SnapshotNow scans have the undesirable property that, in the face of
concurrent updates, the scan can fail to see either the old or the new
versions of the row. In many cases, we work around this by requiring
DDL operations to hold AccessExclusiveLock on the object being
modified; in some cases, the existing locking is inadequate and random
failures occur as a result. This commit doesn't change anything
related to locking, but will hopefully pave the way to allowing lock
strength reductions in the future.
The major issue has held us back from making this change in the past
is that taking an MVCC snapshot is significantly more expensive than
using a static special snapshot such as SnapshotNow. However, testing
of various worst-case scenarios reveals that this problem is not
severe except under fairly extreme workloads. To mitigate those
problems, we avoid retaking the MVCC snapshot for each new scan;
instead, we take a new snapshot only when invalidation messages have
been processed. The catcache machinery already requires that
invalidation messages be sent before releasing the related heavyweight
lock; else other backends might rely on locally-cached data rather
than scanning the catalog at all. Thus, making snapshot reuse
dependent on the same guarantees shouldn't break anything that wasn't
already subtly broken.
Patch by me. Review by Michael Paquier and Andres Freund.
When there's a comment on an index that was created with UNIQUE or PRIMARY
KEY constraint syntax, we need to label the comment as depending on the
constraint not the index, since only the constraint object actually appears
in the dump. This incorrect dependency can lead to parallel pg_restore
trying to restore the comment before the index has been created, per bug
#8257 from Lloyd Albin.
This patch fixes pg_dump to produce the right dependency in dumps made
in the future. Usually we also try to hack pg_restore to work around
bogus dependencies, so that existing (wrong) dumps can still be restored in
parallel mode; but that doesn't seem practical here since there's no easy
way to relate the constraint dump entry to the comment after the fact.
Andres Freund
In binary upgrade mode, we need to recreate and then drop dropped
columns so that all the columns get the right attribute number. This is
true for foreign tables as well as for native tables. For foreign
tables we have been getting the first part right but not the second,
leading to bogus columns in the upgraded database. Fix this all the way
back to 9.1, where foreign tables were introduced.
pg_isready displays the host name and the port number that it uses to connect
to the server. So far, pg_isready didn't use the conninfo specified in -d option
for calculating those host name and port number. This can lead to wrong display
to a user. This commit changes pg_isready so that it uses the conninfo for that
calculation.
Original patch by Phil Sorber, modified by me.
getSchemaData() must identify extension member objects and mark them
as not to be dumped. This must happen after reading all objects that can be
direct members of extensions, but before we begin to process table subsidiary
objects. Both rules and event triggers were wrong in this regard.
Backport rules portion of patch to 9.1 -- event triggers do not exist prior to 9.3.
Suggested fix by Tom Lane, initial complaint and patch by me.
In the primary_conninfo line that "pg_basebackup -R" generates, single
quotes in parameter values need to be escaped into \\'; the libpq parser
requires the quotes to be escaped into \', and recovery.conf parser requires
the \ to be escaped into \\.
Also, don't quote parameter values unnecessarily, to make the connection
string prettier. Most options in a libpq connection string don't need
quoting.
Reported by Hari Babu, closer analysis by Zoltan Boszormenyi, although I
didn't use his patch.
If a standby server has a cascading standby server connected to it, it's
possible that WAL has already been sent up to the next WAL page boundary,
splitting a WAL record in the middle, when the first standby server is
promoted. Don't throw an assertion failure or error in walsender if that
happens.
Also, fix a variant of the same bug in pg_receivexlog: if it had already
received WAL on previous timeline up to a segment boundary, when the
upstream standby server is promoted so that the timeline switch record falls
on the previous segment, pg_receivexlog would miss the segment containing
the timeline switch. To fix that, have walsender send the position of the
timeline switch at end-of-streaming, in addition to the next timeline's ID.
It was previously assumed that the switch happened exactly where the
streaming stopped.
Note: this is an incompatible change in the streaming protocol. You might
get an error if you try to stream over timeline switches, if the client is
running 9.3beta1 and the server is more recent. It should be fine after a
reconnect, however.
Reported by Fujii Masao.
Previously this state was represented by whether the view's disk file had
zero or nonzero size, which is problematic for numerous reasons, since it's
breaking a fundamental assumption about heap storage. This was done to
allow unlogged matviews to revert to unpopulated status after a crash
despite our lack of any ability to update catalog entries post-crash.
However, this poses enough risk of future problems that it seems better to
not support unlogged matviews until we can find another way. Accordingly,
revert that choice as well as a number of existing kluges forced by it
in favor of creating a pg_class.relispopulated flag column.
Very old versions of msgfmt choke on these specific messages, for reasons
that are unclear at the moment. Remove them so that we can ship a beta
release and not get complaints from testers (these messages will just go
untranslated, instead, and we're hardly at 100% coverage anyway).
Peter Eisentraut will look for a better fix later.
Print the command tag if we get PGRES_COMMAND_OK, and throw an error for
other cases. Per gripe from Michael Paquier.
In passing, add an fflush(), just to be real sure the output appears
before we sleep.
Previously, libpq and the backend had opposite ideas about whether
it was necessary for the client to send a CopyDone message after
receiving an ErrorResponse, making it impossible to cleanly exit
COPY BOTH mode. Fix libpq so that works correctly, adopting the
backend's notion that an ErrorResponse kills the copy in both
directions.
