feat(online): support last join (window) (#3565)

Author: aceforeverd

cases/query/last_join_subquery_window.yml

@@ -136,22 +136,6 @@ RowIterator* TabletTableHandler::GetRawIterator() {
     return new storage::FullTableIterator(table_->GetSegments(),
                                           table_->GetSegCnt(), table_);
 }
-const uint64_t TabletTableHandler::GetCount() {
-    auto iter = GetIterator();
-    uint64_t cnt = 0;
-    while (iter->Valid()) {
-        iter->Next();
-        cnt++;
-    }
-    return cnt;
-}
-Row TabletTableHandler::At(uint64_t pos) {
-    auto iter = GetIterator();
-    while (pos-- > 0 && iter->Valid()) {
-        iter->Next();
-    }
-    return iter->Valid() ? iter->GetValue() : Row();
-}
 
 TabletCatalog::TabletCatalog() : tables_(), db_() {}
 
@@ -249,22 +233,6 @@ std::unique_ptr<WindowIterator> TabletSegmentHandler::GetWindowIterator(
     const std::string& idx_name) {
     return std::unique_ptr<WindowIterator>();
 }
-const uint64_t TabletSegmentHandler::GetCount() {
-    auto iter = GetIterator();
-    uint64_t cnt = 0;
-    while (iter->Valid()) {
-        cnt++;
-        iter->Next();
-    }
-    return cnt;
-}
-Row TabletSegmentHandler::At(uint64_t pos) {
-    auto iter = GetIterator();
-    while (pos-- > 0 && iter->Valid()) {
-        iter->Next();
-    }
-    return iter->Valid() ? iter->GetValue() : Row();
-}
 
 const uint64_t TabletPartitionHandler::GetCount() {
     auto iter = GetWindowIterator();
@@ -275,5 +243,6 @@ const uint64_t TabletPartitionHandler::GetCount() {
     }
     return cnt;
 }
+
 }  // namespace tablet
 }  // namespace hybridse

hybridse/examples/toydb/src/tablet/tablet_catalog.cc

@@ -68,8 +68,6 @@ class TabletSegmentHandler : public TableHandler {
     std::unique_ptr<vm::RowIterator> GetIterator() override;
     RowIterator* GetRawIterator() override;
     std::unique_ptr<codec::WindowIterator> GetWindowIterator(const std::string& idx_name) override;
-    const uint64_t GetCount() override;
-    Row At(uint64_t pos) override;
     const std::string GetHandlerTypeName() override {
         return "TabletSegmentHandler";
     }
@@ -104,6 +102,7 @@ class TabletPartitionHandler
     std::unique_ptr<codec::WindowIterator> GetWindowIterator() override {
         return table_handler_->GetWindowIterator(index_name_);
     }
+
     const uint64_t GetCount() override;
 
     std::shared_ptr<TableHandler> GetSegment(const std::string& key) override {
@@ -152,8 +151,6 @@ class TabletTableHandler
     RowIterator* GetRawIterator() override;
     std::unique_ptr<codec::WindowIterator> GetWindowIterator(
         const std::string& idx_name);
-    virtual const uint64_t GetCount();
-    Row At(uint64_t pos) override;
 
     virtual std::shared_ptr<PartitionHandler> GetPartition(
         const std::string& index_name) {

hybridse/examples/toydb/src/tablet/tablet_catalog.h

@@ -15,8 +15,9 @@
  */
 
 #include "testing/toydb_engine_test_base.h"
+
+#include "absl/strings/str_join.h"
 #include "gtest/gtest.h"
-#include "gtest/internal/gtest-param-util.h"
 
 using namespace llvm;       // NOLINT (build/namespaces)
 using namespace llvm::orc;  // NOLINT (build/namespaces)
@@ -141,18 +142,12 @@ std::shared_ptr<tablet::TabletCatalog> BuildOnePkTableStorage(
     }
     return catalog;
 }
-void BatchRequestEngineCheckWithCommonColumnIndices(
-    const SqlCase& sql_case, const EngineOptions options,
-    const std::set<size_t>& common_column_indices) {
-    std::ostringstream oss;
-    for (size_t index : common_column_indices) {
-        oss << index << ",";
-    }
-    LOG(INFO) << "BatchRequestEngineCheckWithCommonColumnIndices: "
-                 "common_column_indices = ["
-              << oss.str() << "]";
-    ToydbBatchRequestEngineTestRunner engine_test(sql_case, options,
-                                                  common_column_indices);
+// Run check with common column index info
+void BatchRequestEngineCheckWithCommonColumnIndices(const SqlCase& sql_case, const EngineOptions options,
+                                                    const std::set<size_t>& common_column_indices) {
+    LOG(INFO) << "BatchRequestEngineCheckWithCommonColumnIndices: common_column_indices = ["
+              << absl::StrJoin(common_column_indices, ",") << "]";
+    ToydbBatchRequestEngineTestRunner engine_test(sql_case, options, common_column_indices);
     engine_test.RunCheck();
 }
 

