Columnar Store Storage Id

src/backend/columnar/cstore_customscan.c

@@ -215,12 +215,18 @@ static Cost
 CStoreScanCost(RangeTblEntry *rte)
 {
 	Relation rel = RelationIdGetRelation(rte->relid);
-	DataFileMetadata *metadata = ReadDataFileMetadata(rel->rd_node.relNode, false);
+	if (UninitializedDatafile(rel->rd_node))
+	{
+		RelationClose(rel);
+		return 0;
+	}
+
+	DataFileMetadata *metadata = ReadDataFileMetadata(rel->rd_node, false);
+	RelationClose(rel);
+
 	uint32 maxColumnCount = 0;
 	uint64 totalStripeSize = 0;
 	ListCell *stripeMetadataCell = NULL;
-
-	RelationClose(rel);
 	rel = NULL;
 
 	foreach(stripeMetadataCell, metadata->stripeMetadataList)

src/backend/columnar/cstore_metadata_tables.c

@@ -41,6 +41,23 @@
 #include "utils/memutils.h"
 #include "utils/lsyscache.h"
 #include "utils/rel.h"
+#include "utils/relfilenodemap.h"
+
+
+/*
+ * Content of the first page in main fork, which stores metadata at file
+ * level.
+ */
+typedef struct ColumnarMetapage
+{
+	/*
+	 * Each of the metadata table rows are identified by a storageId.
+	 * We store it also in the main fork so we can link metadata rows
+	 * with data files.
+	 */
+	uint64 storageId;
+} ColumnarMetapage;
+
 
 typedef struct
 {
@@ -48,11 +65,11 @@ typedef struct
 	EState *estate;
 } ModifyState;
 
-static void InsertStripeMetadataRow(Oid relfilenode, StripeMetadata *stripe);
+static void InsertStripeMetadataRow(uint64 storageId, StripeMetadata *stripe);
-static void GetHighestUsedAddressAndId(Oid relfilenode,
+static void GetHighestUsedAddressAndId(uint64 storageId,
 									   uint64 *highestUsedAddress,
 									   uint64 *highestUsedId);
-static List * ReadDataFileStripeList(Oid relfilenode, Snapshot snapshot);
+static List * ReadDataFileStripeList(uint64 storageId, Snapshot snapshot);
 static Oid CStoreStripesRelationId(void);
 static Oid CStoreStripesIndexRelationId(void);
 static Oid CStoreDataFilesRelationId(void);
@@ -60,7 +77,7 @@ static Oid CStoreDataFilesIndexRelationId(void);
 static Oid CStoreSkipNodesRelationId(void);
 static Oid CStoreSkipNodesIndexRelationId(void);
 static Oid CStoreNamespaceId(void);
-static bool ReadCStoreDataFiles(Oid relfilenode, DataFileMetadata *metadata);
+static bool ReadCStoreDataFiles(uint64 storageId, DataFileMetadata *metadata);
 static ModifyState * StartModifyRelation(Relation rel);
 static void InsertTupleAndEnforceConstraints(ModifyState *state, Datum *values,
 											 bool *nulls);
@@ -69,10 +86,16 @@ static void FinishModifyRelation(ModifyState *state);
 static EState * create_estate_for_relation(Relation rel);
 static bytea * DatumToBytea(Datum value, Form_pg_attribute attrForm);
 static Datum ByteaToDatum(bytea *bytes, Form_pg_attribute attrForm);
+static ColumnarMetapage InitMetapage(Relation relation);
+static bool ReadMetapage(RelFileNode relfilenode, ColumnarMetapage *metapage);
+static uint64 GetNextStorageId(void);
+
+PG_FUNCTION_INFO_V1(columnar_relation_storageid);
+
 
 /* constants for cstore_table */
 #define Natts_cstore_data_files 6
-#define Anum_cstore_data_files_relfilenode 1
+#define Anum_cstore_data_files_storageid 1
 #define Anum_cstore_data_files_block_row_count 2
 #define Anum_cstore_data_files_stripe_row_count 3
 #define Anum_cstore_data_files_compression 4
@@ -85,7 +108,7 @@ static Datum ByteaToDatum(bytea *bytes, Form_pg_attribute attrForm);
  */
 typedef struct FormData_cstore_data_files
 {
-	Oid relfilenode;
+	uint64 storageid;
 	int32 block_row_count;
 	int32 stripe_row_count;
 	NameData compression;
@@ -99,7 +122,7 @@ typedef FormData_cstore_data_files *Form_cstore_data_files;
 
 /* constants for cstore_stripe */
 #define Natts_cstore_stripes 8
-#define Anum_cstore_stripes_relfilenode 1
+#define Anum_cstore_stripes_storageid 1
 #define Anum_cstore_stripes_stripe 2
 #define Anum_cstore_stripes_file_offset 3
 #define Anum_cstore_stripes_data_length 4
@@ -110,7 +133,7 @@ typedef FormData_cstore_data_files *Form_cstore_data_files;
 
 /* constants for cstore_skipnodes */
 #define Natts_cstore_skipnodes 12
-#define Anum_cstore_skipnodes_relfilenode 1
+#define Anum_cstore_skipnodes_storageid 1
 #define Anum_cstore_skipnodes_stripe 2
 #define Anum_cstore_skipnodes_attr 3
 #define Anum_cstore_skipnodes_block 4
@@ -125,18 +148,19 @@ typedef FormData_cstore_data_files *Form_cstore_data_files;
 
 
 /*
- * InitCStoreDataFileMetadata adds a record for the given relfilenode
+ * InitCStoreDataFileMetadata adds a record for the given relation
  * in cstore_data_files.
  */
 void
-InitCStoreDataFileMetadata(Oid relfilenode, int blockRowCount, int stripeRowCount,
+InitCStoreDataFileMetadata(Relation relation, int blockRowCount,
+						   int stripeRowCount,
 						   CompressionType compression)
 {
-	NameData compressionName = { 0 };
+	ColumnarMetapage metapage = InitMetapage(relation);
 
 	bool nulls[Natts_cstore_data_files] = { 0 };
 	Datum values[Natts_cstore_data_files] = {
-		ObjectIdGetDatum(relfilenode),
+		UInt64GetDatum(metapage.storageId),
 		Int32GetDatum(blockRowCount),
 		Int32GetDatum(stripeRowCount),
 		0, /* to be filled below */
@@ -144,10 +168,11 @@ InitCStoreDataFileMetadata(Oid relfilenode, int blockRowCount, int stripeRowCoun
 		Int32GetDatum(CSTORE_VERSION_MINOR)
 	};
 
+	NameData compressionName = { 0 };
 	namestrcpy(&compressionName, CompressionTypeStr(compression));
 	values[Anum_cstore_data_files_compression - 1] = NameGetDatum(&compressionName);
 
-	DeleteDataFileMetadataRowIfExists(relfilenode);
+	DeleteDataFileMetadataRowIfExists(relation->rd_node);
 
 	Oid cstoreDataFilesOid = CStoreDataFilesRelationId();
 	Relation cstoreDataFiles = heap_open(cstoreDataFilesOid, RowExclusiveLock);
@@ -163,7 +188,8 @@ InitCStoreDataFileMetadata(Oid relfilenode, int blockRowCount, int stripeRowCoun
 
 
 void
-UpdateCStoreDataFileMetadata(Oid relfilenode, int blockRowCount, int stripeRowCount,
+UpdateCStoreDataFileMetadata(RelFileNode relfilenode, int blockRowCount, int
+							 stripeRowCount,
 							 CompressionType compression)
 {
 	const int scanKeyCount = 1;
@@ -174,11 +200,17 @@ UpdateCStoreDataFileMetadata(Oid relfilenode, int blockRowCount, int stripeRowCo
 	bool replace[Natts_cstore_data_files] = { 0 };
 	bool changed = false;
 
