The existing APIs for creating and parsing command and status messages are
rather messy; for example, archive-format modules have to provide code
for constructing command messages, which is entirely pointless since
the code to read them is hard-wired in WaitForCommands() and hence
no format-specific variation is actually possible. But there's little
foreseeable reason to need format-specific variation anyway.
The situation for status messages is no better; at least those are both
constructed and parsed by format-specific code, but said code is quite
redundant since there's no actual need for format-specific variation.
To add insult to injury, the first API involves returning pointers to
static buffers, which is bad, while the second involves returning pointers
to malloc'd strings, which is safer but randomly inconsistent.
Hence, get rid of the MasterStartParallelItem and MasterEndParallelItem
APIs, and instead write centralized functions that construct and parse
command and status messages. If we ever do need more flexibility, these
functions can be the standard implementations of format-specific
callback methods, but that's a long way off if it ever happens.
Tom Lane, reviewed by Kevin Grittner
Discussion: <17340.1464465717@sss.pgh.pa.us>
The ListenToWorkers/ReapWorkerStatus APIs were messy and hard to use.
Instead, make DispatchJobForTocEntry register a callback function that
will take care of state cleanup, doing whatever had been done by the caller
of ReapWorkerStatus in the old design. (This callback is essentially just
the old mark_work_done function in the restore case, and a trivial test for
worker failure in the dump case.) Then we can have ListenToWorkers call
the callback immediately on receipt of a status message, and return the
worker to WRKR_IDLE state; so the WRKR_FINISHED state goes away.
This allows us to design a unified wait-for-worker-messages loop:
WaitForWorkers replaces EnsureIdleWorker and EnsureWorkersFinished as well
as the mess in restore_toc_entries_parallel. Also, we no longer need the
fragile API spec that the caller of DispatchJobForTocEntry is responsible
for ensuring there's an idle worker, since DispatchJobForTocEntry can just
wait until there is one.
In passing, I got rid of the ParallelArgs struct, which was a net negative
in terms of notational verboseness, and didn't seem to be providing any
noticeable amount of abstraction either.
Tom Lane, reviewed by Kevin Grittner
Discussion: <1188.1464544443@sss.pgh.pa.us>
Formerly, Unix builds of pg_dump/pg_restore would trap SIGINT and similar
signals and set a flag that was tested in various data-transfer loops.
This was prone to errors of omission (cf commit 3c8aa6654); and even if
the client-side response was prompt, we did nothing that would cause
long-running SQL commands (e.g. CREATE INDEX) to terminate early.
Also, the master process would effectively do nothing at all upon receipt
of SIGINT; the only reason it seemed to work was that in typical scenarios
the signal would also be delivered to the child processes. We should
support termination when a signal is delivered only to the master process,
though.
Windows builds had no console interrupt handler, so they would just fall
over immediately at control-C, again leaving long-running SQL commands to
finish unmolested.
To fix, remove the flag-checking approach altogether. Instead, allow the
Unix signal handler to send a cancel request directly and then exit(1).
In the master process, also have it forward the signal to the children.
On Windows, add a console interrupt handler that behaves approximately
the same. The main difference is that a single execution of the Windows
handler can send all the cancel requests since all the info is available
in one process, whereas on Unix each process sends a cancel only for its
own database connection.
In passing, fix an old problem that DisconnectDatabase tends to send a
cancel request before exiting a parallel worker, even if nothing went
wrong. This is at least a waste of cycles, and could lead to unexpected
log messages, or maybe even data loss if it happened in pg_restore (though
in the current code the problem seems to affect only pg_dump). The cause
was that after a COPY step, pg_dump was leaving libpq in PGASYNC_BUSY
state, causing PQtransactionStatus() to report PQTRANS_ACTIVE. That's
normally harmless because the next PQexec() will silently clear the
PGASYNC_BUSY state; but in a parallel worker we might exit without any
additional SQL commands after a COPY step. So add an extra PQgetResult()
call after a COPY to allow libpq to return to PGASYNC_IDLE state.
