duckdb 0.7.2-dev0.0 → 0.7.2-dev1138.0

package/src/duckdb/src/include/duckdb/storage/table/scan_state.hpp

@@ -25,10 +25,12 @@ class RowGroupCollection;
 class UpdateSegment;
 class TableScanState;
 class ColumnSegment;
+class ColumnSegmentTree;
 class ValiditySegment;
 class TableFilterSet;
 class ColumnData;
 class DuckTransaction;
+class RowGroupSegmentTree;
 
 struct SegmentScanState {
 	virtual ~SegmentScanState() {
@@ -45,6 +47,8 @@ typedef unordered_map<block_id_t, BufferHandle> buffer_handle_set_t;
 struct ColumnScanState {
 	//! The column segment that is currently being scanned
 	ColumnSegment *current = nullptr;
+	//! Column segment tree
+	ColumnSegmentTree *segment_tree = nullptr;
 	//! The current row index of the scan
 	idx_t row_index = 0;
 	//! The internal row index (i.e. the position of the SegmentScanState)
@@ -64,14 +68,15 @@ struct ColumnScanState {
 	//! We initialize one SegmentScanState per segment, however, if scanning a DataChunk requires us to scan over more
 	//! than one Segment, we need to keep the scan states of the previous segments around
 	vector<unique_ptr<SegmentScanState>> previous_states;
+	//! The last read offset in the child state (used for LIST columns only)
+	idx_t last_offset = 0;
 
 public:
+	void Initialize(const LogicalType &type);
 	//! Move the scan state forward by "count" rows (including all child states)
 	void Next(idx_t count);
 	//! Move ONLY this state forward by "count" rows (i.e. not the child states)
 	void NextInternal(idx_t count);
-	//! Move the scan state forward by STANDARD_VECTOR_SIZE rows
-	void NextVector();
 };
 
 struct ColumnFetchState {
@@ -83,50 +88,33 @@ struct ColumnFetchState {
 	BufferHandle &GetOrInsertHandle(ColumnSegment &segment);
 };
 
-class RowGroupScanState {
+class CollectionScanState {
 public:
-	RowGroupScanState(CollectionScanState &parent_p)
-	    : row_group(nullptr), vector_index(0), max_row(0), parent(parent_p) {
-	}
+	CollectionScanState(TableScanState &parent_p);
 
 	//! The current row_group we are scanning
-	RowGroup *row_group = nullptr;
+	RowGroup *row_group;
 	//! The vector index within the row_group
-	idx_t vector_index = 0;
-	//! The maximum row index of this row_group scan
-	idx_t max_row = 0;
+	idx_t vector_index;
+	//! The maximum row within the row group
+	idx_t max_row_group_row;
 	//! Child column scans
 	unique_ptr<ColumnScanState[]> column_scans;
-
-public:
-	const vector<column_t> &GetColumnIds();
-	TableFilterSet *GetFilters();
-	AdaptiveFilter *GetAdaptiveFilter();
-	idx_t GetParentMaxRow();
-
-private:
-	//! The parent scan state
-	CollectionScanState &parent;
-};
-
-class CollectionScanState {
-public:
-	CollectionScanState(TableScanState &parent_p)
-	    : row_group_state(*this), max_row(0), batch_index(0), parent(parent_p) {};
-
-	//! The row_group scan state
-	RowGroupScanState row_group_state;
+	//! Row group segment tree
+	RowGroupSegmentTree *row_groups;
 	//! The total maximum row index
 	idx_t max_row;
 	//! The current batch index
 	idx_t batch_index;
 
 public:
+	void Initialize(const vector<LogicalType> &types);
 	const vector<column_t> &GetColumnIds();
 	TableFilterSet *GetFilters();
 	AdaptiveFilter *GetAdaptiveFilter();
 	bool Scan(DuckTransaction &transaction, DataChunk &result);
 	bool ScanCommitted(DataChunk &result, TableScanType type);
+	bool ScanCommitted(DataChunk &result, SegmentLock &l, TableScanType type);
 
 private:
 	TableScanState &parent;
@@ -158,12 +146,16 @@ private:
 };
 
 struct ParallelCollectionScanState {
+	ParallelCollectionScanState();
+
 	//! The row group collection we are scanning
 	RowGroupCollection *collection;
 	RowGroup *current_row_group;
 	idx_t vector_index;
 	idx_t max_row;
 	idx_t batch_index;
+	atomic<idx_t> processed_rows;
+	mutex lock;
 };
 
