be/src/exprs/virtual_slot_ref.cpp
Line | Count | Source |
1 | | // Licensed to the Apache Software Foundation (ASF) under one |
2 | | // or more contributor license agreements. See the NOTICE file |
3 | | // distributed with this work for additional information |
4 | | // regarding copyright ownership. The ASF licenses this file |
5 | | // to you under the Apache License, Version 2.0 (the |
6 | | // "License"); you may not use this file except in compliance |
7 | | // with the License. You may obtain a copy of the License at |
8 | | // |
9 | | // http://www.apache.org/licenses/LICENSE-2.0 |
10 | | // |
11 | | // Unless required by applicable law or agreed to in writing, |
12 | | // software distributed under the License is distributed on an |
13 | | // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
14 | | // KIND, either express or implied. See the License for the |
15 | | // specific language governing permissions and limitations |
16 | | // under the License. |
17 | | |
18 | | #include "exprs/virtual_slot_ref.h" |
19 | | |
20 | | #include <gen_cpp/Exprs_types.h> |
21 | | #include <glog/logging.h> |
22 | | #include <thrift/protocol/TDebugProtocol.h> |
23 | | |
24 | | #include <ostream> |
25 | | #include <vector> |
26 | | |
27 | | #include "common/exception.h" |
28 | | #include "common/logging.h" |
29 | | #include "common/status.h" |
30 | | #include "core/block/block.h" |
31 | | #include "core/block/column_with_type_and_name.h" |
32 | | #include "core/column/column.h" |
33 | | #include "core/column/column_nothing.h" |
34 | | #include "exprs/vectorized_fn_call.h" |
35 | | #include "exprs/vexpr_context.h" |
36 | | #include "exprs/vexpr_fwd.h" |
37 | | #include "runtime/descriptors.h" |
38 | | #include "runtime/runtime_state.h" |
39 | | namespace doris { |
40 | | #include "common/compile_check_begin.h" |
41 | | VirtualSlotRef::VirtualSlotRef(const doris::TExprNode& node) |
42 | 279 | : VExpr(node), |
43 | 279 | _column_id(-1), |
44 | 279 | _slot_id(node.slot_ref.slot_id), |
45 | 279 | _column_name(nullptr), |
46 | 279 | _column_label(node.label) {} |
47 | | |
48 | | VirtualSlotRef::VirtualSlotRef(const SlotDescriptor* desc) |
49 | 6 | : VExpr(desc->type(), false), _column_id(-1), _slot_id(desc->id()), _column_name(nullptr) {} |
50 | | |
51 | | Status VirtualSlotRef::prepare(doris::RuntimeState* state, const doris::RowDescriptor& desc, |
52 | 246 | VExprContext* context) { |
53 | 246 | RETURN_IF_ERROR_OR_PREPARED(VExpr::prepare(state, desc, context)); |
54 | 246 | DCHECK_EQ(_children.size(), 0); |
55 | 246 | if (_slot_id == -1) { |
56 | 3 | _prepare_finished = true; |
57 | 3 | return Status::OK(); |
58 | 3 | } |
59 | | |
60 | 243 | const SlotDescriptor* slot_desc = state->desc_tbl().get_slot_descriptor(_slot_id); |
61 | 243 | if (slot_desc == nullptr) { |
62 | 0 | return Status::Error<ErrorCode::INTERNAL_ERROR>( |
63 | 0 | "couldn't resolve slot descriptor {}, desc: {}", _slot_id, |
64 | 0 | state->desc_tbl().debug_string()); |
65 | 0 | } |
66 | | |
67 | 243 | if (slot_desc->get_virtual_column_expr() == nullptr) { |
68 | 0 | return Status::InternalError( |
69 | 0 | "VirtualSlotRef {} has no virtual column expr, slot_id: {}, desc: {}, " |
70 | 0 | "slot_desc: {}, desc_tbl: {}", |
71 | 0 | *_column_name, _slot_id, desc.debug_string(), slot_desc->debug_string(), |
72 | 0 | state->desc_tbl().debug_string()); |
73 | 0 | } |
74 | | |
75 | 243 | _column_name = &slot_desc->col_name(); |
76 | 243 | _column_data_type = slot_desc->get_data_type_ptr(); |
77 | 243 | DCHECK(_column_data_type != nullptr); |
78 | 243 | _column_id = desc.get_column_id(_slot_id); |
79 | 243 | if (_column_id < 0) { |
80 | 0 | return Status::Error<ErrorCode::INTERNAL_ERROR>( |
81 | 0 | "VirtualSlotRef {} has invalid slot id: " |
82 | 0 | "{}.\nslot_desc:\n{},\ndesc:\n{},\ndesc_tbl:\n{}", |
83 | 0 | *_column_name, _slot_id, slot_desc->debug_string(), desc.debug_string(), |
