/*
* _hash_getbucketbuf_from_hashkey() -- Get the bucket's buffer for the given
* hashkey.
*
* Bucket pages do not move or get removed once they are allocated. This give
* us an opportunity to use the previously saved metapage contents to reach
* the target bucket buffer, instead of reading from the metapage every time.
* This saves one buffer access every time we want to reach the target bucket
* buffer, which is very helpful savings in bufmgr traffic and contention.
*
* The access type parameter (HASH_READ or HASH_WRITE) indicates whether the
* bucket buffer has to be locked for reading or writing.
*
* The out parameter cachedmetap is set with metapage contents used for
* hashkey to bucket buffer mapping. Some callers need this info to reach the
* old bucket in case of bucket split, see _hash_doinsert().
*/
Buffer
_hash_getbucketbuf_from_hashkey(Relation rel, uint32 hashkey, int access,
HashMetaPage *cachedmetap)
{
HashMetaPage metap;
Buffer buf;
Buffer metabuf = InvalidBuffer;
Page page;
Bucket bucket;
BlockNumber blkno;
HashPageOpaque opaque;
/* We read from target bucket buffer, hence locking is must. */
Assert(access == HASH_READ || access == HASH_WRITE);
metap = _hash_getcachedmetap(rel, &metabuf, false);
Assert(metap != NULL);
/*
* Loop until we get a lock on the correct target bucket.
*/
for (;;)
{
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, access, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(opaque->hasho_bucket == bucket);
Assert(opaque->hasho_prevblkno != InvalidBlockNumber);
/*
* If this bucket hasn't been split, we're done.
*/
if (opaque->hasho_prevblkno <= metap->hashm_maxbucket)
break;
/* Drop lock on this buffer, update cached metapage, and retry. */
_hash_relbuf(rel, buf);
metap = _hash_getcachedmetap(rel, &metabuf, true);
Assert(metap != NULL);
}
if (BufferIsValid(metabuf))
_hash_dropbuf(rel, metabuf);
if (cachedmetap)
*cachedmetap = metap;
return buf;
}
/*
* _hash_first() -- Find the first item in a scan.
*
* Find the first item in the index that
* satisfies the qualification associated with the scan descriptor. On
* success, the page containing the current index tuple is read locked
* and pinned, and the scan's opaque data entry is updated to
* include the buffer.
*/
bool
_hash_first(IndexScanDesc scan, ScanDirection dir)
{
Relation rel = scan->indexRelation;
HashScanOpaque so = (HashScanOpaque) scan->opaque;
uint32 hashkey;
Bucket bucket;
BlockNumber blkno;
Buffer buf;
Buffer metabuf;
Page page;
HashPageOpaque opaque;
HashMetaPage metap;
IndexTuple itup;
ItemPointer current;
OffsetNumber offnum;
MIRROREDLOCK_BUFMGR_MUST_ALREADY_BE_HELD;
pgstat_count_index_scan(rel);
current = &(scan->currentItemData);
ItemPointerSetInvalid(current);
/*
* We do not support hash scans with no index qualification, because we
* would have to read the whole index rather than just one bucket. That
* creates a whole raft of problems, since we haven't got a practical way
* to lock all the buckets against splits or compactions.
*/
if (scan->numberOfKeys < 1)
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("hash indexes do not support whole-index scans")));
/*
* If the constant in the index qual is NULL, assume it cannot match any
* items in the index.
*/
if (scan->keyData[0].sk_flags & SK_ISNULL)
return false;
/*
* Okay to compute the hash key. We want to do this before acquiring any
* locks, in case a user-defined hash function happens to be slow.
*/
hashkey = _hash_datum2hashkey(rel, scan->keyData[0].sk_argument);
/*
* Acquire shared split lock so we can compute the target bucket safely
* (see README).
*/
_hash_getlock(rel, 0, HASH_SHARE);
/* Read the metapage */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ);
_hash_checkpage(rel, metabuf, LH_META_PAGE);
metap = (HashMetaPage) BufferGetPage(metabuf);
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* done with the metapage */
_hash_relbuf(rel, metabuf);
/*
* Acquire share lock on target bucket; then we can release split lock.
*/
_hash_getlock(rel, blkno, HASH_SHARE);
_hash_droplock(rel, 0, HASH_SHARE);
/* Update scan opaque state to show we have lock on the bucket */
so->hashso_bucket = bucket;
so->hashso_bucket_valid = true;
so->hashso_bucket_blkno = blkno;
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, HASH_READ);
_hash_checkpage(rel, buf, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(opaque->hasho_bucket == bucket);
/* If a backwards scan is requested, move to the end of the chain */
if (ScanDirectionIsBackward(dir))
{
while (BlockNumberIsValid(opaque->hasho_nextblkno))
_hash_readnext(rel, &buf, &page, &opaque);
}
/* Now find the first tuple satisfying the qualification */
if (!_hash_step(scan, &buf, dir))
return false;
/* if we're here, _hash_step found a valid tuple */
offnum = ItemPointerGetOffsetNumber(current);
_hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
page = BufferGetPage(buf);
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
scan->xs_ctup.t_self = itup->t_tid;
return true;
}
/*
* Helper function to perform deletion of index entries from a bucket.
*
* This function expects that the caller has acquired a cleanup lock on the
* primary bucket page, and will return with a write lock again held on the
* primary bucket page. The lock won't necessarily be held continuously,
* though, because we'll release it when visiting overflow pages.
*
* It would be very bad if this function cleaned a page while some other
* backend was in the midst of scanning it, because hashgettuple assumes
* that the next valid TID will be greater than or equal to the current
* valid TID. There can't be any concurrent scans in progress when we first
* enter this function because of the cleanup lock we hold on the primary
* bucket page, but as soon as we release that lock, there might be. We
* handle that by conspiring to prevent those scans from passing our cleanup
* scan. To do that, we lock the next page in the bucket chain before
* releasing the lock on the previous page. (This type of lock chaining is
* not ideal, so we might want to look for a better solution at some point.)
*
* We need to retain a pin on the primary bucket to ensure that no concurrent
* split can start.
*/
void
hashbucketcleanup(Relation rel, Bucket cur_bucket, Buffer bucket_buf,
BlockNumber bucket_blkno, BufferAccessStrategy bstrategy,
uint32 maxbucket, uint32 highmask, uint32 lowmask,
double *tuples_removed, double *num_index_tuples,
bool split_cleanup,
IndexBulkDeleteCallback callback, void *callback_state)
{
BlockNumber blkno;
Buffer buf;
Bucket new_bucket PG_USED_FOR_ASSERTS_ONLY = InvalidBucket;
bool bucket_dirty = false;
blkno = bucket_blkno;
buf = bucket_buf;
if (split_cleanup)
new_bucket = _hash_get_newbucket_from_oldbucket(rel, cur_bucket,
lowmask, maxbucket);
/* Scan each page in bucket */
for (;;)
{
HashPageOpaque opaque;
OffsetNumber offno;
OffsetNumber maxoffno;
Buffer next_buf;
Page page;
OffsetNumber deletable[MaxOffsetNumber];
int ndeletable = 0;
bool retain_pin = false;
bool clear_dead_marking = false;
vacuum_delay_point();
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
/* Scan each tuple in page */
maxoffno = PageGetMaxOffsetNumber(page);
for (offno = FirstOffsetNumber;
offno <= maxoffno;
offno = OffsetNumberNext(offno))
{
ItemPointer htup;
IndexTuple itup;
Bucket bucket;
bool kill_tuple = false;
itup = (IndexTuple) PageGetItem(page,
PageGetItemId(page, offno));
htup = &(itup->t_tid);
/*
* To remove the dead tuples, we strictly want to rely on results
* of callback function. refer btvacuumpage for detailed reason.