 struct ParallelTableScanState {

package/src/duckdb/src/include/duckdb/storage/table/segment_base.hpp

@@ -13,13 +13,12 @@
 
 namespace duckdb {
 
+template <class T>
 class SegmentBase {
 public:
 	SegmentBase(idx_t start, idx_t count) : start(start), count(count), next(nullptr) {
 	}
-	virtual ~SegmentBase() {
-	}
-	SegmentBase *Next() {
+	T *Next() {
 #ifndef DUCKDB_R_BUILD
 		return next.load();
 #else
@@ -32,12 +31,13 @@ public:
 	//! The amount of entries in this storage chunk
 	atomic<idx_t> count;
 	//! The next segment after this one
-
 #ifndef DUCKDB_R_BUILD
-	atomic<SegmentBase *> next;
+	atomic<T *> next;
 #else
-	SegmentBase *next;
+	T *next;
 #endif
+	//! The index within the segment tree
+	idx_t index;
 };
 
 } // namespace duckdb

package/src/duckdb/src/include/duckdb/storage/table/segment_tree.hpp

@@ -10,71 +10,326 @@
 
 #include "duckdb/common/constants.hpp"
 #include "duckdb/storage/storage_lock.hpp"
-#include "duckdb/storage/table/segment_base.hpp"
 #include "duckdb/storage/table/segment_lock.hpp"
 #include "duckdb/common/vector.hpp"
 #include "duckdb/common/mutex.hpp"
+#include "duckdb/common/string_util.hpp"
 
 namespace duckdb {
 
+template <class T>
 struct SegmentNode {
 	idx_t row_start;
-	unique_ptr<SegmentBase> node;
+	unique_ptr<T> node;
 };
 
 //! The SegmentTree maintains a list of all segments of a specific column in a table, and allows searching for a segment
 //! by row number
+template <class T, bool SUPPORTS_LAZY_LOADING = false>
 class SegmentTree {
+private:
+	class SegmentIterationHelper;
+
 public:
+	explicit SegmentTree() : finished_loading(true) {
+	}
+	virtual ~SegmentTree() {
+	}
+
 	//! Locks the segment tree. All methods to the segment tree either lock the segment tree, or take an already
 	//! obtained lock.
-	SegmentLock Lock();
+	SegmentLock Lock() {
+		return SegmentLock(node_lock);
+	}
 
-	bool IsEmpty(SegmentLock &);
+	bool IsEmpty(SegmentLock &l) {
+		return GetRootSegment(l) == nullptr;
+	}
 
 	//! Gets a pointer to the first segment. Useful for scans.
-	SegmentBase *GetRootSegment();
-	SegmentBase *GetRootSegment(SegmentLock &);
+	T *GetRootSegment() {
+		auto l = Lock();
+		return GetRootSegment(l);
+	}
+
+	T *GetRootSegment(SegmentLock &l) {
+		if (nodes.empty()) {
+			LoadNextSegment(l);
+		}
+		return GetRootSegmentInternal();
+	}
 	//! Obtains ownership of the data of the segment tree
-	vector<SegmentNode> MoveSegments(SegmentLock &);
+	vector<SegmentNode<T>> MoveSegments(SegmentLock &l) {
+		LoadAllSegments(l);
+		return std::move(nodes);
+	}
 	//! Gets a pointer to the nth segment. Negative numbers start from the back.
-	SegmentBase *GetSegmentByIndex(int64_t index);
-	SegmentBase *GetSegmentByIndex(SegmentLock &, int64_t index);
+	T *GetSegmentByIndex(int64_t index) {
+		auto l = Lock();
+		return GetSegmentByIndex(l, index);
+	}
+	T *GetSegmentByIndex(SegmentLock &l, int64_t index) {
+		if (index < 0) {
+			// load all segments
+			LoadAllSegments(l);
+			index = nodes.size() + index;
+			if (index < 0) {
+				return nullptr;
+			}
+			return nodes[index].node.get();
+		} else {
+			// lazily load segments until we reach the specific segment
+			while (idx_t(index) >= nodes.size() && LoadNextSegment(l)) {
+			}
+			if (idx_t(index) >= nodes.size()) {
+				return nullptr;
+			}
+			return nodes[index].node.get();
+		}
+	}
+	//! Gets the next segment
+	T *GetNextSegment(T *segment) {
+		if (!SUPPORTS_LAZY_LOADING) {
+			return segment->Next();
+		}
+		if (finished_loading) {
+			return segment->Next();
+		}
+		auto l = Lock();
+		return GetNextSegment(l, segment);
+	}
+	T *GetNextSegment(SegmentLock &l, T *segment) {
+		if (!segment) {
+			return nullptr;
+		}
+#ifdef DEBUG
+		D_ASSERT(nodes[segment->index].node.get() == segment);
+#endif
+		return GetSegmentByIndex(l, segment->index + 1);
+	}
 
