WalSndWakeup() currently loops through all the walsenders slots, with a
spinlock acquisition and release for every iteration, to wake up waiting
walsenders.
This commonly was not a problem before e101dfac3a. But, to allow logical
decoding on standbys, we need to wake up logical walsenders after every WAL
record is applied on the standby, rather just when flushing WAL or switching
timelines. This causes a performance regression for workloads replaying a lot
of WAL records.
To solve this, we use condition variable (CV) to efficiently wake up
walsenders in WalSndWakeup().
Every walsender prepares to sleep on a shared memory CV. Note that it just
prepares to sleep on the CV (i.e., adds itself to the CV's waitlist), but does
not actually wait on the CV (IOW, it never calls ConditionVariableSleep()). It
still uses WaitEventSetWait() for waiting, because CV infrastructure doesn't
handle FeBe socket events currently. The processes (startup process,
walreceiver etc.) wanting to wake up walsenders use
ConditionVariableBroadcast(), which in turn calls SetLatch(), helping
walsenders come out of WaitEventSetWait().
We use separate shared memory CVs for physical and logical walsenders for
selective wake ups, see WalSndWakeup() for more details.
This approach is simple and reasonably efficient. But not very elegant. But
for 16 it seems to be a better path than a larger redesign of the CV
mechanism. A desirable future improvement would be to add support for CVs
into WaitEventSetWait().
This still leaves us with a small regression in very extreme workloads (due to
the spinlock acquisition in ConditionVariableBroadcast() when there are no
waiters) - but that seems acceptable.
Reported-by: Andres Freund <andres@anarazel.de>
Suggested-by: Andres Freund <andres@anarazel.de>
Author: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>
Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com>
Reviewed-by: Zhijie Hou <houzj.fnst@fujitsu.com>
Discussion: https://www.postgresql.org/message-id/20230509190247.3rrplhdgem6su6cg%40awork3.anarazel.de
Physical walsenders can't send data until it's been flushed; logical
walsenders can't decode and send data until it's been applied. On the
standby, the WAL is flushed first, which will only wake up physical
walsenders; and then applied, which will only wake up logical
walsenders.
Previously, all walsenders were awakened when the WAL was flushed. That
was fine for logical walsenders on the primary; but on the standby the
flushed WAL would have been not applied yet, so logical walsenders were
awakened too early.
Per idea from Jeff Davis and Amit Kapila.
Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com>
Reviewed-By: Jeff Davis <pgsql@j-davis.com>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com>
Discussion: https://postgr.es/m/CAA4eK1+zO5LUeisabX10c81LU-fWMKO4M9Wyg1cdkbW7Hqh6vQ@mail.gmail.com
Make sure that function declarations use names that exactly match the
corresponding names from function definitions for several "lexer
adjacent" backend functions.
These functions were missed by recent commits because they were obscured
by clang-tidy warnings about functions whose signature is directly under
the control of the lexer (flex seems to always generate function
declarations with unnamed parameters). We probably can't fix most of
the warnings it generates for translation units that get built from .l
and .y files, but we can at least do this much.
Author: Peter Geoghegan <pg@bowt.ie>
Discussion: https://postgr.es/m/CAH2-WznJt9CMM9KJTMjJh_zbL5hD9oX44qdJ4aqZtjFi-zA3Tg@mail.gmail.com
Up until now, we've had a policy of only marking certain variables
in the PostgreSQL header files with PGDLLIMPORT, but now we've
decided to mark them all. This means that extensions running on
Windows should no longer operate at a disadvantage as compared to
extensions running on Linux: if the variable is present in a header
file, it should be accessible.
Discussion: http://postgr.es/m/CA+TgmoYanc1_FSfimhgiWSqVyP5KKmh5NP2BWNwDhO8Pg2vGYQ@mail.gmail.com
In logical replication mode, a WalSender is supposed to be able
to execute any regular SQL command, as well as the special
replication commands. Poor design of the replication-command
parser caused it to fail in various cases, notably:
* semicolons embedded in a command, or multiple SQL commands
sent in a single message;
* dollar-quoted literals containing odd numbers of single
or double quote marks;
* commands starting with a comment.
The basic problem here is that we're trying to run repl_scanner.l
across the entire input string even when it's not a replication
command. Since repl_scanner.l does not understand all of the
token types known to the core lexer, this is doomed to have
failure modes.