84 | 0 | state->desc_tbl().debug_string()); |
85 | 0 | } |
86 | 243 | const TExpr& expr = *slot_desc->get_virtual_column_expr(); |
87 | | // Create a temp_ctx only for create_expr_tree. |
88 | 243 | VExprContextSPtr temp_ctx; |
89 | 243 | RETURN_IF_ERROR(VExpr::create_expr_tree(expr, temp_ctx)); |
90 | 243 | _virtual_column_expr = temp_ctx->root(); |
91 | | // Virtual column expr should do prepare with original context. |
92 | 243 | RETURN_IF_ERROR(_virtual_column_expr->prepare(state, desc, context)); |
93 | 243 | _prepare_finished = true; |
94 | 243 | return Status::OK(); |
95 | 243 | } |
96 | | |
97 | | Status VirtualSlotRef::open(RuntimeState* state, VExprContext* context, |
98 | 1.11k | FunctionContext::FunctionStateScope scope) { |
99 | 1.11k | DCHECK(_prepare_finished); |
100 | 1.11k | RETURN_IF_ERROR(_virtual_column_expr->open(state, context, scope)); |
101 | 1.11k | RETURN_IF_ERROR(VExpr::open(state, context, scope)); |
102 | 1.11k | _open_finished = true; |
103 | 1.11k | return Status::OK(); |
104 | 1.11k | } |
105 | | |
106 | | Status VirtualSlotRef::execute_column(VExprContext* context, const Block* block, Selector* selector, |
107 | 319 | size_t count, ColumnPtr& result_column) const { |
108 | 320 | if (_column_id >= 0 && _column_id >= block->columns()) { |
109 | 0 | return Status::Error<ErrorCode::INTERNAL_ERROR>( |
110 | 0 | "input block not contain slot column {}, column_id={}, block={}", *_column_name, |
111 | 0 | _column_id, block->dump_structure()); |
112 | 0 | } |
113 | | |
114 | 319 | ColumnWithTypeAndName col_type_name = block->get_by_position(_column_id); |
115 | | |
116 | 319 | if (!col_type_name.column) { |
117 | | // Maybe we need to create a column in this situation. |
118 | 0 | return Status::InternalError( |
119 | 0 | "VirtualSlotRef column is null, column_id: {}, column_name: {}", _column_id, |
120 | 0 | *_column_name); |
121 | 0 | } |
122 | | |
123 | 319 | const auto* col_nothing = check_and_get_column<ColumnNothing>(col_type_name.column.get()); |
124 | 319 | bool column_from_virtual_column_expr = false; |
125 | 320 | if (this->_virtual_column_expr != nullptr) { |
126 | 320 | if (col_nothing != nullptr) { |
127 | | // Virtual column is not materialized, so we need to materialize it. |
128 | | // Note: After executing 'execute', we cannot use the column from line 120 in subsequent code, |
129 | | // because the vector might be resized during execution, causing previous references to become invalid. |
130 | 4 | ColumnPtr tmp_column; |
131 | 4 | RETURN_IF_ERROR(_virtual_column_expr->execute_column(context, block, selector, count, |
132 | 4 | tmp_column)); |
133 | 4 | result_column = std::move(tmp_column); |
134 | 4 | column_from_virtual_column_expr = true; |
135 | | |
136 | 4 | VLOG_DEBUG << fmt::format( |
137 | 0 | "Materialization of virtual column, slot_id {}, column_id {}, " |
138 | 0 | "column_name {}, column size {}", |
139 | 0 | _slot_id, _column_id, *_column_name, |
140 | 0 | block->get_by_position(_column_id).column->size()); |
141 | 4 | } |
142 | | |
143 | 320 | #ifndef NDEBUG |
144 | | // get_by_position again since vector in block may be resized |
145 | 320 | col_type_name = block->get_by_position(_column_id); |
146 | 320 | DCHECK(col_type_name.type != nullptr); |
147 | 320 | if (!_column_data_type->equals(*col_type_name.type)) { |
148 | 0 | throw doris::Exception(doris::ErrorCode::FATAL_ERROR, |
149 | 0 | "Virtual column type not match, column_id: {}, " |
150 | 0 | "column_name: {}, column_type: {}, virtual_column_type: {}", |
151 | 0 | _column_id, *_column_name, col_type_name.type->get_name(), |
152 | 0 | _column_data_type->get_name()); |
153 | 0 | } |
154 | 320 | #endif |
155 | 18.4E | } else { |
156 | | // This is a virtual slot ref that not pushed to segment_iterator |
157 | 18.4E | if (col_nothing == nullptr) { |
158 | 0 | return Status::InternalError("Logical error, virtual column can not be materialized"); |