hybridse/examples/toydb/src/testing/toydb_engine_test_base.cc

@@ -54,7 +54,7 @@ class Row {
 
     inline int32_t size() const { return slice_.size(); }
     inline int32_t size(int32_t pos) const {
-        return 0 == pos ? slice_.size() : slices_[pos - 1].size();
+        return 0 == pos ? slice_.size() : slices_.at(pos - 1).size();
     }
 
     // Return true if the length of the referenced data is zero

hybridse/include/codec/row.h

@@ -71,7 +71,14 @@ class WindowIterator {
     virtual bool Valid() = 0;
     /// Return the RowIterator of current segment
     /// of dataset if Valid() return `true`.
-    virtual std::unique_ptr<RowIterator> GetValue() = 0;
+    virtual std::unique_ptr<RowIterator> GetValue() {
+        auto p = GetRawValue();
+        if (!p) {
+            return nullptr;
+        }
+
+        return std::unique_ptr<RowIterator>(p);
+    }
     /// Return the RowIterator of current segment
     /// of dataset if Valid() return `true`.
     virtual RowIterator *GetRawValue() = 0;

hybridse/include/codec/row_iterator.h

@@ -76,7 +76,7 @@ class ListV {
     virtual const uint64_t GetCount() {
         auto iter = GetIterator();
         uint64_t cnt = 0;
-        while (iter->Valid()) {
+        while (iter && iter->Valid()) {
             iter->Next();
             cnt++;
         }

hybridse/include/codec/row_list.h

@@ -217,6 +217,7 @@ class TableHandler : public DataHandler {
     virtual ~TableHandler() {}
 
     /// Return table column Types information.
+    /// TODO: rm it, never used
     virtual const Types& GetTypes() = 0;
 
     /// Return the index information

hybridse/include/vm/catalog.h

@@ -25,8 +25,6 @@
 #include <string>
 #include <utility>
 #include <vector>
-#include "base/fe_slice.h"
-#include "codec/list_iterator_codec.h"
 #include "glog/logging.h"
 #include "vm/catalog.h"
 
@@ -674,6 +672,7 @@ class MemPartitionHandler
     IndexHint index_hint_;
     OrderType order_type_;
 };
+
 class ConcatTableHandler : public MemTimeTableHandler {
  public:
     ConcatTableHandler(std::shared_ptr<TableHandler> left, size_t left_slices,
@@ -692,19 +691,19 @@ class ConcatTableHandler : public MemTimeTableHandler {
         status_ = SyncValue();
         return MemTimeTableHandler::At(pos);
     }
-    std::unique_ptr<RowIterator> GetIterator() {
+    std::unique_ptr<RowIterator> GetIterator() override {
         if (status_.isRunning()) {
             status_ = SyncValue();
         }
         return MemTimeTableHandler::GetIterator();
     }
-    RowIterator* GetRawIterator() {
+    RowIterator* GetRawIterator() override {
         if (status_.isRunning()) {
             status_ = SyncValue();
         }
         return MemTimeTableHandler::GetRawIterator();
     }
-    virtual const uint64_t GetCount() {
+    const uint64_t GetCount() override {
         if (status_.isRunning()) {
             status_ = SyncValue();
         }

hybridse/include/vm/mem_catalog.h

@@ -785,7 +785,11 @@ class PhysicalAggregationNode : public PhysicalProjectNode {
  public:
     PhysicalAggregationNode(PhysicalOpNode *node, const ColumnProjects &project, const node::ExprNode *condition)
         : PhysicalProjectNode(node, kAggregation, project, true), having_condition_(condition) {
-        output_type_ = kSchemaTypeRow;
+        if (node->GetOutputType() == kSchemaTypeGroup) {
+            output_type_ = kSchemaTypeGroup;
+        } else {
+            output_type_ = kSchemaTypeRow;
+        }
         fn_infos_.push_back(&having_condition_.fn_info());
     }
     virtual ~PhysicalAggregationNode() {}
@@ -1065,7 +1069,7 @@ class RequestWindowUnionList {
     RequestWindowUnionList() : window_unions_() {}
     virtual ~RequestWindowUnionList() {}
     void AddWindowUnion(PhysicalOpNode *node, const RequestWindowOp &window) {
-        window_unions_.push_back(std::make_pair(node, window));
+        window_unions_.emplace_back(node, window);
     }
     const PhysicalOpNode *GetKey(uint32_t index) {
         auto iter = window_unions_.begin();
@@ -1415,7 +1419,7 @@ class PhysicalRequestUnionNode : public PhysicalBinaryNode {
           instance_not_in_window_(false),
           exclude_current_time_(false),
           output_request_row_(true) {
-        output_type_ = kSchemaTypeTable;
+        InitOuptput();
 