We certainly don't want to make repl_scanner.l as big as scan.l,
so instead rejigger stuff so that we only lex the first token of
a non-replication command. That will usually look like an IDENT
to repl_scanner.l, though a comment would end up getting reported
as a '-' or '/' single-character token. If the token is a replication
command keyword, we push it back and proceed normally with repl_gram.y
parsing. Otherwise, we can drop out of exec_replication_command()
without examining the rest of the string.
(It's still theoretically possible for repl_scanner.l to fail on
the first token; but that could only happen if it's an unterminated
single- or double-quoted string, in which case you'd have gotten
largely the same error from the core lexer too.)
In this way, repl_gram.y isn't involved at all in handling general
SQL commands, so we can get rid of the SQLCmd node type. (In
the back branches, we can't remove it because renumbering enum
NodeTag would be an ABI break; so just leave it sit there unused.)
I failed to resist the temptation to clean up some other sloppy
coding in repl_scanner.l while at it. The only externally-visible
behavior change from that is it now accepts \r and \f as whitespace,
same as the core lexer.
Per bug #17379 from Greg Rychlewski. Back-patch to all supported
branches.
Discussion: https://postgr.es/m/17379-6a5c6cfb3f1f5e77@postgresql.org
The stats with this commit was available only for WALSenders, however,
users might want to see for backends doing logical decoding via SQL API.
Then, users might want to reset and access these stats across server
restart which was not possible with the current patch.
List of commits reverted:
caa3c4242c Don't call elog() while holding spinlock.
e641b2a995 Doc: Update the documentation for spilled transaction
statistics.
5883f5fe27 Fix unportable printf format introduced in commit 9290ad198.
9290ad198b Track statistics for spilling of changes from ReorderBuffer.
Additionaly, remove the release notes entry for this feature.
Backpatch-through: 13, where it was introduced
Discussion: https://postgr.es/m/CA+fd4k5_pPAYRTDrO2PbtTOe0eHQpBvuqmCr8ic39uTNmR49Eg@mail.gmail.com
We have repeatedly seen the buildfarm reach the Assert(false) in
SyncRepGetSyncStandbysPriority. This apparently is due to failing to
consider the possibility that the sync_standby_priority values in
shared memory might be inconsistent; but they will be whenever only
some of the walsenders have updated their values after a change in
the synchronous_standby_names setting. That function is vastly too
complex for what it does, anyway, so rewriting it seems better than
trying to apply a band-aid fix.
Furthermore, the API of SyncRepGetSyncStandbys is broken by design:
it returns a list of WalSnd array indexes, but there is nothing
guaranteeing that the contents of the WalSnd array remain stable.
Thus, if some walsender exits and then a new walsender process
takes over that WalSnd array slot, a caller might make use of
WAL position data that it should not, potentially leading to
incorrect decisions about whether to release transactions that
are waiting for synchronous commit.
To fix, replace SyncRepGetSyncStandbys with a new function
SyncRepGetCandidateStandbys that copies all the required data
from shared memory while holding the relevant mutexes. If the
associated walsender process then exits, this data is still safe to
make release decisions with, since we know that that much WAL *was*
sent to a valid standby server. This incidentally means that we no
longer need to treat sync_standby_priority as protected by the
SyncRepLock rather than the per-walsender mutex.
SyncRepGetSyncStandbys is no longer used by the core code, so remove
it entirely in HEAD. However, it seems possible that external code is
relying on that function, so do not remove it from the back branches.
Instead, just remove the known-incorrect Assert. When the bug occurs,
the function will return a too-short list, which callers should treat
as meaning there are not enough sync standbys, which seems like a
reasonably safe fallback until the inconsistent state is resolved.
Moreover it's bug-compatible with what has been happening in non-assert
builds. We cannot do anything about the walsender-replacement race
condition without an API/ABI break.
The bogus assertion exists back to 9.6, but 9.6 is sufficiently
different from the later branches that the patch doesn't apply at all.
I chose to just remove the bogus assertion in 9.6, feeling that the
probability of a bad outcome from the walsender-replacement race
condition is too low to justify rewriting the whole patch for 9.6.
Discussion: https://postgr.es/m/21519.1585272409@sss.pgh.pa.us
This adds the statistics about transactions spilled to disk from
ReorderBuffer. Users can query the pg_stat_replication view to check
these stats.
