be/src/storage/predicate/accept_null_predicate.h
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 | | #pragma once |
19 | | |
20 | | #include <cstdint> |
21 | | #include <memory> |
22 | | |
23 | | #include "common/factory_creator.h" |
24 | | #include "core/column/column_dictionary.h" |
25 | | #include "storage/index/bloom_filter/bloom_filter.h" |
26 | | #include "storage/index/inverted/inverted_index_cache.h" |
27 | | #include "storage/index/inverted/inverted_index_reader.h" |
28 | | #include "storage/predicate/column_predicate.h" |
29 | | |
30 | | namespace doris { |
31 | | |
32 | | /** |
33 | | * A wrapper predicate that delegate to nested predicate |
34 | | * but pass (set/return true) for NULL value rows. |
35 | | * |
36 | | * At parent, it's used for topn runtime predicate. |
37 | | * Eg: original input indexs is '1,2,3,7,8,9' and value of index9 is null, we get nested predicate output index is '1,2,3', but we finally output '1,2,3,9' |
38 | | */ |
39 | | class AcceptNullPredicate : public ColumnPredicate { |
40 | | ENABLE_FACTORY_CREATOR(AcceptNullPredicate); |
41 | | |
42 | | public: |
43 | | AcceptNullPredicate(const std::shared_ptr<ColumnPredicate>& nested) |
44 | 7 | : ColumnPredicate(nested->column_id(), nested->col_name(), nested->primitive_type(), |
45 | 7 | nested->opposite()), |
46 | 7 | _nested {nested} {} |
47 | | AcceptNullPredicate(const AcceptNullPredicate& other, uint32_t col_id) |
48 | 1 | : ColumnPredicate(other, col_id), |
49 | 1 | _nested(assert_cast<const AcceptNullPredicate&>(other)._nested |
50 | 1 | ? assert_cast<const AcceptNullPredicate&>(other)._nested->clone( |
51 | 1 | col_id) |
52 | 1 | : nullptr) {} |
53 | | AcceptNullPredicate(const AcceptNullPredicate& other) = delete; |
54 | 8 | ~AcceptNullPredicate() override = default; |
55 | 1 | std::shared_ptr<ColumnPredicate> clone(uint32_t col_id) const override { |
56 | 1 | return AcceptNullPredicate::create_shared(*this, col_id); |
57 | 1 | } |
58 | 1 | std::string debug_string() const override { |
59 | 1 | auto n = _nested; |
60 | 1 | fmt::memory_buffer debug_string_buffer; |
61 | 1 | fmt::format_to(debug_string_buffer, "AcceptNullPredicate({}, nested={})", |
62 | 1 | ColumnPredicate::debug_string(), n ? n->debug_string() : "null"); |
63 | 1 | return fmt::to_string(debug_string_buffer); |
64 | 1 | } |
65 | | |
66 | 2 | PredicateType type() const override { return _nested->type(); } |
67 | | |
68 | | Status evaluate(const IndexFieldNameAndTypePair& name_with_type, IndexIterator* iterator, |
69 | 4 | uint32_t num_rows, roaring::Roaring* bitmap) const override { |
70 | 4 | roaring::Roaring null_rows_in_bitmap; |
71 | 4 | if (iterator != nullptr) { |
72 | 3 | bool has_null = DORIS_TRY(iterator->has_null()); |
73 | 3 | if (has_null) { |
74 | 2 | InvertedIndexQueryCacheHandle null_bitmap_cache_handle; |
75 | 2 | RETURN_IF_ERROR(iterator->read_null_bitmap(&null_bitmap_cache_handle)); |
76 | 2 | auto null_bitmap = null_bitmap_cache_handle.get_bitmap(); |
77 | 2 | if (null_bitmap) { |
78 | 2 | null_rows_in_bitmap = *bitmap & *null_bitmap; |
79 | 2 | } |
80 | 2 | } |
81 | 3 | } |
82 | 4 | RETURN_IF_ERROR(_nested->evaluate(name_with_type, iterator, num_rows, bitmap)); |
83 | 4 | *bitmap |= null_rows_in_bitmap; |
84 | 4 | return Status::OK(); |
85 | 4 | } |
86 | | |
87 | | void evaluate_and(const IColumn& column, const uint16_t* sel, uint16_t size, |
88 | 0 | bool* flags) const override { |
89 | 0 | if (column.has_null()) { |
90 | 0 | std::vector<uint8_t> original_flags(size); |
91 | 0 | memcpy(original_flags.data(), flags, size); |
92 | |
|
93 | 0 | const auto& nullable_col = assert_cast<const ColumnNullable&>(column); |
94 | 0 | _nested->evaluate_and(nullable_col.get_nested_column(), sel, size, flags); |
95 | 0 | const auto& nullmap = nullable_col.get_null_map_data(); |
96 | 0 | for (uint16_t i = 0; i < size; ++i) { |
97 | 0 | flags[i] |= (original_flags[i] && nullmap[sel[i]]); |
98 | 0 | } |
99 | 0 | } else { |
100 | 0 | _nested->evaluate_and(column, sel, size, flags); |
101 | 0 | } |
102 | 0 | } |
103 | | |
104 | | void evaluate_or(const IColumn& column, const uint16_t* sel, uint16_t size, |
105 | 0 | bool* flags) const override { |
106 | 0 | DCHECK(false) << "should not reach here"; |
107 | 0 | } |
108 | | |
109 | 0 | bool evaluate_and(const segment_v2::ZoneMap& zone_map) const override { |
110 | | // there is null in range, accept it |