159 | 18.4E | } else { |
160 | 18.4E | return Status::OK(); |
161 | 18.4E | } |
162 | 18.4E | } |
163 | | |
164 | 320 | if (!column_from_virtual_column_expr) { |
165 | | // if the column is from virtual column expr, result_column has been filter already |
166 | 316 | result_column = filter_column_with_selector(col_type_name.column, selector, count); |
167 | 316 | } |
168 | | |
169 | 320 | DCHECK_EQ(result_column->size(), count); |
170 | 320 | return Status::OK(); |
171 | 319 | } |
172 | | |
173 | 303 | const std::string& VirtualSlotRef::expr_name() const { |
174 | 303 | return *_column_name; |
175 | 303 | } |
176 | 2 | std::string VirtualSlotRef::expr_label() { |
177 | 2 | return _column_label; |
178 | 2 | } |
179 | | |
180 | 2 | std::string VirtualSlotRef::debug_string() const { |
181 | 2 | std::stringstream out; |
182 | 2 | out << "VirtualSlotRef(slot_id=" << _slot_id << VExpr::debug_string() << ")"; |
183 | 2 | return out.str(); |
184 | 2 | } |
185 | | |
186 | 25 | bool VirtualSlotRef::equals(const VExpr& other) { |
187 | 25 | const auto* other_ptr = dynamic_cast<const VirtualSlotRef*>(&other); |
188 | 25 | if (!other_ptr) { |
189 | 3 | return false; |
190 | 3 | } |
191 | | |
192 | | // Compare slot_id and column_id |
193 | 22 | if (this->_slot_id != other_ptr->_slot_id || this->_column_id != other_ptr->_column_id) { |
194 | 7 | return false; |
195 | 7 | } |
196 | | |
197 | | // Compare column_name pointers properly |
198 | 15 | if (this->_column_name == nullptr && other_ptr->_column_name == nullptr) { |
199 | | // Both are null, they are equal |
200 | 8 | } else if (this->_column_name == nullptr || other_ptr->_column_name == nullptr) { |
201 | | // One is null, the other is not, they are not equal |
202 | 2 | return false; |
203 | 5 | } else if (*this->_column_name != *other_ptr->_column_name) { |
204 | | // Both are not null, compare the string contents |
205 | 3 | return false; |
206 | 3 | } |
207 | | |
208 | | // Compare column_label |
209 | 10 | if (this->_column_label != other_ptr->_column_label) { |
210 | 3 | return false; |
211 | 3 | } |
212 | | |
213 | 7 | return true; |
214 | 10 | } |
215 | | |
216 | | /** |
217 | | * @brief Implements ANN range search evaluation for virtual slot references. |
218 | | * |
219 | | * This method handles the case where a virtual slot reference wraps a distance |
220 | | * function call that can be optimized using ANN index range search. Instead of |
221 | | * computing distances for all rows, it delegates to the underlying virtual |
222 | | * expression to perform the optimized search. |
223 | | * |
224 | | * @param range_search_runtime Runtime parameters for the range search |
225 | | * @param cid_to_index_iterators Index iterators for each column |
226 | | * @param idx_to_cid Column ID mapping |
227 | | * @param column_iterators Data column iterators |
228 | | * @param row_bitmap Result bitmap to be updated with matching rows |
229 | | * @param ann_index_stats Performance statistics collector |
230 | | * @return Status::OK() if successful, error status otherwise |
231 | | */ |
232 | | Status VirtualSlotRef::evaluate_ann_range_search( |
233 | | const segment_v2::AnnRangeSearchRuntime& range_search_runtime, |
234 | | const std::vector<std::unique_ptr<segment_v2::IndexIterator>>& cid_to_index_iterators, |
235 | | const std::vector<ColumnId>& idx_to_cid, |
236 | | const std::vector<std::unique_ptr<segment_v2::ColumnIterator>>& column_iterators, |
237 | 0 | roaring::Roaring& row_bitmap, segment_v2::AnnIndexStats& ann_index_stats) { |
238 | 0 | return _virtual_column_expr->evaluate_ann_range_search( |
239 | 0 | range_search_runtime, cid_to_index_iterators, idx_to_cid, column_iterators, row_bitmap, |
240 | 0 | ann_index_stats); |
241 | | |
242 | 0 | return Status::OK(); |
243 | 0 | } |
244 | | #include "common/compile_check_end.h" |
245 | | } // namespace doris |