/*
* given a spool loaded by successive calls to _bt_spool,
* create an entire btree.
*/
void
_bt_leafbuild(BTSpool *btspool, BTSpool *btspool2)
{
BTWriteState wstate;
#ifdef BTREE_BUILD_STATS
if (log_btree_build_stats)
{
ShowUsage("BTREE BUILD (Spool) STATISTICS");
ResetUsage();
}
#endif /* BTREE_BUILD_STATS */
if(gp_enable_mk_sort)
{
tuplesort_performsort_mk((Tuplesortstate_mk *) btspool->sortstate);
if (btspool2)
tuplesort_performsort_mk((Tuplesortstate_mk *) btspool2->sortstate);
}
else
{
tuplesort_performsort((Tuplesortstate *) btspool->sortstate);
if (btspool2)
tuplesort_performsort((Tuplesortstate *) btspool2->sortstate);
}
wstate.index = btspool->index;
/*
* We need to log index creation in WAL iff WAL archiving is enabled AND
* it's not a temp index.
*/
wstate.btws_use_wal = !XLog_UnconvertedCanBypassWal() && !wstate.index->rd_istemp;
/* reserve the metapage */
wstate.btws_pages_alloced = BTREE_METAPAGE + 1;
wstate.btws_pages_written = 0;
wstate.btws_zeropage = NULL; /* until needed */
_bt_load(&wstate, btspool, btspool2);
}
/*
* _bitmap_init_buildstate() -- initialize the build state before building
* a bitmap index.
*/
void
_bitmap_init_buildstate(Relation index, BMBuildState *bmstate)
{
MIRROREDLOCK_BUFMGR_DECLARE;
BMMetaPage mp;
HASHCTL hash_ctl;
int hash_flags;
int i;
Buffer metabuf;
/* initialize the build state */
bmstate->bm_tupDesc = RelationGetDescr(index);
bmstate->bm_tidLocsBuffer = (BMTidBuildBuf *)
palloc(sizeof(BMTidBuildBuf));
bmstate->bm_tidLocsBuffer->byte_size = 0;
bmstate->bm_tidLocsBuffer->lov_blocks = NIL;
bmstate->bm_tidLocsBuffer->max_lov_block = InvalidBlockNumber;
// -------- MirroredLock ----------
MIRROREDLOCK_BUFMGR_LOCK;
metabuf = _bitmap_getbuf(index, BM_METAPAGE, BM_READ);
mp = _bitmap_get_metapage_data(index, metabuf);
_bitmap_open_lov_heapandindex(index, mp, &(bmstate->bm_lov_heap),
&(bmstate->bm_lov_index),
RowExclusiveLock);
_bitmap_relbuf(metabuf);
MIRROREDLOCK_BUFMGR_UNLOCK;
// -------- MirroredLock ----------
cur_bmbuild = (BMBuildHashData *)palloc(sizeof(BMBuildHashData));
cur_bmbuild->hash_funcs = (FmgrInfo *)
palloc(sizeof(FmgrInfo) * bmstate->bm_tupDesc->natts);
cur_bmbuild->eq_funcs = (FmgrInfo *)
palloc(sizeof(FmgrInfo) * bmstate->bm_tupDesc->natts);
cur_bmbuild->hash_func_is_strict = (bool *)
palloc(sizeof(bool) * bmstate->bm_tupDesc->natts);
for (i = 0; i < bmstate->bm_tupDesc->natts; i++)
{
Oid typid = bmstate->bm_tupDesc->attrs[i]->atttypid;
Operator optup;
Oid eq_opr;
Oid eq_function;
Oid left_hash_function;
Oid right_hash_function;
optup = equality_oper(typid, false);
eq_opr = oprid(optup);
eq_function = oprfuncid(optup);
ReleaseOperator(optup);
if (!get_op_hash_functions(eq_opr,
&left_hash_function,
&right_hash_function))
{
pfree(cur_bmbuild);
cur_bmbuild = NULL;
break;
}
Assert(left_hash_function == right_hash_function);
fmgr_info(eq_function, &cur_bmbuild->eq_funcs[i]);
fmgr_info(right_hash_function, &cur_bmbuild->hash_funcs[i]);
cur_bmbuild->hash_func_is_strict[i] = func_strict(right_hash_function);
}
if (cur_bmbuild)
{
cur_bmbuild->natts = bmstate->bm_tupDesc->natts;
