Until now, our Serializable mode has in fact been what's called Snapshot
Isolation, which allows some anomalies that could not occur in any
serialized ordering of the transactions. This patch fixes that using a
method called Serializable Snapshot Isolation, based on research papers by
Michael J. Cahill (see README-SSI for full references). In Serializable
Snapshot Isolation, transactions run like they do in Snapshot Isolation,
but a predicate lock manager observes the reads and writes performed and
aborts transactions if it detects that an anomaly might occur. This method
produces some false positives, ie. it sometimes aborts transactions even
though there is no anomaly.
To track reads we implement predicate locking, see storage/lmgr/predicate.c.
Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared
memory is finite, so when a transaction takes many tuple-level locks on a
page, the locks are promoted to a single page-level lock, and further to a
single relation level lock if necessary. To lock key values with no matching
tuple, a sequential scan always takes a relation-level lock, and an index
scan acquires a page-level lock that covers the search key, whether or not
there are any matching keys at the moment.
A predicate lock doesn't conflict with any regular locks or with another
predicate locks in the normal sense. They're only used by the predicate lock
manager to detect the danger of anomalies. Only serializable transactions
participate in predicate locking, so there should be no extra overhead for
for other transactions.
Predicate locks can't be released at commit, but must be remembered until
all the transactions that overlapped with it have completed. That means that
we need to remember an unbounded amount of predicate locks, so we apply a
lossy but conservative method of tracking locks for committed transactions.
If we run short of shared memory, we overflow to a new "pg_serial" SLRU
pool.
We don't currently allow Serializable transactions in Hot Standby mode.
That would be hard, because even read-only transactions can cause anomalies
that wouldn't otherwise occur.
Serializable isolation mode now means the new fully serializable level.
Repeatable Read gives you the old Snapshot Isolation level that we have
always had.
Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and
Anssi Kääriäinen
relation using the general PARAM_EXEC executor parameter mechanism, rather
than the ad-hoc kluge of passing the outer tuple down through ExecReScan.
The previous method was hard to understand and could never be extended to
handle parameters coming from multiple join levels. This patch doesn't
change the set of possible plans nor have any significant performance effect,
but it's necessary infrastructure for future generalization of the concept
of an inner indexscan plan.
ExecReScan's second parameter is now unused, so it's removed.
a lot of strange behaviors that occurred in join cases. We now identify the
"current" row for every joined relation in UPDATE, DELETE, and SELECT FOR
UPDATE/SHARE queries. If an EvalPlanQual recheck is necessary, we jam the
appropriate row into each scan node in the rechecking plan, forcing it to emit
only that one row. The former behavior could rescan the whole of each joined
relation for each recheck, which was terrible for performance, and what's much
worse could result in duplicated output tuples.
Also, the original implementation of EvalPlanQual could not re-use the recheck
execution tree --- it had to go through a full executor init and shutdown for
every row to be tested. To avoid this overhead, I've associated a special
runtime Param with each LockRows or ModifyTable plan node, and arranged to
make every scan node below such a node depend on that Param. Thus, by
signaling a change in that Param, the EPQ machinery can just rescan the
already-built test plan.
This patch also adds a prohibition on set-returning functions in the
targetlist of SELECT FOR UPDATE/SHARE. This is needed to avoid the
duplicate-output-tuple problem. It seems fairly reasonable since the
other restrictions on SELECT FOR UPDATE are meant to ensure that there
is a unique correspondence between source tuples and result tuples,
which an output SRF destroys as much as anything else does.
corresponding struct definitions. This allows other headers to avoid including
certain highly-loaded headers such as rel.h and relscan.h, instead using just
relcache.h, heapam.h or genam.h, which are more lightweight and thus cause less
unnecessary dependencies.
ps_TupFromTlist in plan nodes that make use of it. This was being done
correctly in join nodes and Result nodes but not in any relation-scan nodes.
