2016-04-01 21:42:24 +08:00
|
|
|
/*-------------------------------------------------------------------------
|
|
|
|
*
|
|
|
|
* blinsert.c
|
|
|
|
* Bloom index build and insert functions.
|
|
|
|
*
|
2018-01-03 12:30:12 +08:00
|
|
|
* Copyright (c) 2016-2018, PostgreSQL Global Development Group
|
2016-04-01 21:42:24 +08:00
|
|
|
*
|
|
|
|
* IDENTIFICATION
|
|
|
|
* contrib/bloom/blinsert.c
|
|
|
|
*
|
|
|
|
*-------------------------------------------------------------------------
|
|
|
|
*/
|
|
|
|
#include "postgres.h"
|
|
|
|
|
|
|
|
#include "access/genam.h"
|
|
|
|
#include "access/generic_xlog.h"
|
|
|
|
#include "catalog/index.h"
|
|
|
|
#include "miscadmin.h"
|
|
|
|
#include "storage/bufmgr.h"
|
|
|
|
#include "storage/indexfsm.h"
|
2016-05-25 09:04:23 +08:00
|
|
|
#include "storage/smgr.h"
|
2016-04-01 21:42:24 +08:00
|
|
|
#include "utils/memutils.h"
|
|
|
|
#include "utils/rel.h"
|
|
|
|
|
|
|
|
#include "bloom.h"
|
|
|
|
|
|
|
|
PG_MODULE_MAGIC;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* State of bloom index build. We accumulate one page data here before
|
|
|
|
* flushing it to buffer manager.
|
|
|
|
*/
|
|
|
|
typedef struct
|
|
|
|
{
|
|
|
|
BloomState blstate; /* bloom index state */
|
2016-06-10 06:02:36 +08:00
|
|
|
MemoryContext tmpCtx; /* temporary memory context reset after each
|
|
|
|
* tuple */
|
2016-04-01 21:42:24 +08:00
|
|
|
char data[BLCKSZ]; /* cached page */
|
|
|
|
int64 count; /* number of tuples in cached page */
|
2016-06-10 06:02:36 +08:00
|
|
|
} BloomBuildState;
|
2016-04-01 21:42:24 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Flush page cached in BloomBuildState.
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
flushCachedPage(Relation index, BloomBuildState *buildstate)
|
|
|
|
{
|
|
|
|
Page page;
|
|
|
|
Buffer buffer = BloomNewBuffer(index);
|
|
|
|
GenericXLogState *state;
|
|
|
|
|
|
|
|
state = GenericXLogStart(index);
|
2016-04-12 23:42:06 +08:00
|
|
|
page = GenericXLogRegisterBuffer(state, buffer, GENERIC_XLOG_FULL_IMAGE);
|
2016-04-01 21:42:24 +08:00
|
|
|
memcpy(page, buildstate->data, BLCKSZ);
|
|
|
|
GenericXLogFinish(state);
|
|
|
|
UnlockReleaseBuffer(buffer);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* (Re)initialize cached page in BloomBuildState.
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
initCachedPage(BloomBuildState *buildstate)
|
|
|
|
{
|
|
|
|
memset(buildstate->data, 0, BLCKSZ);
|
|
|
|
BloomInitPage(buildstate->data, 0);
|
|
|
|
buildstate->count = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Per-tuple callback from IndexBuildHeapScan.
