If we have Limit->Result->Sort, the Result might be projecting a tlist
that contains a set-returning function. If so, it's possible for the
SRF to sometimes return zero rows, which means we could need to fetch
more than N rows from the Sort in order to satisfy LIMIT N.
So top-N sorting cannot be used in this scenario.
Fix things so that top-N sorting can be used in child Sort nodes of a
MergeAppend node, when there is a LIMIT and no intervening joins or
grouping. Actually doing this on the executor side isn't too bad,
but it's a bit messier to get the planner to cost it properly.
Per gripe from Robert Haas.
In passing, fix an oversight in the original top-N-sorting patch:
query_planner should not assume that a LIMIT can be used to make an
explicit sort cheaper when there will be grouping or aggregation in
between. Possibly this should be back-patched, but I'm not sure the
mistake is serious enough to be a real problem in practice.
In the previous coding, we simply issued ALTER SEQUENCE RESTART commands,
which do not roll back on error. This meant that an error between
truncating and committing left the sequences out of sync with the table
contents, with potentially bad consequences as were noted in a Warning on
the TRUNCATE man page.
To fix, create a new storage file (relfilenode) for a sequence that is to
be reset due to RESTART IDENTITY. If the transaction aborts, we'll
automatically revert to the old storage file. This acts just like a
rewriting ALTER TABLE operation. A penalty is that we have to take
exclusive lock on the sequence, but since we've already got exclusive lock
on its owning table, that seems unlikely to be much of a problem.
The interaction of this with usual nontransactional behaviors of sequence
operations is a bit weird, but it's hard to see what would be completely
consistent. Our choice is to discard cached-but-unissued sequence values
both when the RESTART is executed, and at rollback if any; but to not touch
the currval() state either time.
In passing, move the sequence reset operations to happen before not after
any AFTER TRUNCATE triggers are fired. The previous ordering was not
logically sensible, but was forced by the need to minimize inconsistency
if the triggers caused an error. Transactional rollback is a much better
solution to that.
Patch by Steve Singer, rather heavily adjusted by me.
Add some additional dependencies to constrain the build order to prevent
parallel make from failing. In the case of src/Makefile, this is likely to be
too complicated to be worth maintaining, so just add .NOTPARALLEL to get the
old for-loop-like behavior.
More fine-tuning might be necessary for some platforms or configurations.
The handle to the shared memory segment containing startup
parameters was sent as 32-bit even on 64-bit systems. Since
HANDLEs appear to be allocated sequentially this shouldn't
be a problem until we reach 2^32 open handles in the postmaster,
but a 64-bit value should be sent across as 64-bit, and not
zero out the top 32 bits.
Noted by Tom Lane.
temporary indexes are not WAL-logged. We used a constant LSN for temporary
indexes, on the assumption that we don't need to worry about concurrent page
splits in temporary indexes because they're only visible to the current
session. But that assumption is wrong, it's possible to insert rows and
split pages in the same session, while a scan is in progress. For example,
by opening a cursor and fetching some rows, and INSERTing new rows before
fetching some more.
Fix by generating fake increasing LSNs, used in place of real LSNs in
temporary GiST indexes.
We must stay in the function's SPI context until done calling the iterator
that returns the set result. Otherwise, any attempt to invoke SPI features
in the python code called by the iterator will malfunction. Diagnosis and
patch by Jan Urbanski, per bug report from Jean-Baptiste Quenot.
Back-patch to 8.2; there was no support for SRFs in previous versions of
plpython.
This new field counts the number of times that a backend which writes a
buffer out to the OS must also fsync() it. This happens when the
bgwriter fsync request queue is full, and is generally detrimental to
performance, so it's good to know when it's happening. Along the way,
log a new message at level DEBUG1 whenever we fail to hand off an fsync,
so that the problem can also be seen in examination of log files
(if the logging level is cranked up high enough).
Greg Smith, with minor tweaks by me.
Similar conflicts were already avoided for related record types.
Massive over-caution resulted in a usability bug. Clear theoretical
basis for doing this is now confirmed by me.
Request to remove from Heikki (twice), over-caution by me.
We must not return any "okay to proceed" result code without having checked
for too many children, else we might fail later on when trying to add the
new child to one of the per-child state arrays. It's not clear whether
this oversight explains Stefan Kaltenbrunner's recent report, but it could
certainly produce a similar symptom.
