/*
* Per-heap-tuple callback for IndexBuildHeapScan.
*
* Note we don't worry about the page range at the end of the table here; it is
* present in the build state struct after we're called the last time, but not
* inserted into the index. Caller must ensure to do so, if appropriate.
*/
static void
brinbuildCallback(Relation index,
HeapTuple htup,
Datum *values,
bool *isnull,
bool tupleIsAlive,
void *brstate)
{
BrinBuildState *state = (BrinBuildState *) brstate;
BlockNumber thisblock;
int i;
thisblock = ItemPointerGetBlockNumber(&htup->t_self);
/*
* If we're in a block that belongs to a future range, summarize what
* we've got and start afresh. Note the scan might have skipped many
* pages, if they were devoid of live tuples; make sure to insert index
* tuples for those too.
*/
while (thisblock > state->bs_currRangeStart + state->bs_pagesPerRange - 1)
{
BRIN_elog((DEBUG2,
"brinbuildCallback: completed a range: %u--%u",
state->bs_currRangeStart,
state->bs_currRangeStart + state->bs_pagesPerRange));
/* create the index tuple and insert it */
form_and_insert_tuple(state);
/* set state to correspond to the next range */
state->bs_currRangeStart += state->bs_pagesPerRange;
/* re-initialize state for it */
brin_memtuple_initialize(state->bs_dtuple, state->bs_bdesc);
}
/* Accumulate the current tuple into the running state */
for (i = 0; i < state->bs_bdesc->bd_tupdesc->natts; i++)
{
FmgrInfo *addValue;
BrinValues *col;
Form_pg_attribute attr = TupleDescAttr(state->bs_bdesc->bd_tupdesc, i);
col = &state->bs_dtuple->bt_columns[i];
addValue = index_getprocinfo(index, i + 1,
BRIN_PROCNUM_ADDVALUE);
/*
* Update dtuple state, if and as necessary.
*/
FunctionCall4Coll(addValue,
attr->attcollation,
PointerGetDatum(state->bs_bdesc),
PointerGetDatum(col),
values[i], isnull[i]);
}
}
/*
* Initialize a page as an empty regular BRIN page, WAL-log this, and record
* the page in FSM.
*
* There are several corner situations in which we extend the relation to
* obtain a new page and later find that we cannot use it immediately. When
* that happens, we don't want to leave the page go unrecorded in FSM, because
* there is no mechanism to get the space back and the index would bloat.
* Also, because we would not WAL-log the action that would initialize the
* page, the page would go uninitialized in a standby (or after recovery).
*/
static void
brin_initialize_empty_new_buffer(Relation idxrel, Buffer buffer)
{
Page page;
BRIN_elog((DEBUG2,
"brin_initialize_empty_new_buffer: initializing blank page %u",
BufferGetBlockNumber(buffer)));
START_CRIT_SECTION();
page = BufferGetPage(buffer);
brin_page_init(page, BRIN_PAGETYPE_REGULAR);
MarkBufferDirty(buffer);
log_newpage_buffer(buffer, true);
END_CRIT_SECTION();
/*
* We update the FSM for this page, but this is not WAL-logged. This is
* acceptable because VACUUM will scan the index and update the FSM with
* pages whose FSM records were forgotten in a crash.
*/
RecordPageWithFreeSpace(idxrel, BufferGetBlockNumber(buffer),
br_page_get_freespace(page));
}
/*
* Return a pinned and exclusively locked buffer which can be used to insert an
* index item of size itemsz (caller must ensure not to request sizes
* impossible to fulfill). If oldbuf is a valid buffer, it is also locked (in
* an order determined to avoid deadlocks.)
*
* If we find that the old page is no longer a regular index page (because
* of a revmap extension), the old buffer is unlocked and we return
* InvalidBuffer.
*
* If there's no existing page with enough free space to accommodate the new
* item, the relation is extended. If this happens, *extended is set to true,
* and it is the caller's responsibility to initialize the page (and WAL-log
* that fact) prior to use.
*
* Note that in some corner cases it is possible for this routine to extend the
* relation and then not return the buffer. It is this routine's
* responsibility to WAL-log the page initialization and to record the page in
* FSM if that happens. Such a buffer may later be reused by this routine.