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
bloomBuildCallback(Relation index, HeapTuple htup, Datum *values,
|
|
|
|
bool *isnull, bool tupleIsAlive, void *state)
|
|
|
|
{
|
|
|
|
BloomBuildState *buildstate = (BloomBuildState *) state;
|
|
|
|
MemoryContext oldCtx;
|
|
|
|
BloomTuple *itup;
|
|
|
|
|
|
|
|
oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
|
|
|
|
|
|
|
|
itup = BloomFormTuple(&buildstate->blstate, &htup->t_self, values, isnull);
|
|
|
|
|
|
|
|
/* Try to add next item to cached page */
|
|
|
|
if (BloomPageAddItem(&buildstate->blstate, buildstate->data, itup))
|
|
|
|
{
|
|
|
|
/* Next item was added successfully */
|
|
|
|
buildstate->count++;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
/* Cached page is full, flush it out and make a new one */
|
|
|
|
flushCachedPage(index, buildstate);
|
|
|
|
|
|
|
|
CHECK_FOR_INTERRUPTS();
|
|
|
|
|
|
|
|
initCachedPage(buildstate);
|
|
|
|
|
2016-04-04 02:17:20 +08:00
|
|
|
if (!BloomPageAddItem(&buildstate->blstate, buildstate->data, itup))
|
2016-04-01 21:42:24 +08:00
|
|
|
{
|
|
|
|
/* We shouldn't be here since we're inserting to the empty page */
|
2016-04-04 02:17:20 +08:00
|
|
|
elog(ERROR, "could not add new bloom tuple to empty page");
|
2016-04-01 21:42:24 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
MemoryContextSwitchTo(oldCtx);
|
|
|
|
MemoryContextReset(buildstate->tmpCtx);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Build a new bloom index.
|
|
|
|
*/
|
|
|
|
IndexBuildResult *
|
|
|
|
blbuild(Relation heap, Relation index, IndexInfo *indexInfo)
|
|
|
|
{
|
|
|
|
IndexBuildResult *result;
|
|
|
|
double reltuples;
|
|
|
|
BloomBuildState buildstate;
|
|
|
|
|
|
|
|
if (RelationGetNumberOfBlocks(index) != 0)
|
|
|
|
elog(ERROR, "index \"%s\" already contains data",
|
|
|
|
RelationGetRelationName(index));
|
|
|
|
|
|
|
|
/* Initialize the meta page */
|
|
|
|
BloomInitMetapage(index);
|
|
|
|
|
|
|
|
/* Initialize the bloom build state */
|
|
|
|
memset(&buildstate, 0, sizeof(buildstate));
|
|
|
|
initBloomState(&buildstate.blstate, index);
|
|
|
|
buildstate.tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
|
|
|
|
"Bloom build temporary context",
|
Add macros to make AllocSetContextCreate() calls simpler and safer.
I found that half a dozen (nearly 5%) of our AllocSetContextCreate calls
had typos in the context-sizing parameters. While none of these led to
especially significant problems, they did create minor inefficiencies,
and it's now clear that expecting people to copy-and-paste those calls
accurately is not a great idea. Let's reduce the risk of future errors
by introducing single macros that encapsulate the common use-cases.
Three such macros are enough to cover all but two special-purpose contexts;
those two calls can be left as-is, I think.
While this patch doesn't in itself improve matters for third-party
extensions, it doesn't break anything for them either, and they can
gradually adopt the simplified notation over time.
In passing, change TopMemoryContext to use the default allocation
parameters. Formerly it could only be extended 8K at a time. That was
probably reasonable when this code was written; but nowadays we create
many more contexts than we did then, so that it's not unusual to have a
couple hundred K in TopMemoryContext, even without considering various
dubious code that sticks other things there. There seems no good reason
not to let it use growing blocks like most other contexts.
Back-patch to 9.6, mostly because that's still close enough to HEAD that
it's easy to do so, and keeping the branches in sync can be expected to
avoid some future back-patching pain. The bugs fixed by these changes
don't seem to be significant enough to justify fixing them further back.
Discussion: <21072.1472321324@sss.pgh.pa.us>
2016-08-28 05:50:38 +08:00
|
|
|
ALLOCSET_DEFAULT_SIZES);
|
2016-04-01 21:42:24 +08:00
|
|
|
initCachedPage(&buildstate);
|
|
|
|
|
|
|
|
/* Do the heap scan */
|
|
|
|
reltuples = IndexBuildHeapScan(heap, index, indexInfo, true,
|
Support parallel btree index builds.
To make this work, tuplesort.c and logtape.c must also support
parallelism, so this patch adds that infrastructure and then applies
it to the particular case of parallel btree index builds. Testing
to date shows that this can often be 2-3x faster than a serial
index build.
The model for deciding how many workers to use is fairly primitive
at present, but it's better than not having the feature. We can
refine it as we get more experience.