This is a bug fix, IMO, so back-patch to 9.3 where parallel dump/restore
were introduced.
Thanks to Kyotaro Horiguchi for Windows testing and code suggestions.
Original-Patch: <7005.1464657274@sss.pgh.pa.us>
Discussion: <20160602.174941.256342236.horiguchi.kyotaro@lab.ntt.co.jp>
Parallel restore from directory format failed to respond to control-C
in a timely manner, because there were no checkAborting() calls in the
code path that reads data from a file and sends it to the backend.
If any worker was in the midst of restoring data for a large table,
you'd just have to wait.
This fix doesn't do anything for the problem of aborting a long-running
server-side command, but at least it fixes things for data transfers.
Back-patch to 9.3 where parallel restore was introduced.
It was using %u to read a string that was earlier produced by snprintf with %d
into a signed integer variable. This seems to work in practice but is
incorrect.
found by cppcheck
Rather than passing around DumpOptions and RestoreOptions as separate
arguments, add fields to struct Archive to carry pointers to these objects,
and access them through those fields when needed. There already was a
RestoreOptions pointer in Archive, though for no obvious reason it was part
of the "private" struct rather than out where pg_dump.c could see it.
Doing this allows reversion of quite a lot of parameter-addition changes
made in commit 0eea8047bf, which is a good thing IMO because this will
reduce the code delta between 9.4 and 9.5, probably easing a few future
back-patch efforts. Moreover, the previous commit only added a DumpOptions
argument to functions that had to have it at the time, which means we could
anticipate still more code churn (and more back-patch hazard) as the
requirement spread further. I'd hit exactly that problem in my upcoming
patch to fix extension membership marking, which is what motivated me to
do this.
Most pg_dump.c global variables, which were passed down individually to
dumping routines, are now grouped as members of the new DumpOptions
struct, which is used as a local variable and passed down into routines
that need it. This helps future development efforts; in particular it
is said to enable a mode in which a parallel pg_dump run can output
multiple streams, and have them restored in parallel.
Also take the opportunity to clean up the pg_dump header files somewhat,
to avoid circularity.
Author: Joachim Wieland, revised by Álvaro Herrera
Reviewed by Peter Eisentraut
Clear errno before calling readdir() and handle old MinGW errno bug
while adding full test coverage for readdir/closedir failures.
Backpatch through 8.4.
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.
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.
libpgcommon is a new static library to allow sharing code among the
various frontend programs and backend; this lets us eliminate duplicate
implementations of common routines. We avoid libpgport, because that's
intended as a place for porting issues; per discussion, it seems better
to keep them separate.
The first use case, and the only implemented by this patch, is pg_malloc
and friends, which many frontend programs were already using.
At the same time, we can use this to provide palloc emulation functions
for the frontend; this way, some palloc-using files in the backend can
also be used by the frontend cleanly. To do this, we change palloc() in
the backend to be a function instead of a macro on top of
MemoryContextAlloc(). This was previously believed to cause loss of
performance, but this implementation has been tweaked by Tom and Andres
so that on modern compilers it provides a slight improvement over the
previous one.
This lets us clean up some places that were already with
localized hacks.
Most of the pg_malloc/palloc changes in this patch were authored by
Andres Freund. Zoltán Böszörményi also independently provided a form of
that. libpgcommon infrastructure was authored by Álvaro.
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.)
The old code was using exit_horribly or die_horribly other depending on
whether it had an ArchiveHandle on which to close the connection or not;
but there were places that were passing a NULL ArchiveHandle to
die_horribly, and other places that used exit_horribly while having an
AH available. So there wasn't all that much consistency.
Improve the situation by keeping only one of the routines, and instead
of having to pass the AH down from the caller, arrange for it to be
present for an on_exit_nicely callback to operate on.
Author: Joachim Wieland
Some tweaks by me
Per a suggestion from Robert Haas, in the ongoing "parallel pg_dump"
saga.
Andrew Dunstan