is that there is some communication between the participating
CPUs which might outweigh the benefits of parallelism in the case of small
hash tables. This is avoided by the planner's existing reluctance to supply
partial plans for small scans, but it may be necessary to estimate
synchronization costs in future if that situation changes. Another is that
outer batch 0 must be written to disk if multiple batches are required.
A potential future advantage of parallel-aware hash joins is that right and
full outer joins could be supported, since there is a single set of matched
bits for each hashtable, but that is not yet implemented.
A new GUC enable_parallel_hash is defined to control the feature, defaulting
to on.
Author: Thomas Munro
Reviewed-By: Andres Freund, Robert Haas
Tested-By: Rafia Sabih, Prabhat Sahu
Discussion:
https://postgr.es/m/CAEepm=2W=cOkiZxcg6qiFQP-dHUe09aqTrEMM7yJDrHMhDv_RA@mail.gmail.comhttps://postgr.es/m/CAEepm=37HKyJ4U6XOLi=JgfSHM3o6B-GaeO-6hkOmneTDkH+Uw@mail.gmail.com
SharedTuplestore allows multiple participants to write into it and
then read the tuples back from it in parallel. Each reader receives
partial results.
For now it always uses disk files, but other buffering policies and
other kinds of scans (ie each reader receives complete results) may be
useful in future.
The upcoming parallel hash join feature will use this facility.
Author: Thomas Munro
Reviewed-By: Peter Geoghegan, Andres Freund, Robert Haas
Discussion: https://postgr.es/m/CAEepm=2W=cOkiZxcg6qiFQP-dHUe09aqTrEMM7yJDrHMhDv_RA@mail.gmail.com
Previously, executor nodes running in parallel worker processes didn't
have access to the dsm_segment object used for parallel execution. In
order to support resource management based on DSM segment lifetime,
they need that. So create a ParallelWorkerContext object to hold it
and pass it to all InitializeWorker functions.
Author: Thomas Munro
Reviewed-By: Andres Freund
Discussion: https://postgr.es/m/CAEepm=2W=cOkiZxcg6qiFQP-dHUe09aqTrEMM7yJDrHMhDv_RA@mail.gmail.com
Hash partitioning is useful when you want to partition a growing data
set evenly. This can be useful to keep table sizes reasonable, which
makes maintenance operations such as VACUUM faster, or to enable
partition-wise join.
At present, we still depend on constraint exclusion for partitioning
pruning, and the shape of the partition constraints for hash
partitioning is such that that doesn't work. Work is underway to fix
that, which should both improve performance and make partitioning
pruning work with hash partitioning.
Amul Sul, reviewed and tested by Dilip Kumar, Ashutosh Bapat, Yugo
Nagata, Rajkumar Raghuwanshi, Jesper Pedersen, and by me. A few
final tweaks also by me.
Discussion: http://postgr.es/m/CAAJ_b96fhpJAP=ALbETmeLk1Uni_GFZD938zgenhF49qgDTjaQ@mail.gmail.com
The following are the individual improvements:
1) Avoidance of FunctionCallInfo based function calls, replaced by
more efficient functions with a native C argument interface.
2) Don't extract columns from a cache entry's tuple whenever matching
entries - instead store them as a Datum array. This also allows to
get rid of having to build dummy tuples for negative & list
entries, and of a hack for dealing with cstring vs. text weirdness.
3) Reorder members of catcache.h struct, so imortant entries are more
likely to be on one cacheline.
4) Allowing the compiler to specialize critical SearchCatCache for a
specific number of attributes allows to unroll loops and avoid
other nkeys dependant initialization.
5) Only initializing the ScanKey when necessary, i.e. catcache misses,
greatly reduces cache unnecessary cpu cache misses.
6) Split of the cache-miss case from the hash lookup, reducing stack
allocations etc in the common case.
7) CatCTup and their corresponding heaptuple are allocated in one
piece.
This results in making cache lookups themselves roughly three times as
fast - full-system benchmarks obviously improve less than that.
I've also evaluated further techniques:
- replace open coded hash with simplehash - the list walk right now
shows up in profiles. Unfortunately it's not easy to do so safely as
an entry's memory location can change at various times, which
doesn't work well with the refcounting and cache invalidation.
- Cacheline-aligning CatCTup entries - helps some with performance,
but the win isn't big and the code for it is ugly, because the
tuples have to be freed as well.
- add more proper functions, rather than macros for
SearchSysCacheCopyN etc., but right now they don't show up in
profiles.
The reason the macro wrapper for syscache.c/h have to be changed,
rather than just catcache, is that doing otherwise would require
exposing the SysCache array to the outside. That might be a good idea
anyway, but it's for another day.
Author: Andres Freund
Reviewed-By: Robert Haas
Discussion: https://postgr.es/m/20170914061207.zxotvyopetm7lrrp@alap3.anarazel.de
Tuples can have type RECORDOID and a typmod number that identifies a blessed
TupleDesc in a backend-private cache. To support the sharing of such tuples
through shared memory and temporary files, provide a typmod registry in
shared memory.
To achieve that, introduce per-session DSM segments, created on demand when a
backend first runs a parallel query. The per-session DSM segment has a
table-of-contents just like the per-query DSM segment, and initially the
contents are a shared record typmod registry and a DSA area to provide the
space it needs to grow.
State relating to the current session is accessed via a Session object
reached through global variable CurrentSession that may require significant
redesign further down the road as we figure out what else needs to be shared
or remodelled.