Bug would lead to bogus results if a set-returning function appeared in the
targetlist of a subquery that could be rescanned after partial execution,
for example a subquery within EXISTS(). Bug has been around forever :-(
... surprising it wasn't reported before.
by creating a reference-count mechanism, similar to what we did a long time
ago for catcache entries. The back branches have an ugly solution involving
lots of extra copies, but this way is more efficient. Reference counting is
only applied to tupdescs that are actually in caches --- there seems no need
to use it for tupdescs that are generated in the executor, since they'll go
away during plan shutdown by virtue of being in the per-query memory context.
Neil Conway and Tom Lane
bits indicating which optional capabilities can actually be exercised
at runtime. This will allow Sort and Material nodes, and perhaps later
other nodes, to avoid unnecessary overhead in common cases.
This commit just adds the infrastructure and arranges to pass the correct
flag values down to plan nodes; none of the actual optimizations are here
yet. I'm committing this separately in case anyone wants to measure the
added overhead. (It should be negligible.)
Simon Riggs and Tom Lane
relation if it's already been locked by execMain.c as either a result
relation or a FOR UPDATE/SHARE relation. This avoids an extra trip to
the shared lock manager state. Per my suggestion yesterday.
a TupleTableSlot: instead of calling ExecClearTuple, inline the needed
operations, so that we can avoid redundant steps. In particular, when
the old and new tuples are both on the same disk page, avoid releasing
and re-acquiring the buffer pin --- this saves work in both the bufmgr
and ResourceOwner modules. To make this improvement actually useful,
partially revert a change I made on 2004-04-21 that caused SeqNext
et al to call ExecClearTuple before ExecStoreTuple. The motivation
for that, to avoid grabbing the BufMgrLock separately for releasing
the old buffer and grabbing the new one, no longer applies. My
profiling says that this saves about 5% of the CPU time for an
all-in-memory seqscan.
in an inconsistent state. (This is only latent because in reality
ExecSeqRestrPos is dead code at the moment ... but someday maybe it won't
be.) Add some comments about what the API for plan node mark/restore
actually is, because it's not immediately obvious.
of tuples when passing data up through multiple plan nodes. A slot can now
hold either a normal "physical" HeapTuple, or a "virtual" tuple consisting
of Datum/isnull arrays. Upper plan levels can usually just copy the Datum
arrays, avoiding heap_formtuple() and possible subsequent nocachegetattr()
calls to extract the data again. This work extends Atsushi Ogawa's earlier
patch, which provided the key idea of adding Datum arrays to TupleTableSlots.
(I believe however that something like this was foreseen way back in Berkeley
days --- see the old comment on ExecProject.) A test case involving many
levels of join of fairly wide tables (about 80 columns altogether) showed
about 3x overall speedup, though simple queries will probably not be
helped very much.
I have also duplicated some code in heaptuple.c in order to provide versions
of heap_formtuple and friends that use "bool" arrays to indicate null
attributes, instead of the old convention of "char" arrays containing either
'n' or ' '. This provides a better match to the convention used by
ExecEvalExpr. While I have not made a concerted effort to get rid of uses
of the old routines, I think they should be deprecated and eventually removed.
the next are handled by ReleaseAndReadBuffer rather than separate
ReleaseBuffer and ReadBuffer calls. This cuts the number of acquisitions
of the BufMgrLock by a factor of 2 (possibly more, if an indexscan happens
to pull successive rows from the same heap page). Unfortunately this
doesn't seem enough to get us out of the recently discussed context-switch
storm problem, but it's surely worth doing anyway.
nodes where it's not really necessary. In many cases where the scan node
is not the topmost plan node (eg, joins, aggregation), it's possible to
just return the table tuple directly instead of generating an intermediate
projection tuple. In preliminary testing, this reduced the CPU time
needed for 'SELECT COUNT(*) FROM foo' by about 10%.
a per-query memory context created by CreateExecutorState --- and destroyed
by FreeExecutorState. This provides a final solution to the longstanding
problem of memory leaked by various ExecEndNode calls.
execution state trees, and ExecEvalExpr takes an expression state tree
not an expression plan tree. The plan tree is now read-only as far as
the executor is concerned. Next step is to begin actually exploiting
this property.
to plan nodes, not vice-versa. All executor state nodes now inherit from
struct PlanState. Copying of plan trees has been simplified by not
storing a list of SubPlans in Plan nodes (eliminating duplicate links).
