ORCA优化器浅析——IMDRelation Storage type of a relation GP6与GP7对比

如上图所示IMDRelation作为Interface for relations in the metadata cache，其定义了Storage type of a relation表的存储类型，如下所示：

enum Erelstoragetype {
		ErelstorageHeap,
		ErelstorageAppendOnlyCols,
		ErelstorageAppendOnlyRows,
		ErelstorageAppendOnlyParquet,
		ErelstorageExternal,
		ErelstorageVirtual,
		ErelstorageSentinel
	};

下面是ErelstorageHeap存储类型所涉及到的代码：

• CTranslatorQueryToDXL::TranslateCTASToDXL Translate a CTAS 设定CDXLLogicalCTAS类的存储类型
• CTranslatorRelcacheToDXL::RetrieveRelStorageType Get relation storage type根据存储类型字符获取其Erelstoragetype枚举类型
• CXformUtils::PexprBuildIndexPlan Construct an expression representing a new access path using the given functors for operator constructors and rewritten access path. Non-heap tables not supported for GiST 非heap表不支持GiST判定

下面是ErelstorageAppendOnlyCols存储类型所涉及到的代码：

• CTranslatorQueryToDXL::GetDXLCtasOptionArray Translate CTAS storage options 根据存储选项判定存储类型
• CTranslatorRelcacheToDXL::RetrieveRelColumns Get relation columns AO表需要添加系统列
• CTranslatorRelcacheToDXL::RetrieveRelStorageType Get relation storage type根据存储类型字符获取其Erelstoragetype枚举类型
• CTableDescriptor::IsAORowOrColTable 判定是否是AO表

下面是ErelstorageAppendOnlyRows存储类型所涉及到的代码：

• CTranslatorQueryToDXL::GetDXLCtasOptionArray Translate CTAS storage options 根据存储选项判定存储类型
• CTranslatorRelcacheToDXL::RetrieveRelColumns Get relation columns AO表需要添加系统列
• CTranslatorRelcacheToDXL::RetrieveRelStorageType Get relation storage type根据存储类型字符获取其Erelstoragetype枚举类型
• CTableDescriptor::IsAORowOrColTable 判定是否是AO表
• CUtils::FGeneratePartOid generate part oid 判定InsertSortOnRows
• CPhysicalDML::FInsertSortOnRows Do we need to sort on insert

下面是ErelstorageAppendOnlyParquet存储类型所涉及到的代码：

• CTranslatorQueryToDXL::GetDXLCtasOptionArray Translate CTAS storage options 根据存储选项判定存储类型
• CUtils::FGeneratePartOid generate part oid 判定InsertSortOnParquet
• CPhysicalDML::FInsertSortOnParquet Do we need to sort on parquet table

下面是ErelstorageExternal存储类型所涉及到的代码：

• CTranslatorDXLToPlStmt::TranslateDXLTblScan Translates a DXL table scan node into a TableScan node
• CTranslatorQueryToDXL::TranslateRTEToDXLLogicalGet Returns a CDXLNode representing a from relation range table entry
• CTranslatorRelcacheToDXL::RetrieveRel Retrieve a relation from the relcache given its metadata id 不为外部表的情况下get partition keys；Retrieve full part constraints partitioned tables with indexes or external partitions, returns NULL for non-partitioned tables
• CTranslatorRelcacheToDXL::RetrieveRelStorageType Get relation storage type根据存储类型字符获取其Erelstoragetype枚举类型
• COptTasks::PrintMissingStatsWarning Print warning messages for columns with missing statistics 不为外部表时打印missing statistic信息

下面是ErelstorageVirtual存储类型所涉及到的代码：CTranslatorRelcacheToDXL::RetrieveRelStorageType Get relation storage type根据存储类型字符获取其Erelstoragetype枚举类型。该枚举类型应该是不用的。

GP6与GP7存储类型对比

enum Erelstoragetype{
		ErelstorageHeap,
		ErelstorageAppendOnlyCols,
		ErelstorageAppendOnlyRows,
		ErelstorageForeign,
		ErelstorageMixedPartitioned,
		ErelstorageCompositeType,
		ErelstorageSentinel
	};