Author: Tomas Vondra, with bug-fixes and minor changes by Dilip Kumar
Reviewed-by: Amit Kapila
Discussion: https://postgr.es/m/688b0b7f-2f6c-d827-c27b-216a8e3ea700@2ndquadrant.com
The timestamp generated by the standby at message transmission has been
included in the protocol since its introduction for both the status
update message and hot standby feedback message, but it has never
appeared in pg_stat_replication. Seeing this timestamp does not matter
much with a cluster which has a lot of activity, but on a mostly-idle
cluster, this makes monitoring able to react faster than the configured
timeouts.
Author: MyungKyu LIM
Reviewed-by: Michael Paquier, Masahiko Sawada
Discussion: https://postgr.es/m/1657809367.407321.1533027417725.JavaMail.jboss@ep2ml404
In WAL receiver and WAL server, some accesses to their corresponding
shared memory control structs were done without holding any kind of
lock, which could lead to inconsistent and possibly insecure results.
In walsender, fix by clarifying the locking rules and following them
correctly, as documented in the new comment in walsender_private.h;
namely that some members can be read in walsender itself without a lock,
because the only writes occur in the same process. The rest of the
struct requires spinlock for accesses, as usual.
In walreceiver, fix by always holding spinlock while accessing the
struct.
While there is potentially a problem in all branches, it is minor in
stable ones. This only became a real problem in pg10 because of quorum
commit in synchronous replication (commit 3901fd70cc), and a potential
security problem in walreceiver because a superuser() check was removed
by default monitoring roles (commit 25fff40798). Thus, no backpatch.
In passing, clean up some leftover braces which were used to create
unconditional blocks. Once upon a time these were used for
volatile-izing accesses to those shmem structs, which is no longer
required. Many other occurrences of this pattern remain.
Author: Michaël Paquier
Reported-by: Michaël Paquier
Reviewed-by: Masahiko Sawada, Kyotaro Horiguchi, Thomas Munro,
Robert Haas
Discussion: https://postgr.es/m/CAB7nPqTWYqtzD=LN_oDaf9r-hAjUEPAy0B9yRkhcsLdRN8fzrw@mail.gmail.com
Change pg_bsd_indent to follow upstream rules for placement of comments
to the right of code, and remove pgindent hack that caused comments
following #endif to not obey the general rule.
Commit e3860ffa4d wasn't actually using
the published version of pg_bsd_indent, but a hacked-up version that
tried to minimize the amount of movement of comments to the right of
code. The situation of interest is where such a comment has to be
moved to the right of its default placement at column 33 because there's
code there. BSD indent has always moved right in units of tab stops
in such cases --- but in the previous incarnation, indent was working
in 8-space tab stops, while now it knows we use 4-space tabs. So the
net result is that in about half the cases, such comments are placed
one tab stop left of before. This is better all around: it leaves
more room on the line for comment text, and it means that in such
cases the comment uniformly starts at the next 4-space tab stop after
the code, rather than sometimes one and sometimes two tabs after.
Also, ensure that comments following #endif are indented the same
as comments following other preprocessor commands such as #else.
That inconsistency turns out to have been self-inflicted damage
from a poorly-thought-through post-indent "fixup" in pgindent.
This patch is much less interesting than the first round of indent
changes, but also bulkier, so I thought it best to separate the effects.
Discussion: https://postgr.es/m/E1dAmxK-0006EE-1r@gemulon.postgresql.org
Discussion: https://postgr.es/m/30527.1495162840@sss.pgh.pa.us
When the checkpointer writes the shutdown checkpoint, it checks
afterwards whether any WAL has been written since it started and
throws a PANIC if so. At that point, only walsenders are still
active, so one might think this could not happen, but walsenders can
also generate WAL, for instance in BASE_BACKUP and logical decoding
related commands (e.g. via hint bits). So they can trigger this panic
if such a command is run while the shutdown checkpoint is being
written.
To fix this, divide the walsender shutdown into two phases. First,
checkpointer, itself triggered by postmaster, sends a
PROCSIG_WALSND_INIT_STOPPING signal to all walsenders. If the backend
is idle or runs an SQL query this causes the backend to shutdown, if
logical replication is in progress all existing WAL records are
processed followed by a shutdown. Otherwise this causes the walsender
to switch to the "stopping" state. In this state, the walsender will
reject any further replication commands. The checkpointer begins the
shutdown checkpoint once all walsenders are confirmed as
stopping. When the shutdown checkpoint finishes, the postmaster sends
us SIGUSR2. This instructs walsender to send any outstanding WAL,
including the shutdown checkpoint record, wait for it to be replicated
to the standby, and then exit.
