/*
* Create append-only auxiliary relations for target relation rel.
* Returns true if they are newly created. If pg_appendonly has already
* known those tables, don't create them and returns false.
*/
bool
CreateAOAuxiliaryTable(
Relation rel,
const char *auxiliaryNamePrefix,
char relkind,
TupleDesc tupledesc,
IndexInfo *indexInfo,
Oid *classObjectId,
int16 *coloptions)
{
char aoauxiliary_relname[NAMEDATALEN];
char aoauxiliary_idxname[NAMEDATALEN];
bool shared_relation;
Oid relOid, aoauxiliary_relid = InvalidOid;
Oid aoauxiliary_idxid = InvalidOid;
ObjectAddress baseobject;
ObjectAddress aoauxiliaryobject;
Assert(RelationIsValid(rel));
Assert(RelationIsAoRows(rel) || RelationIsAoCols(rel));
Assert(auxiliaryNamePrefix);
Assert(tupledesc);
Assert(classObjectId);
if (relkind != RELKIND_AOSEGMENTS)
Assert(indexInfo);
shared_relation = rel->rd_rel->relisshared;
/*
* We cannot allow creating an auxiliary table for a shared relation
* after initdb (because there's no way to let other databases know
* this visibility map.
*/
if (shared_relation && !IsBootstrapProcessingMode())
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("shared tables cannot have append-only auxiliary relations after initdb")));
relOid = RelationGetRelid(rel);
switch(relkind)
{
case RELKIND_AOVISIMAP:
GetAppendOnlyEntryAuxOids(relOid, SnapshotNow, NULL,
NULL, NULL, &aoauxiliary_relid, &aoauxiliary_idxid);
break;
case RELKIND_AOBLOCKDIR:
GetAppendOnlyEntryAuxOids(relOid, SnapshotNow, NULL,
&aoauxiliary_relid, &aoauxiliary_idxid, NULL, NULL);
break;
case RELKIND_AOSEGMENTS:
GetAppendOnlyEntryAuxOids(relOid, SnapshotNow,
&aoauxiliary_relid,
NULL, NULL, NULL, NULL);
break;
default:
elog(ERROR, "unsupported auxiliary relkind '%c'", relkind);
}
/*
* Does it have the auxiliary relation?
*/
if (OidIsValid(aoauxiliary_relid))
{
return false;
}
snprintf(aoauxiliary_relname, sizeof(aoauxiliary_relname),
"%s_%u", auxiliaryNamePrefix, relOid);
snprintf(aoauxiliary_idxname, sizeof(aoauxiliary_idxname),
"%s_%u_index", auxiliaryNamePrefix, relOid);
/*
* We place auxiliary relation in the pg_aoseg namespace
* even if its master relation is a temp table. There cannot be
* any naming collision, and the auxiliary relation will be
* destroyed when its master is, so there is no need to handle
* the aovisimap relation as temp.
*/
aoauxiliary_relid = heap_create_with_catalog(aoauxiliary_relname,
PG_AOSEGMENT_NAMESPACE,
rel->rd_rel->reltablespace,
InvalidOid,
rel->rd_rel->relowner,
tupledesc,
/* relam */ InvalidOid,
relkind,
RELSTORAGE_HEAP,
shared_relation,
true,
/* bufferPoolBulkLoad */ false,
0,
ONCOMMIT_NOOP,
NULL, /* GP Policy */
(Datum) 0,
true,
/* valid_opts */ false,
/* persistentTid */ NULL,
/* persistentSerialNum */ NULL);
/* Make this table visible, else index creation will fail */
CommandCounterIncrement();
/* Create an index on AO auxiliary tables (like visimap) except for pg_aoseg table */
if (relkind != RELKIND_AOSEGMENTS)
{
aoauxiliary_idxid = index_create(aoauxiliary_relid,
aoauxiliary_idxname,
InvalidOid,
indexInfo,
BTREE_AM_OID,
rel->rd_rel->reltablespace,
classObjectId, coloptions, (Datum) 0,
true, false, true, false,
false, NULL);
/* Unlock target table -- no one can see it */
UnlockRelationOid(aoauxiliary_relid, ShareLock);
/* Unlock the index -- no one can see it anyway */
UnlockRelationOid(aoauxiliary_idxid, AccessExclusiveLock);
}
/*
* Store the auxiliary table's OID in the parent relation's pg_appendonly row.
* TODO (How to generalize this?)