Author: Thomas Munro
Reviewed-By: Andres Freund
Discussion: https://postgr.es/m/CAEepm=0ZtQ-SpsgCyzzYpsXS6e=kZWqk3g5Ygn3MDV7A8dabUA@mail.gmail.com
Add general purpose chaining hash tables for DSA memory. Unlike
DynaHash in shared memory mode, these hash tables can grow as
required, and cope with being mapped into different addresses in
different backends.
There is a wide range of potential users for such a hash table, though
it's very likely the interface will need to evolve as we come to
understand the needs of different kinds of users. E.g support for
iterators and incremental resizing is planned for later commits and
the details of the callback signatures are likely to change.
Author: Thomas Munro
Reviewed-By: John Gorman, Andres Freund, Dilip Kumar, Robert Haas
Discussion:
https://postgr.es/m/CAEepm=3d8o8XdVwYT6O=bHKsKAM2pu2D6sV1S_=4d+jStVCE7w@mail.gmail.comhttps://postgr.es/m/CAEepm=0ZtQ-SpsgCyzzYpsXS6e=kZWqk3g5Ygn3MDV7A8dabUA@mail.gmail.com
Periodically while the server is running, and at shutdown, write out a
list of blocks in shared buffers. When the server reaches consistency
-- unfortunatey, we can't do it before that point without breaking
things -- reload those blocks into any still-unused shared buffers.
Mithun Cy and Robert Haas, reviewed and tested by Beena Emerson,
Amit Kapila, Jim Nasby, and Rafia Sabih.
Discussion: http://postgr.es/m/CAD__OugubOs1Vy7kgF6xTjmEqTR4CrGAv8w+ZbaY_+MZeitukw@mail.gmail.com
Previously, UNBOUNDED meant no lower bound when used in the FROM list,
and no upper bound when used in the TO list, which was OK for
single-column range partitioning, but problematic with multiple
columns. For example, an upper bound of (10.0, UNBOUNDED) would not be
collocated with a lower bound of (10.0, UNBOUNDED), thus making it
difficult or impossible to define contiguous multi-column range
partitions in some cases.
Fix this by using MINVALUE and MAXVALUE instead of UNBOUNDED to
represent a partition column that is unbounded below or above
respectively. This syntax removes any ambiguity, and ensures that if
one partition's lower bound equals another partition's upper bound,
then the partitions are contiguous.
Also drop the constraint prohibiting finite values after an unbounded
column, and just document the fact that any values after MINVALUE or
MAXVALUE are ignored. Previously it was necessary to repeat UNBOUNDED
multiple times, which was needlessly verbose.
Note: Forces a post-PG 10 beta2 initdb.
Report by Amul Sul, original patch by Amit Langote with some
additional hacking by me.
Discussion: https://postgr.es/m/CAAJ_b947mowpLdxL3jo3YLKngRjrq9+Ej4ymduQTfYR+8=YAYQ@mail.gmail.com
This replaces the old, recursive tree-walk based evaluation, with
non-recursive, opcode dispatch based, expression evaluation.
Projection is now implemented as part of expression evaluation.
This both leads to significant performance improvements, and makes
future just-in-time compilation of expressions easier.
The speed gains primarily come from:
- non-recursive implementation reduces stack usage / overhead
- simple sub-expressions are implemented with a single jump, without
function calls
- sharing some state between different sub-expressions
- reduced amount of indirect/hard to predict memory accesses by laying
out operation metadata sequentially; including the avoidance of
nearly all of the previously used linked lists
- more code has been moved to expression initialization, avoiding
constant re-checks at evaluation time
Future just-in-time compilation (JIT) has become easier, as
demonstrated by released patches intended to be merged in a later
release, for primarily two reasons: Firstly, due to a stricter split
between expression initialization and evaluation, less code has to be
handled by the JIT. Secondly, due to the non-recursive nature of the
generated "instructions", less performance-critical code-paths can
easily be shared between interpreted and compiled evaluation.
The new framework allows for significant future optimizations. E.g.:
- basic infrastructure for to later reduce the per executor-startup
overhead of expression evaluation, by caching state in prepared
statements. That'd be helpful in OLTPish scenarios where
initialization overhead is measurable.
- optimizing the generated "code". A number of proposals for potential
work has already been made.
- optimizing the interpreter. Similarly a number of proposals have
been made here too.
The move of logic into the expression initialization step leads to some
backward-incompatible changes:
- Function permission checks are now done during expression
initialization, whereas previously they were done during
execution. In edge cases this can lead to errors being raised that
previously wouldn't have been, e.g. a NULL array being coerced to a
different array type previously didn't perform checks.
- The set of domain constraints to be checked, is now evaluated once
during expression initialization, previously it was re-built
every time a domain check was evaluated. For normal queries this
doesn't change much, but e.g. for plpgsql functions, which caches
ExprStates, the old set could stick around longer. The behavior
around might still change.
Author: Andres Freund, with significant changes by Tom Lane,
changes by Heikki Linnakangas
Reviewed-By: Tom Lane, Heikki Linnakangas
Discussion: https://postgr.es/m/20161206034955.bh33paeralxbtluv@alap3.anarazel.de
This is the beginning of a collection of SQL-callable functions to
verify the integrity of data files. For now it only contains code to
verify B-Tree indexes.
This adds two SQL-callable functions, validating B-Tree consistency to
a varying degree. Check the, extensive, docs for details.
The goal is to later extend the coverage of the module to further
access methods, possibly including the heap. Once checks for
additional access methods exist, we'll likely add some "dispatch"
functions that cover multiple access methods.