Peter Geoghegan with some help from Rushabh Lathia. While Heikki
Linnakangas is not an author of this patch, he wrote other patches
without which this feature would not have been possible, and
therefore the release notes should possibly credit him as an author
of this feature. Reviewed by Claudio Freire, Heikki Linnakangas,
Thomas Munro, Tels, Amit Kapila, me.
Discussion: http://postgr.es/m/CAM3SWZQKM=Pzc=CAHzRixKjp2eO5Q0Jg1SoFQqeXFQ647JiwqQ@mail.gmail.com
Discussion: http://postgr.es/m/CAH2-Wz=AxWqDoVvGU7dq856S4r6sJAj6DBn7VMtigkB33N5eyg@mail.gmail.com
2018-02-03 02:25:55 +08:00
|
|
|
bloomBuildCallback, (void *) &buildstate,
|
|
|
|
NULL);
|
2016-04-01 21:42:24 +08:00
|
|
|
|
|
|
|
/*
|
2016-06-10 06:02:36 +08:00
|
|
|
* There are could be some items in cached page. Flush this page if
|
|
|
|
* needed.
|
2016-04-01 21:42:24 +08:00
|
|
|
*/
|
|
|
|
if (buildstate.count > 0)
|
|
|
|
flushCachedPage(index, &buildstate);
|
|
|
|
|
|
|
|
MemoryContextDelete(buildstate.tmpCtx);
|
|
|
|
|
|
|
|
result = (IndexBuildResult *) palloc(sizeof(IndexBuildResult));
|
|
|
|
result->heap_tuples = result->index_tuples = reltuples;
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Build an empty bloom index in the initialization fork.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
blbuildempty(Relation index)
|
|
|
|
{
|
2016-05-25 09:04:23 +08:00
|
|
|
Page metapage;
|
2016-04-01 21:42:24 +08:00
|
|
|
|
2016-05-25 09:04:23 +08:00
|
|
|
/* Construct metapage. */
|
|
|
|
metapage = (Page) palloc(BLCKSZ);
|
|
|
|
BloomFillMetapage(index, metapage);
|
|
|
|
|
2016-12-09 03:09:09 +08:00
|
|
|
/*
|
2017-05-18 04:31:56 +08:00
|
|
|
* Write the page and log it. It might seem that an immediate sync would
|
|
|
|
* be sufficient to guarantee that the file exists on disk, but recovery
|
|
|
|
* itself might remove it while replaying, for example, an
|
|
|
|
* XLOG_DBASE_CREATE or XLOG_TBLSPC_CREATE record. Therefore, we need
|
|
|
|
* this even when wal_level=minimal.
|
2016-12-09 03:09:09 +08:00
|
|
|
*/
|
2016-05-25 09:04:23 +08:00
|
|
|
PageSetChecksumInplace(metapage, BLOOM_METAPAGE_BLKNO);
|
|
|
|
smgrwrite(index->rd_smgr, INIT_FORKNUM, BLOOM_METAPAGE_BLKNO,
|
|
|
|
(char *) metapage, true);
|
2016-12-09 03:09:09 +08:00
|
|
|
log_newpage(&index->rd_smgr->smgr_rnode.node, INIT_FORKNUM,
|
2017-11-04 04:31:32 +08:00
|
|
|
BLOOM_METAPAGE_BLKNO, metapage, true);
|
2016-05-25 09:04:23 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* An immediate sync is required even if we xlog'd the page, because the
|
|
|
|
* write did not go through shared_buffers and therefore a concurrent
|
|
|
|
* checkpoint may have moved the redo pointer past our xlog record.
|
|
|
|
*/
|
|
|
|
smgrimmedsync(index->rd_smgr, INIT_FORKNUM);
|
2016-04-01 21:42:24 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Insert new tuple to the bloom index.
|
|
|
|
*/
|
|
|
|
bool
|
|
|
|
blinsert(Relation index, Datum *values, bool *isnull,
|
Allow index AMs to cache data across aminsert calls within a SQL command.
It's always been possible for index AMs to cache data across successive
amgettuple calls within a single SQL command: the IndexScanDesc.opaque
field is meant for precisely that. However, no comparable facility
exists for amortizing setup work across successive aminsert calls.
This patch adds such a feature and teaches GIN, GIST, and BRIN to use it
to amortize catalog lookups they'd previously been doing on every call.