*/
if (callback && callback(htup, callback_state))
{
kill_tuple = true;
if (tuples_removed)
*tuples_removed += 1;
}
else if (split_cleanup)
{
/* delete the tuples that are moved by split. */
bucket = _hash_hashkey2bucket(_hash_get_indextuple_hashkey(itup),
maxbucket,
highmask,
lowmask);
/* mark the item for deletion */
if (bucket != cur_bucket)
{
/*
* We expect tuples to either belong to current bucket or
* new_bucket. This is ensured because we don't allow
* further splits from bucket that contains garbage. See
* comments in _hash_expandtable.
*/
Assert(bucket == new_bucket);
kill_tuple = true;
}
}
if (kill_tuple)
{
/* mark the item for deletion */
deletable[ndeletable++] = offno;
}
else
{
/* we're keeping it, so count it */
if (num_index_tuples)
*num_index_tuples += 1;
}
}
/* retain the pin on primary bucket page till end of bucket scan */
if (blkno == bucket_blkno)
retain_pin = true;
else
retain_pin = false;
blkno = opaque->hasho_nextblkno;
/*
* Apply deletions, advance to next page and write page if needed.
*/
if (ndeletable > 0)
{
/* No ereport(ERROR) until changes are logged */
START_CRIT_SECTION();
PageIndexMultiDelete(page, deletable, ndeletable);
bucket_dirty = true;
/*
* Let us mark the page as clean if vacuum removes the DEAD tuples
* from an index page. We do this by clearing
* LH_PAGE_HAS_DEAD_TUPLES flag.
*/
if (tuples_removed && *tuples_removed > 0 &&
H_HAS_DEAD_TUPLES(opaque))
{
opaque->hasho_flag &= ~LH_PAGE_HAS_DEAD_TUPLES;
clear_dead_marking = true;
}
MarkBufferDirty(buf);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
xl_hash_delete xlrec;
XLogRecPtr recptr;
xlrec.clear_dead_marking = clear_dead_marking;
xlrec.is_primary_bucket_page = (buf == bucket_buf) ? true : false;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHashDelete);
/*
* bucket buffer needs to be registered to ensure that we can
* acquire a cleanup lock on it during replay.
*/
if (!xlrec.is_primary_bucket_page)
XLogRegisterBuffer(0, bucket_buf, REGBUF_STANDARD | REGBUF_NO_IMAGE);
XLogRegisterBuffer(1, buf, REGBUF_STANDARD);
XLogRegisterBufData(1, (char *) deletable,
ndeletable * sizeof(OffsetNumber));
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_DELETE);
PageSetLSN(BufferGetPage(buf), recptr);
}
END_CRIT_SECTION();
}
/* bail out if there are no more pages to scan. */
if (!BlockNumberIsValid(blkno))
break;
next_buf = _hash_getbuf_with_strategy(rel, blkno, HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
/*
* release the lock on previous page after acquiring the lock on next
* page
*/
if (retain_pin)
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
else
_hash_relbuf(rel, buf);
buf = next_buf;
}
/*
* lock the bucket page to clear the garbage flag and squeeze the bucket.
* if the current buffer is same as bucket buffer, then we already have
* lock on bucket page.
*/
if (buf != bucket_buf)
{
_hash_relbuf(rel, buf);
LockBuffer(bucket_buf, BUFFER_LOCK_EXCLUSIVE);
}
/*
* Clear the garbage flag from bucket after deleting the tuples that are
* moved by split. We purposefully clear the flag before squeeze bucket,
* so that after restart, vacuum shouldn't again try to delete the moved
* by split tuples.
*/
if (split_cleanup)
{
HashPageOpaque bucket_opaque;
Page page;
page = BufferGetPage(bucket_buf);
bucket_opaque = (HashPageOpaque) PageGetSpecialPointer(page);
/* No ereport(ERROR) until changes are logged */
START_CRIT_SECTION();
bucket_opaque->hasho_flag &= ~LH_BUCKET_NEEDS_SPLIT_CLEANUP;
MarkBufferDirty(bucket_buf);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
XLogRecPtr recptr;
XLogBeginInsert();
XLogRegisterBuffer(0, bucket_buf, REGBUF_STANDARD);
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_CLEANUP);
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
}
/*
* If we have deleted anything, try to compact free space. For squeezing
* the bucket, we must have a cleanup lock, else it can impact the
* ordering of tuples for a scan that has started before it.
*/
if (bucket_dirty && IsBufferCleanupOK(bucket_buf))
_hash_squeezebucket(rel, cur_bucket, bucket_blkno, bucket_buf,
bstrategy);
else
LockBuffer(bucket_buf, BUFFER_LOCK_UNLOCK);
}
/*
* _hash_first() -- Find the first item in a scan.
*
* Find the first item in the index that
* satisfies the qualification associated with the scan descriptor. On
* success, the page containing the current index tuple is read locked
* and pinned, and the scan's opaque data entry is updated to
* include the buffer.
*/
bool
_hash_first(IndexScanDesc scan, ScanDirection dir)
{
Relation rel = scan->indexRelation;
HashScanOpaque so = (HashScanOpaque) scan->opaque;
ScanKey cur;
uint32 hashkey;
Bucket bucket;
BlockNumber blkno;
Buffer buf;
Buffer metabuf;
Page page;
HashPageOpaque opaque;
HashMetaPage metap;
IndexTuple itup;
ItemPointer current;
OffsetNumber offnum;
pgstat_count_index_scan(rel);
current = &(so->hashso_curpos);
ItemPointerSetInvalid(current);
/*
* We do not support hash scans with no index qualification, because we
* would have to read the whole index rather than just one bucket. That
* creates a whole raft of problems, since we haven't got a practical way
* to lock all the buckets against splits or compactions.
*/
if (scan->numberOfKeys < 1)
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("hash indexes do not support whole-index scans")));
/* There may be more than one index qual, but we hash only the first */
cur = &scan->keyData[0];
/* We support only single-column hash indexes */
Assert(cur->sk_attno == 1);
/* And there's only one operator strategy, too */
Assert(cur->sk_strategy == HTEqualStrategyNumber);
/*
* If the constant in the index qual is NULL, assume it cannot match any
* items in the index.
*/
if (cur->sk_flags & SK_ISNULL)
return false;
/*
* Okay to compute the hash key. We want to do this before acquiring any
* locks, in case a user-defined hash function happens to be slow.
*
* If scankey operator is not a cross-type comparison, we can use the
* cached hash function; otherwise gotta look it up in the catalogs.
*
* We support the convention that sk_subtype == InvalidOid means the
* opclass input type; this is a hack to simplify life for ScanKeyInit().
*/
if (cur->sk_subtype == rel->rd_opcintype[0] ||
cur->sk_subtype == InvalidOid)
hashkey = _hash_datum2hashkey(rel, cur->sk_argument);
else
hashkey = _hash_datum2hashkey_type(rel, cur->sk_argument,
cur->sk_subtype);
so->hashso_sk_hash = hashkey;
/*
* Acquire shared split lock so we can compute the target bucket safely
* (see README).
*/
_hash_getlock(rel, 0, HASH_SHARE);
/* Read the metapage */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* done with the metapage */
_hash_relbuf(rel, metabuf);
/*
* Acquire share lock on target bucket; then we can release split lock.