Author: Peter Geoghegan, editorialized by Andres Freund
Reviewed-By: Andres Freund, Tomas Vondra, Thomas Munro,
Anastasia Lubennikova, Robert Haas, Amit Langote
Discussion: CAM3SWZQzLMhMwmBqjzK+pRKXrNUZ4w90wYMUWfkeV8mZ3Debvw@mail.gmail.com
Like Gather, we spawn multiple workers and run the same plan in each
one; however, Gather Merge is used when each worker produces the same
output ordering and we want to preserve that output ordering while
merging together the streams of tuples from various workers. (In a
way, Gather Merge is like a hybrid of Gather and MergeAppend.)
This works out to a win if it saves us from having to perform an
expensive Sort. In cases where only a small amount of data would need
to be sorted, it may actually be faster to use a regular Gather node
and then sort the results afterward, because Gather Merge sometimes
needs to wait synchronously for tuples whereas a pure Gather generally
doesn't. But if this avoids an expensive sort then it's a win.
Rushabh Lathia, reviewed and tested by Amit Kapila, Thomas Munro,
and Neha Sharma, and reviewed and revised by me.
Discussion: http://postgr.es/m/CAGPqQf09oPX-cQRpBKS0Gq49Z+m6KBxgxd_p9gX8CKk_d75HoQ@mail.gmail.com
This makes the handling of operators similar to that of functions and
aggregates.
Rename node FuncWithArgs to ObjectWithArgs, to reflect the expanded use.
Reviewed-by: Jim Nasby <Jim.Nasby@BlueTreble.com>
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
The new slab allocator needs different per-allocation information than
the classical aset.c. The definition in 58b25e981 wasn't sufficiently
careful on 32 platforms with 8 byte alignment, leading to buildfarm
failures. That's not entirely easy to fix by just adjusting the
definition.
As slab.c doesn't actually need the size part(s) of the common header,
all chunks are equally sized after all, it seems better to instead
reduce the header to the part needed by all allocators, namely which
context an allocation belongs to. That has the advantage of reducing
the overhead of slab allocations, and also allows for more flexibility
in future allocators.
To avoid spreading the logic about accessing a chunk's context around,
centralize it in GetMemoryChunkContext(), which allows to delete a
good number of lines.
A followup commit will revise the mmgr/README portion about
StandardChunkHeader, and more.
Author: Andres Freund
Discussion: https://postgr.es/m/20170228074420.aazv4iw6k562mnxg@alap3.anarazel.de
The default general purpose aset.c style memory context is not a great
choice for allocations that are all going to be evenly sized,
especially when those objects aren't small, and have varying
lifetimes. There tends to be a lot of fragmentation, larger
allocations always directly go to libc rather than have their cost
amortized over several pallocs.
These problems lead to the introduction of ad-hoc slab allocators in
reorderbuffer.c. But it turns out that the simplistic implementation
leads to problems when a lot of objects are allocated and freed, as
aset.c is still the underlying implementation. Especially freeing can
easily run into O(n^2) behavior in aset.c.
While the O(n^2) behavior in aset.c can, and probably will, be
addressed, custom allocators for this behavior are more efficient
both in space and time.
This allocator is for evenly sized allocations, and supports both
cheap allocations and freeing, without fragmenting significantly. It
does so by allocating evenly sized blocks via malloc(), and carves
them into chunks that can be used for allocations. In order to
release blocks to the OS as early as possible, chunks are allocated
from the fullest block that still has free objects, increasing the
likelihood of a block being entirely unused.
A subsequent commit uses this in reorderbuffer.c, but a further
allocator is needed to resolve the performance problems triggering
this work.
There likely are further potentialy uses of this allocator besides
reorderbuffer.c.
There's potential further optimizations of the new slab.c, in
particular the array of freelists could be replaced by a more
intelligent structure - but for now this looks more than good enough.
Author: Tomas Vondra, editorialized by Andres Freund
Reviewed-By: Andres Freund, Petr Jelinek, Robert Haas, Jim Nasby
Discussion: https://postgr.es/m/d15dff83-0b37-28ed-0809-95a5cc7292ad@2ndquadrant.com
This isn't exposed to the optimizer or the executor yet; we'll add
support for those things in a separate patch. But this puts the
basic mechanism in place: several processes can attach to a parallel
btree index scan, and each one will get a subset of the tuples that
would have been produced by a non-parallel scan. Each index page
becomes the responsibility of a single worker, which then returns
all of the TIDs on that page.
Rahila Syed, Amit Kapila, Robert Haas, reviewed and tested by
Anastasia Lubennikova, Tushar Ahuja, and Haribabu Kommi.
This patch doesn't actually make any index AM parallel-aware, but it
provides the necessary functions at the AM layer to do so.
Rahila Syed, Amit Kapila, Robert Haas
Table partitioning is like table inheritance and reuses much of the
existing infrastructure, but there are some important differences.
The parent is called a partitioned table and is always empty; it may
not have indexes or non-inherited constraints, since those make no
sense for a relation with no data of its own. The children are called
partitions and contain all of the actual data. Each partition has an
implicit partitioning constraint. Multiple inheritance is not
allowed, and partitioning and inheritance can't be mixed. Partitions
can't have extra columns and may not allow nulls unless the parent
does. Tuples inserted into the parent are automatically routed to the
correct partition, so tuple-routing ON INSERT triggers are not needed.
Tuple routing isn't yet supported for partitions which are foreign
tables, and it doesn't handle updates that cross partition boundaries.
Currently, tables can be range-partitioned or list-partitioned. List
partitioning is limited to a single column, but range partitioning can
involve multiple columns. A partitioning "column" can be an
expression.
