When there are cost-delay-related storage options set for a table,
trying to make that table participate in the autovacuum cost-limit
balancing algorithm produces undesirable results: instead of using the
configured values, the global values are always used,
as illustrated by Mark Kirkwood in
http://www.postgresql.org/message-id/52FACF15.8020507@catalyst.net.nz
Since the mechanism is already complicated, just disable it for those
cases rather than trying to make it cope. There are undesirable
side-effects from this too, namely that the total I/O impact on the
system will be higher whenever such tables are vacuumed. However, this
is seen as less harmful than slowing down vacuum, because that would
cause bloat to accumulate. Anyway, in the new system it is possible to
tweak options to get the precise behavior one wants, whereas with the
previous system one was simply hosed.
This has been broken forever, so backpatch to all supported branches.
This might affect systems where cost_limit and cost_delay have been set
for individual tables.
DetermineSleepTime() was previously called without blocked
signals. That's not good, because it allows signal handlers to
interrupt its workings.
DetermineSleepTime() was added in 9.3 with the addition of background
workers (da07a1e856511), where it only read from
BackgroundWorkerList.
Since 9.4, where dynamic background workers were added (7f7485a0cde),
the list is also manipulated in DetermineSleepTime(). That's bad
because the list now can be persistently corrupted if modified by both
a signal handler and DetermineSleepTime().
This was discovered during the investigation of hangs on buildfarm
member anole. It's unclear whether this bug is the source of these
hangs or not, but it's worth fixing either way. I have confirmed that
it can cause crashes.
It luckily looks like this only can cause problems when bgworkers are
actively used.
Discussion: 20140929193733.GB14400@awork2.anarazel.de
Backpatch to 9.3 where background workers were introduced.
When autovacuum is nominally off, we will still launch autovac workers
to vacuum tables that are at risk of XID wraparound. But after we'd done
that, an autovac worker would proceed to autovacuum every table in the
targeted database, if they meet the usual thresholds for autovacuuming.
This is at best pretty unexpected; at worst it delays response to the
wraparound threat. Fix it so that if autovacuum is nominally off, we
*only* do forced vacuums and not any other work.
Per gripe from Andrey Zhidenkov. This has been like this all along,
so back-patch to all supported branches.
InitProcess() relies on IsBackgroundWorker to decide whether the PGPROC
for a new backend should be taken from ProcGlobal's freeProcs or from
bgworkerFreeProcs. In EXEC_BACKEND builds, InitProcess() is called
sooner than in non-EXEC_BACKEND builds, and IsBackgroundWorker wasn't
getting initialized soon enough.
Report by Noah Misch. Diagnosis and fix by me.
In yesterday's commit 2dc4f011fd61501cce507be78c39a2677690d44b, I tried
to force buffering of stdout/stderr in initdb to be what it is by
default when the program is run interactively on Unix (since that's how
most manual testing is done). This tripped over the fact that Windows
doesn't support _IOLBF mode. We dealt with that a long time ago in
syslogger.c by falling back to unbuffered mode on Windows. Export that
solution in port.h and use it in initdb.
Back-patch to 8.4, like the previous commit.
This was not changed in HEAD, but will be done later as part of a
pgindent run. Future pgindent runs will also do this.
Report by Tom Lane
Backpatch through all supported branches, but not HEAD
According to the Single Unix Spec and assorted man pages, you're supposed
to use the constants named AF_xxx when setting ai_family for a getaddrinfo
call. In a few places we were using PF_xxx instead. Use of PF_xxx
appears to be an ancient BSD convention that was not adopted by later
standardization. On BSD and most later Unixen, it doesn't matter much
because those constants have equivalent values anyway; but nonetheless
this code is not per spec.
In the same vein, replace PF_INET by AF_INET in one socket() call, which
wasn't even consistent with the other socket() call in the same function
let alone the remainder of our code.
Per investigation of a Cygwin trouble report from Marco Atzeri. It's
probably a long shot that this will fix his issue, but it's wrong in
any case.
Previously, in some places, socket creation errors were checked for
negative values, which is not true for Windows because sockets are
unsigned. This masked socket creation errors on Windows.
Backpatch through 9.0. 8.4 doesn't have the infrastructure to fix this.