+	ColumnarMetapage metapage;
+	if (!ReadMetapage(relfilenode, &metapage))
+	{
+		elog(ERROR, "metapage was not found");
+	}
+
 	Relation cstoreDataFiles = heap_open(CStoreDataFilesRelationId(), RowExclusiveLock);
 	TupleDesc tupleDescriptor = RelationGetDescr(cstoreDataFiles);
 
-	ScanKeyInit(&scanKey[0], Anum_cstore_data_files_relfilenode, BTEqualStrategyNumber,
+	ScanKeyInit(&scanKey[0], Anum_cstore_data_files_storageid, BTEqualStrategyNumber,
-				F_INT8EQ, ObjectIdGetDatum(relfilenode));
+				F_INT8EQ, UInt64GetDatum(metapage.storageId));
 
 	SysScanDesc scanDescriptor = systable_beginscan(cstoreDataFiles,
 													CStoreDataFilesIndexRelationId(),
@@ -189,7 +221,7 @@ UpdateCStoreDataFileMetadata(Oid relfilenode, int blockRowCount, int stripeRowCo
 	if (heapTuple == NULL)
 	{
 		ereport(ERROR, (errmsg("relfilenode %d doesn't belong to a cstore table",
-							   relfilenode)));
+							   relfilenode.relNode)));
 	}
 
 	Form_cstore_data_files metadata = (Form_cstore_data_files) GETSTRUCT(heapTuple);
@@ -242,13 +274,19 @@ UpdateCStoreDataFileMetadata(Oid relfilenode, int blockRowCount, int stripeRowCo
  * of cstore_skipnodes.
  */
 void
-SaveStripeSkipList(Oid relfilenode, uint64 stripe, StripeSkipList *stripeSkipList,
+SaveStripeSkipList(RelFileNode relfilenode, uint64 stripe, StripeSkipList *stripeSkipList,
 				   TupleDesc tupleDescriptor)
 {
 	uint32 columnIndex = 0;
 	uint32 blockIndex = 0;
 	uint32 columnCount = stripeSkipList->columnCount;
 
+	ColumnarMetapage metapage;
+	if (!ReadMetapage(relfilenode, &metapage))
+	{
+		elog(WARNING, "metapage was not found");
+	}
+
 	Oid cstoreSkipNodesOid = CStoreSkipNodesRelationId();
 	Relation cstoreSkipNodes = heap_open(cstoreSkipNodesOid, RowExclusiveLock);
 	ModifyState *modifyState = StartModifyRelation(cstoreSkipNodes);
@@ -261,7 +299,7 @@ SaveStripeSkipList(Oid relfilenode, uint64 stripe, StripeSkipList *stripeSkipLis
 				&stripeSkipList->blockSkipNodeArray[columnIndex][blockIndex];
 
 			Datum values[Natts_cstore_skipnodes] = {
-				ObjectIdGetDatum(relfilenode),
+				UInt64GetDatum(metapage.storageId),
 				Int64GetDatum(stripe),
 				Int32GetDatum(columnIndex + 1),
 				Int32GetDatum(blockIndex),
@@ -307,7 +345,7 @@ SaveStripeSkipList(Oid relfilenode, uint64 stripe, StripeSkipList *stripeSkipLis
  * ReadStripeSkipList fetches StripeSkipList for a given stripe.
  */
 StripeSkipList *
-ReadStripeSkipList(Oid relfilenode, uint64 stripe, TupleDesc tupleDescriptor,
+ReadStripeSkipList(RelFileNode relfilenode, uint64 stripe, TupleDesc tupleDescriptor,
 				   uint32 blockCount)
 {
 	int32 columnIndex = 0;
@@ -315,12 +353,18 @@ ReadStripeSkipList(Oid relfilenode, uint64 stripe, TupleDesc tupleDescriptor,
 	uint32 columnCount = tupleDescriptor->natts;
 	ScanKeyData scanKey[2];
 
+	ColumnarMetapage metapage;
+	if (!ReadMetapage(relfilenode, &metapage))
+	{
+		elog(WARNING, "metapage was not found");
+	}
+
 	Oid cstoreSkipNodesOid = CStoreSkipNodesRelationId();
 	Relation cstoreSkipNodes = heap_open(cstoreSkipNodesOid, AccessShareLock);
 	Relation index = index_open(CStoreSkipNodesIndexRelationId(), AccessShareLock);
 
-	ScanKeyInit(&scanKey[0], Anum_cstore_skipnodes_relfilenode,
+	ScanKeyInit(&scanKey[0], Anum_cstore_skipnodes_storageid,
-				BTEqualStrategyNumber, F_OIDEQ, Int32GetDatum(relfilenode));
+				BTEqualStrategyNumber, F_OIDEQ, UInt64GetDatum(metapage.storageId));
 	ScanKeyInit(&scanKey[1], Anum_cstore_skipnodes_stripe,
 				BTEqualStrategyNumber, F_OIDEQ, Int32GetDatum(stripe));
 
@@ -410,11 +454,11 @@ ReadStripeSkipList(Oid relfilenode, uint64 stripe, TupleDesc tupleDescriptor,
  * InsertStripeMetadataRow adds a row to cstore_stripes.
  */
 static void
-InsertStripeMetadataRow(Oid relfilenode, StripeMetadata *stripe)
+InsertStripeMetadataRow(uint64 storageId, StripeMetadata *stripe)
 {
 	bool nulls[Natts_cstore_stripes] = { 0 };
 	Datum values[Natts_cstore_stripes] = {
-		ObjectIdGetDatum(relfilenode),
+		UInt64GetDatum(storageId),
 		Int64GetDatum(stripe->id),
 		Int64GetDatum(stripe->fileOffset),
 		Int64GetDatum(stripe->dataLength),
@@ -444,16 +488,23 @@ InsertStripeMetadataRow(Oid relfilenode, StripeMetadata *stripe)
  * from cstore_data_files and cstore_stripes.
  */
 DataFileMetadata *
-ReadDataFileMetadata(Oid relfilenode, bool missingOk)
+ReadDataFileMetadata(RelFileNode relfilenode, bool missingOk)
 {
+	ColumnarMetapage metapage;
 	DataFileMetadata *datafileMetadata = palloc0(sizeof(DataFileMetadata));
-	bool found = ReadCStoreDataFiles(relfilenode, datafileMetadata);
+
+	bool found = ReadMetapage(relfilenode, &metapage);
+	if (found)
+	{
+		found = ReadCStoreDataFiles(metapage.storageId, datafileMetadata);
+	}
+
 	if (!found)
 	{
 		if (!missingOk)
 		{
 			ereport(ERROR, (errmsg("Relfilenode %d doesn't belong to a cstore table.",
-								   relfilenode)));
+								   relfilenode.relNode)));
 		}
 		else
 		{
@@ -462,7 +513,7 @@ ReadDataFileMetadata(Oid relfilenode, bool missingOk)
 	}
 
 	datafileMetadata->stripeMetadataList =
-		ReadDataFileStripeList(relfilenode, GetTransactionSnapshot());
+		ReadDataFileStripeList(metapage.storageId, GetTransactionSnapshot());
 
 	return datafileMetadata;
 }
@@ -473,12 +524,17 @@ ReadDataFileMetadata(Oid relfilenode, bool missingOk)
  * relfilenode across all active and inactive transactions.
  */
 uint64
-GetHighestUsedAddress(Oid relfilenode)
+GetHighestUsedAddress(RelFileNode relfilenode)
 {
 	uint64 highestUsedAddress = 0;
 	uint64 highestUsedId = 0;
+	ColumnarMetapage metapage;
+	if (!ReadMetapage(relfilenode, &metapage))
+	{
+		elog(ERROR, "metapage was not found");
+	}
 