*/
static Buffer
brin_getinsertbuffer(Relation irel, Buffer oldbuf, Size itemsz,
bool *extended)
{
BlockNumber oldblk;
BlockNumber newblk;
Page page;
int freespace;
/* callers must have checked */
Assert(itemsz <= BrinMaxItemSize);
*extended = false;
if (BufferIsValid(oldbuf))
oldblk = BufferGetBlockNumber(oldbuf);
else
oldblk = InvalidBlockNumber;
/*
* Loop until we find a page with sufficient free space. By the time we
* return to caller out of this loop, both buffers are valid and locked;
* if we have to restart here, neither buffer is locked and buf is not a
* pinned buffer.
*/
newblk = RelationGetTargetBlock(irel);
if (newblk == InvalidBlockNumber)
newblk = GetPageWithFreeSpace(irel, itemsz);
for (;;)
{
Buffer buf;
bool extensionLockHeld = false;
CHECK_FOR_INTERRUPTS();
if (newblk == InvalidBlockNumber)
{
/*
* There's not enough free space in any existing index page,
* according to the FSM: extend the relation to obtain a shiny new
* page.
*/
if (!RELATION_IS_LOCAL(irel))
{
LockRelationForExtension(irel, ExclusiveLock);
extensionLockHeld = true;
}
buf = ReadBuffer(irel, P_NEW);
newblk = BufferGetBlockNumber(buf);
*extended = true;
BRIN_elog((DEBUG2, "brin_getinsertbuffer: extending to page %u",
BufferGetBlockNumber(buf)));
}
else if (newblk == oldblk)
{
/*
* There's an odd corner-case here where the FSM is out-of-date,
* and gave us the old page.
*/
buf = oldbuf;
}
else
{
buf = ReadBuffer(irel, newblk);
}
/*
* We lock the old buffer first, if it's earlier than the new one; but
* before we do, we need to check that it hasn't been turned into a
* revmap page concurrently; if we detect that it happened, give up
* and tell caller to start over.
*/
if (BufferIsValid(oldbuf) && oldblk < newblk)
{
LockBuffer(oldbuf, BUFFER_LOCK_EXCLUSIVE);
if (!BRIN_IS_REGULAR_PAGE(BufferGetPage(oldbuf)))
{
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
/*
* It is possible that the new page was obtained from
* extending the relation. In that case, we must be sure to
* record it in the FSM before leaving, because otherwise the
* space would be lost forever. However, we cannot let an
* uninitialized page get in the FSM, so we need to initialize
* it first.
*/
if (*extended)
{
brin_initialize_empty_new_buffer(irel, buf);
/* shouldn't matter, but don't confuse caller */
*extended = false;
}
if (extensionLockHeld)
UnlockRelationForExtension(irel, ExclusiveLock);
ReleaseBuffer(buf);
return InvalidBuffer;
}
}
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
if (extensionLockHeld)
UnlockRelationForExtension(irel, ExclusiveLock);
page = BufferGetPage(buf);
/*
* We have a new buffer to insert into. Check that the new page has
* enough free space, and return it if it does; otherwise start over.
* Note that we allow for the FSM to be out of date here, and in that
* case we update it and move on.
*
* (br_page_get_freespace also checks that the FSM didn't hand us a
* page that has since been repurposed for the revmap.)
*/
freespace = *extended ?
BrinMaxItemSize : br_page_get_freespace(page);
if (freespace >= itemsz)
{
RelationSetTargetBlock(irel, BufferGetBlockNumber(buf));
/*
* Since the target block specification can get lost on cache
* invalidations, make sure we update the more permanent FSM with
* data about it before going away.
*/
if (*extended)
RecordPageWithFreeSpace(irel, BufferGetBlockNumber(buf),
freespace);
/*
* Lock the old buffer if not locked already. Note that in this
* case we know for sure it's a regular page: it's later than the
* new page we just got, which is not a revmap page, and revmap
* pages are always consecutive.