 	//! Gets a pointer to the last segment. Useful for appends.
-	SegmentBase *GetLastSegment();
-	SegmentBase *GetLastSegment(SegmentLock &);
+	T *GetLastSegment(SegmentLock &l) {
+		LoadAllSegments(l);
+		if (nodes.empty()) {
+			return nullptr;
+		}
+		return nodes.back().node.get();
+	}
 	//! Gets a pointer to a specific column segment for the given row
-	SegmentBase *GetSegment(idx_t row_number);
-	SegmentBase *GetSegment(SegmentLock &, idx_t row_number);
+	T *GetSegment(idx_t row_number) {
+		auto l = Lock();
+		return GetSegment(l, row_number);
+	}
+	T *GetSegment(SegmentLock &l, idx_t row_number) {
+		return nodes[GetSegmentIndex(l, row_number)].node.get();
+	}
 
 	//! Append a column segment to the tree
-	void AppendSegment(unique_ptr<SegmentBase> segment);
-	void AppendSegment(SegmentLock &, unique_ptr<SegmentBase> segment);
+	void AppendSegmentInternal(SegmentLock &l, unique_ptr<T> segment) {
+		D_ASSERT(segment);
+		// add the node to the list of nodes
+		if (!nodes.empty()) {
+			nodes.back().node->next = segment.get();
+		}
+		SegmentNode<T> node;
+		segment->index = nodes.size();
+		node.row_start = segment->start;
+		node.node = std::move(segment);
+		nodes.push_back(std::move(node));
+	}
+	void AppendSegment(unique_ptr<T> segment) {
+		auto l = Lock();
+		AppendSegment(l, std::move(segment));
+	}
+	void AppendSegment(SegmentLock &l, unique_ptr<T> segment) {
+		LoadAllSegments(l);
+		AppendSegmentInternal(l, std::move(segment));
+	}
 	//! Debug method, check whether the segment is in the segment tree
-	bool HasSegment(SegmentBase *segment);
-	bool HasSegment(SegmentLock &, SegmentBase *segment);
+	bool HasSegment(T *segment) {
+		auto l = Lock();
+		return HasSegment(l, segment);
+	}
+	bool HasSegment(SegmentLock &, T *segment) {
+		return segment->index < nodes.size() && nodes[segment->index].node.get() == segment;
+	}
 
 	//! Replace this tree with another tree, taking over its nodes in-place
-	void Replace(SegmentTree &other);
-	void Replace(SegmentLock &, SegmentTree &other);
+	void Replace(SegmentTree<T> &other) {
+		auto l = Lock();
+		Replace(l, other);
+	}
+	void Replace(SegmentLock &l, SegmentTree<T> &other) {
+		other.LoadAllSegments(l);
+		nodes = std::move(other.nodes);
+	}
 
 	//! Erase all segments after a specific segment
-	void EraseSegments(SegmentLock &, idx_t segment_start);
+	void EraseSegments(SegmentLock &l, idx_t segment_start) {
+		LoadAllSegments(l);
+		if (segment_start >= nodes.size() - 1) {
+			return;
+		}
+		nodes.erase(nodes.begin() + segment_start + 1, nodes.end());
+	}
 