-	GetHighestUsedAddressAndId(relfilenode, &highestUsedAddress, &highestUsedId);
+	GetHighestUsedAddressAndId(metapage.storageId, &highestUsedAddress, &highestUsedId);
 
 	return highestUsedAddress;
 }
@@ -489,7 +545,7 @@ GetHighestUsedAddress(Oid relfilenode)
  * the given relfilenode across all active and inactive transactions.
  */
 static void
-GetHighestUsedAddressAndId(Oid relfilenode,
+GetHighestUsedAddressAndId(uint64 storageId,
 						   uint64 *highestUsedAddress,
 						   uint64 *highestUsedId)
 {
@@ -498,10 +554,12 @@ GetHighestUsedAddressAndId(Oid relfilenode,
 	SnapshotData SnapshotDirty;
 	InitDirtySnapshot(SnapshotDirty);
 
-	List *stripeMetadataList = ReadDataFileStripeList(relfilenode, &SnapshotDirty);
+	List *stripeMetadataList = ReadDataFileStripeList(storageId, &SnapshotDirty);
 
 	*highestUsedId = 0;
-	*highestUsedAddress = 0;
+
+	/* file starts with metapage */
+	*highestUsedAddress = sizeof(ColumnarMetapage);
 
 	foreach(stripeMetadataCell, stripeMetadataList)
 	{
@@ -535,8 +593,14 @@ ReserveStripe(Relation rel, uint64 sizeBytes,
 	 */
 	LockRelation(rel, ShareUpdateExclusiveLock);
 
-	Oid relfilenode = rel->rd_node.relNode;
+	RelFileNode relfilenode = rel->rd_node;
-	GetHighestUsedAddressAndId(relfilenode, &currLogicalHigh, &highestId);
+	ColumnarMetapage metapage;
+	if (!ReadMetapage(relfilenode, &metapage))
+	{
+		elog(WARNING, "metapage was not found");
+	}
+
+	GetHighestUsedAddressAndId(metapage.storageId, &currLogicalHigh, &highestId);
 	SmgrAddr currSmgrHigh = logical_to_smgr(currLogicalHigh);
 
 	SmgrAddr resSmgrStart = next_block_start(currSmgrHigh);
@@ -565,7 +629,7 @@ ReserveStripe(Relation rel, uint64 sizeBytes,
 	stripe.rowCount = rowCount;
 	stripe.id = highestId + 1;
 
-	InsertStripeMetadataRow(relfilenode, &stripe);
+	InsertStripeMetadataRow(metapage.storageId, &stripe);
 
 	UnlockRelation(rel, ShareUpdateExclusiveLock);
 
@@ -574,18 +638,18 @@ ReserveStripe(Relation rel, uint64 sizeBytes,
 
 
 /*
- * ReadDataFileStripeList reads the stripe list for a given relfilenode
+ * ReadDataFileStripeList reads the stripe list for a given storageId
  * in the given snapshot.
  */
 static List *
-ReadDataFileStripeList(Oid relfilenode, Snapshot snapshot)
+ReadDataFileStripeList(uint64 storageId, Snapshot snapshot)
 {
 	List *stripeMetadataList = NIL;
 	ScanKeyData scanKey[1];
 	HeapTuple heapTuple;
 
-	ScanKeyInit(&scanKey[0], Anum_cstore_stripes_relfilenode,
+	ScanKeyInit(&scanKey[0], Anum_cstore_stripes_storageid,
-				BTEqualStrategyNumber, F_OIDEQ, Int32GetDatum(relfilenode));
+				BTEqualStrategyNumber, F_OIDEQ, Int32GetDatum(storageId));
 
 	Oid cstoreStripesOid = CStoreStripesRelationId();
 	Relation cstoreStripes = heap_open(cstoreStripesOid, AccessShareLock);
@@ -634,13 +698,13 @@ ReadDataFileStripeList(Oid relfilenode, Snapshot snapshot)
  * false if table was not found in cstore_data_files.
  */
 static bool
-ReadCStoreDataFiles(Oid relfilenode, DataFileMetadata *metadata)
+ReadCStoreDataFiles(uint64 storageid, DataFileMetadata *metadata)
 {
 	bool found = false;
 	ScanKeyData scanKey[1];
 
-	ScanKeyInit(&scanKey[0], Anum_cstore_data_files_relfilenode,
+	ScanKeyInit(&scanKey[0], Anum_cstore_data_files_storageid,
-				BTEqualStrategyNumber, F_OIDEQ, Int32GetDatum(relfilenode));
+				BTEqualStrategyNumber, F_OIDEQ, UInt64GetDatum(storageid));
 
 	Oid cstoreDataFilesOid = CStoreDataFilesRelationId();
 	Relation cstoreDataFiles = try_relation_open(cstoreDataFilesOid, AccessShareLock);
@@ -684,6 +748,7 @@ ReadCStoreDataFiles(Oid relfilenode, DataFileMetadata *metadata)
 				datumArray[Anum_cstore_data_files_compression - 1]);
 			metadata->compression = ParseCompressionType(NameStr(*compressionName));
 		}
+
 		found = true;
 	}
 
@@ -699,7 +764,7 @@ ReadCStoreDataFiles(Oid relfilenode, DataFileMetadata *metadata)
  * DeleteDataFileMetadataRowIfExists removes the row with given relfilenode from cstore_stripes.
  */
 void
-DeleteDataFileMetadataRowIfExists(Oid relfilenode)
+DeleteDataFileMetadataRowIfExists(RelFileNode relfilenode)
 {
 	ScanKeyData scanKey[1];
 
@@ -712,8 +777,14 @@ DeleteDataFileMetadataRowIfExists(Oid relfilenode)
 		return;
 	}
 
-	ScanKeyInit(&scanKey[0], Anum_cstore_data_files_relfilenode,
+	ColumnarMetapage metapage;
-				BTEqualStrategyNumber, F_OIDEQ, Int32GetDatum(relfilenode));
+	if (!ReadMetapage(relfilenode, &metapage))
+	{
+		return;
+	}
+
+	ScanKeyInit(&scanKey[0], Anum_cstore_data_files_storageid,
+				BTEqualStrategyNumber, F_INT8EQ, UInt64GetDatum(metapage.storageId));
 
 	Oid cstoreDataFilesOid = CStoreDataFilesRelationId();
 	Relation cstoreDataFiles = try_relation_open(cstoreDataFilesOid, AccessShareLock);
@@ -996,3 +1067,90 @@ CStoreNamespaceId(void)
 {
 	return get_namespace_oid("cstore", false);
 }
+
+
+static bool
+ReadMetapage(RelFileNode relfilenode, ColumnarMetapage *metapage)
+{
+	Oid relationId = RelidByRelfilenode(relfilenode.spcNode,
+										relfilenode.relNode);
+	if (!OidIsValid(relationId))
+	{
+		return false;
+	}
+
+	Relation relation = relation_open(relationId, NoLock);
+
+	RelationOpenSmgr(relation);
+	int nblocks = smgrnblocks(relation->rd_smgr, MAIN_FORKNUM);
+	RelationCloseSmgr(relation);
+
+	if (nblocks == 0)
+	{
+		relation_close(relation, NoLock);
+		return false;
+	}
+
+	StringInfo metapageBuffer = ReadFromSmgr(relation, 0, sizeof(ColumnarMetapage));
+	relation_close(relation, NoLock);
+
+	memcpy((void *) metapage, metapageBuffer->data, sizeof(ColumnarMetapage));
+
+	return true;
+}
+
+
+static ColumnarMetapage
+InitMetapage(Relation relation)
+{
+	ColumnarMetapage metapage;
+	metapage.storageId = GetNextStorageId();
+
+	/* create the first block */
+	Buffer newBuffer = ReadBuffer(relation, P_NEW);
+	ReleaseBuffer(newBuffer);
+
+	Assert(sizeof(ColumnarMetapage) <= BLCKSZ - SizeOfPageHeaderData);
+	WriteToSmgr(relation, 0, (char *) &metapage, sizeof(ColumnarMetapage));
+
+	return metapage;
+}
+
+
+static uint64
+GetNextStorageId(void)
+{
+	Oid sequenceId = get_relname_relid("cstore_storageid_seq", CStoreNamespaceId());
+	Datum sequenceIdDatum = ObjectIdGetDatum(sequenceId);
+
+	/*
+	 * Generate new and unique storage id from sequence.
+	 * TODO: should we restrict access to the sequence, which might require
+	 * switching security context here?
+	 */
+	Datum storageIdDatum = DirectFunctionCall1(nextval_oid, sequenceIdDatum);
+
+	uint64 storageId = DatumGetInt64(storageIdDatum);
+
+	return storageId;
+}
+
+
+Datum
+columnar_relation_storageid(PG_FUNCTION_ARGS)
+{
+	Oid relationId = PG_GETARG_OID(0);
+	uint64 storageId = -1;
+
+	Relation relation = relation_open(relationId, AccessShareLock);
+	ColumnarMetapage metadata;
+	if (IsCStoreTableAmTable(relationId) &&
+		ReadMetapage(relation->rd_node, &metadata))
+	{
+		storageId = metadata.storageId;
+	}
+
+	relation_close(relation, AccessShareLock);
+
+	PG_RETURN_INT64(storageId);
+}