*/
if (BufferIsValid(oldbuf) && oldblk > newblk)
{
LockBuffer(oldbuf, BUFFER_LOCK_EXCLUSIVE);
Assert(BRIN_IS_REGULAR_PAGE(BufferGetPage(oldbuf)));
}
return buf;
}
/* This page is no good. */
/*
* If an entirely new page does not contain enough free space for the
* new item, then surely that item is oversized. Complain loudly; but
* first make sure we initialize the page and record it as free, for
* next time.
*/
if (*extended)
{
brin_initialize_empty_new_buffer(irel, buf);
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds maximum %lu for index \"%s\"",
(unsigned long) itemsz,
(unsigned long) freespace,
RelationGetRelationName(irel))));
return InvalidBuffer; /* keep compiler quiet */
}
if (newblk != oldblk)
UnlockReleaseBuffer(buf);
if (BufferIsValid(oldbuf) && oldblk <= newblk)
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
newblk = RecordAndGetPageWithFreeSpace(irel, newblk, freespace, itemsz);
}
}
/*
* Insert an index tuple into the index relation. The revmap is updated to
* mark the range containing the given page as pointing to the inserted entry.
* A WAL record is written.
*
* The buffer, if valid, is first checked for free space to insert the new
* entry; if there isn't enough, a new buffer is obtained and pinned. No
* buffer lock must be held on entry, no buffer lock is held on exit.
*
* Return value is the offset number where the tuple was inserted.
*/
OffsetNumber
brin_doinsert(Relation idxrel, BlockNumber pagesPerRange,
BrinRevmap *revmap, Buffer *buffer, BlockNumber heapBlk,
BrinTuple *tup, Size itemsz)
{
Page page;
BlockNumber blk;
OffsetNumber off;
Buffer revmapbuf;
ItemPointerData tid;
bool extended;
Assert(itemsz == MAXALIGN(itemsz));
/* If the item is oversized, don't even bother. */
if (itemsz > BrinMaxItemSize)
{
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds maximum %lu for index \"%s\"",
(unsigned long) itemsz,
(unsigned long) BrinMaxItemSize,
RelationGetRelationName(idxrel))));
return InvalidOffsetNumber; /* keep compiler quiet */
}
/* Make sure the revmap is long enough to contain the entry we need */
brinRevmapExtend(revmap, heapBlk);
/*
* Acquire lock on buffer supplied by caller, if any. If it doesn't have
* enough space, unpin it to obtain a new one below.
*/
if (BufferIsValid(*buffer))
{
/*
* It's possible that another backend (or ourselves!) extended the
* revmap over the page we held a pin on, so we cannot assume that
* it's still a regular page.
*/
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
if (br_page_get_freespace(BufferGetPage(*buffer)) < itemsz)
{
UnlockReleaseBuffer(*buffer);
*buffer = InvalidBuffer;
}
}
/*
* If we still don't have a usable buffer, have brin_getinsertbuffer
* obtain one for us.
*/
if (!BufferIsValid(*buffer))
{
do
*buffer = brin_getinsertbuffer(idxrel, InvalidBuffer, itemsz, &extended);
while (!BufferIsValid(*buffer));
}
else
extended = false;
/* Now obtain lock on revmap buffer */
revmapbuf = brinLockRevmapPageForUpdate(revmap, heapBlk);
page = BufferGetPage(*buffer);
blk = BufferGetBlockNumber(*buffer);
/* Execute the actual insertion */
START_CRIT_SECTION();
if (extended)
brin_page_init(BufferGetPage(*buffer), BRIN_PAGETYPE_REGULAR);
off = PageAddItem(page, (Item) tup, itemsz, InvalidOffsetNumber,
false, false);
if (off == InvalidOffsetNumber)
elog(ERROR, "could not insert new index tuple to page");
MarkBufferDirty(*buffer);
BRIN_elog((DEBUG2, "inserted tuple (%u,%u) for range starting at %u",
blk, off, heapBlk));
ItemPointerSet(&tid, blk, off);
brinSetHeapBlockItemptr(revmapbuf, pagesPerRange, heapBlk, tid);
MarkBufferDirty(revmapbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_insert xlrec;
XLogRecPtr recptr;
uint8 info;
info = XLOG_BRIN_INSERT | (extended ? XLOG_BRIN_INIT_PAGE : 0);
xlrec.heapBlk = heapBlk;
xlrec.pagesPerRange = pagesPerRange;
xlrec.offnum = off;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfBrinInsert);
XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD | (extended ? REGBUF_WILL_INIT : 0));
XLogRegisterBufData(0, (char *) tup, itemsz);
XLogRegisterBuffer(1, revmapbuf, 0);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(page, recptr);
PageSetLSN(BufferGetPage(revmapbuf), recptr);
}
END_CRIT_SECTION();
/* Tuple is firmly on buffer; we can release our locks */
LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
LockBuffer(revmapbuf, BUFFER_LOCK_UNLOCK);
if (extended)
FreeSpaceMapVacuum(idxrel);
return off;
}
/*
* Insert an index tuple into the index relation. The revmap is updated to
* mark the range containing the given page as pointing to the inserted entry.