cur_bmbuild->tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
"Bitmap build temp space",
ALLOCSET_DEFAULT_MINSIZE,
ALLOCSET_DEFAULT_INITSIZE,
ALLOCSET_DEFAULT_MAXSIZE);
/* setup the hash table */
MemSet(&hash_ctl, 0, sizeof(hash_ctl));
/**
* Reserve enough space for the hash key header and then the data segments (values followed by nulls)
*/
hash_ctl.keysize = MAXALIGN(sizeof(BMBuildHashKey)) +
MAXALIGN(sizeof(Datum) * cur_bmbuild->natts) +
MAXALIGN(sizeof(bool) * cur_bmbuild->natts);
hash_ctl.entrysize = hash_ctl.keysize + sizeof(BMBuildLovData) + 200;
hash_ctl.hash = build_hash_key;
hash_ctl.match = build_match_key;
hash_ctl.keycopy = build_keycopy;
hash_ctl.hcxt = AllocSetContextCreate(CurrentMemoryContext,
"Bitmap build hash table",
ALLOCSET_DEFAULT_MINSIZE,
ALLOCSET_DEFAULT_INITSIZE,
ALLOCSET_DEFAULT_MAXSIZE);
cur_bmbuild->hash_cxt = hash_ctl.hcxt;
hash_flags = HASH_ELEM | HASH_FUNCTION | HASH_COMPARE | HASH_CONTEXT | HASH_KEYCOPY;
bmstate->lovitem_hash = hash_create("Bitmap index build lov item hash",
100, &hash_ctl, hash_flags);
bmstate->lovitem_hashKeySize = hash_ctl.keysize;
}
else
{
int attno;
bmstate->lovitem_hash = NULL;
bmstate->lovitem_hashKeySize = 0;
bmstate->bm_lov_scanKeys =
(ScanKey)palloc0(bmstate->bm_tupDesc->natts * sizeof(ScanKeyData));
for (attno = 0; attno < bmstate->bm_tupDesc->natts; attno++)
{
RegProcedure opfuncid;
Oid atttypid;
atttypid = bmstate->bm_tupDesc->attrs[attno]->atttypid;
opfuncid = equality_oper_funcid(atttypid);
ScanKeyEntryInitialize(&(bmstate->bm_lov_scanKeys[attno]), SK_ISNULL,
attno + 1, BTEqualStrategyNumber, InvalidOid,
opfuncid, 0);
}
bmstate->bm_lov_scanDesc = index_beginscan(bmstate->bm_lov_heap,
bmstate->bm_lov_index, ActiveSnapshot,
bmstate->bm_tupDesc->natts,
bmstate->bm_lov_scanKeys);
}
/*
* We need to log index creation in WAL iff WAL archiving is enabled
* AND it's not a temp index. Currently, since building an index
* writes page to the shared buffer, we can't disable WAL archiving.
* We will add this shortly.
*/
bmstate->use_wal = !XLog_UnconvertedCanBypassWal() && !index->rd_istemp;
}
void
_bitmap_create_lov_heapandindex(Relation rel,
Oid lovComptypeOid,
Oid *lovHeapOid,
Oid *lovIndexOid,
Oid lovHeapRelfilenode,
Oid lovIndexRelfilenode)
{
char lovHeapName[NAMEDATALEN];
char lovIndexName[NAMEDATALEN];
TupleDesc tupDesc;
IndexInfo *indexInfo;
ObjectAddress objAddr, referenced;
Oid *classObjectId;
int16 *coloptions;
Oid heapid;
Oid idxid;
int indattrs;
int i;
Oid unusedArrayOid = InvalidOid;
Assert(rel != NULL);
/* create the new names for the new lov heap and index */
snprintf(lovHeapName, sizeof(lovHeapName),
"pg_bm_%u", RelationGetRelid(rel));
snprintf(lovIndexName, sizeof(lovIndexName),
"pg_bm_%u_index", RelationGetRelid(rel));
heapid = get_relname_relid(lovHeapName, PG_BITMAPINDEX_NAMESPACE);
/*
* If heapid exists, then this is happening during re-indexing.
* We allocate new relfilenodes for lov heap and lov index.
*
* XXX Each segment db may have different relfilenodes for lov heap and
* lov index, which should not be an issue now. Ideally, we would like each
* segment db use the same oids.