 	//! Get the segment index of the column segment for the given row
-	idx_t GetSegmentIndex(idx_t row_number);
-	idx_t GetSegmentIndex(SegmentLock &, idx_t row_number);
-	bool TryGetSegmentIndex(SegmentLock &, idx_t row_number, idx_t &);
+	idx_t GetSegmentIndex(SegmentLock &l, idx_t row_number) {
+		idx_t segment_index;
+		if (TryGetSegmentIndex(l, row_number, segment_index)) {
+			return segment_index;
+		}
+		string error;
+		error = StringUtil::Format("Attempting to find row number \"%lld\" in %lld nodes\n", row_number, nodes.size());
+		for (idx_t i = 0; i < nodes.size(); i++) {
+			error += StringUtil::Format("Node %lld: Start %lld, Count %lld", i, nodes[i].row_start,
+			                            nodes[i].node->count.load());
+		}
+		throw InternalException("Could not find node in column segment tree!\n%s%s", error, Exception::GetStackTrace());
+	}
+
+	bool TryGetSegmentIndex(SegmentLock &l, idx_t row_number, idx_t &result) {
+		// load segments until the row number is within bounds
+		while (nodes.empty() || (row_number >= (nodes.back().row_start + nodes.back().node->count))) {
+			if (!LoadNextSegment(l)) {
+				break;
+			}
+		}
+		if (nodes.empty()) {
+			return false;
+		}
+		D_ASSERT(!nodes.empty());
+		D_ASSERT(row_number >= nodes[0].row_start);
+		D_ASSERT(row_number < nodes.back().row_start + nodes.back().node->count);
+		idx_t lower = 0;
+		idx_t upper = nodes.size() - 1;
+		// binary search to find the node
+		while (lower <= upper) {
+			idx_t index = (lower + upper) / 2;
+			D_ASSERT(index < nodes.size());
+			auto &entry = nodes[index];
+			D_ASSERT(entry.row_start == entry.node->start);
+			if (row_number < entry.row_start) {
+				upper = index - 1;
+			} else if (row_number >= entry.row_start + entry.node->count) {
+				lower = index + 1;
+			} else {
+				result = index;
+				return true;
+			}
+		}
+		return false;
+	}
 
-	void Verify(SegmentLock &);
-	void Verify();
+	void Verify(SegmentLock &) {
+#ifdef DEBUG
+		idx_t base_start = nodes.empty() ? 0 : nodes[0].node->start;
+		for (idx_t i = 0; i < nodes.size(); i++) {
+			D_ASSERT(nodes[i].row_start == nodes[i].node->start);
+			D_ASSERT(nodes[i].node->start == base_start);
+			base_start += nodes[i].node->count;
+		}
+#endif
+	}
+	void Verify() {
+#ifdef DEBUG
+		auto l = Lock();
+		Verify(l);
+#endif
+	}
+
+	SegmentIterationHelper Segments() {
+		return SegmentIterationHelper(*this);
+	}
+
+protected:
+	atomic<bool> finished_loading;
+
+	//! Load the next segment - only used when lazily loading
+	virtual unique_ptr<T> LoadSegment() {
+		return nullptr;
+	}
 
 private:
 	//! The nodes in the tree, can be binary searched
-	vector<SegmentNode> nodes;
+	vector<SegmentNode<T>> nodes;
 	//! Lock to access or modify the nodes
 	mutex node_lock;
+
+private:
+	T *GetRootSegmentInternal() {
+		return nodes.empty() ? nullptr : nodes[0].node.get();
+	}
+
+	class SegmentIterationHelper {
+	public:
+		explicit SegmentIterationHelper(SegmentTree &tree) : tree(tree) {
+		}
+
+	private:
+		SegmentTree &tree;
+
+	private:
+		class SegmentIterator {
+		public:
+			SegmentIterator(SegmentTree &tree_p, T *current_p) : tree(tree_p), current(current_p) {
+			}
+
+			SegmentTree &tree;
+			T *current;
+
+		public:
+			void Next() {
+				current = tree.GetNextSegment(current);
+			}
+
+			SegmentIterator &operator++() {
+				Next();
+				return *this;
+			}
+			bool operator!=(const SegmentIterator &other) const {
+				return current != other.current;
+			}
+			T &operator*() const {
+				D_ASSERT(current);
+				return *current;
+			}
+		};
+
+	public:
+		SegmentIterator begin() {
+			return SegmentIterator(tree, tree.GetRootSegment());
+		}
+		SegmentIterator end() {
+			return SegmentIterator(tree, nullptr);
+		}
+	};
+
+	//! Load the next segment, if there are any left to load
+	bool LoadNextSegment(SegmentLock &l) {
+		if (!SUPPORTS_LAZY_LOADING) {
+			return false;
+		}
+		if (finished_loading) {
+			return false;
+		}
+		auto result = LoadSegment();
+		if (result) {
+			AppendSegmentInternal(l, std::move(result));
+			return true;
+		}
+		return false;
+	}
+
+	//! Load all segments, if there are any left to load
+	void LoadAllSegments(SegmentLock &l) {
+		if (!SUPPORTS_LAZY_LOADING) {
+			return;
+		}
+		while (LoadNextSegment(l))
+			;
+	}
 };
 