Formerly, we set up the postmaster's signal handling only when we were
about to start launching subprocesses. This is a bad idea though, as
it means that for example a SIGINT arriving before that will kill the
postmaster instantly, perhaps leaving lockfiles, socket files, shared
memory, etc laying about. We'd rather that such a signal caused orderly
postmaster termination including releasing of those resources. A simple
fix is to move the PostmasterMain stanza that initializes signal handling
to an earlier point, before we've created any such resources. Then, an
early-arriving signal will be blocked until we're ready to deal with it
in the usual way. (The only part that really needs to be moved up is
blocking of signals, but it seems best to keep the signal handler
installation calls together with that; for one thing this ensures the
kernel won't drop any signals we wished to get. The handlers won't get
invoked in any case until we unblock signals in ServerLoop.)
Per a report from MauMau. He proposed changing the way "pg_ctl stop"
works to deal with this, but that'd just be masking one symptom not
fixing the core issue.
It's been like this since forever, so back-patch to all supported branches.
The code for matching clients to pg_hba.conf lines that specify host names
(instead of IP address ranges) failed to complain if reverse DNS lookup
failed; instead it silently didn't match, so that you might end up getting
a surprising "no pg_hba.conf entry for ..." error, as seen in bug #9518
from Mike Blackwell. Since we don't want to make this a fatal error in
situations where pg_hba.conf contains a mixture of host names and IP
addresses (clients matching one of the numeric entries should not have to
have rDNS data), remember the lookup failure and mention it as DETAIL if
we get to "no pg_hba.conf entry". Apply the same approach to forward-DNS
lookup failures, too, rather than treating them as immediate hard errors.
Along the way, fix a couple of bugs that prevented us from detecting an
rDNS lookup error reliably, and make sure that we make only one rDNS lookup
attempt; formerly, if the lookup attempt failed, the code would try again
for each host name entry in pg_hba.conf. Since more or less the whole
point of this design is to ensure there's only one lookup attempt not one
per entry, the latter point represents a performance bug that seems
sufficient justification for back-patching.
Also, adjust src/port/getaddrinfo.c so that it plays as well as it can
with this code. Which is not all that well, since it does not have actual
support for rDNS lookup, but at least it should return the expected (and
required by spec) error codes so that the main code correctly perceives the
lack of functionality as a lookup failure. It's unlikely that PG is still
being used in production on any machines that require our getaddrinfo.c,
so I'm not excited about working harder than this.
To keep the code in the various branches similar, this includes
back-patching commits c424d0d1052cb4053c8712ac44123f9b9a9aa3f2 and
1997f34db4687e671690ed054c8f30bb501b1168 into 9.2 and earlier.
Back-patch to 9.1 where the facility for hostnames in pg_hba.conf was
introduced.
Initialization of this field was not being done according to the
st_changecount protocol (it has to be done within the changecount increment
range, not outside). And the test to see if the value should be reported
as null was wrong. Noted while perusing uses of Port.remote_hostname.
This was wrong from the introduction of this code (commit 4a25bc145),
so back-patch to 9.1.
We were unlinking the permanent file, not the non-permanent one. But
since the stat collector already unlinks all permanent files on startup,
there was nothing for it to unlink. The non-permanent file remained in
place, and was copied to the permanent directory on shutdown, so in
effect no file was ever dropped.
Backpatch to 9.3, where the issue was introduced by commit 187492b6c2e8.
Before that, there were no per-database files and thus no file to drop
on DROP DATABASE.
Per report from Thom Brown.
Author: Tomáš Vondra
We should not assume that struct timeval.tv_sec is a long, because
it ain't necessarily. (POSIX says that it's a time_t, which might
well be 64 bits now or in the future; or for that matter might be
32 bits on machines with 64-bit longs.) Per buildfarm member panther.
Back-patch to 9.3 where the dubious coding was introduced.
Previously we were piggybacking on transaction ID parameters to freeze
multixacts; but since there isn't necessarily any relationship between
rates of Xid and multixact consumption, this turns out not to be a good
idea.