src/backend/columnar/cstore_reader.c

@@ -76,7 +76,6 @@ static BlockData * DeserializeBlockData(StripeBuffers *stripeBuffers, uint64 blo
 										List *projectedColumnList);
 static Datum ColumnDefaultValue(TupleConstr *tupleConstraints,
 								Form_pg_attribute attributeForm);
-static StringInfo ReadFromSmgr(Relation rel, uint64 offset, uint32 size);
 
 /*
  * CStoreBeginRead initializes a cstore read operation. This function returns a
@@ -86,9 +85,7 @@ TableReadState *
 CStoreBeginRead(Relation relation, TupleDesc tupleDescriptor,
 				List *projectedColumnList, List *whereClauseList)
 {
-	Oid relNode = relation->rd_node.relNode;
+	DataFileMetadata *datafileMetadata = ReadDataFileMetadata(relation->rd_node, true);
-
-	DataFileMetadata *datafileMetadata = ReadDataFileMetadata(relNode, false);
 
 	/*
 	 * We allocate all stripe specific data in the stripeReadContext, and reset
@@ -127,6 +124,11 @@ CStoreReadNextRow(TableReadState *readState, Datum *columnValues, bool *columnNu
 	StripeMetadata *stripeMetadata = readState->currentStripeMetadata;
 	MemoryContext oldContext = NULL;
 
+	if (readState->datafileMetadata == NULL)
+	{
+		return false;
+	}
+
 	/*
 	 * If no stripes are loaded, load the next non-empty stripe. Note that when
 	 * loading stripes, we skip over blocks whose contents can be filtered with
@@ -238,8 +240,13 @@ void
 CStoreEndRead(TableReadState *readState)
 {
 	MemoryContextDelete(readState->stripeReadContext);
+
+	if (readState->datafileMetadata)
+	{
 		list_free_deep(readState->datafileMetadata->stripeMetadataList);
 		pfree(readState->datafileMetadata);
+	}
+
 	pfree(readState);
 }
 
@@ -317,7 +324,12 @@ CStoreTableRowCount(Relation relation)
 	ListCell *stripeMetadataCell = NULL;
 	uint64 totalRowCount = 0;
 
-	DataFileMetadata *datafileMetadata = ReadDataFileMetadata(relation->rd_node.relNode,
+	if (UninitializedDatafile(relation->rd_node))
+	{
+		return 0;
+	}
+
+	DataFileMetadata *datafileMetadata = ReadDataFileMetadata(relation->rd_node,
 															  false);
 
 	foreach(stripeMetadataCell, datafileMetadata->stripeMetadataList)
@@ -345,7 +357,7 @@ LoadFilteredStripeBuffers(Relation relation, StripeMetadata *stripeMetadata,
 
 	bool *projectedColumnMask = ProjectedColumnMask(columnCount, projectedColumnList);
 
-	StripeSkipList *stripeSkipList = ReadStripeSkipList(relation->rd_node.relNode,
+	StripeSkipList *stripeSkipList = ReadStripeSkipList(relation->rd_node,
 														stripeMetadata->id,
 														tupleDescriptor,
 														stripeMetadata->blockCount);
@@ -1009,7 +1021,7 @@ ColumnDefaultValue(TupleConstr *tupleConstraints, Form_pg_attribute attributeFor
 }
 
 
-static StringInfo
+StringInfo
 ReadFromSmgr(Relation rel, uint64 offset, uint32 size)
 {
 	StringInfo resultBuffer = makeStringInfo();

src/backend/columnar/cstore_tableam.c

@@ -106,7 +106,6 @@ static void CStoreTableAMProcessUtility(PlannedStmt *plannedStatement,
 										char *completionTag);
 #endif
 
-static bool IsCStoreTableAmTable(Oid relationId);
 static bool ConditionalLockRelationWithTimeout(Relation rel, LOCKMODE lockMode,
 											   int timeout, int retryInterval);
 static void LogRelationStats(Relation rel, int elevel);
@@ -117,8 +116,8 @@ static void TruncateCStore(Relation rel, int elevel);
  * CStoreTableAMDefaultOptions returns the default options for a cstore table am table.
  * These options are based on the GUC's controlling the defaults.
  */
-static CStoreOptions *
+CStoreOptions *
-CStoreTableAMDefaultOptions()
+CStoreTableAMDefaultOptions(void)
 {
 	CStoreOptions *cstoreOptions = palloc0(sizeof(CStoreOptions));
 	cstoreOptions->compressionType = cstore_compression;
@@ -133,11 +132,14 @@ CStoreTableAMDefaultOptions()
  * relation is a cstore table am table, if not it will raise an error
  */
 CStoreOptions *
-CStoreTableAMGetOptions(Oid relfilenode)
+CStoreTableAMGetOptions(RelFileNode relfilenode)
 {
-	Assert(OidIsValid(relfilenode));
-
 	CStoreOptions *cstoreOptions = palloc0(sizeof(CStoreOptions));
+	if (UninitializedDatafile(relfilenode))
+	{
+		return CStoreTableAMDefaultOptions();
+	}
+
 	DataFileMetadata *metadata = ReadDataFileMetadata(relfilenode, false);
 	cstoreOptions->compressionType = metadata->compression;
 	cstoreOptions->stripeRowCount = metadata->stripeRowCount;
@@ -570,50 +572,32 @@ cstore_relation_set_new_filenode(Relation rel,
 								 TransactionId *freezeXid,
 								 MultiXactId *minmulti)
 {
-	DataFileMetadata *metadata = ReadDataFileMetadata(rel->rd_node.relNode, true);
-	uint64 blockRowCount = 0;
-	uint64 stripeRowCount = 0;
-	CompressionType compression = 0;
 	Oid oldRelfilenode = rel->rd_node.relNode;
 
 	MarkRelfilenodeDropped(oldRelfilenode, GetCurrentSubTransactionId());
 
-	if (metadata != NULL)
-	{
-		/* existing table (e.g. TRUNCATE), use existing blockRowCount */
-		blockRowCount = metadata->blockRowCount;
-		stripeRowCount = metadata->stripeRowCount;
-		compression = metadata->compression;
-	}
-	else
-	{
-		/* new table, use options */
-		CStoreOptions *options = CStoreTableAMDefaultOptions();
-		blockRowCount = options->blockRowCount;
-		stripeRowCount = options->stripeRowCount;
-		compression = options->compressionType;
-	}
-
 	/* delete old relfilenode metadata */
-	DeleteDataFileMetadataRowIfExists(rel->rd_node.relNode);
+	DeleteDataFileMetadataRowIfExists(rel->rd_node);
 
