Several places weren't careful about fsyncing in the way. See 1d4a0ab1
and 606e0f98 for details about required fsyncs.
This adds a couple of functions in src/common/ that have an equivalent
in the backend: durable_rename(), fsync_parent_path()
From: Michael Paquier <michael.paquier@gmail.com>
<sys/select.h> is required by POSIX.1-2001 to get the prototype of
select(2), but nearly no systems enforce that because older standards
let you get away with including some other headers. Recent OpenBSD
hacking has removed that frail touch of friendliness, however, which
broke some compiles; fix all the way back to 9.1 by adding the required
standard. Only vacuumdb.c was reported to fail, but it seems easier to
fix the whole lot in a fell swoop.
Per bug #14334 by Sean Farrell.
Much cruft had accumulated over time with a large number of parameters
passed down between functions very deep. With this refactoring, instead
introduce a StreamCtl structure that holds the parameters, and pass around
a pointer to this structure instead. This makes it much easier to add or
remove fields that are needed deeper down in the implementation without
having to modify every function header in the file.
Patch by me after much nagging from Andres
Reviewed by Craig Ringer and Daniel Gustafsson
A oversight in 2c0a485896 causes 'could not create archive status file
"...": No such file or directory' errors in pg_receivexlog if the
target directory doesn't happen to contain a archive_status
directory. That's due to a stupidly left over 'true' constant instead
of mark_done being passed down to ProcessXLogDataMsg().
The bug is only present in the master branch, and luckily wasn't
released.
Spotted by Fujii Masao.
As every error in mark_file_as_archived() will lead to a failure of
pg_basebackup the FD leak couldn't ever lead to a real problem. It
seems better to fix the leak anyway though, rather than silence
Coverity, as the usage of the function might get extended or copied at
some point in the future.
Pointed out by Coverity.
Backpatch to 9.2, like the relevant part of the previous patch.
WAL (and timeline history) files created by pg_basebackup did not
maintain the new base backup's archive status. That's currently not a
problem if the new node is used as a standby - but if that node is
promoted all still existing files can get archived again. With a high
wal_keep_segment settings that can happen a significant time later -
which is quite confusing.
Change both the backend (for the -x/-X fetch case) and pg_basebackup
(for -X stream) itself to always mark WAL/timeline files included in
the base backup as .done. That's in line with walreceiver.c doing so.
The verbosity of the pg_basebackup changes show pretty clearly that it
needs some refactoring, but that'd result in not be backpatchable
changes.
Backpatch to 9.1 where pg_basebackup was introduced.
Discussion: 20141205002854.GE21964@awork2.anarazel.de
In pg_receivexlog, in order to check whether the current WAL file is
being opened or not, its file descriptor has to be checked against -1
as an invalid value. But, oops, 7900e94 added the incorrect test
checking the descriptor against 1. This commit fixes that bug.
Back-patch to 9.4 where the bug was added.
Spotted by Magnus Hagander
When pg_receivexlog --slot is connecting to the server, at the shutdown
of the server, walsender keeps waiting for the last WAL record to be
replicated and flushed in pg_receivexlog. But previously pg_receivexlog
issued sync command only when WAL file was switched. So there was
the case where the last WAL was never flushed and walsender had to
keep waiting infinitely. This caused the server shutdown to get stuck.
pg_recvlogical handles this problem by calling fsync() when it receives
the request of immediate reply from the server. That is, at shutdown,
walsender sends the request, pg_recvlogical receives it, flushes the last
WAL record, and sends the flush location back to the server. Since
walsender can see that the last WAL record is successfully flushed, it can
exit cleanly.
This commit introduces the same logic as pg_recvlogical has,
to pg_receivexlog.
Back-patch to 9.4 where pg_receivexlog was changed so that it can use
the replication slot.
Original patch by Michael Paquier, rewritten by me.
Bug report by Furuya Osamu.
Previously pg_receivexlog flushed WAL data only when WAL file was switched.
Then 3dad73e added -F option to pg_receivexlog so that users could control
how frequently sync commands were issued to WAL files. It also allowed users
to make pg_receivexlog flush WAL data immediately after writing by
specifying 0 in -F option. However feedback messages were not sent back
immediately even after a flush location was updated. So even if WAL data
was flushed in real time, the server could not see that for a while.
This commit removes -F option from and adds --synchronous to pg_receivexlog.
If --synchronous is specified, like the standby's wal receiver, pg_receivexlog
flushes WAL data as soon as there is WAL data which has not been flushed yet.
Then it sends back the feedback message identifying the latest flush location
to the server. This option is useful to make pg_receivexlog behave as sync
standby by using replication slot, for example.
Original patch by Furuya Osamu, heavily rewritten by me.
Reviewed by Heikki Linnakangas, Alvaro Herrera and Sawada Masahiko.
This reverts commit 083d29c65b.
The commit changed the code so that it causes an errors when
IDENTIFY_SYSTEM returns three columns. But which prevents us
from using the replication-related utilities against the server
with older version. This is not what we want. For that
compatibility, we allow the utilities to receive three columns
as the result of IDENTIFY_SYSTEM eventhough it actually returns
four columns in 9.4 or later.