(The other standard index AMs keep everything they need in the relcache,
so there's little to improve there.)
For GIN, the overall improvement in a statement that inserts many rows
can be as much as 10%, though it seems a bit less for the other two.
In addition, this makes a really significant difference in runtime
for CLOBBER_CACHE_ALWAYS tests, since in those builds the repeated
catalog lookups are vastly more expensive.
The reason this has been hard up to now is that the aminsert function is
not passed any useful place to cache per-statement data. What I chose to
do is to add suitable fields to struct IndexInfo and pass that to aminsert.
That's not widening the index AM API very much because IndexInfo is already
within the ken of ambuild; in fact, by passing the same info to aminsert
as to ambuild, this is really removing an inconsistency in the AM API.
Discussion: https://postgr.es/m/27568.1486508680@sss.pgh.pa.us
2017-02-10 00:52:12 +08:00
|
|
|
ItemPointer ht_ctid, Relation heapRel,
|
|
|
|
IndexUniqueCheck checkUnique,
|
|
|
|
IndexInfo *indexInfo)
|
2016-04-01 21:42:24 +08:00
|
|
|
{
|
|
|
|
BloomState blstate;
|
|
|
|
BloomTuple *itup;
|
|
|
|
MemoryContext oldCtx;
|
|
|
|
MemoryContext insertCtx;
|
|
|
|
BloomMetaPageData *metaData;
|
|
|
|
Buffer buffer,
|
|
|
|
metaBuffer;
|
|
|
|
Page page,
|
|
|
|
metaPage;
|
|
|
|
BlockNumber blkno = InvalidBlockNumber;
|
|
|
|
OffsetNumber nStart;
|
|
|
|
GenericXLogState *state;
|
|
|
|
|
|
|
|
insertCtx = AllocSetContextCreate(CurrentMemoryContext,
|
|
|
|
"Bloom insert temporary context",
|
Add macros to make AllocSetContextCreate() calls simpler and safer.
I found that half a dozen (nearly 5%) of our AllocSetContextCreate calls
had typos in the context-sizing parameters. While none of these led to
especially significant problems, they did create minor inefficiencies,
and it's now clear that expecting people to copy-and-paste those calls
accurately is not a great idea. Let's reduce the risk of future errors
by introducing single macros that encapsulate the common use-cases.
Three such macros are enough to cover all but two special-purpose contexts;
those two calls can be left as-is, I think.
While this patch doesn't in itself improve matters for third-party
extensions, it doesn't break anything for them either, and they can
gradually adopt the simplified notation over time.
In passing, change TopMemoryContext to use the default allocation
parameters. Formerly it could only be extended 8K at a time. That was
probably reasonable when this code was written; but nowadays we create
many more contexts than we did then, so that it's not unusual to have a
couple hundred K in TopMemoryContext, even without considering various
dubious code that sticks other things there. There seems no good reason
not to let it use growing blocks like most other contexts.
Back-patch to 9.6, mostly because that's still close enough to HEAD that
it's easy to do so, and keeping the branches in sync can be expected to
avoid some future back-patching pain. The bugs fixed by these changes
don't seem to be significant enough to justify fixing them further back.
Discussion: <21072.1472321324@sss.pgh.pa.us>
2016-08-28 05:50:38 +08:00
|
|
|
ALLOCSET_DEFAULT_SIZES);
|
2016-04-01 21:42:24 +08:00
|
|
|
|
|
|
|
oldCtx = MemoryContextSwitchTo(insertCtx);
|
|
|
|
|
|
|
|
initBloomState(&blstate, index);
|
|
|
|
itup = BloomFormTuple(&blstate, ht_ctid, values, isnull);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* At first, try to insert new tuple to the first page in notFullPage
|
2016-04-10 03:39:14 +08:00
|
|
|
* array. If successful, we don't need to modify the meta page.