Author: Andres Freund, based on an earlier patch by Michael Paquier
Reported-By: Fujii Masao, Andres Freund
Reviewed-By: Michael Paquier
Discussion: https://postgr.es/m/20170602002912.tqlwn4gymzlxpvs2@alap3.anarazel.de
Backpatch: 9.4, where logical decoding was introduced
This reverts commit 086221cf6b, which
was made to master only.
The approach implemented in the above commit has some issues. While
those could easily be fixed incrementally, doing so would make
backpatching considerably harder, so instead first revert this patch.
Discussion: https://postgr.es/m/20170602002912.tqlwn4gymzlxpvs2@alap3.anarazel.de
When the checkpointer writes the shutdown checkpoint, it checks
afterwards whether any WAL has been written since it started and throws
a PANIC if so. At that point, only walsenders are still active, so one
might think this could not happen, but walsenders can also generate WAL,
for instance in BASE_BACKUP and certain variants of
CREATE_REPLICATION_SLOT. So they can trigger this panic if such a
command is run while the shutdown checkpoint is being written.
To fix this, divide the walsender shutdown into two phases. First, the
postmaster sends a SIGUSR2 signal to all walsenders. The walsenders
then put themselves into the "stopping" state. In this state, they
reject any new commands. (For simplicity, we reject all new commands,
so that in the future we do not have to track meticulously which
commands might generate WAL.) The checkpointer waits for all walsenders
to reach this state before proceeding with the shutdown checkpoint.
After the shutdown checkpoint is done, the postmaster sends
SIGINT (previously unused) to the walsenders. This triggers the
existing shutdown behavior of sending out the shutdown checkpoint record
and then terminating.
Author: Michael Paquier <michael.paquier@gmail.com>
Reported-by: Fujii Masao <masao.fujii@gmail.com>
Adds write_lag, flush_lag and replay_lag cols to pg_stat_replication.
Implements a lag tracker module that reports the lag times based upon
measurements of the time taken for recent WAL to be written, flushed and
replayed and for the sender to hear about it. These times
represent the commit lag that was (or would have been) introduced by each
synchronous commit level, if the remote server was configured as a
synchronous standby. For an asynchronous standby, the replay_lag column
approximates the delay before recent transactions became visible to queries.
If the standby server has entirely caught up with the sending server and
there is no more WAL activity, the most recently measured lag times will
continue to be displayed for a short time and then show NULL.
Physical replication lag tracking is automatic. Logical replication tracking
is possible but is the responsibility of the logical decoding plugin.
Tracking is a private module operating within each walsender individually,
with values reported to shared memory. Module not used outside of walsender.
Design and code is good enough now to commit - kudos to the author.
In many ways a difficult topic, with important and subtle behaviour so this
shoudl be expected to generate discussion and multiple open items: Test now!
Author: Thomas Munro, following designs by Fujii Masao and Simon Riggs
Review: Simon Riggs, Ian Barwick and Craig Ringer
This improves on commit bbfd7edae5 by
making two simple changes:
* pg_attribute_noreturn now takes parentheses, ie pg_attribute_noreturn().
Likewise pg_attribute_unused(), pg_attribute_packed(). This reduces
pgindent's tendency to misformat declarations involving them.
* attributes are now always attached to function declarations, not
definitions. Previously some places were taking creative shortcuts,
which were not merely candidates for bad misformatting by pgindent
but often were outright wrong anyway. (It does little good to put a
noreturn annotation where callers can't see it.) In any case, if
we would like to believe that these macros can be used with non-gcc
compilers, we should avoid gratuitous variance in usage patterns.
I also went through and manually improved the formatting of a lot of
declarations, and got rid of excessively repetitive (and now obsolete
anyway) comments informing the reader what pg_attribute_printf is for.
Replace some bogus "x[1]" declarations with "x[FLEXIBLE_ARRAY_MEMBER]".
Aside from being more self-documenting, this should help prevent bogus
warnings from static code analyzers and perhaps compiler misoptimizations.