Back-patch to 8.4; the logic was not broken before that.
3.80 breaks if the expansion of $(eval) is long enough to require expansion
of its internal variable_buffer. For the purposes of $(recurse) that means
it'll work so long as no single evaluation of _create_recursive_target
produces more than 195 bytes. We can manage that by looping over
subdirectories outside the call instead of complicating the generated rule.
This coding is simpler and more readable anyway.
Or at least, this works for me. We'll see if the buildfarm likes it.
This is needed to support debug_print_parse, per report from Jon Nelson.
Cursory testing via the regression tests suggests we aren't missing
anything else.
Having this in src/include/port.h makes no sense, now that copydir.c lives
in src/backend/strorage rather than src/port. Along the way, remove an
obsolete comment from contrib/pg_upgrade that makes reference to the old
location.
Once we have found a non-null constant argument, there is no need to
examine additional arguments of the COALESCE. The previous coding got it
right only if the constant was in the first argument position; otherwise
it tried to simplify following arguments too, leading to unexpected
behavior like this:
regression=# select coalesce(f1, 42, 1/0) from int4_tbl;
ERROR: division by zero
It's a minor corner case, but a bug is a bug, so back-patch all the way.
Replace for loops in makefiles with proper dependencies. Parallel
make can now span across directories. Also, make -k and make -q work
properly.
GNU make 3.80 or newer is now required.
belonging to a user at DROP OWNED BY. Foreign data wrappers and servers
don't do anything useful yet, which is why no-one has noticed, but since we
have them, seems prudent to fix this. Per report from Chetan Suttraway.
Backpatch to 9.0, 8.4 has the same problem but this patch didn't apply
there so I'm not going to bother.
location read from backup label file can be found: wasShutdown was set
incorrectly when a backup label file was found.
Jeff Davis, with a little tweaking by me.
This code was just plain wrong: what you got was not a line through the
given point but a line almost indistinguishable from the Y-axis, although
not truly vertical. The only caller that tries to use this function with
m == DBL_MAX is dist_ps_internal for the case where the lseg is horizontal;
it would end up producing the distance from the given point to the place
where the lseg's line crosses the Y-axis. That function is used by other
operators too, so there are several operators that could compute wrong
distances from a line segment to something else. Per bug #5745 from
jindiax.
Back-patch to all supported branches.
Explicitly document that the -o options of pg_ctl init mode are meant
for initdb, not postgres (Euler Taveira de Oliveira). Assorted other
copy-editing (Tom).
The general design of memory management in Postgres is that intermediate
results computed by an expression are not freed until the end of the tuple
cycle. For expression indexes, ANALYZE has to re-evaluate each expression
for each of its sample rows, and it wasn't bothering to free intermediate
results until the end of processing of that index. This could lead to very
substantial leakage if the intermediate results were large, as in a recent
example from Jakub Ouhrabka. Fix by doing ResetExprContext for each sample
row. This necessitates adding a datumCopy step to ensure that the final
expression value isn't recycled too. Some quick testing suggests that this
change adds at worst about 10% to the time needed to analyze a table with
an expression index; which is annoying, but seems a tolerable price to pay
to avoid unexpected out-of-memory problems.
Back-patch to all supported branches.
length stored in the line pointer the same way it's calculated in the normal
heap_insert() codepath. As noted by Jeff Davis, the length stored by
raw_heap_insert() included padding but the one stored by the normal codepath
did not. While the mismatch seems to be harmless, inconsistency isn't good,
and the normal codepath has received a lot more testing over the years.
Backpatch to 8.3 where the heap rewrite code was introduced.
The original coding in FileClose() reset the file-is-temp flag before
unlinking the file, so that if control came back through due to an error,
it wouldn't try to unlink the file twice. This was correct when written,
but when the log_temp_files feature was added, the logging action was put
in between those two steps. An error occurring during the logging action
--- such as a query cancel --- would result in the unlink not getting done
at all, as in recent report from Michael Glaesemann.
To fix this, make sure that we do both the stat and the unlink before doing
anything that could conceivably CHECK_FOR_INTERRUPTS. There is a judgment
call here, which is which log message to emit first: if you can see only
one, which should it be? I chose to log unlink failure at the risk of
losing the log_temp_files log message --- after all, if the unlink does
fail, the temp file is still there for you to see.