Pointed out by Andres Freund.
5a991ef869 added new column into
the result of IDENTIFY_SYSTEM command. But it was not reflected into
several codes checking that result. Specifically though the number of
columns in the result was increased to 4, it was still compared with 3
in some replication codes.
Back-patch to 9.4 where the number of columns in IDENTIFY_SYSTEM
result was increased.
Report from Michael Paquier
This allows us to specify the maximum time to issue fsync to ensure
the received WAL file is safely flushed to disk. Without this,
pg_receivexlog always flushes WAL file only when it's closed and
which can cause WAL data to be lost at the event of a crash.
Furuya Osamu, heavily modified by me.
Previously the source codes for processing the received data and handling
the end of stream were included in pg_receivexlog main loop. This commit
splits out them as separate functions. This is useful for improving the
readability of main loop code and making the future pg_receivexlog-related
patch simpler.
In 9.2, pg_receivexlog with verbose option has emitted the messages
at the end of each WAL file. But the commit 0b63291 suppressed such
messages by mistake. This commit fixes the bug so that pg_receivexlog
--verbose outputs such messages again.
Back-patch to 9.3 where the bug was added.
The problem is that pg_receivexlog calls select(2) with timeout=0 and
goes into busy loop when --status-interval option is set to 0. This bug
was introduced by the commit,
74cbe966fe.
Per report from Sawada Masahiko
Previously the source codes for receiving the data and for
polling the socket were included in pg_receivexlog main loop.
This commit splits out them as separate functions. This is
useful for improving the readability of main loop code and
making the future pg_receivexlog-related patch simpler.
The leak is fairly small and rare, but a leak nevertheless.
Per Coverity report. Backpatch to 9.2, where pg_receivexlog was added.
pg_basebackup shares the code, but it always exits on error, so there is
no real leak.
Commit d298b50a3b by Heikki Linnakangas
requested that the version check message be updated at next release, suggesting
that the appropriate text would be “9.3 or later”. The logic used for the check
indicates that the correct text for 9.4 is “9.3 or 9.4”, since the logic would
cause this to fail for later releases.
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.
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 is a similar fix as c6ec8793aa
9.2. This should never happen in 9.3 and newer since the special case
cannot happen there, but this patch synchronizes up the code so there
is no confusion on why they're different. An empty block is as harmless
in 9.3 as it was in 9.2, and can safely be ignored.
When a timeline history file is fetched from server, it is initially created
with a temporary file name, and renamed to place. However, the temporary
file name was constructed using an uninitialized buffer. Usually that meant
that the file was created in current directory instead of the target, which
usually goes unnoticed, but if the target is on a different filesystem than
the current dir, the rename() would fail. Fix that.
The second issue is that pg_receivexlog would not take .partial files into
account when determining when scanning the target directory for existing
WAL files. If the timeline has switched in the server several times in the
last WAL segment, and pg_receivexlog is restarted, it would choose a too
old starting point. That's not a problem as long as the old WAL segment
exists in the server and can be streamed over, but will cause a failure if
it's not.
Backpatch to 9.3, where this timeline handling code was written.
Analysed by Andrew Gierth, bug #8453, based on a bug report on IRC.
In receivelog.c:writeTimeLineHistoryFile(), we were not properly
closing the open'd file descriptor in error cases. While this
wouldn't matter much if we were about to exit due to such an
error, that's not the case with pg_receivexlog as it can be a
long-running process and these errors are non-fatal.
This resource leak was found by the Coverity scanner.
Back-patch to 9.3 where this issue first appeared.
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, 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.
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.
This mirrors the changes done earlier to the server in standby mode. When
receivelog reaches the end of a timeline, as reported by the server, it
fetches the timeline history file of the next timeline, and restarts
streaming from the new timeline by issuing a new START_STREAMING command.
When pg_receivexlog crosses a timeline, it leaves the .partial suffix on the
last segment on the old timeline. This helps you to tell apart a partial
segment left in the directory because of a timeline switch, and a completed
segment. If you just follow a single server, it won't make a difference, but
it can be significant in more complicated scenarios where new WAL is still
generated on the old timeline.
This includes two small changes to the streaming replication protocol:
First, when you reach the end of timeline while streaming, the server now
sends the TLI of the next timeline in the server's history to the client.
pg_receivexlog uses that as the next timeline, so that it doesn't need to
parse the timeline history file like a standby server does. Second, when
BASE_BACKUP command sends the begin and end WAL positions, it now also sends
the timeline IDs corresponding the positions.
This gets rid of XLByteLT, XLByteLE, XLByteEQ and XLByteAdvance.
These were useful for brevity when XLogRecPtrs were split in
xlogid/xrecoff; but now that they are simple uint64's, they are just
clutter. The only downside to making this change would be ease of
backporting patches, but that has been negated by other substantive
changes to the involved code anyway. The clarity of simpler expressions
makes the change worthwhile.