*/
_hash_getlock(rel, blkno, HASH_SHARE);
_hash_droplock(rel, 0, HASH_SHARE);
/* Update scan opaque state to show we have lock on the bucket */
so->hashso_bucket = bucket;
so->hashso_bucket_valid = true;
so->hashso_bucket_blkno = blkno;
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, HASH_READ, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(opaque->hasho_bucket == bucket);
/* If a backwards scan is requested, move to the end of the chain */
if (ScanDirectionIsBackward(dir))
{
while (BlockNumberIsValid(opaque->hasho_nextblkno))
_hash_readnext(rel, &buf, &page, &opaque);
}
/* Now find the first tuple satisfying the qualification */
if (!_hash_step(scan, &buf, dir))
return false;
/* if we're here, _hash_step found a valid tuple */
offnum = ItemPointerGetOffsetNumber(current);
_hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
page = BufferGetPage(buf);
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
so->hashso_heappos = itup->t_tid;
return true;
}
/*
* _hash_splitbucket -- split 'obucket' into 'obucket' and 'nbucket'
*
* We are splitting a bucket that consists of a base bucket page and zero
* or more overflow (bucket chain) pages. We must relocate tuples that
* belong in the new bucket, and compress out any free space in the old
* bucket.
*
* The caller must hold exclusive locks on both buckets to ensure that
* no one else is trying to access them (see README).
*
* The caller must hold a pin, but no lock, on the metapage buffer.
* The buffer is returned in the same state. (The metapage is only
* touched if it becomes necessary to add or remove overflow pages.)
*/
static void
_hash_splitbucket(Relation rel,
Buffer metabuf,
Bucket obucket,
Bucket nbucket,
BlockNumber start_oblkno,
BlockNumber start_nblkno,
uint32 maxbucket,
uint32 highmask,
uint32 lowmask)
{
Bucket bucket;
Buffer obuf;
Buffer nbuf;
BlockNumber oblkno;
BlockNumber nblkno;
bool null;
Datum datum;
HashItem hitem;
HashPageOpaque oopaque;
HashPageOpaque nopaque;
IndexTuple itup;
Size itemsz;
OffsetNumber ooffnum;
OffsetNumber noffnum;
OffsetNumber omaxoffnum;
Page opage;
Page npage;
TupleDesc itupdesc = RelationGetDescr(rel);
/*
* It should be okay to simultaneously write-lock pages from each
* bucket, since no one else can be trying to acquire buffer lock
* on pages of either bucket.
*/
oblkno = start_oblkno;
nblkno = start_nblkno;
obuf = _hash_getbuf(rel, oblkno, HASH_WRITE);
nbuf = _hash_getbuf(rel, nblkno, HASH_WRITE);
opage = BufferGetPage(obuf);
npage = BufferGetPage(nbuf);
_hash_checkpage(rel, opage, LH_BUCKET_PAGE);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
/* initialize the new bucket's primary page */
_hash_pageinit(npage, BufferGetPageSize(nbuf));
nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
nopaque->hasho_prevblkno = InvalidBlockNumber;
nopaque->hasho_nextblkno = InvalidBlockNumber;
nopaque->hasho_bucket = nbucket;
nopaque->hasho_flag = LH_BUCKET_PAGE;
nopaque->hasho_filler = HASHO_FILL;
/*
* Partition the tuples in the old bucket between the old bucket and the
* new bucket, advancing along the old bucket's overflow bucket chain
* and adding overflow pages to the new bucket as needed.
*/
ooffnum = FirstOffsetNumber;
omaxoffnum = PageGetMaxOffsetNumber(opage);
for (;;)
{
/*
* at each iteration through this loop, each of these variables
* should be up-to-date: obuf opage oopaque ooffnum omaxoffnum
*/
/* check if we're at the end of the page */
if (ooffnum > omaxoffnum)
{
/* at end of page, but check for an(other) overflow page */
oblkno = oopaque->hasho_nextblkno;
if (!BlockNumberIsValid(oblkno))
break;
/*
* we ran out of tuples on this particular page, but we
* have more overflow pages; advance to next page.
*/
_hash_wrtbuf(rel, obuf);
obuf = _hash_getbuf(rel, oblkno, HASH_WRITE);
opage = BufferGetPage(obuf);
_hash_checkpage(rel, opage, LH_OVERFLOW_PAGE);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
ooffnum = FirstOffsetNumber;
omaxoffnum = PageGetMaxOffsetNumber(opage);
continue;
}
/*
* Re-hash the tuple to determine which bucket it now belongs in.
*
* It is annoying to call the hash function while holding locks,
* but releasing and relocking the page for each tuple is unappealing
* too.
*/
hitem = (HashItem) PageGetItem(opage, PageGetItemId(opage, ooffnum));
itup = &(hitem->hash_itup);
datum = index_getattr(itup, 1, itupdesc, &null);
Assert(!null);
bucket = _hash_hashkey2bucket(_hash_datum2hashkey(rel, datum),
maxbucket, highmask, lowmask);
if (bucket == nbucket)
{
/*
* insert the tuple into the new bucket. if it doesn't fit on
* the current page in the new bucket, we must allocate a new
* overflow page and place the tuple on that page instead.
*/
itemsz = IndexTupleDSize(hitem->hash_itup)
+ (sizeof(HashItemData) - sizeof(IndexTupleData));
itemsz = MAXALIGN(itemsz);
if (PageGetFreeSpace(npage) < itemsz)
{
/* write out nbuf and drop lock, but keep pin */
_hash_chgbufaccess(rel, nbuf, HASH_WRITE, HASH_NOLOCK);
/* chain to a new overflow page */
nbuf = _hash_addovflpage(rel, metabuf, nbuf);
npage = BufferGetPage(nbuf);
_hash_checkpage(rel, npage, LH_OVERFLOW_PAGE);
/* we don't need nopaque within the loop */
}
noffnum = OffsetNumberNext(PageGetMaxOffsetNumber(npage));
if (PageAddItem(npage, (Item) hitem, itemsz, noffnum, LP_USED)
== InvalidOffsetNumber)
elog(ERROR, "failed to add index item to \"%s\"",
RelationGetRelationName(rel));
/*
* now delete the tuple from the old bucket. after this
* section of code, 'ooffnum' will actually point to the
* ItemId to which we would point if we had advanced it before
* the deletion (PageIndexTupleDelete repacks the ItemId
* array). this also means that 'omaxoffnum' is exactly one
* less than it used to be, so we really can just decrement it
* instead of calling PageGetMaxOffsetNumber.
*/
PageIndexTupleDelete(opage, ooffnum);
omaxoffnum = OffsetNumberPrev(omaxoffnum);
}
else
{
/*
* the tuple stays on this page. we didn't move anything, so
* we didn't delete anything and therefore we don't have to
* change 'omaxoffnum'.
*/
Assert(bucket == obucket);
ooffnum = OffsetNumberNext(ooffnum);
}
}
/*
* We're at the end of the old bucket chain, so we're done partitioning
* the tuples. Before quitting, call _hash_squeezebucket to ensure the
* tuples remaining in the old bucket (including the overflow pages) are
* packed as tightly as possible. The new bucket is already tight.
*/
_hash_wrtbuf(rel, obuf);
_hash_wrtbuf(rel, nbuf);
_hash_squeezebucket(rel, obucket, start_oblkno);
}
/*
* Helper function to perform deletion of index entries from a bucket.
*
* This function expects that the caller has acquired a cleanup lock on the
* primary bucket page, and will return with a write lock again held on the
* primary bucket page. The lock won't necessarily be held continuously,
* though, because we'll release it when visiting overflow pages.
*
* It would be very bad if this function cleaned a page while some other
* backend was in the midst of scanning it, because hashgettuple assumes
* that the next valid TID will be greater than or equal to the current
* valid TID. There can't be any concurrent scans in progress when we first
* enter this function because of the cleanup lock we hold on the primary
* bucket page, but as soon as we release that lock, there might be. We
* handle that by conspiring to prevent those scans from passing our cleanup
* scan. To do that, we lock the next page in the bucket chain before
* releasing the lock on the previous page. (This type of lock chaining is
* not ideal, so we might want to look for a better solution at some point.)
*
* We need to retain a pin on the primary bucket to ensure that no concurrent
* split can start.