 	Assert(persistence == RELPERSISTENCE_PERMANENT);
 	*freezeXid = RecentXmin;
 	*minmulti = GetOldestMultiXactId();
 	SMgrRelation srel = RelationCreateStorage(*newrnode, persistence);
-	InitCStoreDataFileMetadata(newrnode->relNode, blockRowCount, stripeRowCount,
-							   compression);
 	smgrclose(srel);
+
+	/* we will lazily initialize metadata in first write */
 }
 
 
 static void
 cstore_relation_nontransactional_truncate(Relation rel)
 {
-	Oid relfilenode = rel->rd_node.relNode;
+	RelFileNode relfilenode = rel->rd_node;
-	DataFileMetadata *metadata = ReadDataFileMetadata(relfilenode, false);
 
-	NonTransactionDropWriteState(relfilenode);
+	NonTransactionDropWriteState(relfilenode.relNode);
+
+	/* Delete old relfilenode metadata and recreate it */
+	DeleteDataFileMetadataRowIfExists(relfilenode);
 
 	/*
 	 * No need to set new relfilenode, since the table was created in this
@@ -624,10 +608,7 @@ cstore_relation_nontransactional_truncate(Relation rel)
 	 */
 	RelationTruncate(rel, 0);
 
-	/* Delete old relfilenode metadata and recreate it */
+	/* we will lazily initialize new metadata in first write */
-	DeleteDataFileMetadataRowIfExists(rel->rd_node.relNode);
-	InitCStoreDataFileMetadata(rel->rd_node.relNode, metadata->blockRowCount,
-							   metadata->stripeRowCount, metadata->compression);
 }
 
 
@@ -675,14 +656,9 @@ cstore_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
 	 * relation first.
 	 */
 
-	CStoreOptions *cstoreOptions = CStoreTableAMGetOptions(OldHeap->rd_node.relNode);
+	CStoreOptions *cstoreOptions = CStoreTableAMGetOptions(OldHeap->rd_node);
 
-	UpdateCStoreDataFileMetadata(NewHeap->rd_node.relNode,
+	cstoreOptions = CStoreTableAMGetOptions(NewHeap->rd_node);
-								 cstoreOptions->blockRowCount,
-								 cstoreOptions->stripeRowCount,
-								 cstoreOptions->compressionType);
-
-	cstoreOptions = CStoreTableAMGetOptions(NewHeap->rd_node.relNode);
 
 	TableWriteState *writeState = CStoreBeginWrite(NewHeap->rd_node,
 												   cstoreOptions->compressionType,
@@ -740,7 +716,7 @@ static void
 LogRelationStats(Relation rel, int elevel)
 {
 	ListCell *stripeMetadataCell = NULL;
-	Oid relfilenode = rel->rd_node.relNode;
+	RelFileNode relfilenode = rel->rd_node;
 	StringInfo infoBuf = makeStringInfo();
 
 	int compressionStats[COMPRESSION_COUNT] = { 0 };
@@ -866,7 +842,7 @@ TruncateCStore(Relation rel, int elevel)
 	 * we're truncating.
 	 */
 	SmgrAddr highestPhysicalAddress =
-		logical_to_smgr(GetHighestUsedAddress(rel->rd_node.relNode));
+		logical_to_smgr(GetHighestUsedAddress(rel->rd_node));
 
 	BlockNumber new_rel_pages = highestPhysicalAddress.blockno + 1;
 	if (new_rel_pages == old_rel_pages)
@@ -1234,10 +1210,10 @@ CStoreTableAMObjectAccessHook(ObjectAccessType access, Oid classId, Oid objectId
 		 * tableam tables storage is managed by postgres.
 		 */
 		Relation rel = table_open(objectId, AccessExclusiveLock);
-		Oid relfilenode = rel->rd_node.relNode;
+		RelFileNode relfilenode = rel->rd_node;
 		DeleteDataFileMetadataRowIfExists(relfilenode);
 
-		MarkRelfilenodeDropped(relfilenode, GetCurrentSubTransactionId());
+		MarkRelfilenodeDropped(relfilenode.relNode, GetCurrentSubTransactionId());
 
 		/* keep the lock since we did physical changes to the relation */
 		table_close(rel, NoLock);
@@ -1249,7 +1225,7 @@ CStoreTableAMObjectAccessHook(ObjectAccessType access, Oid classId, Oid objectId
  * IsCStoreTableAmTable returns true if relation has cstore_tableam
  * access method. This can be called before extension creation.
  */
-static bool
+bool
 IsCStoreTableAmTable(Oid relationId)
 {
 	if (!OidIsValid(relationId))
@@ -1367,7 +1343,7 @@ alter_columnar_table_set(PG_FUNCTION_ARGS)
 	Oid relationId = PG_GETARG_OID(0);
 
 	Relation rel = table_open(relationId, AccessExclusiveLock); /* ALTER TABLE LOCK */
-	DataFileMetadata *metadata = ReadDataFileMetadata(rel->rd_node.relNode, true);
+	DataFileMetadata *metadata = ReadDataFileMetadata(rel->rd_node, true);
 	if (!metadata)
 	{
 		ereport(ERROR, (errmsg("table %s is not a cstore table",
@@ -1406,7 +1382,7 @@ alter_columnar_table_set(PG_FUNCTION_ARGS)
 								CompressionTypeStr(compression))));
 	}
 
-	UpdateCStoreDataFileMetadata(rel->rd_node.relNode, blockRowCount, stripeRowCount,
+	UpdateCStoreDataFileMetadata(rel->rd_node, blockRowCount, stripeRowCount,
 								 compression);
 
 	table_close(rel, NoLock);
@@ -1422,7 +1398,7 @@ alter_columnar_table_reset(PG_FUNCTION_ARGS)
 	Oid relationId = PG_GETARG_OID(0);
 
 	Relation rel = table_open(relationId, AccessExclusiveLock); /* ALTER TABLE LOCK */
-	DataFileMetadata *metadata = ReadDataFileMetadata(rel->rd_node.relNode, true);
+	DataFileMetadata *metadata = ReadDataFileMetadata(rel->rd_node, true);
 	if (!metadata)
 	{
 		ereport(ERROR, (errmsg("table %s is not a cstore table",
@@ -1455,7 +1431,7 @@ alter_columnar_table_reset(PG_FUNCTION_ARGS)
 								CompressionTypeStr(compression))));
 	}
 
-	UpdateCStoreDataFileMetadata(rel->rd_node.relNode, blockRowCount, stripeRowCount,
+	UpdateCStoreDataFileMetadata(rel->rd_node, blockRowCount, stripeRowCount,
 								 compression);
 
 	table_close(rel, NoLock);

src/backend/columnar/cstore_writer.c

@@ -51,6 +51,7 @@ static void UpdateBlockSkipNodeMinMax(ColumnBlockSkipNode *blockSkipNode,
 									  FmgrInfo *comparisonFunction);
 static Datum DatumCopy(Datum datum, bool datumTypeByValue, int datumTypeLength);
 static StringInfo CopyStringInfo(StringInfo sourceString);
+static void InitializeMetadata(RelFileNode relfilenode);
 
 /*
  * CStoreBeginWrite initializes a cstore data load operation and returns a table
@@ -65,12 +66,16 @@ CStoreBeginWrite(RelFileNode relfilenode,
 				 uint64 stripeMaxRowCount, uint32 blockRowCount,
 				 TupleDesc tupleDescriptor)
 {
-	uint32 columnIndex = 0;
+	/* initialize metadata if this is the first write */
+	if (UninitializedDatafile(relfilenode))
+	{
+		InitializeMetadata(relfilenode);
+	}
 