The executor still needs such a list, but it can build it during
ExecutorStart since it has to scan the plan tree anyway.
No initdb forced since no stored-on-disk structures changed, but you
will need a full recompile because of node-numbering changes.
yesterday's proposal to pghackers. Also remove unnecessary parameters
to heap_beginscan, heap_rescan. I modified pg_proc.h to reflect the
new numbers of parameters for the AM interface routines, but did not
force an initdb because nothing actually looks at those fields.
Improve 'pg_internal.init' relcache entry preload mechanism so that it is
safe to use for all system catalogs, and arrange to preload a realistic
set of system-catalog entries instead of only the three nailed-in-cache
indexes that were formerly loaded this way. Fix mechanism for deleting
out-of-date pg_internal.init files: this must be synchronized with transaction
commit, not just done at random times within transactions. Drive it off
relcache invalidation mechanism so that no special-case tests are needed.
Cache additional information in relcache entries for indexes (their pg_index
tuples and index-operator OIDs) to eliminate repeated lookups. Also cache
index opclass info at the per-opclass level to avoid repeated lookups during
relcache load.
Generalize 'systable scan' utilities originally developed by Hiroshi,
move them into genam.c, use in a number of places where there was formerly
ugly code for choosing either heap or index scan. In particular this allows
simplification of the logic that prevents infinite recursion between syscache
and relcache during startup: we can easily switch to heapscans in relcache.c
when and where needed to avoid recursion, so IndexScanOK becomes simpler and
does not need any expensive initialization.
Eliminate useless opening of a heapscan data structure while doing an indexscan
(this saves an mdnblocks call and thus at least one kernel call).
per previous discussion on pghackers. Most of the duplicate code in
different AMs' ambuild routines has been moved out to a common routine
in index.c; this means that all index types now do the right things about
inserting recently-dead tuples, etc. (I also removed support for EXTEND
INDEX in the ambuild routines, since that's about to go away anyway, and
it cluttered the code a lot.) The retail indextuple deletion routines have
been replaced by a "bulk delete" routine in which the indexscan is inside
the access method. I haven't pushed this change as far as it should go yet,
but it should allow considerable simplification of the internal bookkeeping
for deletions. Also, add flag columns to pg_am to eliminate various
hardcoded tests on AM OIDs, and remove unused pg_am columns.
Fix rtree and gist index types to not attempt to store NULLs; before this,
gist usually crashed, while rtree managed not to crash but computed wacko
bounding boxes for NULL entries (which might have had something to do with
the performance problems we've heard about occasionally).
Add AtEOXact routines to hash, rtree, and gist, all of which have static
state that needs to be reset after an error. We discovered this need long
ago for btree, but missed the other guys.
Oh, one more thing: concurrent VACUUM is now the default.
allocated by plan nodes are not leaked at end of query. This doesn't
really matter for normal queries, but it sure does for queries invoked
repetitively inside SQL functions. Clean up some other grotty code
associated with tupdescs, and fix a few other memory leaks exposed by
tests with simple SQL functions.
joins, and clean things up a good deal at the same time. Append plan node
no longer hacks on rangetable at runtime --- instead, all child tables are
given their own RT entries during planning. Concept of multiple target
tables pushed up into execMain, replacing bug-prone implementation within
nodeAppend. Planner now supports generating Append plans for inheritance
sets either at the top of the plan (the old way) or at the bottom. Expanding
at the bottom is appropriate for tables used as sources, since they may
appear inside an outer join; but we must still expand at the top when the
target of an UPDATE or DELETE is an inheritance set, because we actually need
a different targetlist and junkfilter for each target table in that case.
Fortunately a target table can't be inside an outer join... Bizarre mutual
recursion between union_planner and prepunion.c is gone --- in fact,
union_planner doesn't really have much to do with union queries anymore,
so I renamed it grouping_planner.
Heikki Linnakangas