*/
void
hashbucketcleanup(Relation rel, Bucket cur_bucket, Buffer bucket_buf,
BlockNumber bucket_blkno, BufferAccessStrategy bstrategy,
uint32 maxbucket, uint32 highmask, uint32 lowmask,
double *tuples_removed, double *num_index_tuples,
bool split_cleanup,
IndexBulkDeleteCallback callback, void *callback_state)
{
BlockNumber blkno;
Buffer buf;
Bucket new_bucket PG_USED_FOR_ASSERTS_ONLY = InvalidBucket;
bool bucket_dirty = false;
blkno = bucket_blkno;
buf = bucket_buf;
if (split_cleanup)
new_bucket = _hash_get_newbucket_from_oldbucket(rel, cur_bucket,
lowmask, maxbucket);
/* Scan each page in bucket */
for (;;)
{
HashPageOpaque opaque;
OffsetNumber offno;
OffsetNumber maxoffno;
Buffer next_buf;
Page page;
OffsetNumber deletable[MaxOffsetNumber];
int ndeletable = 0;
bool retain_pin = false;
bool curr_page_dirty = false;
vacuum_delay_point();
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
/* Scan each tuple in page */
maxoffno = PageGetMaxOffsetNumber(page);
for (offno = FirstOffsetNumber;
offno <= maxoffno;
offno = OffsetNumberNext(offno))
{
ItemPointer htup;
IndexTuple itup;
Bucket bucket;
bool kill_tuple = false;
itup = (IndexTuple) PageGetItem(page,
PageGetItemId(page, offno));
htup = &(itup->t_tid);
/*
* To remove the dead tuples, we strictly want to rely on results
* of callback function. refer btvacuumpage for detailed reason.
*/
if (callback && callback(htup, callback_state))
{
kill_tuple = true;
if (tuples_removed)
*tuples_removed += 1;
}
else if (split_cleanup)
{
/* delete the tuples that are moved by split. */
bucket = _hash_hashkey2bucket(_hash_get_indextuple_hashkey(itup),
maxbucket,
highmask,
lowmask);
/* mark the item for deletion */
if (bucket != cur_bucket)
{
/*
* We expect tuples to either belong to curent bucket or
* new_bucket. This is ensured because we don't allow
* further splits from bucket that contains garbage. See
* comments in _hash_expandtable.
*/
Assert(bucket == new_bucket);
kill_tuple = true;
}
}
if (kill_tuple)
{
/* mark the item for deletion */
deletable[ndeletable++] = offno;
}
else
{
/* we're keeping it, so count it */
if (num_index_tuples)
*num_index_tuples += 1;
}
}
/* retain the pin on primary bucket page till end of bucket scan */
if (blkno == bucket_blkno)
retain_pin = true;
else
retain_pin = false;
blkno = opaque->hasho_nextblkno;
/*
* Apply deletions, advance to next page and write page if needed.
*/
if (ndeletable > 0)
{
PageIndexMultiDelete(page, deletable, ndeletable);
bucket_dirty = true;
curr_page_dirty = true;
}
/* bail out if there are no more pages to scan. */
if (!BlockNumberIsValid(blkno))
break;
next_buf = _hash_getbuf_with_strategy(rel, blkno, HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
/*
* release the lock on previous page after acquiring the lock on next
* page
*/
if (curr_page_dirty)
{
if (retain_pin)
_hash_chgbufaccess(rel, buf, HASH_WRITE, HASH_NOLOCK);
else
_hash_wrtbuf(rel, buf);
curr_page_dirty = false;
}
else if (retain_pin)
_hash_chgbufaccess(rel, buf, HASH_READ, HASH_NOLOCK);
else
_hash_relbuf(rel, buf);
buf = next_buf;
}
/*
* lock the bucket page to clear the garbage flag and squeeze the bucket.
* if the current buffer is same as bucket buffer, then we already have
* lock on bucket page.
*/
if (buf != bucket_buf)
{
_hash_relbuf(rel, buf);
_hash_chgbufaccess(rel, bucket_buf, HASH_NOLOCK, HASH_WRITE);
}
/*
* Clear the garbage flag from bucket after deleting the tuples that are
* moved by split. We purposefully clear the flag before squeeze bucket,
* so that after restart, vacuum shouldn't again try to delete the moved
* by split tuples.
*/
if (split_cleanup)
{
HashPageOpaque bucket_opaque;
Page page;
page = BufferGetPage(bucket_buf);
bucket_opaque = (HashPageOpaque) PageGetSpecialPointer(page);
bucket_opaque->hasho_flag &= ~LH_BUCKET_NEEDS_SPLIT_CLEANUP;
}
/*
* If we have deleted anything, try to compact free space. For squeezing
* the bucket, we must have a cleanup lock, else it can impact the
* ordering of tuples for a scan that has started before it.
*/
if (bucket_dirty && IsBufferCleanupOK(bucket_buf))
_hash_squeezebucket(rel, cur_bucket, bucket_blkno, bucket_buf,
bstrategy);
else
_hash_chgbufaccess(rel, bucket_buf, HASH_WRITE, HASH_NOLOCK);
}
/*
* _hash_splitbucket -- split 'obucket' into 'obucket' and 'nbucket'
*
* We are splitting a bucket that consists of a base bucket page and zero
* or more overflow (bucket chain) pages. We must relocate tuples that
* belong in the new bucket, and compress out any free space in the old
* bucket.
*
* The caller must hold exclusive locks on both buckets to ensure that
* no one else is trying to access them (see README).
*
* The caller must hold a pin, but no lock, on the metapage buffer.
* The buffer is returned in the same state. (The metapage is only
* touched if it becomes necessary to add or remove overflow pages.)
*/
static void
_hash_splitbucket(Relation rel,
Buffer metabuf,
Bucket obucket,
Bucket nbucket,
BlockNumber start_oblkno,
BlockNumber start_nblkno,
uint32 maxbucket,
uint32 highmask,
uint32 lowmask)
{
BlockNumber oblkno;
BlockNumber nblkno;
Buffer obuf;
Buffer nbuf;
Page opage;
Page npage;
HashPageOpaque oopaque;
HashPageOpaque nopaque;
/*
* It should be okay to simultaneously write-lock pages from each bucket,
* since no one else can be trying to acquire buffer lock on pages of
* either bucket.
*/
oblkno = start_oblkno;
obuf = _hash_getbuf(rel, oblkno, HASH_WRITE, LH_BUCKET_PAGE);
opage = BufferGetPage(obuf);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
nblkno = start_nblkno;
nbuf = _hash_getnewbuf(rel, nblkno, MAIN_FORKNUM);
npage = BufferGetPage(nbuf);
/* initialize the new bucket's primary page */
nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
nopaque->hasho_prevblkno = InvalidBlockNumber;
nopaque->hasho_nextblkno = InvalidBlockNumber;
nopaque->hasho_bucket = nbucket;
nopaque->hasho_flag = LH_BUCKET_PAGE;
nopaque->hasho_page_id = HASHO_PAGE_ID;
/*
* Partition the tuples in the old bucket between the old bucket and the
* new bucket, advancing along the old bucket's overflow bucket chain and
* adding overflow pages to the new bucket as needed. Outer loop iterates
* once per page in old bucket.
*/
for (;;)
{
OffsetNumber ooffnum;
OffsetNumber omaxoffnum;
OffsetNumber deletable[MaxOffsetNumber];
int ndeletable = 0;
/* Scan each tuple in old page */
omaxoffnum = PageGetMaxOffsetNumber(opage);
for (ooffnum = FirstOffsetNumber;
ooffnum <= omaxoffnum;
ooffnum = OffsetNumberNext(ooffnum))
{
IndexTuple itup;
Size itemsz;
Bucket bucket;
/*
* Fetch the item's hash key (conveniently stored in the item) and
* determine which bucket it now belongs in.