 } // namespace duckdb

package/src/duckdb/src/include/duckdb/storage/table/standard_column_data.hpp

@@ -59,10 +59,6 @@ public:
 	void GetStorageInfo(idx_t row_group_index, vector<idx_t> col_path, TableStorageInfo &result) override;
 
 	void Verify(RowGroup &parent) override;
-
-private:
-	template <bool SCAN_COMMITTED, bool ALLOW_UPDATES>
-	void TemplatedScan(Transaction *transaction, ColumnScanState &state, Vector &result);
 };
 
 } // namespace duckdb

package/src/duckdb/src/include/duckdb/storage/table/table_statistics.hpp

@@ -14,6 +14,7 @@
 #include "duckdb/storage/statistics/column_statistics.hpp"
 
 namespace duckdb {
+class ColumnList;
 class PersistentTableData;
 
 class TableStatisticsLock {
@@ -38,6 +39,7 @@ public:
 	void MergeStats(idx_t i, BaseStatistics &stats);
 	void MergeStats(TableStatisticsLock &lock, idx_t i, BaseStatistics &stats);
 
+	void CopyStats(TableStatistics &other);
 	unique_ptr<BaseStatistics> CopyStats(idx_t i);
 	ColumnStatistics &GetStats(idx_t i);
 
@@ -45,6 +47,9 @@ public:
 
 	unique_ptr<TableStatisticsLock> GetLock();
 
+	void Serialize(Serializer &serializer);
+	void Deserialize(Deserializer &source, ColumnList &columns);
+
 private:
 	//! The statistics lock
 	mutex stats_lock;

package/src/duckdb/src/include/duckdb/storage/table/update_segment.hpp

@@ -34,7 +34,6 @@ public:
 	bool HasUncommittedUpdates(idx_t vector_index);
 	bool HasUpdates(idx_t vector_index) const;
 	bool HasUpdates(idx_t start_row_idx, idx_t end_row_idx);
-	void ClearUpdates();
 
 	void FetchUpdates(TransactionData transaction, idx_t vector_index, Vector &result);
 	void FetchCommitted(idx_t vector_index, Vector &result);

package/src/duckdb/src/include/duckdb/storage/write_ahead_log.hpp

@@ -137,7 +137,7 @@ public:
 	//! Sets the table used for subsequent insert/delete/update commands
 	void WriteSetTable(string &schema, string &table);
 
-	void WriteAlter(AlterInfo &info);
+	void WriteAlter(data_ptr_t ptr, idx_t data_size);
 
 	void WriteInsert(DataChunk &chunk);
 	void WriteDelete(DataChunk &chunk);

package/src/duckdb/src/include/duckdb/transaction/local_storage.hpp

@@ -9,7 +9,6 @@
 #pragma once
 
 #include "duckdb/storage/table/row_group_collection.hpp"
-#include "duckdb/storage/table/scan_state.hpp"
 #include "duckdb/storage/table/table_index_list.hpp"
 #include "duckdb/storage/table/table_statistics.hpp"
 
@@ -18,6 +17,7 @@ class AttachedDatabase;
 class DataTable;
 class Transaction;
 class WriteAheadLog;
+struct LocalAppendState;
 struct TableAppendState;
 
 class OptimisticDataWriter {
@@ -88,8 +88,8 @@ public:
 
 	void AppendToIndexes(DuckTransaction &transaction, TableAppendState &append_state, idx_t append_count,
 	                     bool append_to_table);
-	bool AppendToIndexes(DuckTransaction &transaction, RowGroupCollection &source, TableIndexList &index_list,
-	                     const vector<LogicalType> &table_types, row_t &start_row);
+	PreservedError AppendToIndexes(DuckTransaction &transaction, RowGroupCollection &source, TableIndexList &index_list,
+	                               const vector<LogicalType> &table_types, row_t &start_row);
 
 	//! Creates an optimistic writer for this table
 	OptimisticDataWriter *CreateOptimisticWriter();
@@ -118,6 +118,9 @@ public:
 
 public:
 	struct CommitState {
+		CommitState();
+		~CommitState();
+
 		unordered_map<DataTable *, unique_ptr<TableAppendState>> append_states;
 	};