下面是ErelstorageHeap存储类型所涉及到的代码：

• CTranslatorQueryToDXL::TranslateCTASToDXL Translate a CTAS 设定CDXLLogicalCTAS类的存储类型
• CTranslatorRelcacheToDXL::RetrieveRelStorageType Get relation storage type
• CTranslatorRelcacheToDXL::RetrieveStorageTypeForPartitionedTable 如果rel->rd_partdesc->nparts == 0，返回ErelstorageHeap存储类型
• CXformUtils::PexprBuildIndexPlan Construct an expression representing a new access path using the given functors for operator constructors and rewritten access path. Non-heap tables not supported for GiST 非heap表不支持GiST判定

下面是ErelstorageAppendOnlyCols存储类型所涉及到的代码：

• CTranslatorQueryToDXL::GetDXLCtasOptionArray Translate CTAS storage options 根据存储选项判定存储类型
• CTranslatorRelcacheToDXL::RetrieveRelStorageType Get relation storage type根据AM获取其Erelstoragetype枚举类型
• CTableDescriptor::IsAORowOrColTable 判定是否是AO表

下面是ErelstorageAppendOnlyRows存储类型所涉及到的代码：

• CTranslatorQueryToDXL::GetDXLCtasOptionArray Translate CTAS storage options 根据存储选项判定存储类型
• CTranslatorRelcacheToDXL::RetrieveRelStorageType Get relation storage type根据AM获取其Erelstoragetype枚举类型
• CTableDescriptor::IsAORowOrColTable 判定是否是AO表

下面是ErelstorageForeign存储类型所涉及到的代码：

• CTranslatorDXLToPlStmt::TranslateDXLTblScan Translates a DXL table scan node into a TableScan node
• TranslatorQueryToDXL::TranslateRTEToDXLLogicalGet Returns a CDXLNode representing a from relation range table entry 新建CDXLLogicalForeignGet分支
• CTranslatorRelcacheToDXL::RetrieveRel Retrieve a relation from the relcache given its metadata id. root partitions don’t have a foreign server 重要
• CTranslatorRelcacheToDXL::RetrieveRelStorageType rel->rd_rel->relkind == RELKIND_FOREIGN_TABLE分支
• CTranslatorRelcacheToDXL::RetrieveStorageTypeForPartitionedTable for partitioned tables with foreign partitions, we want to ignore the foreign partitions for determining the storage-type (unless all of the partitions are foreign) as we’ll be separating them out to different scans later in CXformExpandDynamicGetWithForeignPartitions
• CTranslatorUtils::GetTableDescr Create a DXL table descriptor from a GPDB range table entry fall back to the planner for queries on coordinator-only table if they are disabled with Orca
• CTranslatorUtils::RelContainsForeignPartitions
• COptTasks::PrintMissingStatsWarning Print warning messages for columns with missing statistics 不为外部表时打印missing statistic信息
• CTranslatorDXLToExpr::PexprLogicalGet Create a Expr logical get from a DXL logical get store array of foreign partitions

下面是ErelstorageMixedPartitioned存储类型所涉及到的代码：

• CTranslatorRelcacheToDXL::RetrieveStorageTypeForPartitionedTable mark any partitioned table with supported partitions of mixed storage types, this is more conservative for certain skans (eg: we can’t do an index scan if any partition is ao, we must only do a sequential or bitmap scan)
• CXformIndexGet2IndexOnlyScan::Exfp 判定possible_ao_table
• CXformIndexGet2IndexScan::Exfp 判定possible_ao_table
• CXformUtils::FIndexApplicable Check if an index is applicable given the required, output and scalar expression columns 判定possible_ao_table
• CXformUtils::PexprBitmapSelectBestIndex Given conjuncts of supported predicates, select the best index and construct a bitmap index path expression. Return unused predicates as residuals 判定possible_ao_table

下面是ErelstorageCompositeType存储类型所涉及到的代码：CTranslatorRelcacheToDXL::RetrieveRelStorageType中对应RELKIND_COMPOSITE_TYPE获取存储类型ErelstorageCompositeType

从上面的对比可以看出AP7将分区表子分区混合存储这种情况单独作为了一种存储类型ErelstorageMixedPartitioned。