*/
itup = (IndexTuple) PageGetItem(opage,
PageGetItemId(opage, ooffnum));
bucket = _hash_hashkey2bucket(_hash_get_indextuple_hashkey(itup),
maxbucket, highmask, lowmask);
if (bucket == nbucket)
{
/*
* insert the tuple into the new bucket. if it doesn't fit on
* the current page in the new bucket, we must allocate a new
* overflow page and place the tuple on that page instead.
*/
itemsz = IndexTupleDSize(*itup);
itemsz = MAXALIGN(itemsz);
if (PageGetFreeSpace(npage) < itemsz)
{
/* write out nbuf and drop lock, but keep pin */
_hash_chgbufaccess(rel, nbuf, HASH_WRITE, HASH_NOLOCK);
/* chain to a new overflow page */
nbuf = _hash_addovflpage(rel, metabuf, nbuf);
npage = BufferGetPage(nbuf);
/* we don't need nblkno or nopaque within the loop */
}
/*
* Insert tuple on new page, using _hash_pgaddtup to ensure
* correct ordering by hashkey. This is a tad inefficient
* since we may have to shuffle itempointers repeatedly.
* Possible future improvement: accumulate all the items for
* the new page and qsort them before insertion.
*/
(void) _hash_pgaddtup(rel, nbuf, itemsz, itup);
/*
* Mark tuple for deletion from old page.
*/
deletable[ndeletable++] = ooffnum;
}
else
{
/*
* the tuple stays on this page, so nothing to do.
*/
Assert(bucket == obucket);
}
}
oblkno = oopaque->hasho_nextblkno;
/*
* Done scanning this old page. If we moved any tuples, delete them
* from the old page.
*/
if (ndeletable > 0)
{
PageIndexMultiDelete(opage, deletable, ndeletable);
_hash_wrtbuf(rel, obuf);
}
else
_hash_relbuf(rel, obuf);
/* Exit loop if no more overflow pages in old bucket */
if (!BlockNumberIsValid(oblkno))
break;
/* Else, advance to next old page */
obuf = _hash_getbuf(rel, oblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
opage = BufferGetPage(obuf);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
}
/*
* We're at the end of the old bucket chain, so we're done partitioning
* the tuples. Before quitting, call _hash_squeezebucket to ensure the
* tuples remaining in the old bucket (including the overflow pages) are
* packed as tightly as possible. The new bucket is already tight.
*/
_hash_wrtbuf(rel, nbuf);
_hash_squeezebucket(rel, obucket, start_oblkno, NULL);
}
/*
* _hash_doinsert() -- Handle insertion of a single index tuple.
*
* This routine is called by the public interface routines, hashbuild
* and hashinsert. By here, itup is completely filled in.
*/
void
_hash_doinsert(Relation rel, IndexTuple itup)
{
Buffer buf;
Buffer metabuf;
HashMetaPage metap;
BlockNumber blkno;
Page page;
HashPageOpaque pageopaque;
Size itemsz;
bool do_expand;
uint32 hashkey;
Bucket bucket;
/*
* Get the hash key for the item (it's stored in the index tuple itself).
*/
hashkey = _hash_get_indextuple_hashkey(itup);
/* compute item size too */
itemsz = IndexTupleDSize(*itup);
itemsz = MAXALIGN(itemsz); /* be safe, PageAddItem will do this but we
* need to be consistent */
/*
* Acquire shared split lock so we can compute the target bucket safely
* (see README).
*/
_hash_getlock(rel, 0, HASH_SHARE);
/* Read the metapage */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/*
* Check whether the item can fit on a hash page at all. (Eventually, we
* ought to try to apply TOAST methods if not.) Note that at this point,
* itemsz doesn't include the ItemId.
*
* XXX this is useless code if we are only storing hash keys.
*/
if (itemsz > HashMaxItemSize((Page) metap))
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds hash maximum %lu",
(unsigned long) itemsz,
(unsigned long) HashMaxItemSize((Page) metap)),
errhint("Values larger than a buffer page cannot be indexed.")));
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* release lock on metapage, but keep pin since we'll need it again */
_hash_chgbufaccess(rel, metabuf, HASH_READ, HASH_NOLOCK);
/*
* Acquire share lock on target bucket; then we can release split lock.
*/
_hash_getlock(rel, blkno, HASH_SHARE);
_hash_droplock(rel, 0, HASH_SHARE);
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, HASH_WRITE, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(pageopaque->hasho_bucket == bucket);
/* Do the insertion */
while (PageGetFreeSpace(page) < itemsz)
{
/*
* no space on this page; check for an overflow page
*/
BlockNumber nextblkno = pageopaque->hasho_nextblkno;
if (BlockNumberIsValid(nextblkno))
{
/*
* ovfl page exists; go get it. if it doesn't have room, we'll
* find out next pass through the loop test above.
*/
_hash_relbuf(rel, buf);
buf = _hash_getbuf(rel, nextblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
page = BufferGetPage(buf);
}
else
{
/*
* we're at the end of the bucket chain and we haven't found a
* page with enough room. allocate a new overflow page.
*/
/* release our write lock without modifying buffer */
_hash_chgbufaccess(rel, buf, HASH_READ, HASH_NOLOCK);
/* chain to a new overflow page */
buf = _hash_addovflpage(rel, metabuf, buf);
page = BufferGetPage(buf);
/* should fit now, given test above */
Assert(PageGetFreeSpace(page) >= itemsz);
}
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(pageopaque->hasho_flag == LH_OVERFLOW_PAGE);
Assert(pageopaque->hasho_bucket == bucket);
}
/* found page with enough space, so add the item here */
(void) _hash_pgaddtup(rel, buf, itemsz, itup);
/* write and release the modified page */
_hash_wrtbuf(rel, buf);
/* We can drop the bucket lock now */
_hash_droplock(rel, blkno, HASH_SHARE);
/*
* Write-lock the metapage so we can increment the tuple count. After
* incrementing it, check to see if it's time for a split.
*/
_hash_chgbufaccess(rel, metabuf, HASH_NOLOCK, HASH_WRITE);
metap->hashm_ntuples += 1;
/* Make sure this stays in sync with _hash_expandtable() */
do_expand = metap->hashm_ntuples >
(double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1);
/* Write out the metapage and drop lock, but keep pin */
_hash_chgbufaccess(rel, metabuf, HASH_WRITE, HASH_NOLOCK);
/* Attempt to split if a split is needed */
if (do_expand)
_hash_expandtable(rel, metabuf);
/* Finally drop our pin on the metapage */
_hash_dropbuf(rel, metabuf);
}
/*
* _hash_splitbucket -- split 'obucket' into 'obucket' and 'nbucket'
*
* This routine is used to partition the tuples between old and new bucket and
* is used to finish the incomplete split operations. To finish the previously
* interrupted split operation, the caller needs to fill htab. If htab is set,
* then we skip the movement of tuples that exists in htab, otherwise NULL
* value of htab indicates movement of all the tuples that belong to the new
* bucket.
*
* We are splitting a bucket that consists of a base bucket page and zero
* or more overflow (bucket chain) pages. We must relocate tuples that
* belong in the new bucket.
*
* The caller must hold cleanup locks on both buckets to ensure that
* no one else is trying to access them (see README).
*
* The caller must hold a pin, but no lock, on the metapage buffer.
* The buffer is returned in the same state. (The metapage is only
* touched if it becomes necessary to add or remove overflow pages.)
*
* Split needs to retain pin on primary bucket pages of both old and new
* buckets till end of operation. This is to prevent vacuum from starting
* while a split is in progress.
*
* In addition, the caller must have created the new bucket's base page,
* which is passed in buffer nbuf, pinned and write-locked. The lock will be
* released here and pin must be released by the caller. (The API is set up
* this way because we must do _hash_getnewbuf() before releasing the metapage
* write lock. So instead of passing the new bucket's start block number, we
* pass an actual buffer.)