Because table partitioning is less general than table inheritance, it
is hoped that it will be easier to reason about properties of
partitions, and therefore that this will serve as a better foundation
for a variety of possible optimizations, including query planner
optimizations. The tuple routing based which this patch does based on
the implicit partitioning constraints is an example of this, but it
seems likely that many other useful optimizations are also possible.
Amit Langote, reviewed and tested by Robert Haas, Ashutosh Bapat,
Amit Kapila, Rajkumar Raghuwanshi, Corey Huinker, Jaime Casanova,
Rushabh Lathia, Erik Rijkers, among others. Minor revisions by me.
Programmers discovered decades ago that it was useful to have a simple
interface for allocating and freeing memory, which is why malloc() and
free() were invented. Unfortunately, those handy tools don't work
with dynamic shared memory segments because those are specific to
PostgreSQL and are not necessarily mapped at the same address in every
cooperating process. So invent our own allocator instead. This makes
it possible for processes cooperating as part of parallel query
execution to allocate and free chunks of memory without having to
reserve them prior to the start of execution. It could also be used
for longer lived objects; for example, we could consider storing data
for pg_stat_statements or the stats collector in shared memory using
these interfaces, rather than writing them to files. Basically,
anything that needs shared memory but can't predict in advance how
much it's going to need might find this useful.
Thomas Munro and Robert Haas. The original code (of mine) on which
Thomas based his work was actually designed to be a new backend-local
memory allocator for PostgreSQL, but that hasn't gone anywhere - or
not yet, anyway. Thomas took that work and performed major
refactoring and extensive modifications to make it work with dynamic
shared memory, including the addition of appropriate locking.
Discussion: CA+TgmobkeWptGwiNa+SGFWsTLzTzD-CeLz0KcE-y6LFgoUus4A@mail.gmail.com
Discussion: CAEepm=1z5WLuNoJ80PaCvz6EtG9dN0j-KuHcHtU6QEfcPP5-qA@mail.gmail.com
This is intended as infrastructure for a full-fledged allocator for
dynamic shared memory. The interface looks a bit like a real
allocator, but only supports allocating and freeing memory in
multiples of the 4kB page size. Further, to free memory, you must
know the size of the span you wish to free, in pages. While these are
make it unsuitable as an allocator in and of itself, it still serves
as very useful scaffolding for a full-fledged allocator.
Robert Haas and Thomas Munro. This code is mostly the same as my 2014
submission, but Thomas fixed quite a few bugs and made some changes to
the interface.
Discussion: CA+TgmobkeWptGwiNa+SGFWsTLzTzD-CeLz0KcE-y6LFgoUus4A@mail.gmail.com
Discussion: CAEepm=1z5WLuNoJ80PaCvz6EtG9dN0j-KuHcHtU6QEfcPP5-qA@mail.gmail.com
The more efficient hashtable speeds up hash-aggregations with more than
a few hundred groups significantly. Improvements of over 120% have been
measured.
Due to the the different hash table queries that not fully
determined (e.g. GROUP BY without ORDER BY) may change their result
order.
The conversion is largely straight-forward, except that, due to the
static element types of simplehash.h type hashes, the additional data
some users store in elements (e.g. the per-group working data for hash
aggregaters) is now stored in TupleHashEntryData->additional. The
meaning of BuildTupleHashTable's entrysize (renamed to additionalsize)
has been changed to only be about the additionally stored size. That
size is only used for the initial sizing of the hash-table.
Reviewed-By: Tomas Vondra
Discussion: <20160727004333.r3e2k2y6fvk2ntup@alap3.anarazel.de>
dynahash.c hash tables aren't quite fast enough for some
use-cases. There are several reasons for lacking performance:
- the use of chaining for collision handling makes them cache
inefficient, that's especially an issue when the tables get bigger.
- as the element sizes for dynahash are only determined at runtime,
offset computations are somewhat expensive
- hash and element comparisons are indirect function calls, causing
unnecessary pipeline stalls
- it's two level structure has some benefits (somewhat natural
partitioning), but increases the number of indirections
to fix several of these the hash tables have to be adjusted to the
individual use-case at compile-time. C unfortunately doesn't provide a
good way to do compile code generation (like e.g. c++'s templates for
all their weaknesses do). Thus the somewhat ugly approach taken here is
to allow for code generation using a macro-templatized header file,
which generates functions and types based on a prefix and other
parameters.
Later patches use this infrastructure to use such hash tables for
tidbitmap.c (bitmap scans) and execGrouping.c (hash aggregation,
...). In queries where these use up a large fraction of the time, this
has been measured to lead to performance improvements of over 100%.
There are other cases where this could be useful (e.g. catcache.c).
The hash table design chosen is a variant of linear open-addressing. The
biggest disadvantage of simple linear addressing schemes are highly
variable lookup times due to clustering, and deletions leaving a lot of
tombstones around. To address these issues a variant of "robin hood"
hashing is employed. Robin hood hashing optimizes chaining lengths by
moving elements close to their optimal bucket ("rich" elements), out of
the way if a to-be-inserted element is further away from its optimal
position (i.e. it's "poor"). While that can make insertions slower, the
average lookup performance is a lot better, and higher fill factors can
be used in a still performant manner. To avoid tombstones - which
normally solve the issue that a deleted node's presence is relevant to
determine whether a lookup needs to continue looking or is done -
buckets following a deleted element are shifted backwards, unless
they're empty or already at their optimal position.
There's further possible improvements that can be made to this
implementation. Amongst others:
- Use distance as a termination criteria during searches. This is
generally a good idea, but I've been able to see the overhead of
distance calculations in some cases.
- Consider combining the 'empty' status into the hashvalue, and enforce
storing the hashvalue. That could, in some cases, increase memory
density and remove a few instructions.