*/
switch (relkind)
{
case RELKIND_AOVISIMAP:
UpdateAppendOnlyEntryAuxOids(relOid, InvalidOid,
InvalidOid, InvalidOid,
aoauxiliary_relid, aoauxiliary_idxid);
break;
case RELKIND_AOBLOCKDIR:
UpdateAppendOnlyEntryAuxOids(relOid, InvalidOid,
aoauxiliary_relid, aoauxiliary_idxid,
InvalidOid, InvalidOid);
break;
case RELKIND_AOSEGMENTS:
UpdateAppendOnlyEntryAuxOids(relOid,
aoauxiliary_relid,
InvalidOid, InvalidOid,
InvalidOid, InvalidOid);
break;
default:
elog(ERROR, "unsupported auxiliary relkind '%c'", relkind);
}
/*
* Register dependency from the auxiliary table to the master, so that the
* aoseg table will be deleted if the master is.
*/
baseobject.classId = RelationRelationId;
baseobject.objectId = relOid;
baseobject.objectSubId = 0;
aoauxiliaryobject.classId = RelationRelationId;
aoauxiliaryobject.objectId = aoauxiliary_relid;
aoauxiliaryobject.objectSubId = 0;
recordDependencyOn(&aoauxiliaryobject, &baseobject, DEPENDENCY_INTERNAL);
/*
* Make changes visible
*/
CommandCounterIncrement();
return true;
}
/*
* create_toast_table --- internal workhorse
*
* rel is already opened and locked
* toastOid and toastIndexOid are normally InvalidOid, but during
* bootstrap they can be nonzero to specify hand-assigned OIDs
*/
static bool
create_toast_table(Relation rel, Oid toastOid, Oid toastIndexOid,
Datum reloptions, LOCKMODE lockmode, bool check)
{
Oid relOid = RelationGetRelid(rel);
HeapTuple reltup;
TupleDesc tupdesc;
bool shared_relation;
bool mapped_relation;
Relation toast_rel;
Relation class_rel;
Oid toast_relid;
Oid toast_typid = InvalidOid;
Oid namespaceid;
char toast_relname[NAMEDATALEN];
char toast_idxname[NAMEDATALEN];
IndexInfo *indexInfo;
Oid collationObjectId[2];
Oid classObjectId[2];
int16 coloptions[2];
ObjectAddress baseobject,
toastobject;
/*
* Toast table is shared if and only if its parent is.
*
* We cannot allow toasting a shared relation after initdb (because
* there's no way to mark it toasted in other databases' pg_class).
*/
shared_relation = rel->rd_rel->relisshared;
if (shared_relation && !IsBootstrapProcessingMode())
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("shared tables cannot be toasted after initdb")));
/* It's mapped if and only if its parent is, too */
mapped_relation = RelationIsMapped(rel);
/*
* Is it already toasted?
*/
if (rel->rd_rel->reltoastrelid != InvalidOid)
return false;
/*
* Check to see whether the table actually needs a TOAST table.
*/
if (!IsBinaryUpgrade)
{
/* Normal mode, normal check */
if (!needs_toast_table(rel))
return false;
}
else
{
/*
* In binary-upgrade mode, create a TOAST table if and only if
* pg_upgrade told us to (ie, a TOAST table OID has been provided).
*
* This indicates that the old cluster had a TOAST table for the
* current table. We must create a TOAST table to receive the old
* TOAST file, even if the table seems not to need one.
*
* Contrariwise, if the old cluster did not have a TOAST table, we
* should be able to get along without one even if the new version's
* needs_toast_table rules suggest we should have one. There is a lot
* of daylight between where we will create a TOAST table and where
* one is really necessary to avoid failures, so small cross-version
* differences in the when-to-create heuristic shouldn't be a problem.
* If we tried to create a TOAST table anyway, we would have the
* problem that it might take up an OID that will conflict with some
* old-cluster table we haven't seen yet.
*/
if (!OidIsValid(binary_upgrade_next_toast_pg_class_oid) ||
!OidIsValid(binary_upgrade_next_toast_pg_type_oid))
return false;
}
/*
* If requested check lockmode is sufficient. This is a cross check in
* case of errors or conflicting decisions in earlier code.