Therefore, we now have multixact-specific freezing parameters:
vacuum_multixact_freeze_min_age: when to remove multis as we come across
them in vacuum (default to 5 million, i.e. early in comparison to Xid's
default of 50 million)
vacuum_multixact_freeze_table_age: when to force whole-table scans
instead of scanning only the pages marked as not all visible in
visibility map (default to 150 million, same as for Xids). Whichever of
both which reaches the 150 million mark earlier will cause a whole-table
scan.
autovacuum_multixact_freeze_max_age: when for cause emergency,
uninterruptible whole-table scans (default to 400 million, double as
that for Xids). This means there shouldn't be more frequent emergency
vacuuming than previously, unless multixacts are being used very
rapidly.
Backpatch to 9.3 where multixacts were made to persist enough to require
freezing. To avoid an ABI break in 9.3, VacuumStmt has a couple of
fields in an unnatural place, and StdRdOptions is split in two so that
the newly added fields can go at the end.
Patch by me, reviewed by Robert Haas, with additional input from Andres
Freund and Tom Lane.
Per reports from Andres Freund and Luke Campbell, a server failure during
set_pglocale_pgservice results in a segfault rather than a useful error
message, because the infrastructure needed to use ereport hasn't been
initialized; specifically, MemoryContextInit hasn't been called.
One known cause of this is starting the server in a directory it
doesn't have permission to read.
We could try to prevent set_pglocale_pgservice from using anything that
depends on palloc or elog, but that would be messy, and the odds of future
breakage seem high. Moreover there are other things being called in main.c
that look likely to use palloc or elog too --- perhaps those things
shouldn't be there, but they are there today. The best solution seems to
be to move the call of MemoryContextInit to very early in the backend's
real main() function. I've verified that an elog or ereport occurring
immediately after that is now capable of sending something useful to
stderr.
I also added code to elog.c to print something intelligible rather than
just crashing if MemoryContextInit hasn't created the ErrorContext.
This could happen if MemoryContextInit itself fails (due to malloc
failure), and provides some future-proofing against someone trying to
sneak in new code even earlier in server startup.
Back-patch to all supported branches. Since we've only heard reports of
this type of failure recently, it may be that some recent change has made
it more likely to see a crash of this kind; but it sure looks like it's
broken all the way back.
Prevent handle_sig_alarm from losing control partway through due to a query
cancel (either an asynchronous SIGINT, or a cancel triggered by one of the
timeout handler functions). That would at least result in failure to
schedule any required future interrupt, and might result in actual
corruption of timeout.c's data structures, if the interrupt happened while
we were updating those.
We could still lose control if an asynchronous SIGINT arrives just as the
function is entered. This wouldn't break any data structures, but it would
have the same effect as if the SIGALRM interrupt had been silently lost:
we'd not fire any currently-due handlers, nor schedule any new interrupt.
To forestall that scenario, forcibly reschedule any pending timer interrupt
during AbortTransaction and AbortSubTransaction. We can avoid any extra
kernel call in most cases by not doing that until we've allowed
LockErrorCleanup to kill the DEADLOCK_TIMEOUT and LOCK_TIMEOUT events.
Another hazard is that some platforms (at least Linux and *BSD) block a
signal before calling its handler and then unblock it on return. When we
longjmp out of the handler, the unblock doesn't happen, and the signal is
left blocked indefinitely. Again, we can fix that by forcibly unblocking
signals during AbortTransaction and AbortSubTransaction.
These latter two problems do not manifest when the longjmp reaches
postgres.c, because the error recovery code there kills all pending timeout
events anyway, and it uses sigsetjmp(..., 1) so that the appropriate signal
mask is restored. So errors thrown outside any transaction should be OK
already, and cleaning up in AbortTransaction and AbortSubTransaction should
be enough to fix these issues. (We're assuming that any code that catches
a query cancel error and doesn't re-throw it will do at least a
subtransaction abort to clean up; but that was pretty much required already
by other subsystems.)
Lastly, ProcSleep should not clear the LOCK_TIMEOUT indicator flag when
disabling that event: if a lock timeout interrupt happened after the lock
was granted, the ensuing query cancel is still going to happen at the next
CHECK_FOR_INTERRUPTS, and we want to report it as a lock timeout not a user
cancel.
Per reports from Dan Wood.
Back-patch to 9.3 where the new timeout handling infrastructure was
introduced. We may at some point decide to back-patch the signal
unblocking changes further, but I'll desist from that until we hear
actual field complaints about it.