- Experiment further with the, very conservatively choosen, fillfactor.
- Make maximum size of hashtable configurable, to allow storing very
very large tables. That'd require 64bit hash values to be more common
than now, though.
- some smaller memcpy calls could be optimized to copy larger chunks
But since the new implementation is already considerably faster than
dynahash it seem sensible to start using it.
Reviewed-By: Tomas Vondra
Discussion: <20160727004333.r3e2k2y6fvk2ntup@alap3.anarazel.de>
Prior to commit 7882c3b0b9, it was
possible to use LWLocks within DSM segments, but that commit broke
this use case by switching from a doubly linked list to a circular
linked list. Switch back, using a new bit of general infrastructure
for maintaining lists of PGPROCs.
Thomas Munro, reviewed by me.
The new pg_check_visible() and pg_check_frozen() functions can be used to
verify that the visibility map bits for a relation's data pages match the
actual state of the tuples on those pages.
Amit Kapila and Robert Haas, reviewed (in earlier versions) by Andres
Freund. Additional testing help by Thomas Munro.
This time, use the buildfarm-supplied contents for this file, instead
of trying to update it by eyeballing the pgindent output.
Per discussion with Tom and Bruce.
In addition to adding new typedefs, I also re-sorted the file so that
various entries add piecemeal, mostly or entirely by me, were alphabetized
the same way as other entries in the file.
It was incorrectly declared as a volatile pointer to a non-volatile
structure. Eliminate the OldSnapshotControl struct definition; it
is really not needed. Pointed out by Tom Lane.
While at it, add OldSnapshotControlData to pgindent's list of
structures.
Pinning/Unpinning a buffer is a very frequent operation; especially in
read-mostly cache resident workloads. Benchmarking shows that in various
scenarios the spinlock protecting a buffer header's state becomes a
significant bottleneck. The problem can be reproduced with pgbench -S on
larger machines, but can be considerably worse for queries which touch
the same buffers over and over at a high frequency (e.g. nested loops
over a small inner table).
To allow atomic operations to be used, cram BufferDesc's flags,
usage_count, buf_hdr_lock, refcount into a single 32bit atomic variable;
that allows to manipulate them together using 32bit compare-and-swap
operations. This requires reducing MAX_BACKENDS to 2^18-1 (which could
be lifted by using a 64bit field, but it's not a realistic configuration
atm).
As not all operations can easily implemented in a lockfree manner,
implement the previous buf_hdr_lock via a flag bit in the atomic
variable. That way we can continue to lock the header in places where
it's needed, but can get away without acquiring it in the more frequent
hot-paths. There's some additional operations which can be done without
the lock, but aren't in this patch; but the most important places are
covered.
As bufmgr.c now essentially re-implements spinlocks, abstract the delay
logic from s_lock.c into something more generic. It now has already two
users, and more are coming up; there's a follupw patch for lwlock.c at
least.
This patch is based on a proof-of-concept written by me, which Alexander
Korotkov made into a fully working patch; the committed version is again
revised by me. Benchmarking and testing has, amongst others, been
provided by Dilip Kumar, Alexander Korotkov, Robert Haas.
On a large x86 system improvements for readonly pgbench, with a high
client count, of a factor of 8 have been observed.
Author: Alexander Korotkov and Andres Freund
Discussion: 2400449.GjM57CE0Yg@dinodell
This introduces a new dependency type which marks an object as depending
on an extension, such that if the extension is dropped, the object
automatically goes away; and also, if the database is dumped, the object
is included in the dump output. Currently the grammar supports this for
indexes, triggers, materialized views and functions only, although the
utility code is generic so adding support for more object types is a
matter of touching the parser rules only.
Author: Abhijit Menon-Sen
Reviewed-by: Alexander Korotkov, Álvaro Herrera
Discussion: http://www.postgresql.org/message-id/20160115062649.GA5068@toroid.org
Commit ac1d794 ("Make idle backends exit if the postmaster dies.")
introduced a regression on, at least, large linux systems. Constantly
adding the same postmaster_alive_fds to the OSs internal datastructures
for implementing poll/select can cause significant contention; leading
to a performance regression of nearly 3x in one example.
This can be avoided by using e.g. linux' epoll, which avoids having to
add/remove file descriptors to the wait datastructures at a high rate.
Unfortunately the current latch interface makes it hard to allocate any
persistent per-backend resources.
Replace, with a backward compatibility layer, WaitLatchOrSocket with a
new WaitEventSet API. Users can allocate such a Set across multiple
calls, and add more than one file-descriptor to wait on. The latter has
been added because there's upcoming postgres features where that will be
helpful.
In addition to the previously existing poll(2), select(2),
WaitForMultipleObjects() implementations also provide an epoll_wait(2)
based implementation to address the aforementioned performance
problem. Epoll is only available on linux, but that is the most likely
OS for machines large enough (four sockets) to reproduce the problem.
To actually address the aforementioned regression, create and use a
long-lived WaitEventSet for FE/BE communication. There are additional
places that would benefit from a long-lived set, but that's a task for
another day.
Thanks to Amit Kapila, who helped make the windows code I blindly wrote
actually work.
Reported-By: Dmitry Vasilyev Discussion:
CAB-SwXZh44_2ybvS5Z67p_CDz=XFn4hNAD=CnMEF+QqkXwFrGg@mail.gmail.com20160114143931.GG10941@awork2.anarazel.de
Up to now checkpoints were written in the order they're in the
BufferDescriptors. That's nearly random in a lot of cases, which
performs badly on rotating media, but even on SSDs it causes slowdowns.