         fn_infos_.push_back(&window_.partition_.fn_info());
         fn_infos_.push_back(&window_.index_key_.fn_info());
@@ -1427,7 +1431,7 @@ class PhysicalRequestUnionNode : public PhysicalBinaryNode {
           instance_not_in_window_(w_ptr->instance_not_in_window()),
           exclude_current_time_(w_ptr->exclude_current_time()),
           output_request_row_(true) {
-        output_type_ = kSchemaTypeTable;
+        InitOuptput();
 
         fn_infos_.push_back(&window_.partition_.fn_info());
         fn_infos_.push_back(&window_.sort_.fn_info());
@@ -1443,7 +1447,7 @@ class PhysicalRequestUnionNode : public PhysicalBinaryNode {
           instance_not_in_window_(instance_not_in_window),
           exclude_current_time_(exclude_current_time),
           output_request_row_(output_request_row) {
-        output_type_ = kSchemaTypeTable;
+        InitOuptput();
 
         fn_infos_.push_back(&window_.partition_.fn_info());
         fn_infos_.push_back(&window_.sort_.fn_info());
@@ -1455,7 +1459,8 @@ class PhysicalRequestUnionNode : public PhysicalBinaryNode {
     virtual void Print(std::ostream &output, const std::string &tab) const;
     const bool Valid() { return true; }
     static PhysicalRequestUnionNode *CastFrom(PhysicalOpNode *node);
-    bool AddWindowUnion(PhysicalOpNode *node) {
+    bool AddWindowUnion(PhysicalOpNode *node) { return AddWindowUnion(node, window_); }
+    bool AddWindowUnion(PhysicalOpNode *node, const RequestWindowOp& window) {
         if (nullptr == node) {
             LOG(WARNING) << "Fail to add window union : table is null";
             return false;
@@ -1472,9 +1477,8 @@ class PhysicalRequestUnionNode : public PhysicalBinaryNode {
                 << "Union Table and window input schema aren't consistent";
             return false;
         }
-        window_unions_.AddWindowUnion(node, window_);
-        RequestWindowOp &window_union =
-            window_unions_.window_unions_.back().second;
+        window_unions_.AddWindowUnion(node, window);
+        RequestWindowOp &window_union = window_unions_.window_unions_.back().second;
         fn_infos_.push_back(&window_union.partition_.fn_info());
         fn_infos_.push_back(&window_union.sort_.fn_info());
         fn_infos_.push_back(&window_union.range_.fn_info());
@@ -1484,11 +1488,10 @@ class PhysicalRequestUnionNode : public PhysicalBinaryNode {
 
     std::vector<PhysicalOpNode *> GetDependents() const override;
 
-    const bool instance_not_in_window() const {
-        return instance_not_in_window_;
-    }
-    const bool exclude_current_time() const { return exclude_current_time_; }
-    const bool output_request_row() const { return output_request_row_; }
+    bool instance_not_in_window() const { return instance_not_in_window_; }
+    bool exclude_current_time() const { return exclude_current_time_; }
+    bool output_request_row() const { return output_request_row_; }
+    void set_output_request_row(bool flag) { output_request_row_ = flag; }
     const RequestWindowOp &window() const { return window_; }
     const RequestWindowUnionList &window_unions() const {
         return window_unions_;
@@ -1506,10 +1509,20 @@ class PhysicalRequestUnionNode : public PhysicalBinaryNode {
     }
 
     RequestWindowOp window_;
-    const bool instance_not_in_window_;
-    const bool exclude_current_time_;
-    const bool output_request_row_;
+    bool instance_not_in_window_;
+    bool exclude_current_time_;
+    bool output_request_row_;
     RequestWindowUnionList window_unions_;
+
+ private:
+    void InitOuptput() {
+        auto left = GetProducer(0);
+        if (left->GetOutputType() == kSchemaTypeRow) {
+            output_type_ = kSchemaTypeTable;
+        } else {
+            output_type_ = kSchemaTypeGroup;
+        }
+    }
 };
 
 class PhysicalRequestAggUnionNode : public PhysicalOpNode {