|
2016-04-01 21:42:24 +08:00
|
|
|
*/
|
|
|
|
metaBuffer = ReadBuffer(index, BLOOM_METAPAGE_BLKNO);
|
|
|
|
LockBuffer(metaBuffer, BUFFER_LOCK_SHARE);
|
2016-04-20 21:31:19 +08:00
|
|
|
metaData = BloomPageGetMeta(BufferGetPage(metaBuffer));
|
2016-04-01 21:42:24 +08:00
|
|
|
|
|
|
|
if (metaData->nEnd > metaData->nStart)
|
|
|
|
{
|
|
|
|
Page page;
|
|
|
|
|
|
|
|
blkno = metaData->notFullPage[metaData->nStart];
|
|
|
|
Assert(blkno != InvalidBlockNumber);
|
2016-04-10 03:39:14 +08:00
|
|
|
|
|
|
|
/* Don't hold metabuffer lock while doing insert */
|
2016-04-01 21:42:24 +08:00
|
|
|
LockBuffer(metaBuffer, BUFFER_LOCK_UNLOCK);
|
|
|
|
|
|
|
|
buffer = ReadBuffer(index, blkno);
|
|
|
|
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
|
2016-04-10 03:39:14 +08:00
|
|
|
|
2016-04-01 21:42:24 +08:00
|
|
|
state = GenericXLogStart(index);
|
2016-04-12 23:42:06 +08:00
|
|
|
page = GenericXLogRegisterBuffer(state, buffer, 0);
|
2016-04-01 21:42:24 +08:00
|
|
|
|
Fix assorted bugs in contrib/bloom.
In blinsert(), cope with the possibility that a page we pull from the
notFullPage list is marked BLOOM_DELETED. This could happen if VACUUM
recently marked it deleted but hasn't (yet) updated the metapage.
We can re-use such a page safely, but we *must* reinitialize it so that
it's no longer marked deleted.
Fix blvacuum() so that it updates the notFullPage list even if it's
going to update it to empty. The previous "optimization" of skipping
the update seems pretty dubious, since it means that the next blinsert()
will uselessly visit whatever pages we left in the list.
Uniformly treat PageIsNew pages the same as deleted pages. This should
allow proper recovery if a crash occurs just after relation extension.
Properly use vacuum_delay_point, not assorted ad-hoc CHECK_FOR_INTERRUPTS
calls, in the blvacuum() main loop.
Fix broken tuple-counting logic: blvacuum.c counted the number of live
index tuples over again in each scan, leading to VACUUM VERBOSE reporting
some multiple of the actual number of surviving index tuples after any
vacuum that removed any tuples (since they'd be counted in blvacuum, maybe
more than once, and then again in blvacuumcleanup, without ever zeroing the
counter). It's sufficient to count them in blvacuumcleanup.
stats->estimated_count is a boolean, not a counter, and we don't want
to set it true, so don't add tuple counts to it.
Add a couple of Asserts that we don't overrun available space on a bloom
page. I don't think there's any bug there today, but the way the
FreeBlockNumberArray size calculation is set up is scarily fragile, and
BloomPageGetFreeSpace isn't much better. The Asserts should help catch
any future mistakes.
Per investigation of a report from Jeff Janes. I think the first item
above may explain his report; the other changes were things I noticed
while casting about for an explanation.
Report: <CAMkU=1xEUuBphDwDmB1WjN4+td4kpnEniFaTBxnk1xzHCw8_OQ@mail.gmail.com>
2016-08-14 10:24:48 +08:00
|
|
|
/*
|
|
|
|
* We might have found a page that was recently deleted by VACUUM. If
|
|
|
|
* so, we can reuse it, but we must reinitialize it.
|
|
|
|
*/
|
|
|
|
if (PageIsNew(page) || BloomPageIsDeleted(page))
|
|
|
|
BloomInitPage(page, 0);
|
|
|
|
|
2016-04-01 21:42:24 +08:00
|
|
|
if (BloomPageAddItem(&blstate, page, itup))
|
|
|
|
{
|
2016-04-10 03:39:14 +08:00
|
|
|
/* Success! Apply the change, clean up, and exit */
|
2016-04-01 21:42:24 +08:00
|
|
|
GenericXLogFinish(state);
|
|
|
|
UnlockReleaseBuffer(buffer);
|
|
|
|
ReleaseBuffer(metaBuffer);
|
|
|
|
MemoryContextSwitchTo(oldCtx);
|
|
|
|
MemoryContextDelete(insertCtx);
|
|
|
|
return false;
|
|
|
|
}
|
2016-04-10 03:39:14 +08:00
|
|
|
|
|
|
|
/* Didn't fit, must try other pages */
|
|
|
|
GenericXLogAbort(state);
|
|
|
|
UnlockReleaseBuffer(buffer);
|
2016-04-01 21:42:24 +08:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2016-04-10 03:39:14 +08:00
|
|
|
/* No entries in notFullPage */
|
2016-04-01 21:42:24 +08:00
|
|
|
LockBuffer(metaBuffer, BUFFER_LOCK_UNLOCK);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Try other pages in notFullPage array. We will have to change nStart in
|
|
|
|
* metapage. Thus, grab exclusive lock on metapage.