*/
if (check && lockmode != AccessExclusiveLock)
elog(ERROR, "AccessExclusiveLock required to add toast table.");
/*
* Create the toast table and its index
*/
snprintf(toast_relname, sizeof(toast_relname),
"pg_toast_%u", relOid);
snprintf(toast_idxname, sizeof(toast_idxname),
"pg_toast_%u_index", relOid);
/* this is pretty painful... need a tuple descriptor */
tupdesc = CreateTemplateTupleDesc(3);
TupleDescInitEntry(tupdesc, (AttrNumber) 1,
"chunk_id",
OIDOID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2,
"chunk_seq",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3,
"chunk_data",
BYTEAOID,
-1, 0);
/*
* Ensure that the toast table doesn't itself get toasted, or we'll be
* toast :-(. This is essential for chunk_data because type bytea is
* toastable; hit the other two just to be sure.
*/
TupleDescAttr(tupdesc, 0)->attstorage = 'p';
TupleDescAttr(tupdesc, 1)->attstorage = 'p';
TupleDescAttr(tupdesc, 2)->attstorage = 'p';
/*
* Toast tables for regular relations go in pg_toast; those for temp
* relations go into the per-backend temp-toast-table namespace.
*/
if (isTempOrTempToastNamespace(rel->rd_rel->relnamespace))
namespaceid = GetTempToastNamespace();
else
namespaceid = PG_TOAST_NAMESPACE;
/*
* Use binary-upgrade override for pg_type.oid, if supplied. We might be
* in the post-schema-restore phase where we are doing ALTER TABLE to
* create TOAST tables that didn't exist in the old cluster.
*/
if (IsBinaryUpgrade && OidIsValid(binary_upgrade_next_toast_pg_type_oid))
{
toast_typid = binary_upgrade_next_toast_pg_type_oid;
binary_upgrade_next_toast_pg_type_oid = InvalidOid;
}
toast_relid = heap_create_with_catalog(toast_relname,
namespaceid,
rel->rd_rel->reltablespace,
toastOid,
toast_typid,
InvalidOid,
rel->rd_rel->relowner,
tupdesc,
NIL,
RELKIND_TOASTVALUE,
rel->rd_rel->relpersistence,
shared_relation,
mapped_relation,
ONCOMMIT_NOOP,
reloptions,
false,
true,
true,
InvalidOid,
NULL);
Assert(toast_relid != InvalidOid);
/* make the toast relation visible, else heap_open will fail */
CommandCounterIncrement();
/* ShareLock is not really needed here, but take it anyway */
toast_rel = heap_open(toast_relid, ShareLock);
/*
* Create unique index on chunk_id, chunk_seq.
*
* NOTE: the normal TOAST access routines could actually function with a
* single-column index on chunk_id only. However, the slice access
* routines use both columns for faster access to an individual chunk. In
* addition, we want it to be unique as a check against the possibility of
* duplicate TOAST chunk OIDs. The index might also be a little more
* efficient this way, since btree isn't all that happy with large numbers
* of equal keys.
*/
indexInfo = makeNode(IndexInfo);
indexInfo->ii_NumIndexAttrs = 2;
indexInfo->ii_NumIndexKeyAttrs = 2;
indexInfo->ii_IndexAttrNumbers[0] = 1;
indexInfo->ii_IndexAttrNumbers[1] = 2;
indexInfo->ii_Expressions = NIL;
indexInfo->ii_ExpressionsState = NIL;
indexInfo->ii_Predicate = NIL;
indexInfo->ii_PredicateState = NULL;
indexInfo->ii_ExclusionOps = NULL;
indexInfo->ii_ExclusionProcs = NULL;
indexInfo->ii_ExclusionStrats = NULL;
indexInfo->ii_Unique = true;
indexInfo->ii_ReadyForInserts = true;
indexInfo->ii_Concurrent = false;
indexInfo->ii_BrokenHotChain = false;
indexInfo->ii_ParallelWorkers = 0;
indexInfo->ii_Am = BTREE_AM_OID;
indexInfo->ii_AmCache = NULL;
indexInfo->ii_Context = CurrentMemoryContext;
collationObjectId[0] = InvalidOid;
collationObjectId[1] = InvalidOid;
classObjectId[0] = OID_BTREE_OPS_OID;
classObjectId[1] = INT4_BTREE_OPS_OID;
coloptions[0] = 0;
coloptions[1] = 0;
index_create(toast_rel, toast_idxname, toastIndexOid, InvalidOid,
InvalidOid, InvalidOid,
indexInfo,
list_make2("chunk_id", "chunk_seq"),
BTREE_AM_OID,
rel->rd_rel->reltablespace,
collationObjectId, classObjectId, coloptions, (Datum) 0,
INDEX_CREATE_IS_PRIMARY, 0, true, true, NULL);
heap_close(toast_rel, NoLock);
/*
* Store the toast table's OID in the parent relation's pg_class row
*/
class_rel = heap_open(RelationRelationId, RowExclusiveLock);
reltup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relOid));
if (!HeapTupleIsValid(reltup))
elog(ERROR, "cache lookup failed for relation %u", relOid);
((Form_pg_class) GETSTRUCT(reltup))->reltoastrelid = toast_relid;
if (!IsBootstrapProcessingMode())
{
/* normal case, use a transactional update */
CatalogTupleUpdate(class_rel, &reltup->t_self, reltup);
}
else
{
/* While bootstrapping, we cannot UPDATE, so overwrite in-place */
heap_inplace_update(class_rel, reltup);
}
heap_freetuple(reltup);
heap_close(class_rel, RowExclusiveLock);
/*
* Register dependency from the toast table to the master, so that the
* toast table will be deleted if the master is. Skip this in bootstrap
* mode.