It seems to make more sense to use "cutoff multixact" terminology
throughout the backend code; "freeze" is associated with replacing of an
Xid with FrozenTransactionId, which is not what we do for MultiXactIds.
Andres Freund
Some adjustments by Álvaro Herrera
Instead of deleting all files in stats_temp_directory and the permanent
directory on a crash, only remove those files that match the pattern of
files we actually write in them, to avoid possibly clobbering existing
unrelated contents of the temporary directory. Per complaint from Jeff
Janes, and subsequent discussion, starting at message
CAMkU=1z9+7RsDODnT4=cDFBRBp8wYQbd_qsLcMtKEf-oFwuOdQ@mail.gmail.com
Also, fix a bug in the same routine to avoid removing files from the
permanent directory twice (instead of once from that directory and then
from the temporary directory), also per report from Jeff Janes, in
message
CAMkU=1wbk947=-pAosDMX5VC+sQw9W4ttq6RM9rXu=MjNeEQKA@mail.gmail.com
Commit 2dee7998f93062e2ae7fcc9048ff170e528d1724 renames these
functions in master, for consistency; per discussion, backport
just the renaming portion of that commit to 9.3 to keep the
branches in sync.
Michael Paquier
We've seen multiple cases of people looking at the postmaster's original
stderr output to try to diagnose problems, not realizing/remembering that
their logging configuration is set up to send log messages somewhere else.
This seems particularly likely to happen in prepackaged distributions,
since many packagers patch the code to change the factory-standard logging
configuration to something more in line with their platform conventions.
In hopes of reducing confusion, emit a LOG message about this at the point
in startup where we are about to switch log output away from the original
stderr, providing a pointer to where to look instead. This message will
appear as the last thing in the original stderr output. (We might later
also try to emit such link messages when logging parameters are changed
on-the-fly; but that case seems to be both noticeably harder to do nicely,
and much less frequently a problem in practice.)
Per discussion, back-patch to 9.3 but not further.
Previously one had to use slist_delete(), implying an additional scan of
the list, making this infrastructure considerably less efficient than
traditional Lists when deletion of element(s) in a long list is needed.
Modify the slist_foreach_modify() macro to support deleting the current
element in O(1) time, by keeping a "prev" pointer in addition to "cur"
and "next". Although this makes iteration with this macro a bit slower,
no real harm is done, since in any scenario where you're not going to
delete the current list element you might as well just use slist_foreach
instead. Improve the comments about when to use each macro.
Back-patch to 9.3 so that we'll have consistent semantics in all branches
that provide ilist.h. Note this is an ABI break for callers of
slist_foreach_modify().
Andres Freund and Tom Lane
Commit 7f7485a0cde92aa4ba235a1ffe4dda0ca0b6cc9a made these changes
in master; per discussion, backport the API changes (but not the
functional changes), so that people don't get used to the 9.3 API
only to see it get broken in the next release. There are already
some people coding to the original 9.3 API, and this will cause
minor breakage, but there will be even more if we wait until next
year to roll out these changes.
Per discussion on pgsql-hackers, these aren't really needed. Interim
versions of the background worker patch had the worker starting with
signals already unblocked, which would have made this necessary.
But the final version does not, so we don't really need it; and it
doesn't work well with the new facility for starting dynamic background
workers, so just rip it out.
Also per discussion on pgsql-hackers, back-patch this change to 9.3.
It's best to get the API break out of the way before we do an
official release of this facility, to avoid more pain for extension
authors later.
Latch activity was not being detected by non-database-connected workers; the
SIGUSR1 signal handler which is normally in charge of that was set to SIG_IGN.
Create a simple handler to call latch_sigusr1_handler instead.
Robert Haas (bug report and suggested fix)
Add a SignalUnconnectedWorkers() call so that non-database-connected background
workers are also notified when postmaster is SIGHUPped. Previously, only
database-connected workers were.
Michael Paquier (bug report and fix)
The point of turning off track_activities is to avoid this reporting
overhead, but a thinko in commit 4f42b546fd87a80be30c53a0f2c897acb826ad52
caused pgstat_report_activity() to perform half of its updates anyway.
Fix that, and also make sure that we clear all the now-disabled fields
when transitioning to the non-reporting state.