This patch is just a down payment on eliminating the whole problem, but
it gets rid of a lot of easy-to-fix cases.
Note that the main problem with doing this is that one must no longer rely
on computing sizeof(the containing struct), since the result would be
compiler-dependent. Instead use offsetof(struct, lastfield). Autoconf
also warns against spelling that offsetof(struct, lastfield[0]).
Michael Paquier, review and additional fixes by me.
Relying on the normal shared latch simplifies interrupt/signal
handling because we can rely on all signal handlers setting the proc
latch. That in turn allows us to avoid the use of
ImmediateInterruptOK, which arguably isn't correct because
WaitLatchOrSocket isn't declared to be immediately interruptible.
Also change sections that wait on the walsender's latch to notice
interrupts quicker/more reliably and make them more consistent with
each other.
This is part of a larger "get rid of ImmediateInterruptOK" series.
Discussion: 20150115020335.GZ5245@awork2.anarazel.de
When a standby server follows the master using WAL archive, and it chooses
a new timeline (recovery_target_timeline='latest'), it only fetches the
timeline history file for the chosen target timeline, not any other history
files that might be missing from pg_xlog. For example, if the current
timeline is 2, and we choose 4 as the new recovery target timeline, the
history file for timeline 3 is not fetched, even if it's part of this
server's history. That's enough for the standby itself - the history file
for timeline 4 includes timeline 3 as well - but if a cascading standby
server wants to recover to timeline 3, it needs the history file. To fix,
when a new recovery target timeline is chosen, try to copy any missing
history files from the archive to pg_xlog between the old and new target
timeline.
A second similar issue was with the WAL files. When a standby recovers from
archive, and it reaches a segment that contains a switch to a new timeline,
recovery fetches only the WAL file labelled with the new timeline's ID. The
file from the new timeline contains a copy of the WAL from the old timeline
up to the point where the switch happened, and recovery recovers it from the
new file. But in streaming replication, walsender only tries to read it
from the old timeline's file. To fix, change walsender to read it from the
new file, so that it behaves the same as recovery in that sense, and doesn't
try to open the possibly nonexistent file with the old timeline's ID.
Before this patch, streaming replication would refuse to start replicating
if the timeline in the primary doesn't exactly match the standby. The
situation where it doesn't match is when you have a master, and two
standbys, and you promote one of the standbys to become new master.
Promoting bumps up the timeline ID, and after that bump, the other standby
would refuse to continue.
There's significantly more timeline related logic in streaming replication
now. First of all, when a standby connects to primary, it will ask the
primary for any timeline history files that are missing from the standby.
The missing files are sent using a new replication command TIMELINE_HISTORY,
and stored in standby's pg_xlog directory. Using the timeline history files,
the standby can follow the latest timeline present in the primary
(recovery_target_timeline='latest'), just as it can follow new timelines
appearing in an archive directory.
START_REPLICATION now takes a TIMELINE parameter, to specify exactly which
timeline to stream WAL from. This allows the standby to request the primary
to send over WAL that precedes the promotion. The replication protocol is
changed slightly (in a backwards-compatible way although there's little hope
of streaming replication working across major versions anyway), to allow
replication to stop when the end of timeline reached, putting the walsender
back into accepting a replication command.
Many thanks to Amit Kapila for testing and reviewing various versions of
this patch.
Rename replication_timeout to wal_sender_timeout, and add a new setting
called wal_receiver_timeout that does the same at the walreceiver side.
There was previously no timeout in walreceiver, so if the network went down,
for example, the walreceiver could take a long time to notice that the
connection was lost. Now with the two settings, both sides of a replication
connection will detect a broken connection similarly.
It is no longer necessary to manually set wal_receiver_status_interval to
a value smaller than the timeout. Both wal sender and receiver now
automatically send a "ping" message if more than 1/2 of the configured
timeout has elapsed, and it hasn't received any messages from the other end.
Amit Kapila, heavily edited by me.
Replication occurs only to memory on standby, not to disk,
so provides additional performance if user wishes to
reduce durability level slightly. Adds concept of multiple
independent sync rep queues.
Fujii Masao and Simon Riggs
Andres Freund
Bruce Momjian
Peter Geoghegan
Michael Paquier
Robert Haas
Tom Lane
Amit Kapila
Alvaro Herrera
Peter Eisentraut
Simon Riggs
Heikki Linnakangas