*/
if (!IsBootstrapProcessingMode())
{
baseobject.classId = RelationRelationId;
baseobject.objectId = relOid;
baseobject.objectSubId = 0;
toastobject.classId = RelationRelationId;
toastobject.objectId = toast_relid;
toastobject.objectSubId = 0;
recordDependencyOn(&toastobject, &baseobject, DEPENDENCY_INTERNAL);
}
/*
* Make changes visible
*/
CommandCounterIncrement();
return true;
}
/*
* create_toast_table --- internal workhorse
*
* rel is already opened and exclusive-locked
* toastOid and toastIndexOid are normally InvalidOid, but during
* bootstrap they can be nonzero to specify hand-assigned OIDs
*/
static bool
create_toast_table(Relation rel, Oid toastOid, Oid toastIndexOid,
bool is_part_child)
{
Oid relOid = RelationGetRelid(rel);
HeapTuple reltup;
TupleDesc tupdesc;
bool shared_relation;
Relation class_rel;
Oid toast_relid;
Oid toast_idxid;
Oid namespaceid;
char toast_relname[NAMEDATALEN];
char toast_idxname[NAMEDATALEN];
IndexInfo *indexInfo;
Oid classObjectId[2];
int16 coloptions[2];
ObjectAddress baseobject,
toastobject;
/*
* Is it already toasted?
*/
if (rel->rd_rel->reltoastrelid != InvalidOid)
return false;
/*
* Check to see whether the table actually needs a TOAST table.
*/
if (!RelationNeedsToastTable(rel))
return false;
/*
* Toast table is shared if and only if its parent is.
*
* We cannot allow toasting a shared relation after initdb (because
* there's no way to mark it toasted in other databases' pg_class).
*/
shared_relation = rel->rd_rel->relisshared;
if (shared_relation && !IsBootstrapProcessingMode())
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("shared tables cannot be toasted after initdb")));
/*
* Create the toast table and its index
*/
snprintf(toast_relname, sizeof(toast_relname),
"pg_toast_%u", relOid);
snprintf(toast_idxname, sizeof(toast_idxname),
"pg_toast_%u_index", relOid);
/* this is pretty painful... need a tuple descriptor */
tupdesc = CreateTemplateTupleDesc(3, false);
TupleDescInitEntry(tupdesc, (AttrNumber) 1,
"chunk_id",
OIDOID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2,
"chunk_seq",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3,
"chunk_data",
BYTEAOID,
-1, 0);
/*
* Ensure that the toast table doesn't itself get toasted, or we'll be
* toast :-(. This is essential for chunk_data because type bytea is
* toastable; hit the other two just to be sure.
*/
tupdesc->attrs[0]->attstorage = 'p';
tupdesc->attrs[1]->attstorage = 'p';
tupdesc->attrs[2]->attstorage = 'p';
/*
* Toast tables for regular relations go in pg_toast; those for temp
* relations go into the per-backend temp-toast-table namespace.
*/
if (rel->rd_istemp)
namespaceid = GetTempToastNamespace();
else
namespaceid = PG_TOAST_NAMESPACE;
/*
* XXX would it make sense to apply the master's reloptions to the toast
* table? Or maybe some toast-specific reloptions?
*/
toast_relid = heap_create_with_catalog(toast_relname,
namespaceid,
rel->rd_rel->reltablespace,
toastOid,
rel->rd_rel->relowner,
tupdesc,
/* relam */ InvalidOid,
RELKIND_TOASTVALUE,
RELSTORAGE_HEAP,
shared_relation,
true,
/* bufferPoolBulkLoad */ false,
0,
ONCOMMIT_NOOP,
NULL, /* CDB POLICY */
(Datum) 0,
true,
/* valid_opts */ false,
/* persistentTid */ NULL,
/* persistentSerialNum */ NULL);
/* make the toast relation visible, else index creation will fail */
CommandCounterIncrement();
/*
* Create unique index on chunk_id, chunk_seq.