Adjust receivelog.c to avoid a degradation in the quality of the
resulting error messages. libpqwalreceiver.c is already doing
the right thing, so no adjustment needed there.
Add an explicit statement to the documentation explaining how
this part of the protocol is supposed to work, in the hopes of
avoiding future confusion in this area.
Since the consequences of all this confusion are very limited,
especially in the back-branches where no client ever attempts
to exit COPY BOTH mode without closing the connection entirely,
no back-patch.
On non-Windows systems, sys/time.h was pulled in by portability/instr_time.h,
which pulled in time.h. We certainly should include time.h directly, since
we're using time(2), but the indirect include masked the problem on most
platforms.
Andres Freund
This changes the behavior of the start and stop actions to exit
successfully if the server was already started or stopped.
This changes the default behavior of the start action: Before, if the
server was already running, it would print a message and succeed. Now,
that situation will result in an error. When running in idempotent
mode, no message is printed and pg_ctl exits successfully.
It was considered to just make the idempotent behavior the default and
only option, but pg_upgrade needs the old behavior.
The intent was that being populated would, long term, be just one
of the conditions which could affect whether a matview was
scannable; being populated should be necessary but not always
sufficient to scan the relation. Since only CREATE and REFRESH
currently determine the scannability, names and comments
accidentally conflated these concepts, leading to confusion.
Also add missing locking for the SQL function which allows a
test for scannability, and fix a modularity violatiion.
Per complaints from Tom Lane, although its not clear that these
will satisfy his concerns. Hopefully this will at least better
frame the discussion.
Counting newlines shows that quite a few recent patches have neglected
to update the output-lines count given to PageOutput(). Fortunately
it's not terribly critical that this be exact, since we long since
exceeded the height of most people's terminal windows. Still, maybe
we ought to think of a way to not have to maintain this manually anymore.
9.2 uses a kluge representation of "indislive"; we have to account for
that when examining pg_index. Simplest solution is to check indisready
for 9.0 and 9.1 as well; that's harmless though unnecessary, so it's
not worth making a version distinction for.
Fixes oversight in commit 683abc73df,
as noted by Andres Freund.
Move functions used only by pg_dump and pg_restore from dumputils.c to a new
file, pg_backup_utils.c. dumputils.c is linked into psql and some programs
in bin/scripts, so it seems good to keep it slim. The parallel functionality
is moved to parallel.c, as is exit_horribly, because the interesting code in
exit_horribly is parallel-related.
This refactoring gets rid of the on_exit_msg_func function pointer. It was
problematic, because a modern gcc version with -Wmissing-format-attribute
complained if it wasn't marked with PF_PRINTF_ATTRIBUTE, but the ancient gcc
version that Tom Lane's old HP-UX box has didn't accept that attribute on a
function pointer, and gave an error. We still use a similar function pointer
trick for getLocalPQBuffer() function, to use a thread-local version of that
in parallel mode on Windows, but that dodges the problem because it doesn't
take printf-like arguments.
Dumping invalid indexes can cause problems at restore time, for example
if the reason the index creation failed was because it tried to enforce
a uniqueness condition not satisfied by the table's data. Also, if the
index creation is in fact still in progress, it seems reasonable to
consider it to be an uncommitted DDL change, which pg_dump wouldn't be
expected to dump anyway.
Back-patch to all active versions, and teach them to ignore invalid
indexes in servers back to 8.2, where the concept was introduced.
Michael Paquier
For getting the server's version in numeric form, use PQserverVersion().
It does the exact same parsing as dumputils.c's parse_version(), and has
been around in libpq for a long time. For the client's version, just use
the PG_VERSION_NUM constant.
A new 'starttli' field was added to the response of BASE_BACKUP command.
Make pg_basebackup tolerate the case that it's missing, so that it still
works with older servers.
Add an explicit check for the server version, so that you get a nicer error
message if you try to use it with a pre-9.1 server.
The streaming protocol message format changed in 9.3, so -X stream still won't
work with pre-9.3 servers. I added a version check to ReceiveXLogStream()
earlier, but write that slightly differently, so that in 9.4, it will still
work with a 9.3 server. (In 9.4, the error message needs to be adjusted to
"9.3 or above", though). Also, if the version check fails, don't retry.
New infrastructure is added which creates a set number of workers
(threads on Windows, forked processes on Unix). Jobs are then
handed out to these workers by the master process as needed.
pg_restore is adjusted to use this new infrastructure in place of the
old setup which created a new worker for each step on the fly. Parallel
dumps acquire a snapshot clone in order to stay consistent, if
available.
The parallel option is selected by the -j / --jobs command line
parameter of pg_dump.
Joachim Wieland, lightly editorialized by Andrew Dunstan.
Checksums are set immediately prior to flush out of shared buffers
and checked when pages are read in again. Hint bit setting will
require full page write when block is dirtied, which causes various
infrastructure changes. Extensive comments, docs and README.
WARNING message thrown if checksum fails on non-all zeroes page;
ERROR thrown but can be disabled with ignore_checksum_failure = on.
Feature enabled by an initdb option, since transition from option off
to option on is long and complex and has not yet been implemented.
Default is not to use checksums.
Checksum used is WAL CRC-32 truncated to 16-bits.
Simon Riggs, Jeff Davis, Greg Smith
Wide input and assistance from many community members. Thank you.