* A WAL record is written.
*
* The buffer, if valid, is first checked for free space to insert the new
* entry; if there isn't enough, a new buffer is obtained and pinned. No
* buffer lock must be held on entry, no buffer lock is held on exit.
*
* Return value is the offset number where the tuple was inserted.
*/
OffsetNumber
brin_doinsert(Relation idxrel, BlockNumber pagesPerRange,
BrinRevmap *revmap, Buffer *buffer, BlockNumber heapBlk,
BrinTuple *tup, Size itemsz)
{
Page page;
BlockNumber blk;
OffsetNumber off;
Buffer revmapbuf;
ItemPointerData tid;
bool extended = false;
itemsz = MAXALIGN(itemsz);
/* Make sure the revmap is long enough to contain the entry we need */
brinRevmapExtend(revmap, heapBlk);
/*
* Obtain a locked buffer to insert the new tuple. Note
* brin_getinsertbuffer ensures there's enough space in the returned
* buffer.
*/
if (BufferIsValid(*buffer))
{
/*
* It's possible that another backend (or ourselves!) extended the
* revmap over the page we held a pin on, so we cannot assume that
* it's still a regular page.
*/
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
if (br_page_get_freespace(BufferGetPage(*buffer)) < itemsz)
{
UnlockReleaseBuffer(*buffer);
*buffer = InvalidBuffer;
}
}
if (!BufferIsValid(*buffer))
{
*buffer = brin_getinsertbuffer(idxrel, InvalidBuffer, itemsz, &extended);
Assert(BufferIsValid(*buffer));
Assert(br_page_get_freespace(BufferGetPage(*buffer)) >= itemsz);
}
/* Now obtain lock on revmap buffer */
revmapbuf = brinLockRevmapPageForUpdate(revmap, heapBlk);
page = BufferGetPage(*buffer);
blk = BufferGetBlockNumber(*buffer);
START_CRIT_SECTION();
off = PageAddItem(page, (Item) tup, itemsz, InvalidOffsetNumber,
false, false);
if (off == InvalidOffsetNumber)
elog(ERROR, "could not insert new index tuple to page");
MarkBufferDirty(*buffer);
BRIN_elog((DEBUG2, "inserted tuple (%u,%u) for range starting at %u",
blk, off, heapBlk));
ItemPointerSet(&tid, blk, off);
brinSetHeapBlockItemptr(revmapbuf, pagesPerRange, heapBlk, tid);
MarkBufferDirty(revmapbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_insert xlrec;
XLogRecPtr recptr;
uint8 info;
info = XLOG_BRIN_INSERT | (extended ? XLOG_BRIN_INIT_PAGE : 0);
xlrec.heapBlk = heapBlk;
xlrec.pagesPerRange = pagesPerRange;
xlrec.offnum = off;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfBrinInsert);
XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD | (extended ? REGBUF_WILL_INIT : 0));
XLogRegisterBufData(0, (char *) tup, itemsz);
XLogRegisterBuffer(1, revmapbuf, 0);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(page, recptr);
PageSetLSN(BufferGetPage(revmapbuf), recptr);
}
END_CRIT_SECTION();
/* Tuple is firmly on buffer; we can release our locks */
LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
LockBuffer(revmapbuf, BUFFER_LOCK_UNLOCK);
if (extended)
FreeSpaceMapVacuum(idxrel);
return off;
}