*
* NOTE: the normal TOAST access routines could actually function with a
* single-column index on chunk_id only. However, the slice access
* routines use both columns for faster access to an individual chunk. In
* addition, we want it to be unique as a check against the possibility of
* duplicate TOAST chunk OIDs. The index might also be a little more
* efficient this way, since btree isn't all that happy with large numbers
* of equal keys.
*/
indexInfo = makeNode(IndexInfo);
indexInfo->ii_NumIndexAttrs = 2;
indexInfo->ii_KeyAttrNumbers[0] = 1;
indexInfo->ii_KeyAttrNumbers[1] = 2;
indexInfo->ii_Expressions = NIL;
indexInfo->ii_ExpressionsState = NIL;
indexInfo->ii_Predicate = NIL;
indexInfo->ii_PredicateState = NIL;
indexInfo->ii_Unique = true;
indexInfo->ii_ReadyForInserts = true;
indexInfo->ii_Concurrent = false;
indexInfo->ii_BrokenHotChain = false;
classObjectId[0] = OID_BTREE_OPS_OID;
classObjectId[1] = INT4_BTREE_OPS_OID;
coloptions[0] = 0;
coloptions[1] = 0;
toast_idxid = index_create(toast_relid, toast_idxname, toastIndexOid,
indexInfo,
BTREE_AM_OID,
rel->rd_rel->reltablespace,
classObjectId, coloptions, (Datum) 0,
true, false, true, false, false, NULL);
/*
* If this is a partitioned child, we can unlock since the master is
* already locked.
*/
if (is_part_child)
{
UnlockRelationOid(toast_relid, ShareLock);
UnlockRelationOid(toast_idxid, AccessExclusiveLock);
}
/*
* Store the toast table's OID in the parent relation's pg_class row
*/
class_rel = heap_open(RelationRelationId, RowExclusiveLock);
reltup = SearchSysCacheCopy(RELOID,
ObjectIdGetDatum(relOid),
0, 0, 0);
if (!HeapTupleIsValid(reltup))
elog(ERROR, "cache lookup failed for relation %u", relOid);
((Form_pg_class) GETSTRUCT(reltup))->reltoastrelid = toast_relid;
if (!IsBootstrapProcessingMode())
{
/* normal case, use a transactional update */
simple_heap_update(class_rel, &reltup->t_self, reltup);
/* Keep catalog indexes current */
CatalogUpdateIndexes(class_rel, reltup);
}
else
{
/* While bootstrapping, we cannot UPDATE, so overwrite in-place */
heap_inplace_update(class_rel, reltup);
}
heap_freetuple(reltup);
heap_close(class_rel, RowExclusiveLock);
/*
* Register dependency from the toast table to the master, so that the
* toast table will be deleted if the master is. Skip this in bootstrap
* mode.
*/
if (!IsBootstrapProcessingMode())
{
baseobject.classId = RelationRelationId;
baseobject.objectId = relOid;
baseobject.objectSubId = 0;
toastobject.classId = RelationRelationId;
toastobject.objectId = toast_relid;
toastobject.objectSubId = 0;
recordDependencyOn(&toastobject, &baseobject, DEPENDENCY_INTERNAL);
}
/*
* Make changes visible
*/
CommandCounterIncrement();
return true;
}
/*
* create_toast_table --- internal workhorse
*
* rel is already opened and locked
* toastOid and toastIndexOid are normally InvalidOid, but during
* bootstrap they can be nonzero to specify hand-assigned OIDs
*/
static bool
create_toast_table(Relation rel, Oid toastOid, Oid toastIndexOid,
Datum reloptions, LOCKMODE lockmode, bool check)
{
Oid relOid = RelationGetRelid(rel);
HeapTuple reltup;
TupleDesc tupdesc;
bool shared_relation;
bool mapped_relation;
Relation toast_rel;
Relation class_rel;
Oid toast_relid;
Oid toast_typid = InvalidOid;
Oid namespaceid;
char toast_relname[NAMEDATALEN];
char toast_idxname[NAMEDATALEN];
IndexInfo *indexInfo;
Oid collationObjectId[2];
Oid classObjectId[2];
int16 coloptions[2];
ObjectAddress baseobject,
toastobject;
/*
* Toast table is shared if and only if its parent is.