 	/* get comparison function pointers for each of the columns */
 	uint32 columnCount = tupleDescriptor->natts;
 	FmgrInfo **comparisonFunctionArray = palloc0(columnCount * sizeof(FmgrInfo *));
-	for (columnIndex = 0; columnIndex < columnCount; columnIndex++)
+	for (uint32 columnIndex = 0; columnIndex < columnCount; columnIndex++)
 	{
 		FmgrInfo *comparisonFunction = NULL;
 		FormData_pg_attribute *attributeForm = TupleDescAttr(tupleDescriptor,
@@ -122,6 +127,34 @@ CStoreBeginWrite(RelFileNode relfilenode,
 }
 
 
+/*
+ * UninitializedDatafile returns true if metadata for the
+ * given relfilenode hasn't been initialized.
+ */
+bool
+UninitializedDatafile(RelFileNode relfilenode)
+{
+	return ReadDataFileMetadata(relfilenode, true) == NULL;
+}
+
+
+static void
+InitializeMetadata(RelFileNode relfilenode)
+{
+	Oid relationId = RelidByRelfilenode(relfilenode.spcNode, relfilenode.relNode);
+	Relation relation = relation_open(relationId, NoLock);
+
+	CStoreOptions *options = CStoreTableAMDefaultOptions();
+
+	InitCStoreDataFileMetadata(relation,
+							   options->blockRowCount,
+							   options->stripeRowCount,
+							   options->compressionType);
+
+	relation_close(relation, NoLock);
+}
+
+
 /*
  * CStoreWriteRow adds a row to the cstore file. If the stripe is not initialized,
  * we create structures to hold stripe data and skip list. Then, we serialize and
@@ -323,8 +356,9 @@ CreateEmptyStripeSkipList(uint32 stripeMaxRowCount, uint32 blockRowCount,
 }
 
 
-static void
+void
-WriteToSmgr(Relation rel, uint64 logicalOffset, char *data, uint32 dataLength)
+WriteToSmgr(Relation relation, uint64 logicalOffset, char *data,
+			uint32 dataLength)
 {
 	uint64 remaining = dataLength;
 	Buffer buffer;
@@ -333,13 +367,13 @@ WriteToSmgr(Relation rel, uint64 logicalOffset, char *data, uint32 dataLength)
 	{
 		SmgrAddr addr = logical_to_smgr(logicalOffset);
 
-		RelationOpenSmgr(rel);
+		RelationOpenSmgr(relation);
-		BlockNumber nblocks = smgrnblocks(rel->rd_smgr, MAIN_FORKNUM);
+		BlockNumber nblocks = smgrnblocks(relation->rd_smgr, MAIN_FORKNUM);
 		Assert(addr.blockno < nblocks);
 		(void) nblocks; /* keep compiler quiet */
-		RelationCloseSmgr(rel);
+		RelationCloseSmgr(relation);
 
-		buffer = ReadBuffer(rel, addr.blockno);
+		buffer = ReadBuffer(relation, addr.blockno);
 		LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
 
 		Page page = BufferGetPage(buffer);
@@ -372,7 +406,7 @@ WriteToSmgr(Relation rel, uint64 logicalOffset, char *data, uint32 dataLength)
 
 		MarkBufferDirty(buffer);
 
-		if (RelationNeedsWAL(rel))
+		if (RelationNeedsWAL(relation))
 		{
 			XLogBeginInsert();
 
@@ -521,7 +555,7 @@ FlushStripe(TableWriteState *writeState)
 	}
 
 	/* create skip list and footer buffers */
-	SaveStripeSkipList(relation->rd_node.relNode,
+	SaveStripeSkipList(writeState->relfilenode,
 					   stripeMetadata.id,
 					   stripeSkipList, tupleDescriptor);
 

src/backend/columnar/sql/columnar--9.5-1--10.0-1.sql

@@ -3,20 +3,22 @@
 CREATE SCHEMA cstore;
 SET search_path TO cstore;
 
+CREATE SEQUENCE cstore_storageid_seq NO CYCLE;
+
 CREATE TABLE cstore_data_files (
-    relfilenode oid NOT NULL,
+    storageid bigint NOT NULL,
     block_row_count int NOT NULL,
     stripe_row_count int NOT NULL,
     compression name NOT NULL,
     version_major bigint NOT NULL,
     version_minor bigint NOT NULL,
-    PRIMARY KEY (relfilenode)
+    PRIMARY KEY (storageid)
 ) WITH (user_catalog_table = true);
 
 COMMENT ON TABLE cstore_data_files IS 'CStore data file wide metadata';
 
 CREATE TABLE cstore_stripes (
-    relfilenode oid NOT NULL,
+    storageid bigint NOT NULL,
     stripe bigint NOT NULL,
     file_offset bigint NOT NULL,
     data_length bigint NOT NULL,
@@ -24,14 +26,14 @@ CREATE TABLE cstore_stripes (
     block_count int NOT NULL,
     block_row_count int NOT NULL,
     row_count bigint NOT NULL,
-    PRIMARY KEY (relfilenode, stripe),
+    PRIMARY KEY (storageid, stripe),
-    FOREIGN KEY (relfilenode) REFERENCES cstore_data_files(relfilenode) ON DELETE CASCADE INITIALLY DEFERRED
+    FOREIGN KEY (storageid) REFERENCES cstore_data_files(storageid) ON DELETE CASCADE INITIALLY DEFERRED
 ) WITH (user_catalog_table = true);
 
 COMMENT ON TABLE cstore_stripes IS 'CStore per stripe metadata';
 
 CREATE TABLE cstore_skipnodes (
-    relfilenode oid NOT NULL,
+    storageid bigint NOT NULL,
     stripe bigint NOT NULL,
     attr int NOT NULL,
     block int NOT NULL,
@@ -43,19 +45,23 @@ CREATE TABLE cstore_skipnodes (
     exists_stream_offset bigint NOT NULL,
     exists_stream_length bigint NOT NULL,
     value_compression_type int NOT NULL,
-    PRIMARY KEY (relfilenode, stripe, attr, block),
+    PRIMARY KEY (storageid, stripe, attr, block),
-    FOREIGN KEY (relfilenode, stripe) REFERENCES cstore_stripes(relfilenode, stripe) ON DELETE CASCADE INITIALLY DEFERRED
+    FOREIGN KEY (storageid, stripe) REFERENCES cstore_stripes(storageid, stripe) ON DELETE CASCADE INITIALLY DEFERRED
 ) WITH (user_catalog_table = true);
 
 COMMENT ON TABLE cstore_skipnodes IS 'CStore per block metadata';
 
+CREATE FUNCTION columnar_relation_storageid(relid oid) RETURNS bigint
+    LANGUAGE C STABLE STRICT
+    AS 'MODULE_PATHNAME', $$columnar_relation_storageid$$;
+
 CREATE VIEW columnar_options AS
 SELECT c.oid::regclass regclass,
        d.block_row_count,
        d.stripe_row_count,
        d.compression
-FROM pg_class c
+FROM pg_class c, cstore.cstore_data_files d
-JOIN cstore.cstore_data_files d USING(relfilenode);
+WHERE d.storageid=columnar_relation_storageid(c.oid);
 
 COMMENT ON VIEW columnar_options IS 'CStore per table settings';
 

src/backend/columnar/sql/downgrades/columnar--10.0-1--9.5-1.sql

@@ -31,6 +31,7 @@ DROP VIEW columnar_options;
 DROP TABLE cstore_skipnodes;
 DROP TABLE cstore_stripes;
 DROP TABLE cstore_data_files;
+DROP SEQUENCE cstore_storageid_seq;
 
 DROP FUNCTION citus_internal.cstore_ensure_objects_exist();
 

src/backend/columnar/write_state_management.c

@@ -178,7 +178,7 @@ cstore_init_write_state(RelFileNode relfilenode, TupleDesc tupdesc,
 	 */
 	MemoryContext oldContext = MemoryContextSwitchTo(WriteStateContext);
 