Back-patch to all versions that have log_temp_files. The code was OK
before that.
get_database_list was uselessly allocating its output data, along some
created along the way, in a permanent memory context. This didn't
matter when autovacuum was a single, short-lived process, but now that
the launcher is permanent, it shows up as a permanent leak.
To fix, make get_database list allocate its output data in the caller's
context, which is in charge of freeing it when appropriate; and the
memory leaked by heap_beginscan et al is allocated in a throwaway
transaction context.
Formerly, we could convert a UNION ALL structure inside a subquery-in-FROM
into an appendrel, as a side effect of pulling up the subquery into its
parent; but top-level UNION ALL always caused use of plan_set_operations().
That didn't matter too much because you got an Append-based plan either
way. However, now that the appendrel code can do things with MergeAppend,
it's worthwhile to hack up the top-level case so it also uses appendrels.
This is a bit of a stopgap; but going much further than this will require
a major rewrite of the planner's set-operations support, which I'm not
prepared to undertake now. For the moment let's grab the low-hanging fruit.
PG 8.4 added a built-in feature for casting pretty much any data type to
string types (text, varchar, etc). We allowed this to work in any of the
historically-allowed syntaxes: CAST(x AS text), x::text, text(x), or
x.text. However, multiple complaints have shown that it's too easy to
invoke such casts unintentionally in the latter two styles, particularly
field selection. To cure the problem with the narrowest possible change
of behavior, disallow use of I/O conversion casts from composite types to
string types via functional/attribute syntax. The new functionality is
still available via cast syntax.
In passing, document the equivalence of functional and attribute syntax
in a more visible place.
\dn without "S" now hides all pg_XXX schemas as well as information_schema.
Thus, in a bare database you'll only see "public". ("public" is considered
a user schema, not a system schema, mainly because it's droppable.)
Per discussion back in late September.
Per recent investigation, the register stack can grow faster than the
regular stack depending on compiler and choice of options. To avoid
crashes we must check both stacks in check_stack_depth().
Since this is poorly-tested code, committing only to HEAD for the
moment ... but we might want to consider back-patching later.
Rather than considering this result as meaning "unknown", report LONG_MAX.
This won't change what superusers can set max_stack_depth to, but it will
cause InitializeGUCOptions() to set the built-in default to 2MB not 100kB.
The latter seems like a fairly unreasonable interpretation of "infinity".
Per my investigation of odd buildfarm results as well as an old complaint
from Heikki.
Since this should persuade all the buildfarm animals to use a reasonable
stack depth setting during "make check", revert previous patch that dumbed
down a recursive regression test to only 5 levels.
I'm mainly interested in finding out what it is on buildfarm machines,
but including the active value in the message seems like good practice
in any case. Add the info to the HINT, not the ERROR string, so as not
to change the regression tests' expected output.
The nominally equivalent call appendStringInfo(buf, "%s", str) can be
significantly slower when str is large. In particular, the former usage in
EVALUATE_MESSAGE led to O(N^2) behavior when collecting a large number of
context lines, as I found out while testing recursive functions. The other
changes are just neatnik-ism and seem unlikely to save anything meaningful,
but a cycle shaved is a cycle earned.
Per my recent proposal, get rid of all the direct inspection of indexes
and manual generation of paths in planagg.c. Instead, set up
EquivalenceClasses for the aggregate argument expressions, and let the
regular path generation logic deal with creating paths that can satisfy
those sort orders. This makes planagg.c a bit more visible to the rest
of the planner than it was originally, but the approach is basically a lot
cleaner than before. A major advantage of doing it this way is that we get
MIN/MAX optimization on inheritance trees (using MergeAppend of indexscans)
practically for free, whereas in the old way we'd have had to add a whole
lot more duplicative logic.
One small disadvantage of this approach is that MIN/MAX aggregates can no
longer exploit partial indexes having an "x IS NOT NULL" predicate, unless
that restriction or something that implies it is specified in the query.
The previous implementation was able to use the added "x IS NOT NULL"
condition as an extra predicate proof condition, but in this version we
rely entirely on indexes that are considered usable by the main planning
process. That seems a fair tradeoff for the simplicity and functionality
gained.