*
* We cannot allow toasting a shared relation after initdb (because
* there's no way to mark it toasted in other databases' pg_class).
*/
shared_relation = rel->rd_rel->relisshared;
if (shared_relation && !IsBootstrapProcessingMode())
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("shared tables cannot be toasted after initdb")));
/* It's mapped if and only if its parent is, too */
mapped_relation = RelationIsMapped(rel);
/*
* Is it already toasted?
*/
if (rel->rd_rel->reltoastrelid != InvalidOid)
return false;
if (!IsBinaryUpgrade)
{
if (!needs_toast_table(rel))
return false;
}
else
{
/*
* Check to see whether the table needs a TOAST table.
*
* If an update-in-place TOAST relfilenode is specified, force TOAST file
* creation even if it seems not to need one. This handles the case
* where the old cluster needed a TOAST table but the new cluster
* would not normally create one.
*/
/*
* If a TOAST oid is not specified, skip TOAST creation as we will do
* it later so we don't create a TOAST table whose OID later conflicts
* with a user-supplied OID. This handles cases where the old cluster
* didn't need a TOAST table, but the new cluster does.
*/
if (!OidIsValid(binary_upgrade_next_toast_pg_class_oid))
return false;
/*
* If a special TOAST value has been passed in, it means we are in
* cleanup mode --- we are creating needed TOAST tables after all user
* tables with specified OIDs have been created. We let the system
* assign a TOAST oid for us. The tables are empty so the missing
* TOAST tables were not a problem.
*/
if (binary_upgrade_next_toast_pg_class_oid == OPTIONALLY_CREATE_TOAST_OID)
{
/* clear as it is not to be used; it is just a flag */
binary_upgrade_next_toast_pg_class_oid = InvalidOid;
if (!needs_toast_table(rel))
return false;
}
/* both should be set, or not set */
Assert(OidIsValid(binary_upgrade_next_toast_pg_class_oid) ==
OidIsValid(binary_upgrade_next_toast_pg_type_oid));
}
/*
* If requested check lockmode is sufficient. This is a cross check in
* case of errors or conflicting decisions in earlier code.
*/
if (check && lockmode != AccessExclusiveLock)
elog(ERROR, "AccessExclusiveLock required to add toast table.");
/*
* Create the toast table and its index
*/
snprintf(toast_relname, sizeof(toast_relname),
"pg_toast_%u", relOid);
snprintf(toast_idxname, sizeof(toast_idxname),
"pg_toast_%u_index", relOid);
/* this is pretty painful... need a tuple descriptor */
tupdesc = CreateTemplateTupleDesc(3, false);
TupleDescInitEntry(tupdesc, (AttrNumber) 1,
"chunk_id",
OIDOID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2,
"chunk_seq",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3,
"chunk_data",
BYTEAOID,
-1, 0);
/*
* Ensure that the toast table doesn't itself get toasted, or we'll be
* toast :-(. This is essential for chunk_data because type bytea is
* toastable; hit the other two just to be sure.
*/
tupdesc->attrs[0]->attstorage = 'p';
tupdesc->attrs[1]->attstorage = 'p';
tupdesc->attrs[2]->attstorage = 'p';
/*
* Toast tables for regular relations go in pg_toast; those for temp
* relations go into the per-backend temp-toast-table namespace.
*/
if (isTempOrToastNamespace(rel->rd_rel->relnamespace))
namespaceid = GetTempToastNamespace();
else
namespaceid = PG_TOAST_NAMESPACE;
/* Use binary-upgrade override for pg_type.oid, if supplied. */
if (IsBinaryUpgrade && OidIsValid(binary_upgrade_next_toast_pg_type_oid))
{
toast_typid = binary_upgrade_next_toast_pg_type_oid;
binary_upgrade_next_toast_pg_type_oid = InvalidOid;
}
toast_relid = heap_create_with_catalog(toast_relname,
namespaceid,
rel->rd_rel->reltablespace,
toastOid,
toast_typid,
InvalidOid,
rel->rd_rel->relowner,
tupdesc,
NIL,
RELKIND_TOASTVALUE,
rel->rd_rel->relpersistence,
shared_relation,
mapped_relation,
true,
0,
ONCOMMIT_NOOP,
reloptions,
false,
true,
true);
Assert(toast_relid != InvalidOid);
/* make the toast relation visible, else heap_open will fail */
CommandCounterIncrement();
/* ShareLock is not really needed here, but take it anyway */
toast_rel = heap_open(toast_relid, ShareLock);
/*
* Create unique index on chunk_id, chunk_seq.