*/
static void
_hash_splitbucket(Relation rel,
Buffer metabuf,
Bucket obucket,
Bucket nbucket,
Buffer obuf,
Buffer nbuf,
HTAB *htab,
uint32 maxbucket,
uint32 highmask,
uint32 lowmask)
{
Buffer bucket_obuf;
Buffer bucket_nbuf;
Page opage;
Page npage;
HashPageOpaque oopaque;
HashPageOpaque nopaque;
OffsetNumber itup_offsets[MaxIndexTuplesPerPage];
IndexTuple itups[MaxIndexTuplesPerPage];
Size all_tups_size = 0;
int i;
uint16 nitups = 0;
bucket_obuf = obuf;
opage = BufferGetPage(obuf);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
bucket_nbuf = nbuf;
npage = BufferGetPage(nbuf);
nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
/*
* Partition the tuples in the old bucket between the old bucket and the
* new bucket, advancing along the old bucket's overflow bucket chain and
* adding overflow pages to the new bucket as needed. Outer loop iterates
* once per page in old bucket.
*/
for (;;)
{
BlockNumber oblkno;
OffsetNumber ooffnum;
OffsetNumber omaxoffnum;
/* Scan each tuple in old page */
omaxoffnum = PageGetMaxOffsetNumber(opage);
for (ooffnum = FirstOffsetNumber;
ooffnum <= omaxoffnum;
ooffnum = OffsetNumberNext(ooffnum))
{
IndexTuple itup;
Size itemsz;
Bucket bucket;
bool found = false;
/* skip dead tuples */
if (ItemIdIsDead(PageGetItemId(opage, ooffnum)))
continue;
/*
* Before inserting a tuple, probe the hash table containing TIDs
* of tuples belonging to new bucket, if we find a match, then
* skip that tuple, else fetch the item's hash key (conveniently
* stored in the item) and determine which bucket it now belongs
* in.
*/
itup = (IndexTuple) PageGetItem(opage,
PageGetItemId(opage, ooffnum));
if (htab)
(void) hash_search(htab, &itup->t_tid, HASH_FIND, &found);
if (found)
continue;
bucket = _hash_hashkey2bucket(_hash_get_indextuple_hashkey(itup),
maxbucket, highmask, lowmask);
if (bucket == nbucket)
{
IndexTuple new_itup;
/*
* make a copy of index tuple as we have to scribble on it.
*/
new_itup = CopyIndexTuple(itup);
/*
* mark the index tuple as moved by split, such tuples are
* skipped by scan if there is split in progress for a bucket.
*/
new_itup->t_info |= INDEX_MOVED_BY_SPLIT_MASK;
/*
* insert the tuple into the new bucket. if it doesn't fit on
* the current page in the new bucket, we must allocate a new
* overflow page and place the tuple on that page instead.
*/
itemsz = IndexTupleDSize(*new_itup);
itemsz = MAXALIGN(itemsz);
if (PageGetFreeSpaceForMultipleTuples(npage, nitups + 1) < (all_tups_size + itemsz))
{
/*
* Change the shared buffer state in critical section,
* otherwise any error could make it unrecoverable.
*/
START_CRIT_SECTION();
_hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
MarkBufferDirty(nbuf);
/* log the split operation before releasing the lock */
log_split_page(rel, nbuf);
END_CRIT_SECTION();
/* drop lock, but keep pin */
LockBuffer(nbuf, BUFFER_LOCK_UNLOCK);
/* be tidy */
for (i = 0; i < nitups; i++)
pfree(itups[i]);
nitups = 0;
all_tups_size = 0;
/* chain to a new overflow page */
nbuf = _hash_addovflpage(rel, metabuf, nbuf, (nbuf == bucket_nbuf) ? true : false);
npage = BufferGetPage(nbuf);
nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
}
itups[nitups++] = new_itup;
all_tups_size += itemsz;
}
else
{
/*
* the tuple stays on this page, so nothing to do.
*/
Assert(bucket == obucket);
}
}
oblkno = oopaque->hasho_nextblkno;
/* retain the pin on the old primary bucket */
if (obuf == bucket_obuf)
LockBuffer(obuf, BUFFER_LOCK_UNLOCK);
else
_hash_relbuf(rel, obuf);
/* Exit loop if no more overflow pages in old bucket */
if (!BlockNumberIsValid(oblkno))
{
/*
* Change the shared buffer state in critical section, otherwise
* any error could make it unrecoverable.
*/
START_CRIT_SECTION();
_hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
MarkBufferDirty(nbuf);
/* log the split operation before releasing the lock */
log_split_page(rel, nbuf);
END_CRIT_SECTION();
if (nbuf == bucket_nbuf)
LockBuffer(nbuf, BUFFER_LOCK_UNLOCK);
else
_hash_relbuf(rel, nbuf);
/* be tidy */
for (i = 0; i < nitups; i++)
pfree(itups[i]);
break;
}
/* Else, advance to next old page */
obuf = _hash_getbuf(rel, oblkno, HASH_READ, LH_OVERFLOW_PAGE);
opage = BufferGetPage(obuf);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
}
/*
* We're at the end of the old bucket chain, so we're done partitioning
* the tuples. Mark the old and new buckets to indicate split is
* finished.
*
* To avoid deadlocks due to locking order of buckets, first lock the old
* bucket and then the new bucket.
*/
LockBuffer(bucket_obuf, BUFFER_LOCK_EXCLUSIVE);
opage = BufferGetPage(bucket_obuf);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
LockBuffer(bucket_nbuf, BUFFER_LOCK_EXCLUSIVE);
npage = BufferGetPage(bucket_nbuf);
nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
START_CRIT_SECTION();
oopaque->hasho_flag &= ~LH_BUCKET_BEING_SPLIT;
nopaque->hasho_flag &= ~LH_BUCKET_BEING_POPULATED;
/*
* After the split is finished, mark the old bucket to indicate that it
* contains deletable tuples. We will clear split-cleanup flag after
* deleting such tuples either at the end of split or at the next split
* from old bucket or at the time of vacuum.
*/
oopaque->hasho_flag |= LH_BUCKET_NEEDS_SPLIT_CLEANUP;
/*
* now write the buffers, here we don't release the locks as caller is
* responsible to release locks.
*/
MarkBufferDirty(bucket_obuf);
MarkBufferDirty(bucket_nbuf);
if (RelationNeedsWAL(rel))
{
XLogRecPtr recptr;
xl_hash_split_complete xlrec;
xlrec.old_bucket_flag = oopaque->hasho_flag;
xlrec.new_bucket_flag = nopaque->hasho_flag;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHashSplitComplete);
XLogRegisterBuffer(0, bucket_obuf, REGBUF_STANDARD);
XLogRegisterBuffer(1, bucket_nbuf, REGBUF_STANDARD);
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_COMPLETE);
PageSetLSN(BufferGetPage(bucket_obuf), recptr);
PageSetLSN(BufferGetPage(bucket_nbuf), recptr);
}
END_CRIT_SECTION();
/*
* If possible, clean up the old bucket. We might not be able to do this
* if someone else has a pin on it, but if not then we can go ahead. This
* isn't absolutely necessary, but it reduces bloat; if we don't do it
* now, VACUUM will do it eventually, but maybe not until new overflow
* pages have been allocated. Note that there's no need to clean up the
* new bucket.
*/
if (IsBufferCleanupOK(bucket_obuf))
{
LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
hashbucketcleanup(rel, obucket, bucket_obuf,
BufferGetBlockNumber(bucket_obuf), NULL,
maxbucket, highmask, lowmask, NULL, NULL, true,
NULL, NULL);
}
else
{
LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
LockBuffer(bucket_obuf, BUFFER_LOCK_UNLOCK);
}
}
/*
* _hash_doinsert() -- Handle insertion of a single HashItem in the table.