-	CStoreOptions *cstoreOptions = CStoreTableAMGetOptions(relfilenode.relNode);
+	CStoreOptions *cstoreOptions = CStoreTableAMGetOptions(relfilenode);
 	SubXidWriteState *stackEntry = palloc0(sizeof(SubXidWriteState));
 	stackEntry->writeState = CStoreBeginWrite(relfilenode,
 											  cstoreOptions->compressionType,

src/include/columnar/cstore.h

@@ -263,6 +263,7 @@ extern void CStoreWriteRow(TableWriteState *state, Datum *columnValues,
 extern void CStoreFlushPendingWrites(TableWriteState *state);
 extern void CStoreEndWrite(TableWriteState *state);
 extern bool ContainsPendingWrites(TableWriteState *state);
+extern bool UninitializedDatafile(RelFileNode relfilenode);
 
 /* Function declarations for reading from a cstore file */
 extern TableReadState * CStoreBeginRead(Relation relation,
@@ -285,23 +286,29 @@ extern bool CompressBuffer(StringInfo inputBuffer, StringInfo outputBuffer,
 						   CompressionType compressionType);
 extern StringInfo DecompressBuffer(StringInfo buffer, CompressionType compressionType);
 extern char * CompressionTypeStr(CompressionType type);
-extern CStoreOptions * CStoreTableAMGetOptions(Oid relfilenode);
+extern CStoreOptions * CStoreTableAMGetOptions(RelFileNode relfilenode);
+extern void WriteToSmgr(Relation relation, uint64 logicalOffset,
+						char *data, uint32 dataLength);
+extern StringInfo ReadFromSmgr(Relation rel, uint64 offset, uint32 size);
+extern bool IsCStoreTableAmTable(Oid relationId);
+extern CStoreOptions * CStoreTableAMDefaultOptions(void);
 
 /* cstore_metadata_tables.c */
-extern void DeleteDataFileMetadataRowIfExists(Oid relfilenode);
+extern void DeleteDataFileMetadataRowIfExists(RelFileNode relfilenode);
-extern void InitCStoreDataFileMetadata(Oid relfilenode, int blockRowCount, int
+extern void InitCStoreDataFileMetadata(Relation relation,
+									   int blockRowCount, int stripeRowCount,
+									   CompressionType compression);
+extern void UpdateCStoreDataFileMetadata(RelFileNode relfilenode, int blockRowCount, int
 										 stripeRowCount, CompressionType compression);
-extern void UpdateCStoreDataFileMetadata(Oid relfilenode, int blockRowCount, int
+extern DataFileMetadata * ReadDataFileMetadata(RelFileNode relfilenode, bool missingOk);
-										 stripeRowCount, CompressionType compression);
+extern uint64 GetHighestUsedAddress(RelFileNode relfilenode);
-extern DataFileMetadata * ReadDataFileMetadata(Oid relfilenode, bool missingOk);
-extern uint64 GetHighestUsedAddress(Oid relfilenode);
 extern StripeMetadata ReserveStripe(Relation rel, uint64 size,
 									uint64 rowCount, uint64 columnCount,
 									uint64 blockCount, uint64 blockRowCount);
-extern void SaveStripeSkipList(Oid relfilenode, uint64 stripe,
+extern void SaveStripeSkipList(RelFileNode relfilenode, uint64 stripe,
 							   StripeSkipList *stripeSkipList,
 							   TupleDesc tupleDescriptor);
-extern StripeSkipList * ReadStripeSkipList(Oid relfilenode, uint64 stripe,
+extern StripeSkipList * ReadStripeSkipList(RelFileNode relfilenode, uint64 stripe,
 										   TupleDesc tupleDescriptor,
 										   uint32 blockCount);
 

src/test/regress/expected/am_drop.out

@@ -26,6 +26,7 @@ SELECT :cstore_data_files_before_drop - count(*) FROM cstore.cstore_data_files;
 -- Create a cstore_fdw table under a schema and drop it.
 CREATE SCHEMA test_schema;
 CREATE TABLE test_schema.test_table(data int) USING columnar;
+INSERT INTO test_schema.test_table VALUES (1);
 SELECT count(*) AS cstore_data_files_before_drop FROM cstore.cstore_data_files \gset
 DROP SCHEMA test_schema CASCADE;
 NOTICE:  drop cascades to table test_schema.test_table

src/test/regress/expected/am_matview.out

@@ -35,20 +35,21 @@ SELECT * FROM t_view a ORDER BY a;
 (6 rows)
 
 -- verify that we have created metadata entries for the materialized view
-SELECT relfilenode FROM pg_class WHERE relname='t_view' \gset
+SELECT cstore.columnar_relation_storageid(oid) AS storageid
-SELECT count(*) FROM cstore.cstore_data_files WHERE relfilenode=:relfilenode;
+FROM pg_class WHERE relname='t_view' \gset
+SELECT count(*) FROM cstore.cstore_data_files WHERE storageid=:storageid;
  count
 ---------------------------------------------------------------------
      1
 (1 row)
 
-SELECT count(*) FROM cstore.cstore_stripes WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_stripes WHERE storageid=:storageid;
  count
 ---------------------------------------------------------------------
      1
 (1 row)
 
-SELECT count(*) FROM cstore.cstore_skipnodes WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_skipnodes WHERE storageid=:storageid;
  count
 ---------------------------------------------------------------------
      3
@@ -57,19 +58,19 @@ SELECT count(*) FROM cstore.cstore_skipnodes WHERE relfilenode=:relfilenode;
 DROP TABLE t CASCADE;
 NOTICE:  drop cascades to materialized view t_view
 -- dropping must remove metadata
-SELECT count(*) FROM cstore.cstore_data_files WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_data_files WHERE storageid=:storageid;
  count
 ---------------------------------------------------------------------
      0
 (1 row)
 
-SELECT count(*) FROM cstore.cstore_stripes WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_stripes WHERE storageid=:storageid;
  count
 ---------------------------------------------------------------------
      0
 (1 row)
 
-SELECT count(*) FROM cstore.cstore_skipnodes WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_skipnodes WHERE storageid=:storageid;
  count
 ---------------------------------------------------------------------
      0

src/test/regress/expected/am_recursive.out

@@ -12,7 +12,7 @@ INSERT INTO t2 SELECT i, f(i) FROM generate_series(1, 5) i;
 -- there are no subtransactions, so above statement should batch
 -- INSERTs inside the UDF and create on stripe per table.
 SELECT relname, count(*) FROM cstore.cstore_stripes a, pg_class b
-WHERE a.relfilenode=b.relfilenode AND relname IN ('t1', 't2')
+WHERE cstore.columnar_relation_storageid(b.oid)=a.storageid AND relname IN ('t1', 't2')
 GROUP BY relname
 ORDER BY relname;
  relname | count

src/test/regress/expected/am_rollback.out

@@ -2,6 +2,9 @@
 -- Testing we handle rollbacks properly
 --
 CREATE TABLE t(a int, b int) USING columnar;
+CREATE VIEW t_stripes AS
+SELECT * FROM cstore.cstore_stripes a, pg_class b
+WHERE a.storageid = cstore.columnar_relation_storageid(b.oid) AND b.relname = 't';
 BEGIN;
 INSERT INTO t SELECT i, i+1 FROM generate_series(1, 10) i;
 ROLLBACK;
@@ -12,8 +15,7 @@ SELECT count(*) FROM t;
 (1 row)
 
 -- check stripe metadata also have been rolled-back
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b
+SELECT count(*) FROM t_stripes;
-WHERE a.relfilenode = b.relfilenode AND b.relname = 't';
  count
 ---------------------------------------------------------------------
      0
@@ -26,8 +28,7 @@ SELECT count(*) FROM t;
     10
 (1 row)
 
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b
+SELECT count(*) FROM t_stripes;
-WHERE a.relfilenode = b.relfilenode AND b.relname = 't';
  count
 ---------------------------------------------------------------------
      1
@@ -73,11 +74,11 @@ SELECT count(*) FROM t;
     20
 (1 row)
 