*
* NOTE: the normal TOAST access routines could actually function with a
* single-column index on chunk_id only. However, the slice access
* routines use both columns for faster access to an individual chunk. In
* addition, we want it to be unique as a check against the possibility of
* duplicate TOAST chunk OIDs. The index might also be a little more
* efficient this way, since btree isn't all that happy with large numbers
* of equal keys.
*/
indexInfo = makeNode(IndexInfo);
indexInfo->ii_NumIndexAttrs = 2;
indexInfo->ii_KeyAttrNumbers[0] = 1;
indexInfo->ii_KeyAttrNumbers[1] = 2;
indexInfo->ii_Expressions = NIL;
indexInfo->ii_ExpressionsState = NIL;
indexInfo->ii_Predicate = NIL;
indexInfo->ii_PredicateState = NIL;
indexInfo->ii_ExclusionOps = NULL;
indexInfo->ii_ExclusionProcs = NULL;
indexInfo->ii_ExclusionStrats = NULL;
indexInfo->ii_Unique = true;
indexInfo->ii_ReadyForInserts = true;
indexInfo->ii_Concurrent = false;
indexInfo->ii_BrokenHotChain = false;
collationObjectId[0] = InvalidOid;
collationObjectId[1] = InvalidOid;
classObjectId[0] = OID_BTREE_OPS_OID;
classObjectId[1] = INT4_BTREE_OPS_OID;
coloptions[0] = 0;
coloptions[1] = 0;
index_create(toast_rel, toast_idxname, toastIndexOid, InvalidOid,
indexInfo,
list_make2("chunk_id", "chunk_seq"),
BTREE_AM_OID,
rel->rd_rel->reltablespace,
collationObjectId, classObjectId, coloptions, (Datum) 0,
true, false, false, false,
true, false, false, true);
heap_close(toast_rel, NoLock);
/*
* Store the toast table's OID in the parent relation's pg_class row
*/
class_rel = heap_open(RelationRelationId, RowExclusiveLock);
reltup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relOid));
if (!HeapTupleIsValid(reltup))
elog(ERROR, "cache lookup failed for relation %u", relOid);
((Form_pg_class) GETSTRUCT(reltup))->reltoastrelid = toast_relid;
if (!IsBootstrapProcessingMode())
{
/* normal case, use a transactional update */
simple_heap_update(class_rel, &reltup->t_self, reltup);
/* Keep catalog indexes current */
CatalogUpdateIndexes(class_rel, reltup);
}
else
{
/* While bootstrapping, we cannot UPDATE, so overwrite in-place */
heap_inplace_update(class_rel, reltup);
}
heap_freetuple(reltup);
heap_close(class_rel, RowExclusiveLock);
/*
* Register dependency from the toast table to the master, so that the
* toast table will be deleted if the master is. Skip this in bootstrap
* mode.
*/
if (!IsBootstrapProcessingMode())
{
baseobject.classId = RelationRelationId;
baseobject.objectId = relOid;
baseobject.objectSubId = 0;
toastobject.classId = RelationRelationId;
toastobject.objectId = toast_relid;
toastobject.objectSubId = 0;
recordDependencyOn(&toastobject, &baseobject, DEPENDENCY_INTERNAL);
}
/*
* Make changes visible
*/
CommandCounterIncrement();
return true;
}
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);
}
/*
* create_aoblkdir_table
*
* rel is already opened and exclusive-locked.
* comptypeOid is InvalidOid.
*/
static bool
create_aoblkdir_table(Relation rel, Oid aoblkdirOid,
Oid aoblkdirIndexOid, Oid *comptypeOid)
{
Oid relOid = RelationGetRelid(rel);
Oid aoblkdir_relid;
Oid aoblkdir_idxid;
bool shared_relation = rel->rd_rel->relisshared;
char aoblkdir_relname[NAMEDATALEN];
char aoblkdir_idxname[NAMEDATALEN];
TupleDesc tupdesc;
IndexInfo *indexInfo;
Oid classObjectId[3];
ObjectAddress baseobject;
ObjectAddress aoblkdirobject;
Oid tablespaceOid = ChooseTablespaceForLimitedObject(rel->rd_rel->reltablespace);
if (!RelationIsAoRows(rel))
return false;
/*
* We cannot allow creating a block directory for a shared relation
* after initdb (because there's no way to let other databases know
* this block directory.