*
* This routine is called by the public interface routines, hashbuild
* and hashinsert. By here, hashitem is completely filled in.
* The datum to be used as a "key" is in the hashitem.
*/
InsertIndexResult
_hash_doinsert(Relation rel, HashItem hitem)
{
Buffer buf;
Buffer metabuf;
HashMetaPage metap;
IndexTuple itup;
BlockNumber itup_blkno;
OffsetNumber itup_off;
InsertIndexResult res;
BlockNumber blkno;
Page page;
HashPageOpaque pageopaque;
Size itemsz;
bool do_expand;
uint32 hashkey;
Bucket bucket;
Datum datum;
bool isnull;
/*
* Compute the hash key for the item. We do this first so as not to
* need to hold any locks while running the hash function.
*/
itup = &(hitem->hash_itup);
if (rel->rd_rel->relnatts != 1)
elog(ERROR, "hash indexes support only one index key");
datum = index_getattr(itup, 1, RelationGetDescr(rel), &isnull);
Assert(!isnull);
hashkey = _hash_datum2hashkey(rel, datum);
/* compute item size too */
itemsz = IndexTupleDSize(hitem->hash_itup)
+ (sizeof(HashItemData) - sizeof(IndexTupleData));
itemsz = MAXALIGN(itemsz); /* be safe, PageAddItem will do this but
* we need to be consistent */
/*
* Acquire shared split lock so we can compute the target bucket
* safely (see README).
*/
_hash_getlock(rel, 0, HASH_SHARE);
/* Read the metapage */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ);
metap = (HashMetaPage) BufferGetPage(metabuf);
_hash_checkpage(rel, (Page) metap, LH_META_PAGE);
/*
* Check whether the item can fit on a hash page at all. (Eventually,
* we ought to try to apply TOAST methods if not.) Note that at this
* point, itemsz doesn't include the ItemId.
*/
if (itemsz > HashMaxItemSize((Page) metap))
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds hash maximum %lu",
(unsigned long) itemsz,
(unsigned long) HashMaxItemSize((Page) metap))));
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* release lock on metapage, but keep pin since we'll need it again */
_hash_chgbufaccess(rel, metabuf, HASH_READ, HASH_NOLOCK);
/*
* Acquire share lock on target bucket; then we can release split lock.
*/
_hash_getlock(rel, blkno, HASH_SHARE);
_hash_droplock(rel, 0, HASH_SHARE);
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, HASH_WRITE);
page = BufferGetPage(buf);
_hash_checkpage(rel, page, LH_BUCKET_PAGE);
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(pageopaque->hasho_bucket == bucket);
/* Do the insertion */
while (PageGetFreeSpace(page) < itemsz)
{
/*
* no space on this page; check for an overflow page
*/
BlockNumber nextblkno = pageopaque->hasho_nextblkno;
if (BlockNumberIsValid(nextblkno))
{
/*
* ovfl page exists; go get it. if it doesn't have room,
* we'll find out next pass through the loop test above.
*/
_hash_relbuf(rel, buf);
buf = _hash_getbuf(rel, nextblkno, HASH_WRITE);
page = BufferGetPage(buf);
}
else
{
/*
* we're at the end of the bucket chain and we haven't found a
* page with enough room. allocate a new overflow page.
*/
/* release our write lock without modifying buffer */
_hash_chgbufaccess(rel, buf, HASH_READ, HASH_NOLOCK);
/* chain to a new overflow page */
buf = _hash_addovflpage(rel, metabuf, buf);
page = BufferGetPage(buf);
/* should fit now, given test above */
Assert(PageGetFreeSpace(page) >= itemsz);
}
_hash_checkpage(rel, page, LH_OVERFLOW_PAGE);
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(pageopaque->hasho_bucket == bucket);
}
/* found page with enough space, so add the item here */
itup_off = _hash_pgaddtup(rel, buf, itemsz, hitem);
itup_blkno = BufferGetBlockNumber(buf);
/* write and release the modified page */
_hash_wrtbuf(rel, buf);
/* We can drop the bucket lock now */
_hash_droplock(rel, blkno, HASH_SHARE);
/*
* Write-lock the metapage so we can increment the tuple count.
* After incrementing it, check to see if it's time for a split.
*/
_hash_chgbufaccess(rel, metabuf, HASH_NOLOCK, HASH_WRITE);
metap->hashm_ntuples += 1;
/* Make sure this stays in sync with _hash_expandtable() */
do_expand = metap->hashm_ntuples >
(double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1);
/* Write out the metapage and drop lock, but keep pin */
_hash_chgbufaccess(rel, metabuf, HASH_WRITE, HASH_NOLOCK);
/* Attempt to split if a split is needed */
if (do_expand)
_hash_expandtable(rel, metabuf);
/* Finally drop our pin on the metapage */
_hash_dropbuf(rel, metabuf);
/* Create the return data structure */
res = (InsertIndexResult) palloc(sizeof(InsertIndexResultData));
ItemPointerSet(&(res->pointerData), itup_blkno, itup_off);
return res;
}
/*
* _hash_doinsert() -- Handle insertion of a single index tuple.
*
* This routine is called by the public interface routines, hashbuild
* and hashinsert. By here, itup is completely filled in.
*/
void
_hash_doinsert(Relation rel, IndexTuple itup)
{
Buffer buf = InvalidBuffer;
Buffer bucket_buf;
Buffer metabuf;
HashMetaPage metap;
BlockNumber blkno;
BlockNumber oldblkno;
bool retry;
Page page;
HashPageOpaque pageopaque;
Size itemsz;
bool do_expand;
uint32 hashkey;
Bucket bucket;
uint32 maxbucket;
uint32 highmask;
uint32 lowmask;
/*
* Get the hash key for the item (it's stored in the index tuple itself).
*/
hashkey = _hash_get_indextuple_hashkey(itup);
/* compute item size too */
itemsz = IndexTupleDSize(*itup);
itemsz = MAXALIGN(itemsz); /* be safe, PageAddItem will do this but we
* need to be consistent */
restart_insert:
/* Read the metapage */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/*
* Check whether the item can fit on a hash page at all. (Eventually, we
* ought to try to apply TOAST methods if not.) Note that at this point,
* itemsz doesn't include the ItemId.
*
* XXX this is useless code if we are only storing hash keys.
*/
if (itemsz > HashMaxItemSize((Page) metap))
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %zu exceeds hash maximum %zu",
itemsz, HashMaxItemSize((Page) metap)),
errhint("Values larger than a buffer page cannot be indexed.")));
oldblkno = InvalidBlockNumber;
retry = false;
/*
* Loop until we get a lock on the correct target bucket.
*/
for (;;)
{
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/*
* Copy bucket mapping info now; refer the comment in
* _hash_expandtable where we copy this information before calling
* _hash_splitbucket to see why this is okay.
*/
maxbucket = metap->hashm_maxbucket;
highmask = metap->hashm_highmask;
lowmask = metap->hashm_lowmask;
/* Release metapage lock, but keep pin. */
_hash_chgbufaccess(rel, metabuf, HASH_READ, HASH_NOLOCK);
/*
* If the previous iteration of this loop locked the primary page of
* what is still the correct target bucket, we are done. Otherwise,
* drop any old lock before acquiring the new one.
*/
if (retry)
{
if (oldblkno == blkno)
break;
_hash_relbuf(rel, buf);
}
/* Fetch and lock the primary bucket page for the target bucket */
buf = _hash_getbuf(rel, blkno, HASH_WRITE, LH_BUCKET_PAGE);
/*
* Reacquire metapage lock and check that no bucket split has taken
* place while we were awaiting the bucket lock.