To avoid that, sort checkpoints before writing them out. We currently
sort by tablespace, relfilenode, fork and block number.
One of the major reasons that previously wasn't done, was fear of
imbalance between tablespaces. To address that balance writes between
tablespaces.
The other prime concern was that the relatively large allocation to sort
the buffers in might fail, preventing checkpoints from happening. Thus
pre-allocate the required memory in shared memory, at server startup.
This particularly makes it more efficient to have checkpoint flushing
enabled, because that'll often result in a lot of writes that can be
coalesced into one flush.
Discussion: alpine.DEB.2.10.1506011320000.28433@sto
Author: Fabien Coelho and Andres Freund
Currently writes to the main data files of postgres all go through the
OS page cache. This means that some operating systems can end up
collecting a large number of dirty buffers in their respective page
caches. When these dirty buffers are flushed to storage rapidly, be it
because of fsync(), timeouts, or dirty ratios, latency for other reads
and writes can increase massively. This is the primary reason for
regular massive stalls observed in real world scenarios and artificial
benchmarks; on rotating disks stalls on the order of hundreds of seconds
have been observed.
On linux it is possible to control this by reducing the global dirty
limits significantly, reducing the above problem. But global
configuration is rather problematic because it'll affect other
applications; also PostgreSQL itself doesn't always generally want this
behavior, e.g. for temporary files it's undesirable.
Several operating systems allow some control over the kernel page
cache. Linux has sync_file_range(2), several posix systems have msync(2)
and posix_fadvise(2). sync_file_range(2) is preferable because it
requires no special setup, whereas msync() requires the to-be-flushed
range to be mmap'ed. For the purpose of flushing dirty data
posix_fadvise(2) is the worst alternative, as flushing dirty data is
just a side-effect of POSIX_FADV_DONTNEED, which also removes the pages
from the page cache. Thus the feature is enabled by default only on
linux, but can be enabled on all systems that have any of the above
APIs.
While desirable and likely possible this patch does not contain an
implementation for windows.
With the infrastructure added, writes made via checkpointer, bgwriter
and normal user backends can be flushed after a configurable number of
writes. Each of these sources of writes controlled by a separate GUC,
checkpointer_flush_after, bgwriter_flush_after and backend_flush_after
respectively; they're separate because the number of flushes that are
good are separate, and because the performance considerations of
controlled flushing for each of these are different.
A later patch will add checkpoint sorting - after that flushes from the
ckeckpoint will almost always be desirable. Bgwriter flushes are most of
the time going to be random, which are slow on lots of storage hardware.
Flushing in backends works well if the storage and bgwriter can keep up,
but if not it can have negative consequences. This patch is likely to
have negative performance consequences without checkpoint sorting, but
unfortunately so has sorting without flush control.
Discussion: alpine.DEB.2.10.1506011320000.28433@sto
Author: Fabien Coelho and Andres Freund
The previous RequestAddinLWLocks() method had several disadvantages.
First, the locks would be in the main tranche; we've recently decided
that it's useful for LWLocks used for separate purposes to have
separate tranche IDs. Second, there wasn't any correlation between
what code called RequestAddinLWLocks() and what code called
LWLockAssign(); when multiple modules are in use, it could become
quite difficult to troubleshoot problems where LWLockAssign() ran out
of locks. To fix, create a concept of named LWLock tranches which
can be used either by extension or by core code.
Amit Kapila and Robert Haas
Move the content lock directly into the BufferDesc, so that locking and
pinning a buffer touches only one cache line rather than two. Adjust
the definition of BufferDesc slightly so that this doesn't make the
BufferDesc any larger than one cache line (at least on platforms where
a spinlock is only 1 or 2 bytes).
We can't fit the I/O locks into the BufferDesc and stay within one
cache line, so move those to a completely separate tranche. This
leaves a relatively limited number of LWLocks in the main tranche, so
increase the padding of those remaining locks to a full cache line,
rather than allowing adjacent locks to share a cache line, hopefully
reducing false sharing.
Performance testing shows that these changes make little difference
on laptop-class machines, but help significantly on larger servers,
especially those with more than 2 sockets.
Andres Freund, originally based on an earlier patch by Simon Riggs.
Review and cosmetic adjustments (including heavy rewriting of the
comments) by me.
Commit 4a4e6893aa, which introduced this
mechanism, failed to account for the fact that the RECORD pseudo-type
uses transient typmods that are only meaningful within a single
backend. Transferring such tuples without modification between two
cooperating backends does not work. This commit installs a system
for passing the tuple descriptors over the same shm_mq being used to
send the tuples themselves. The two sides might not assign the same
transient typmod to any given tuple descriptor, so we must also
substitute the appropriate receiver-side typmod for the one used by
the sender. That adds some CPU overhead, but still seems better than
being unable to pass records between cooperating parallel processes.
Along the way, move the logic for handling multiple tuple queues from
tqueue.c to nodeGather.c; tqueue.c now provides a TupleQueueReader,
which reads from a single queue, rather than a TupleQueueFunnel, which
potentially reads from multiple queues. This change was suggested
previously as a way to make sure that nodeGather.c rather than tqueue.c
had policy control over the order in which to read from queues, but
it wasn't clear to me until now how good an idea it was. typmod
mapping needs to be performed separately for each queue, and it is
much simpler if the tqueue.c code handles that and leaves multiplexing
multiple queues to higher layers of the stack.
A Gather executor node runs any number of copies of a plan in an equal
number of workers and merges all of the results into a single tuple
stream. It can also run the plan itself, if the workers are
unavailable or haven't started up yet. It is intended to work with
the Partial Seq Scan node which will be added in future commits.