*/
if (shared_relation && !IsBootstrapProcessingMode())
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("shared tables cannot have block directory after initdb")));
GetAppendOnlyEntryAuxOids(relOid, SnapshotNow, NULL,NULL, &aoblkdir_relid, &aoblkdir_idxid);
/*
* Does it have a block directory?
*/
if (aoblkdir_relid != InvalidOid)
{
return false;
}
snprintf(aoblkdir_relname, sizeof(aoblkdir_relname),
"pg_aoblkdir_%u", relOid);
snprintf(aoblkdir_idxname, sizeof(aoblkdir_idxname),
"pg_aoblkdir_%u_index", relOid);
/* Create a tuple descriptor */
tupdesc = CreateTemplateTupleDesc(4, false);
TupleDescInitEntry(tupdesc, (AttrNumber) 1,
"segno",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 2,
"columngroup_no",
INT4OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 3,
"first_row_no",
INT8OID,
-1, 0);
TupleDescInitEntry(tupdesc, (AttrNumber) 4,
"minipage",
VARBITOID,
-1, 0);
/*
* We don't want any toast columns here.
*/
tupdesc->attrs[0]->attstorage = 'p';
tupdesc->attrs[1]->attstorage = 'p';
tupdesc->attrs[2]->attstorage = 'p';
tupdesc->attrs[2]->attstorage = 'p';
/*
* We place aoblkdir relation in the pg_aoseg namespace
* even if its master relation is a temp table. There cannot be
* any naming collision, and the aoblkdir relation will be
* destroyed when its master is, so there is no need to handle
* the aoblkdir relation as temp.
*/
aoblkdir_relid = heap_create_with_catalog(aoblkdir_relname,
PG_AOSEGMENT_NAMESPACE,
tablespaceOid,
aoblkdirOid,
rel->rd_rel->relowner,
tupdesc,
/* relam */ InvalidOid,
RELKIND_AOBLOCKDIR,
RELSTORAGE_HEAP,
shared_relation,
true,
/* bufferPoolBulkLoad */ false,
0,
ONCOMMIT_NOOP,
NULL, /* GP Policy */
(Datum) 0,
true,
comptypeOid,
/* persistentTid */ NULL,
/* persistentSerialNum */ NULL);
/* Make this table visible, else index creation will fail */
CommandCounterIncrement();
/*
* Create index on segno, first_row_no.
*/
indexInfo = makeNode(IndexInfo);
indexInfo->ii_NumIndexAttrs = 3;
indexInfo->ii_KeyAttrNumbers[0] = 1;
indexInfo->ii_KeyAttrNumbers[1] = 2;
indexInfo->ii_KeyAttrNumbers[2] = 3;
indexInfo->ii_Expressions = NIL;
indexInfo->ii_ExpressionsState = NIL;
indexInfo->ii_Predicate = NIL;
indexInfo->ii_PredicateState = NIL;
indexInfo->ii_Unique = false;
indexInfo->ii_Concurrent = false;
classObjectId[0] = INT4_BTREE_OPS_OID;
classObjectId[1] = INT4_BTREE_OPS_OID;
classObjectId[2] = INT8_BTREE_OPS_OID;
aoblkdir_idxid = index_create(aoblkdirOid, aoblkdir_idxname, aoblkdirIndexOid,
indexInfo,
BTREE_AM_OID,
tablespaceOid,
classObjectId, (Datum) 0,
true, false, (Oid *) NULL, true, false, false, NULL);
/* Unlock target table -- no one can see it */
UnlockRelationOid(aoblkdirOid, ShareLock);
/* Unlock the index -- no one can see it anyway */
UnlockRelationOid(aoblkdirIndexOid, AccessExclusiveLock);
/*
* Store the aoblkdir table's OID in the parent relation's pg_appendonly row.
*/
UpdateAppendOnlyEntryAuxOids(relOid, InvalidOid, InvalidOid,
aoblkdir_relid, aoblkdir_idxid);
/*
* Register dependency from the aoseg table to the master, so that the
* aoseg table will be deleted if the master is.
*/
baseobject.classId = RelationRelationId;
baseobject.objectId = relOid;
baseobject.objectSubId = 0;
aoblkdirobject.classId = RelationRelationId;
aoblkdirobject.objectId = aoblkdirOid;
aoblkdirobject.objectSubId = 0;
recordDependencyOn(&aoblkdirobject, &baseobject, DEPENDENCY_INTERNAL);
/*
* Make changes visible
*/
CommandCounterIncrement();
return true;
}