An oversight in commit e710b65c1c56ca7b91f662c63d37ff2e72862a94 allowed
database names beginning with "-" to be treated as though they were secure
command-line switches; and this switch processing occurs before client
authentication, so that even an unprivileged remote attacker could exploit
the bug, needing only connectivity to the postmaster's port. Assorted
exploits for this are possible, some requiring a valid database login,
some not. The worst known problem is that the "-r" switch can be invoked
to redirect the process's stderr output, so that subsequent error messages
will be appended to any file the server can write. This can for example be
used to corrupt the server's configuration files, so that it will fail when
next restarted. Complete destruction of database tables is also possible.
Fix by keeping the database name extracted from a startup packet fully
separate from command-line switches, as had already been done with the
user name field.
The Postgres project thanks Mitsumasa Kondo for discovering this bug,
Kyotaro Horiguchi for drafting the fix, and Noah Misch for recognizing
the full extent of the danger.
Security: CVE-2013-1899
Previously, if the postmaster initialized OpenSSL's PRNG (which it will do
when ssl=on in postgresql.conf), the same pseudo-random state would be
inherited by each forked child process. The problem is masked to a
considerable extent if the incoming connection uses SSL encryption, but
when it does not, identical pseudo-random state is made available to
functions like contrib/pgcrypto. The process's PID does get mixed into any
requested random output, but on most systems that still only results in 32K
or so distinct random sequences available across all Postgres sessions.
This might allow an attacker who has database access to guess the results
of "secure" operations happening in another session.
To fix, forcibly reset the PRNG after fork(). Each child process that has
need for random numbers from OpenSSL's generator will thereby be forced to
go through OpenSSL's normal initialization sequence, which should provide
much greater variability of the sequences. There are other ways we might
do this that would be slightly cheaper, but this approach seems the most
future-proof against SSL-related code changes.
This has been assigned CVE-2013-1900, but since the issue and the patch
have already been publicized on pgsql-hackers, there's no point in trying
to hide this commit.
Back-patch to all supported branches.
Marko Kreen
This GUC allows limiting the time spent waiting to acquire any one
heavyweight lock.
In support of this, improve the recently-added timeout infrastructure
to permit efficiently enabling or disabling multiple timeouts at once.
That reduces the performance hit from turning on lock_timeout, though
it's still not zero.
Zoltán Böszörményi, reviewed by Tom Lane,
Stephen Frost, and Hari Babu
A materialized view has a rule just like a view and a heap and
other physical properties like a table. The rule is only used to
populate the table, references in queries refer to the
materialized data.
This is a minimal implementation, but should still be useful in
many cases. Currently data is only populated "on demand" by the
CREATE MATERIALIZED VIEW and REFRESH MATERIALIZED VIEW statements.
It is expected that future releases will add incremental updates
with various timings, and that a more refined concept of defining
what is "fresh" data will be developed. At some point it may even
be possible to have queries use a materialized in place of
references to underlying tables, but that requires the other
above-mentioned features to be working first.
Much of the documentation work by Robert Haas.
Review by Noah Misch, Thom Brown, Robert Haas, Marko Tiikkaja
Security review by KaiGai Kohei, with a decision on how best to
implement sepgsql still pending.
We now write one file per database and one global file, instead of
having the whole thing in a single huge file. This reduces the I/O that
must be done when partial data is required -- which is all the time,
because each process only needs information on its own database anyway.
Also, the autovacuum launcher does not need data about tables and
functions in each database; having the global stats for all DBs is
enough.
Catalog version bumped because we have a new subdir under PGDATA.
Author: Tomas Vondra. Some rework by Álvaro
Testing by Jeff Janes
Other discussion by Heikki Linnakangas, Tom Lane.
Normally, we suppress sending a tabstats message to the collector unless
there were some actual table stats to send. However, during backend exit
we should force out the message if there are any transaction commit/abort
counts to send, else the session's last few commit/abort counts will never
get reported at all. We had logic for this, but the short-circuit test
at the top of pgstat_report_stat() ignored the "force" flag, with the
consequence that session-ending transactions that touched no database-local
tables would not get counted. Seems to be an oversight in my commit
641912b4d17fd214a5e5bae4e7bb9ddbc28b144b, which added the "force" flag.