It could also be used to implement parallel query of a different sort
by itself, without help from Partial Seq Scan, if the single_copy mode
is used. In that mode, a worker executes the plan, and the parallel
leader does not, merely collecting the worker's results. So, a Gather
node could be inserted into a plan to split the execution of that plan
across two processes. Nested Gather nodes aren't currently supported,
but we might want to add support for that in the future.
There's nothing in the planner to actually generate Gather nodes yet,
so it's not quite time to break out the champagne. But we're getting
close.
Amit Kapila. Some designs suggestions were provided by me, and I also
reviewed the patch. Single-copy mode, documentation, and other minor
changes also by me.
The fact that multixact truncations are not WAL logged has caused a fair
share of problems. Amongst others it requires to do computations during
recovery while the database is not in a consistent state, delaying
truncations till checkpoints, and handling members being truncated, but
offset not.
We tried to put bandaids on lots of these issues over the last years,
but it seems time to change course. Thus this patch introduces WAL
logging for multixact truncations.
This allows:
1) to perform the truncation directly during VACUUM, instead of delaying it
to the checkpoint.
2) to avoid looking at the offsets SLRU for truncation during recovery,
we can just use the master's values.
3) simplify a fair amount of logic to keep in memory limits straight,
this has gotten much easier
During the course of fixing this a bunch of additional bugs had to be
fixed:
1) Data was not purged from memory the member's SLRU before deleting
segments. This happened to be hard or impossible to hit due to the
interlock between checkpoints and truncation.
2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but
that doesn't work for offsets that haven't yet been flushed to
disk. Add code to flush the SLRUs to fix. Not pretty, but it feels
slightly safer to only make decisions based on actual on-disk state.
3) find_multixact_start() could be called concurrently with a truncation
and thus fail. Via SetOffsetVacuumLimit() that could lead to a round
of emergency vacuuming. The problem remains in
pg_get_multixact_members(), but that's quite harmless.
For now this is going to only get applied to 9.5+, leaving the issues in
the older branches in place. It is quite possible that we need to
backpatch at a later point though.
For the case this gets backpatched we need to handle that an updated
standby may be replaying WAL from a not-yet upgraded primary. We have to
recognize that situation and use "old style" truncation (i.e. looking at
the SLRUs) during WAL replay. In contrast to before, this now happens in
the startup process, when replaying a checkpoint record, instead of the
checkpointer. Doing truncation in the restartpoint is incorrect, they
can happen much later than the original checkpoint, thereby leading to
wraparound. To avoid "multixact_redo: unknown op code 48" errors
standbys would have to be upgraded before primaries.
A later patch will bump the WAL page magic, and remove the legacy
truncation codepaths. Legacy truncation support is just included to make
a possible future backpatch easier.
Discussion: 20150621192409.GA4797@alap3.anarazel.de
Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro
Backpatch: 9.5 for now
The shm_mq mechanism was built to send error (and notice) messages and
tuples between backends. However, shm_mq itself only deals in raw
bytes. Since commit 2bd9e412f9, we have
had infrastructure for one message to redirect protocol messages to a
queue and for another backend to parse them and do useful things with
them. This commit introduces a somewhat analogous facility for tuples
by adding a new type of DestReceiver, DestTupleQueue, which writes
each tuple generated by a query into a shm_mq, and a new
TupleQueueFunnel facility which reads raw tuples out of the queue and
reconstructs the HeapTuple format expected by the executor.
The TupleQueueFunnel abstraction supports reading from multiple tuple
streams at the same time, but only in round-robin fashion. Someone
could imaginably want other policies, but this should be good enough
to meet our short-term needs related to parallel query, and we can
always extend it later.
This also makes one minor addition to the shm_mq API that didn'
seem worth breaking out as a separate patch.
Extracted from Amit Kapila's parallel sequential scan patch. This
code was originally written by me, and then it was revised by Amit,
and then it was revised some more by me.
The meta data of PGLZ symbolized by PGLZ_Header is removed, to make
the compression and decompression code independent on the backend-only
varlena facility. PGLZ_Header is being used to store some meta data
related to the data being compressed like the raw length of the uncompressed
record or some varlena-related data, making it unpluggable once PGLZ is
stored in src/common as it contains some backend-only code paths with
the management of varlena structures. The APIs of PGLZ are reworked
at the same time to do only compression and decompression of buffers
without the meta-data layer, simplifying its use for a more general usage.
On-disk format is preserved as well, so there is no incompatibility with
previous major versions of PostgreSQL for TOAST entries.
Exposing compression and decompression APIs of pglz makes possible its
use by extensions and contrib modules. Especially this commit is required
for upcoming WAL compression feature so that the WAL reader facility can
decompress the WAL data by using pglz_decompress.
Michael Paquier, reviewed by me.
As 'ALTER TABLESPACE .. MOVE ALL' really didn't change the tablespace
but instead changed objects inside tablespaces, it made sense to
rework the syntax and supporting functions to operate under the
'ALTER (TABLE|INDEX|MATERIALIZED VIEW)' syntax and to be in
tablecmds.c.
Pointed out by Alvaro, who also suggested the new syntax.
Back-patch to 9.4.
In order for this to work, walsenders need the optional ability to
connect to a database, so the "replication" keyword now allows true
or false, for backward-compatibility, and the new value "database"
(which causes the "dbname" parameter to be respected).
walsender needs to loop not only when idle but also when sending
decoded data to the user and when waiting for more xlog data to decode.
This means that there are now three separate loops inside walsender.c;
although some refactoring has been done here, this is still a bit ugly.