*/
_hash_chgbufaccess(rel, metabuf, HASH_NOLOCK, HASH_READ);
oldblkno = blkno;
retry = true;
}
/* remember the primary bucket buffer to release the pin on it at end. */
bucket_buf = buf;
page = BufferGetPage(buf);
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(pageopaque->hasho_bucket == bucket);
/*
* If this bucket is in the process of being split, try to finish the
* split before inserting, because that might create room for the
* insertion to proceed without allocating an additional overflow page.
* It's only interesting to finish the split if we're trying to insert
* into the bucket from which we're removing tuples (the "old" bucket),
* not if we're trying to insert into the bucket into which tuples are
* being moved (the "new" bucket).
*/
if (H_BUCKET_BEING_SPLIT(pageopaque) && IsBufferCleanupOK(buf))
{
/* release the lock on bucket buffer, before completing the split. */
_hash_chgbufaccess(rel, buf, HASH_READ, HASH_NOLOCK);
_hash_finish_split(rel, metabuf, buf, pageopaque->hasho_bucket,
maxbucket, highmask, lowmask);
/* release the pin on old and meta buffer. retry for insert. */
_hash_dropbuf(rel, buf);
_hash_dropbuf(rel, metabuf);
goto restart_insert;
}
/* Do the insertion */
while (PageGetFreeSpace(page) < itemsz)
{
/*
* no space on this page; check for an overflow page
*/
BlockNumber nextblkno = pageopaque->hasho_nextblkno;
if (BlockNumberIsValid(nextblkno))
{
/*
* ovfl page exists; go get it. if it doesn't have room, we'll
* find out next pass through the loop test above. we always
* release both the lock and pin if this is an overflow page, but
* only the lock if this is the primary bucket page, since the pin
* on the primary bucket must be retained throughout the scan.
*/
if (buf != bucket_buf)
_hash_relbuf(rel, buf);
else
_hash_chgbufaccess(rel, buf, HASH_READ, HASH_NOLOCK);
buf = _hash_getbuf(rel, nextblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
page = BufferGetPage(buf);
}
else
{
/*
* we're at the end of the bucket chain and we haven't found a
* page with enough room. allocate a new overflow page.
*/
/* release our write lock without modifying buffer */
_hash_chgbufaccess(rel, buf, HASH_READ, HASH_NOLOCK);
/* chain to a new overflow page */
buf = _hash_addovflpage(rel, metabuf, buf, (buf == bucket_buf) ? true : false);
page = BufferGetPage(buf);
/* should fit now, given test above */
Assert(PageGetFreeSpace(page) >= itemsz);
}
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(pageopaque->hasho_flag == LH_OVERFLOW_PAGE);
Assert(pageopaque->hasho_bucket == bucket);
}
/* found page with enough space, so add the item here */
(void) _hash_pgaddtup(rel, buf, itemsz, itup);
/*
* dirty and release the modified page. if the page we modified was an
* overflow page, we also need to separately drop the pin we retained on
* the primary bucket page.
*/
MarkBufferDirty(buf);
_hash_relbuf(rel, buf);
if (buf != bucket_buf)
_hash_dropbuf(rel, bucket_buf);
/*
* Write-lock the metapage so we can increment the tuple count. After
* incrementing it, check to see if it's time for a split.
*/
_hash_chgbufaccess(rel, metabuf, HASH_NOLOCK, HASH_WRITE);
metap->hashm_ntuples += 1;
/* Make sure this stays in sync with _hash_expandtable() */
do_expand = metap->hashm_ntuples >
(double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1);
/* Write out the metapage and drop lock, but keep pin */
_hash_chgbufaccess(rel, metabuf, HASH_WRITE, HASH_NOLOCK);
/* Attempt to split if a split is needed */
if (do_expand)
_hash_expandtable(rel, metabuf);
/* Finally drop our pin on the metapage */
_hash_dropbuf(rel, metabuf);
}
/*
* _hash_first() -- Find the first item in a scan.
*
* Find the first item in the index that
* satisfies the qualification associated with the scan descriptor. On
* success, the page containing the current index tuple is read locked
* and pinned, and the scan's opaque data entry is updated to
* include the buffer.
*/
bool _hash_first(struct index_scan *scan, enum scandir dir)
{
struct relation *rel = scan->indexRelation;
struct hash_scan_opaque_data *so = (struct hash_scan_opaque_data *)scan->opaque;
struct scankey *cur;
uint32 hashkey;
bucket_t bucket;
block_t blkno;
buf_id_t buf;
buf_id_t metabuf;
page_p page;
struct hash_page *opaque;
struct hash_meta_page_data *metap;
struct index_tuple *itup;
struct item_ptr *current;
item_id_t offnum;
stat_index_scan(rel);
current = &(so->hashso_curpos);
ITEM_PTR_SET_INVALID(current);
/*
* We do not support hash scans with no index qualification, because we
* would have to read the whole index rather than just one bucket. That
* creates a whole raft of problems, since we haven't got a practical way
* to lock all the buckets against splits or compactions.
*/
if (scan->numberOfKeys < 1) {
ereport(ERROR, (
errcode(E_FEATURE_NOT_SUPPORTED),
errmsg("hash indexes do not support whole-index scans")));
}
/* There may be more than one index qual, but we hash only the first */
cur = &scan->keyData[0];
/* We support only single-column hash indexes */
ASSERT(cur->sk_attno == 1);
/* And there's only one operator strategy, too */
ASSERT(cur->sk_strategy == HT_EQ_STRATEGY_NR);
/*
* If the constant in the index qual is NULL, assume it cannot match any
* items in the index.
*/
if (cur->sk_flags & SK_ISNULL)
return false;
/*
* Okay to compute the hash key. We want to do this before acquiring any
* locks, in case a user-defined hash function happens to be slow.
*
* If scankey operator is not a cross-type comparison, we can use the
* cached hash function; otherwise gotta look it up in the catalogs.
*
* We support the convention that sk_subtype == INVALID_OID means the
* opclass input type; this is a hack to simplify life for scankey_init().
*/
if (cur->sk_subtype == rel->rd_opcintype[0]
|| cur->sk_subtype == INVALID_OID)
hashkey = _hash_datum2hashkey(rel, cur->sk_argument);
else
hashkey = _hash_datum2hashkey_type(rel, cur->sk_argument, cur->sk_subtype);
so->hashso_sk_hash = hashkey;
/*
* Acquire shared split lock so we can compute the target bucket safely
* (see README).
*/
_hash_getlock(rel, 0, HASH_SHARE);
/* Read the metapage */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HASH_PAGE_GET_META(BUF_PAGE(metabuf));
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(
hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* done with the metapage */
_hash_relbuf(rel, metabuf);
/*
* Acquire share lock on target bucket; then we can release split lock.
*/
_hash_getlock(rel, blkno, HASH_SHARE);
_hash_droplock(rel, 0, HASH_SHARE);
/* Update scan opaque state to show we have lock on the bucket */
so->hashso_bucket = bucket;
so->hashso_bucket_valid = true;
so->hashso_bucket_blkno = blkno;
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, HASH_READ, LH_BUCKET_PAGE);
page = BUF_PAGE(buf);
opaque = (struct hash_page*) PAGE_SPECIAL_PTR(page);
ASSERT(opaque->hasho_bucket == bucket);
/* If a backwards scan is requested, move to the end of the chain */
if (SCANDIR_BACKWARD(dir)) {
while (BLK_NR_VALID(opaque->hasho_nextblkno))
_hash_readnext(rel, &buf, &page, &opaque);
}
/* Now find the first tuple satisfying the qualification */
if (!_hash_step(scan, &buf, dir))
return false;
/* if we're here, _hash_step found a valid tuple */
offnum = ITEM_PTR_OFFSET(current);
_hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
page = BUF_PAGE(buf);
itup = (struct index_tuple *)PAGE_GET_ITEM(page, PAGE_ITEM_ID(page, offnum));
so->hashso_heappos = itup->t_tid;
return true;
}