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b
+SELECT count(*) FROM t_stripes;
-WHERE a.relfilenode = b.relfilenode AND b.relname = 't';
  count
 ---------------------------------------------------------------------
      2
 (1 row)
 
 DROP TABLE t;
+DROP VIEW t_stripes;

src/test/regress/sql/am_drop.sql

@@ -27,6 +27,7 @@ SELECT :cstore_data_files_before_drop - count(*) FROM cstore.cstore_data_files;
 -- Create a cstore_fdw table under a schema and drop it.
 CREATE SCHEMA test_schema;
 CREATE TABLE test_schema.test_table(data int) USING columnar;
+INSERT INTO test_schema.test_table VALUES (1);
 
 SELECT count(*) AS cstore_data_files_before_drop FROM cstore.cstore_data_files \gset
 DROP SCHEMA test_schema CASCADE;

src/test/regress/sql/am_matview.sql

@@ -20,15 +20,16 @@ REFRESH MATERIALIZED VIEW t_view;
 SELECT * FROM t_view a ORDER BY a;
 
 -- verify that we have created metadata entries for the materialized view
-SELECT relfilenode FROM pg_class WHERE relname='t_view' \gset
+SELECT cstore.columnar_relation_storageid(oid) AS storageid
+FROM pg_class WHERE relname='t_view' \gset
 
-SELECT count(*) FROM cstore.cstore_data_files WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_data_files WHERE storageid=:storageid;
-SELECT count(*) FROM cstore.cstore_stripes WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_stripes WHERE storageid=:storageid;
-SELECT count(*) FROM cstore.cstore_skipnodes WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_skipnodes WHERE storageid=:storageid;
 
 DROP TABLE t CASCADE;
 
 -- dropping must remove metadata
-SELECT count(*) FROM cstore.cstore_data_files WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_data_files WHERE storageid=:storageid;
-SELECT count(*) FROM cstore.cstore_stripes WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_stripes WHERE storageid=:storageid;
-SELECT count(*) FROM cstore.cstore_skipnodes WHERE relfilenode=:relfilenode;
+SELECT count(*) FROM cstore.cstore_skipnodes WHERE storageid=:storageid;

src/test/regress/sql/am_recursive.sql

@@ -16,7 +16,7 @@ INSERT INTO t2 SELECT i, f(i) FROM generate_series(1, 5) i;
 -- there are no subtransactions, so above statement should batch
 -- INSERTs inside the UDF and create on stripe per table.
 SELECT relname, count(*) FROM cstore.cstore_stripes a, pg_class b
-WHERE a.relfilenode=b.relfilenode AND relname IN ('t1', 't2')
+WHERE cstore.columnar_relation_storageid(b.oid)=a.storageid AND relname IN ('t1', 't2')
 GROUP BY relname
 ORDER BY relname;
 

src/test/regress/sql/am_rollback.sql

@@ -4,20 +4,22 @@
 
 CREATE TABLE t(a int, b int) USING columnar;
 
+CREATE VIEW t_stripes AS
+SELECT * FROM cstore.cstore_stripes a, pg_class b
+WHERE a.storageid = cstore.columnar_relation_storageid(b.oid) AND b.relname = 't';
+
 BEGIN;
 INSERT INTO t SELECT i, i+1 FROM generate_series(1, 10) i;
 ROLLBACK;
 SELECT count(*) FROM t;
 
 -- check stripe metadata also have been rolled-back
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b
+SELECT count(*) FROM t_stripes;
-WHERE a.relfilenode = b.relfilenode AND b.relname = 't';
 
 INSERT INTO t SELECT i, i+1 FROM generate_series(1, 10) i;
 SELECT count(*) FROM t;
 
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b
+SELECT count(*) FROM t_stripes;
-WHERE a.relfilenode = b.relfilenode AND b.relname = 't';
 
 -- savepoint rollback
 BEGIN;
@@ -36,7 +38,7 @@ COMMIT;
 
 SELECT count(*) FROM t;
 
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b
+SELECT count(*) FROM t_stripes;
-WHERE a.relfilenode = b.relfilenode AND b.relname = 't';
 
 DROP TABLE t;
+DROP VIEW t_stripes;

src/test/regress/sql/am_vacuum.sql

@@ -2,41 +2,49 @@ SELECT count(*) AS columnar_table_count FROM cstore.cstore_data_files \gset
 
 CREATE TABLE t(a int, b int) USING columnar;
 
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b WHERE a.relfilenode=b.relfilenode AND b.relname='t';
+CREATE VIEW t_stripes AS
+SELECT * FROM cstore.cstore_stripes a, pg_class b
+WHERE a.storageid = cstore.columnar_relation_storageid(b.oid) AND b.relname='t';
+
+SELECT count(*) FROM t_stripes;
 
 INSERT INTO t SELECT i, i * i FROM generate_series(1, 10) i;
 INSERT INTO t SELECT i, i * i FROM generate_series(11, 20) i;
 INSERT INTO t SELECT i, i * i FROM generate_series(21, 30) i;
 
 SELECT sum(a), sum(b) FROM t;
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b WHERE a.relfilenode=b.relfilenode AND b.relname='t';
+SELECT count(*) FROM t_stripes;
 
 -- vacuum full should merge stripes together
 VACUUM FULL t;
 
 SELECT sum(a), sum(b) FROM t;
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b WHERE a.relfilenode=b.relfilenode AND b.relname='t';
+SELECT count(*) FROM t_stripes;
 
 -- test the case when all data cannot fit into a single stripe
 SELECT alter_columnar_table_set('t', stripe_row_count => 1000);
 INSERT INTO t SELECT i, 2 * i FROM generate_series(1,2500) i;
 
 SELECT sum(a), sum(b) FROM t;
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b WHERE a.relfilenode=b.relfilenode AND b.relname='t';
+SELECT count(*) FROM t_stripes;
 
 VACUUM FULL t;
 
 SELECT sum(a), sum(b) FROM t;
-SELECT count(*) FROM cstore.cstore_stripes a, pg_class b WHERE a.relfilenode=b.relfilenode AND b.relname='t';
+SELECT count(*) FROM t_stripes;
 
 -- VACUUM FULL doesn't reclaim dropped columns, but converts them to NULLs
 ALTER TABLE t DROP COLUMN a;
 
-SELECT stripe, attr, block, minimum_value IS NULL, maximum_value IS NULL FROM cstore.cstore_skipnodes a, pg_class b WHERE a.relfilenode=b.relfilenode AND b.relname='t' ORDER BY 1, 2, 3;
+SELECT stripe, attr, block, minimum_value IS NULL, maximum_value IS NULL
+FROM cstore.cstore_skipnodes a, pg_class b
+WHERE a.storageid = cstore.columnar_relation_storageid(b.oid) AND b.relname='t' ORDER BY 1, 2, 3;
 
 VACUUM FULL t;
 
-SELECT stripe, attr, block, minimum_value IS NULL, maximum_value IS NULL FROM cstore.cstore_skipnodes a, pg_class b WHERE a.relfilenode=b.relfilenode AND b.relname='t' ORDER BY 1, 2, 3;
+SELECT stripe, attr, block, minimum_value IS NULL, maximum_value IS NULL
+FROM cstore.cstore_skipnodes a, pg_class b
+WHERE a.storageid = cstore.columnar_relation_storageid(b.oid) AND b.relname='t' ORDER BY 1, 2, 3;
 
 -- Make sure we cleaned-up the transient table metadata after VACUUM FULL commands
 SELECT count(*) - :columnar_table_count FROM cstore.cstore_data_files;
@@ -99,6 +107,7 @@ VACUUM FULL t;
 VACUUM VERBOSE t;
 
 DROP TABLE t;
+DROP VIEW t_stripes;
 
 -- Make sure we cleaned the metadata for t too
 SELECT count(*) - :columnar_table_count FROM cstore.cstore_data_files;