*/
if (OidIsValid(heapid))
{
Relation lovHeap;
Relation lovIndex;
Buffer btree_metabuf;
Page btree_metapage;
*lovHeapOid = heapid;
idxid = get_relname_relid(lovIndexName, PG_BITMAPINDEX_NAMESPACE);
Assert(OidIsValid(idxid));
*lovIndexOid = idxid;
lovComptypeOid = get_rel_type_id(heapid);
Assert(OidIsValid(lovComptypeOid));
lovHeap = heap_open(heapid, AccessExclusiveLock);
lovIndex = index_open(idxid, AccessExclusiveLock);
if (OidIsValid(lovHeapRelfilenode))
setNewRelfilenodeToOid(lovHeap, lovHeapRelfilenode);
else
setNewRelfilenode(lovHeap);
if (OidIsValid(lovIndexRelfilenode))
setNewRelfilenodeToOid(lovIndex, lovIndexRelfilenode);
else
setNewRelfilenode(lovIndex);
/*
* After creating the new relfilenode for a btee index, this is not
* a btree anymore. We create the new metapage for this btree.
*/
btree_metabuf = _bt_getbuf(lovIndex, P_NEW, BT_WRITE);
Assert (BTREE_METAPAGE == BufferGetBlockNumber(btree_metabuf));
btree_metapage = BufferGetPage(btree_metabuf);
_bt_initmetapage(btree_metapage, P_NONE, 0);
/* XLOG the metapage */
if (!XLog_UnconvertedCanBypassWal() && !lovIndex->rd_istemp)
{
// Fetch gp_persistent_relation_node information that will be added to XLOG record.
RelationFetchGpRelationNodeForXLog(lovIndex);
_bt_lognewpage(lovIndex,
btree_metapage,
BufferGetBlockNumber(btree_metabuf));
}
/* This cache value is not valid anymore. */
if (lovIndex->rd_amcache)
{
pfree(lovIndex->rd_amcache);
lovIndex->rd_amcache = NULL;
}
MarkBufferDirty(btree_metabuf);
_bt_relbuf(lovIndex, btree_metabuf);
index_close(lovIndex, NoLock);
heap_close(lovHeap, NoLock);
return;
}
/*
* create a new empty heap to store all attribute values with their
* corresponding block number and offset in LOV.
*/
tupDesc = _bitmap_create_lov_heapTupleDesc(rel);
Assert(rel->rd_rel != NULL);
heapid =
heap_create_with_catalog(lovHeapName, PG_BITMAPINDEX_NAMESPACE,
rel->rd_rel->reltablespace,
*lovHeapOid, rel->rd_rel->relowner,
tupDesc,
/* relam */ InvalidOid, RELKIND_RELATION, RELSTORAGE_HEAP,
rel->rd_rel->relisshared, false, /* bufferPoolBulkLoad */ false, 0,
ONCOMMIT_NOOP, NULL /* GP Policy */,
(Datum)0, true,
/* valid_opts */ true,
&lovComptypeOid,
&unusedArrayOid,
/* persistentTid */ NULL,
/* persistentSerialNum */ NULL);
Assert(heapid == *lovHeapOid);
/*
* We must bump the command counter to make the newly-created relation
* tuple visible for opening.
*/
CommandCounterIncrement();
objAddr.classId = RelationRelationId;
objAddr.objectId = *lovHeapOid;
objAddr.objectSubId = 0 ;
referenced.classId = RelationRelationId;
referenced.objectId = RelationGetRelid(rel);
referenced.objectSubId = 0;
recordDependencyOn(&objAddr, &referenced, DEPENDENCY_INTERNAL);
/*
* create a btree index on the newly-created heap.
* The key includes all attributes to be indexed in this bitmap index.
*/
indattrs = tupDesc->natts - 2;
indexInfo = makeNode(IndexInfo);
indexInfo->ii_NumIndexAttrs = indattrs;
indexInfo->ii_Expressions = NIL;
indexInfo->ii_ExpressionsState = NIL;
indexInfo->ii_Predicate = make_ands_implicit(NULL);
indexInfo->ii_PredicateState = NIL;
indexInfo->ii_Unique = true;
indexInfo->opaque = NULL;
classObjectId = (Oid *) palloc(indattrs * sizeof(Oid));
coloptions = (int16 *) palloc(indattrs * sizeof(int16));
for (i = 0; i < indattrs; i++)
{
Oid typid = tupDesc->attrs[i]->atttypid;
indexInfo->ii_KeyAttrNumbers[i] = i + 1;
classObjectId[i] = GetDefaultOpClass(typid, BTREE_AM_OID);
coloptions[i] = 0;
}
idxid = index_create(*lovHeapOid, lovIndexName, *lovIndexOid,
indexInfo, BTREE_AM_OID,
rel->rd_rel->reltablespace,
classObjectId, coloptions, 0, false, false, (Oid *) NULL, true,
false, false, NULL);
Assert(idxid == *lovIndexOid);
}