Andres Freund, with contributions from Álvaro Herrera, and further
review by me.
This feature, building on previous commits, allows the write-ahead log
stream to be decoded into a series of logical changes; that is,
inserts, updates, and deletes and the transactions which contain them.
It is capable of handling decoding even across changes to the schema
of the effected tables. The output format is controlled by a
so-called "output plugin"; an example is included. To make use of
this in a real replication system, the output plugin will need to be
modified to produce output in the format appropriate to that system,
and to perform filtering.
Currently, information can be extracted from the logical decoding
system only via SQL; future commits will add the ability to stream
changes via walsender.
Andres Freund, with review and other contributions from many other
people, including Álvaro Herrera, Abhijit Menon-Sen, Peter Gheogegan,
Kevin Grittner, Robert Haas, Heikki Linnakangas, Fujii Masao, Abhijit
Menon-Sen, Michael Paquier, Simon Riggs, Craig Ringer, and Steve
Singer.
Replication slots are a crash-safe data structure which can be created
on either a master or a standby to prevent premature removal of
write-ahead log segments needed by a standby, as well as (with
hot_standby_feedback=on) pruning of tuples whose removal would cause
replication conflicts. Slots have some advantages over existing
techniques, as explained in the documentation.
In a few places, we refer to the type of replication slots introduced
by this patch as "physical" slots, because forthcoming patches for
logical decoding will also have slots, but with somewhat different
properties.
Andres Freund and Robert Haas
This makes it possible to store lwlocks as part of some other data
structure in the main shared memory segment, or in a dynamic shared
memory segment. There is still a main LWLock array and this patch does
not move anything out of it, but it provides necessary infrastructure
for doing that in the future.
This change is likely to increase the size of LWLockPadded on some
platforms, especially 32-bit platforms where it was previously only
16 bytes.
Patch by me. Review by Andres Freund and KaiGai Kohei.
This adds a 'MOVE' sub-command to ALTER TABLESPACE which allows moving sets of
objects from one tablespace to another. This can be extremely handy and avoids
a lot of error-prone scripting. ALTER TABLESPACE ... MOVE will only move
objects the user owns, will notify the user if no objects were found, and can
be used to move ALL objects or specific types of objects (TABLES, INDEXES, or
MATERIALIZED VIEWS).
When wal_level=logical, we'll log columns from the old tuple as
configured by the REPLICA IDENTITY facility added in commit
07cacba983. This makes it possible
a properly-configured logical replication solution to correctly
follow table updates even if they change the chosen key columns,
or, with REPLICA IDENTITY FULL, even if the table has no key at
all. Note that updates which do not modify the replica identity
column won't log anything extra, making the choice of a good key
(i.e. one that will rarely be changed) important to performance
when wal_level=logical is configured.
Each insert, update, or delete to a catalog table will also log
the CMIN and/or CMAX values of stamped by the current transaction.
This is necessary because logical decoding will require access to
historical snapshots of the catalog in order to decode some data
types, and the CMIN/CMAX values that we may need in order to judge
row visibility may have been overwritten by the time we need them.
Andres Freund, reviewed in various versions by myself, Heikki
Linnakangas, KONDO Mitsumasa, and many others.
Commit 95ef6a3448 removed the
ability to create rules on an individual column as of 7.3, but
left some residual code which has since been useless. This cleans
up that dead code without any change in behavior other than
dropping the useless column from the catalog.
Until now, our Serializable mode has in fact been what's called Snapshot
Isolation, which allows some anomalies that could not occur in any
serialized ordering of the transactions. This patch fixes that using a
method called Serializable Snapshot Isolation, based on research papers by
Michael J. Cahill (see README-SSI for full references). In Serializable
Snapshot Isolation, transactions run like they do in Snapshot Isolation,
but a predicate lock manager observes the reads and writes performed and
aborts transactions if it detects that an anomaly might occur. This method
produces some false positives, ie. it sometimes aborts transactions even
though there is no anomaly.
To track reads we implement predicate locking, see storage/lmgr/predicate.c.
Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared
memory is finite, so when a transaction takes many tuple-level locks on a
page, the locks are promoted to a single page-level lock, and further to a
single relation level lock if necessary. To lock key values with no matching
tuple, a sequential scan always takes a relation-level lock, and an index
scan acquires a page-level lock that covers the search key, whether or not
there are any matching keys at the moment.
A predicate lock doesn't conflict with any regular locks or with another
predicate locks in the normal sense. They're only used by the predicate lock
manager to detect the danger of anomalies. Only serializable transactions
participate in predicate locking, so there should be no extra overhead for
for other transactions.
Predicate locks can't be released at commit, but must be remembered until
all the transactions that overlapped with it have completed. That means that
we need to remember an unbounded amount of predicate locks, so we apply a
lossy but conservative method of tracking locks for committed transactions.
If we run short of shared memory, we overflow to a new "pg_serial" SLRU
pool.
We don't currently allow Serializable transactions in Hot Standby mode.
That would be hard, because even read-only transactions can cause anomalies
that wouldn't otherwise occur.
Serializable isolation mode now means the new fully serializable level.
Repeatable Read gives you the old Snapshot Isolation level that we have
always had.
Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and
Anssi Kääriäinen
Thomas Munro
Alexander Korotkov
Andres Freund
Michael Paquier
Robert Haas
Andrew Dunstan
Tom Lane
Simon Riggs
Dean Rasheed
Bruce Momjian
Peter Eisentraut
Kevin Grittner
Alvaro Herrera
Fujii Masao
Stephen Frost
Heikki Linnakangas