|
|
|
|
*/
|
|
|
|
LockBuffer(metaBuffer, BUFFER_LOCK_EXCLUSIVE);
|
|
|
|
|
2016-04-10 03:39:14 +08:00
|
|
|
/* nStart might have changed while we didn't have lock */
|
2016-04-01 21:42:24 +08:00
|
|
|
nStart = metaData->nStart;
|
2016-04-10 03:39:14 +08:00
|
|
|
|
|
|
|
/* Skip first page if we already tried it above */
|
|
|
|
if (nStart < metaData->nEnd &&
|
2016-04-01 21:42:24 +08:00
|
|
|
blkno == metaData->notFullPage[nStart])
|
|
|
|
nStart++;
|
|
|
|
|
2016-04-10 03:39:14 +08:00
|
|
|
/*
|
|
|
|
* This loop iterates for each page we try from the notFullPage array, and
|
|
|
|
* will also initialize a GenericXLogState for the fallback case of having
|
|
|
|
* to allocate a new page.
|
|
|
|
*/
|
|
|
|
for (;;)
|
2016-04-01 21:42:24 +08:00
|
|
|
{
|
2016-04-10 03:39:14 +08:00
|
|
|
state = GenericXLogStart(index);
|
|
|
|
|
|
|
|
/* get modifiable copy of metapage */
|
2016-04-12 23:42:06 +08:00
|
|
|
metaPage = GenericXLogRegisterBuffer(state, metaBuffer, 0);
|
2016-04-10 03:39:14 +08:00
|
|
|
metaData = BloomPageGetMeta(metaPage);
|
|
|
|
|
|
|
|
if (nStart >= metaData->nEnd)
|
|
|
|
break; /* no more entries in notFullPage array */
|
|
|
|
|
2016-04-01 21:42:24 +08:00
|
|
|
blkno = metaData->notFullPage[nStart];
|
|
|
|
Assert(blkno != InvalidBlockNumber);
|
|
|
|
|
|
|
|
buffer = ReadBuffer(index, blkno);
|
|
|
|
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
|
2016-04-12 23:42:06 +08:00
|
|
|
page = GenericXLogRegisterBuffer(state, buffer, 0);
|
2016-04-01 21:42:24 +08:00
|
|
|
|
Fix assorted bugs in contrib/bloom.
In blinsert(), cope with the possibility that a page we pull from the
notFullPage list is marked BLOOM_DELETED. This could happen if VACUUM
recently marked it deleted but hasn't (yet) updated the metapage.
We can re-use such a page safely, but we *must* reinitialize it so that
it's no longer marked deleted.
Fix blvacuum() so that it updates the notFullPage list even if it's
going to update it to empty. The previous "optimization" of skipping
the update seems pretty dubious, since it means that the next blinsert()
will uselessly visit whatever pages we left in the list.
Uniformly treat PageIsNew pages the same as deleted pages. This should
allow proper recovery if a crash occurs just after relation extension.
Properly use vacuum_delay_point, not assorted ad-hoc CHECK_FOR_INTERRUPTS
calls, in the blvacuum() main loop.
Fix broken tuple-counting logic: blvacuum.c counted the number of live
index tuples over again in each scan, leading to VACUUM VERBOSE reporting
some multiple of the actual number of surviving index tuples after any
vacuum that removed any tuples (since they'd be counted in blvacuum, maybe
more than once, and then again in blvacuumcleanup, without ever zeroing the
counter). It's sufficient to count them in blvacuumcleanup.
stats->estimated_count is a boolean, not a counter, and we don't want
to set it true, so don't add tuple counts to it.