Most of the changes are mechanical, but in a couple of places, the patch
author chose to invert the operator sense, making the code flow more
logical (and more in line with preceding comments).
Author: Andres Freund
Eyeballed by Dimitri Fontaine and Alvaro Herrera
Without this, the connection will be killed after timeout if
wal_sender_timeout is set in the server.
Original patch by Amit Kapila, modified by me to fit recent changes in the
code.
We used to send structs wrapped in CopyData messages, which works as long as
the client and server agree on things like endianess, timestamp format and
alignment. That's good enough for running a standby server, which has to run
on the same platform anyway, but it's useful for tools like pg_receivexlog
to work across platforms.
This breaks protocol compatibility of streaming replication, but we never
promised that to be compatible across versions, anyway.
We had a number of variants on the theme of "malloc or die", with the
majority named like "pg_malloc", but by no means all. Standardize on the
names pg_malloc, pg_malloc0, pg_realloc, pg_strdup. Get rid of pg_calloc
entirely in favor of using pg_malloc0.
This is an essentially cosmetic change, so no back-patch. (I did find
a couple of places where psql and pg_dump were using plain malloc or
strdup instead of the pg_ versions, but they don't look significant
enough to bother back-patching.)
timeval.t_sec is of type time_t, which is not always compatible with long.
I'm not sure if this was just harmless warning or a real bug, but this
fixes it, anyway.
In particular, with a controlled shutdown of the master, pg_basebackup
with streaming log could terminate without an error message, even though
the backup is not consistent.
In passing, fix a few cases where walfile wasn't properly set to -1 after
closing.
Fujii Masao
The most user-visible part of this is to change the long options
--statusint and --noloop to --status-interval and --no-loop,
respectively, per discussion.
Also, consistently enclose file names in double quotes, per our
conventions; and consistently use the term "transaction log file" to
talk about WAL segments. (Someday we may need to go over this
terminology and make it consistent across the whole source code.)
Finally, reflow the code to better fit in 80 columns, and have pgindent
fix it up some more.
When the internal loop mode was added, freeing memory and closing
filedescriptors before returning became important, and a few cases
in the code missed that.
Fujii Masao
This makes it possible for the master to track how much data has
actually been written my pg_receivexlog - and not just how much
has been sent towards it.
This simplifies code that needs to do arithmetic on XLogRecPtrs.
To avoid changing on-disk format of data pages, the LSN on data pages is
still stored in the old format. That should keep pg_upgrade happy. However,
we have XLogRecPtrs embedded in the control file, and in the structs that
are sent over the replication protocol, so this changes breaks compatibility
of pg_basebackup and server. I didn't do anything about this in this patch,
per discussion on -hackers, the right thing to do would to be to change the
replication protocol to be architecture-independent, so that you could use
a newer version of pg_receivexlog, for example, against an older server
version.
The comments claimed that wasting the last segment made it easier to do
calculations with XLogRecPtrs, because you don't have problems representing
last-byte-position-plus-1 that way. In my experience, however, it only made
things more complicated, because the there was two ways to represent the
boundary at the beginning of a logical log file: logid = n+1 and xrecoff = 0,
or as xlogid = n and xrecoff = 4GB - XLOG_SEG_SIZE. Some functions were
picky about which representation was used.
Also, use a 64-bit segment number instead of the log/seg combination, to
point to a certain WAL segment. We assume that all platforms have a working
64-bit integer type nowadays.
This is an incompatible change in WAL format, so bumping WAL version number.
Since the replication protocol deals with TimestampTz, we need to
care for the floating point case as well in the frontend tools.
Fujii Masao, with changes from Magnus Hagander
When backing up from a standby server, the backup process
will not automatically switch xlog segment. So we must
accept a partially transferred xlog file in this case, but
rename it into position anyway.
In passing, merge the two callbacks for segment end and
stop stream into a single callback, since their implementations
were close to identical, and rename this callback to
reflect that it stops streaming rather than continues it.
Patch by Magnus Hagander, review by Fujii Masao
Teach pg_basebackup in streaming mode to deal with keepalive messages.
Also change the order of checks to complain at the message rather than
block size when a new message is introduced.
In passing, switch to using sizeof() instead of hardcoded sizes for
WAL protocol structs.
Instead of filling files as they appear, pre-pad the
WAL files received when streaming xlog the same way
that the server does. Data is streamed into a .partial
file which is then renamed()d into palce when it's complete,
but it will always be 16MB.
This also means that the starting position for pg_receivexlog
is now simply right after the last complete segment, and we
never need to deal with partial segments there.
Patch by me, review by Fujii Masao
Add option for parallel streaming of the transaction log while a
base backup is running, to get the logfiles before the server has
removed them.
Also add a tool called pg_receivexlog, which streams the transaction
log into files, creating a log archive without having to wait for
segments to complete, thus decreasing the window of data loss without
having to waste space using archive_timeout. This works best in
combination with archive_command - suggested usage docs etc coming later.