That was back in 8.3, so back-patch to all supported versions.
When we eliminated "unnecessary" wakeups of the syslogger process, we
broke size-based logfile rotation on Windows, because on that platform
data transfer is done in a separate thread. While non-Windows platforms
would recheck the output file size after every log message, Windows only
did so when the control thread woke up for some other reason, which might
be quite infrequent. Per bug #7814 from Tsunezumi. Back-patch to 9.2
where the problem was introduced.
Jeff Janes
This patch introduces two additional lock modes for tuples: "SELECT FOR
KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each
other, in contrast with already existing "SELECT FOR SHARE" and "SELECT
FOR UPDATE". UPDATE commands that do not modify the values stored in
the columns that are part of the key of the tuple now grab a SELECT FOR
NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently
with tuple locks of the FOR KEY SHARE variety.
Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this
means the concurrency improvement applies to them, which is the whole
point of this patch.
The added tuple lock semantics require some rejiggering of the multixact
module, so that the locking level that each transaction is holding can
be stored alongside its Xid. Also, multixacts now need to persist
across server restarts and crashes, because they can now represent not
only tuple locks, but also tuple updates. This means we need more
careful tracking of lifetime of pg_multixact SLRU files; since they now
persist longer, we require more infrastructure to figure out when they
can be removed. pg_upgrade also needs to be careful to copy
pg_multixact files over from the old server to the new, or at least part
of multixact.c state, depending on the versions of the old and new
servers.
Tuple time qualification rules (HeapTupleSatisfies routines) need to be
careful not to consider tuples with the "is multi" infomask bit set as
being only locked; they might need to look up MultiXact values (i.e.
possibly do pg_multixact I/O) to find out the Xid that updated a tuple,
whereas they previously were assured to only use information readily
available from the tuple header. This is considered acceptable, because
the extra I/O would involve cases that would previously cause some
commands to block waiting for concurrent transactions to finish.
Another important change is the fact that locking tuples that have
previously been updated causes the future versions to be marked as
locked, too; this is essential for correctness of foreign key checks.
This causes additional WAL-logging, also (there was previously a single
WAL record for a locked tuple; now there are as many as updated copies
of the tuple there exist.)
With all this in place, contention related to tuples being checked by
foreign key rules should be much reduced.
As a bonus, the old behavior that a subtransaction grabbing a stronger
tuple lock than the parent (sub)transaction held on a given tuple and
later aborting caused the weaker lock to be lost, has been fixed.
Many new spec files were added for isolation tester framework, to ensure
overall behavior is sane. There's probably room for several more tests.
There were several reviewers of this patch; in particular, Noah Misch
and Andres Freund spent considerable time in it. Original idea for the
patch came from Simon Riggs, after a problem report by Joel Jacobson.
Most code is from me, with contributions from Marti Raudsepp, Alexander
Shulgin, Noah Misch and Andres Freund.
This patch was discussed in several pgsql-hackers threads; the most
important start at the following message-ids:
AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com1290721684-sup-3951@alvh.no-ip.org1294953201-sup-2099@alvh.no-ip.org1320343602-sup-2290@alvh.no-ip.org1339690386-sup-8927@alvh.no-ip.org4FE5FF020200002500048A3D@gw.wicourts.gov4FEAB90A0200002500048B7D@gw.wicourts.gov
Commit da07a1e8 was broken for EXEC_BACKEND because I failed to realize
that the MaxBackends recomputation needed to be duplicated by
subprocesses in SubPostmasterMain. However, instead of having the value
be recomputed at all, it's better to assign the correct value at
postmaster initialization time, and have it be propagated to exec'ed
backends via BackendParameters.
MaxBackends stays as zero until after modules in
shared_preload_libraries have had a chance to register bgworkers, since
the value is going to be untrustworthy till that's finished.
Heikki Linnakangas and Álvaro Herrera
We used to set it to the current recovery target timeline, but the recovery
target timeline can change during recovery, leaving ThisTimeLineID at an
old value. That seems worse than always leaving it at zero to begin with.
AFAICS there was no good reason to set it in the first place. ThisTimeLineID
is not needed in checkpointer or bgwriter process, until it's time to write
the end-of-recovery checkpoint, and at that point ThisTimeLineID is updated
anyway.
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.