Add a couple of Asserts that we don't overrun available space on a bloom
page. I don't think there's any bug there today, but the way the
FreeBlockNumberArray size calculation is set up is scarily fragile, and
BloomPageGetFreeSpace isn't much better. The Asserts should help catch
any future mistakes.
Per investigation of a report from Jeff Janes. I think the first item
above may explain his report; the other changes were things I noticed
while casting about for an explanation.
Report: <CAMkU=1xEUuBphDwDmB1WjN4+td4kpnEniFaTBxnk1xzHCw8_OQ@mail.gmail.com>
2016-08-14 10:24:48 +08:00
|
|
|
/* Basically same logic as above */
|
|
|
|
if (PageIsNew(page) || BloomPageIsDeleted(page))
|
|
|
|
BloomInitPage(page, 0);
|
|
|
|
|
2016-04-01 21:42:24 +08:00
|
|
|
if (BloomPageAddItem(&blstate, page, itup))
|
|
|
|
{
|
2016-04-10 03:39:14 +08:00
|
|
|
/* Success! Apply the changes, clean up, and exit */
|
2016-04-01 21:42:24 +08:00
|
|
|
metaData->nStart = nStart;
|
|
|
|
GenericXLogFinish(state);
|
|
|
|
UnlockReleaseBuffer(buffer);
|
|
|
|
UnlockReleaseBuffer(metaBuffer);
|
|
|
|
MemoryContextSwitchTo(oldCtx);
|
|
|
|
MemoryContextDelete(insertCtx);
|
|
|
|
return false;
|
|
|
|
}
|
2016-04-10 03:39:14 +08:00
|
|
|
|
|
|
|
/* Didn't fit, must try other pages */
|
|
|
|
GenericXLogAbort(state);
|
|
|
|
UnlockReleaseBuffer(buffer);
|
2016-04-01 21:42:24 +08:00
|
|
|
nStart++;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Didn't find place to insert in notFullPage array. Allocate new page.
|
2016-04-10 03:39:14 +08:00
|
|
|
* (XXX is it good to do this while holding ex-lock on the metapage??)
|
2016-04-01 21:42:24 +08:00
|
|
|
*/
|
|
|
|
buffer = BloomNewBuffer(index);
|
|
|
|
|
2016-04-12 23:42:06 +08:00
|
|
|
page = GenericXLogRegisterBuffer(state, buffer, GENERIC_XLOG_FULL_IMAGE);
|
2016-04-01 21:42:24 +08:00
|
|
|
BloomInitPage(page, 0);
|
2016-04-04 02:17:20 +08:00
|
|
|
|
|
|
|
if (!BloomPageAddItem(&blstate, page, itup))
|
|
|
|
{
|
2016-04-10 03:39:14 +08:00
|
|
|
/* We shouldn't be here since we're inserting to an empty page */
|
2016-04-04 02:17:20 +08:00
|
|
|
elog(ERROR, "could not add new bloom tuple to empty page");
|
|
|
|
}
|
2016-04-01 21:42:24 +08:00
|
|
|
|
2016-04-10 03:39:14 +08:00
|
|
|
/* Reset notFullPage array to contain just this new page */
|
2016-04-01 21:42:24 +08:00
|
|
|
metaData->nStart = 0;
|
|
|
|
metaData->nEnd = 1;
|
|
|
|
metaData->notFullPage[0] = BufferGetBlockNumber(buffer);
|
|
|
|
|
2016-04-10 03:39:14 +08:00
|
|
|
/* Apply the changes, clean up, and exit */
|
2016-04-01 21:42:24 +08:00
|
|
|
GenericXLogFinish(state);
|
|
|
|
|
|
|
|
UnlockReleaseBuffer(buffer);
|
|
|
|
UnlockReleaseBuffer(metaBuffer);
|
|
|
|
|
2016-04-10 03:39:14 +08:00
|
|
|
MemoryContextSwitchTo(oldCtx);
|
|
|
|
MemoryContextDelete(insertCtx);
|
|
|
|
|
2016-04-01 21:42:24 +08:00
|
|
|
return false;
|
|
|
|
}
|