111 | 0 | if (zone_map.has_null) { |
112 | 0 | return true; |
113 | 0 | } |
114 | 0 | return _nested->evaluate_and(zone_map); |
115 | 0 | } |
116 | | |
117 | 0 | bool evaluate_and(ParquetPredicate::ColumnStat* statistic) const override { |
118 | 0 | return _nested->evaluate_and(statistic) || statistic->has_null; |
119 | 0 | } |
120 | | |
121 | | bool evaluate_and(ParquetPredicate::CachedPageIndexStat* statistic, |
122 | 0 | RowRanges* row_ranges) const override { |
123 | 0 | _nested->evaluate_and(statistic, row_ranges); |
124 | 0 | ParquetPredicate::PageIndexStat* stat = nullptr; |
125 | 0 | if (!(statistic->get_stat_func)(&stat, column_id())) { |
126 | 0 | return true; |
127 | 0 | } |
128 | | |
129 | 0 | for (int page_id = 0; page_id < stat->num_of_pages; page_id++) { |
130 | 0 | if (stat->has_null[page_id]) { |
131 | 0 | row_ranges->add(stat->ranges[page_id]); |
132 | 0 | } |
133 | 0 | } |
134 | 0 | return row_ranges->count() > 0; |
135 | 0 | } |
136 | | |
137 | 0 | bool evaluate_del(const segment_v2::ZoneMap& zone_map) const override { |
138 | 0 | return _nested->evaluate_del(zone_map); |
139 | 0 | } |
140 | | |
141 | 0 | bool evaluate_and(const BloomFilter* bf) const override { return _nested->evaluate_and(bf); } |
142 | | |
143 | 0 | bool can_do_bloom_filter(bool ngram) const override { |
144 | 0 | return _nested->can_do_bloom_filter(ngram); |
145 | 0 | } |
146 | | |
147 | 0 | void evaluate_vec(const IColumn& column, uint16_t size, bool* flags) const override { |
148 | 0 | if (column.has_null()) { |
149 | 0 | const auto& nullable_col = assert_cast<const ColumnNullable&>(column); |
150 | 0 | _nested->evaluate_vec(nullable_col.get_nested_column(), size, flags); |
151 | 0 | for (uint16_t i = 0; i < size; ++i) { |
152 | 0 | if (!flags[i] && nullable_col.is_null_at(i)) { |
153 | | // set true for NULL rows |
154 | 0 | flags[i] = true; |
155 | 0 | } |
156 | 0 | } |
157 | 0 | } else { |
158 | 0 | _nested->evaluate_vec(column, size, flags); |
159 | 0 | } |
160 | 0 | } |
161 | | |
162 | 0 | void evaluate_and_vec(const IColumn& column, uint16_t size, bool* flags) const override { |
163 | 0 | if (column.has_null()) { |
164 | | // copy original flags |
165 | 0 | std::vector<uint8_t> original_flags(size); |
166 | 0 | memcpy(original_flags.data(), flags, size); |
167 | |
|
168 | 0 | const auto& nullable_col = assert_cast<const ColumnNullable&>(column); |
169 | | // call evaluate_and_vec and restore true for NULL rows |
170 | 0 | _nested->evaluate_and_vec(nullable_col.get_nested_column(), size, flags); |
171 | 0 | for (uint16_t i = 0; i < size; ++i) { |
172 | 0 | if (original_flags[i] && !flags[i] && nullable_col.is_null_at(i)) { |
173 | 0 | flags[i] = true; |
174 | 0 | } |
175 | 0 | } |
176 | 0 | } else { |
177 | 0 | _nested->evaluate_and_vec(column, size, flags); |
178 | 0 | } |
179 | 0 | } |
180 | | |
181 | 0 | std::string get_search_str() const override { return _nested->get_search_str(); } |
182 | | |
183 | | private: |
184 | 0 | uint16_t _evaluate_inner(const IColumn& column, uint16_t* sel, uint16_t size) const override { |
185 | 0 | if (column.has_null()) { |
186 | 0 | if (size == 0) { |
187 | 0 | return 0; |
188 | 0 | } |
189 | | // create selected_flags |
190 | 0 | uint16_t max_idx = sel[size - 1]; |
191 | 0 | std::vector<uint16_t> old_sel(size); |
192 | 0 | memcpy(old_sel.data(), sel, sizeof(uint16_t) * size); |
193 | |
|
194 | 0 | const auto& nullable_col = assert_cast<const ColumnNullable&>(column); |
195 | | // call nested predicate evaluate |
196 | 0 | uint16_t new_size = _nested->evaluate(nullable_col.get_nested_column(), sel, size); |
197 | | |
198 | | // process NULL values |
199 | 0 | if (new_size < size) { |
200 | 0 | std::vector<uint8_t> selected(max_idx + 1, 0); |
201 | 0 | const auto* nullmap = nullable_col.get_null_map_data().data(); |
202 | | // add rows selected by _nested->evaluate |
203 | 0 | for (uint16_t i = 0; i < new_size; ++i) { |
204 | 0 | uint16_t row_idx = sel[i]; |
205 | 0 | selected[row_idx] = true; |
206 | 0 | } |
207 | | // reset null from original data |
208 | 0 | for (uint16_t i = 0; i < size; ++i) { |
209 | 0 | uint16_t row_idx = old_sel[i]; |
210 | 0 | selected[row_idx] |= nullmap[row_idx]; |
211 | 0 | } |
212 | | |
213 | | // recaculate new_size and sel array |
214 | 0 | new_size = 0; |
215 | 0 | for (uint16_t row_idx = 0; row_idx < max_idx + 1; ++row_idx) { |
216 | 0 | if (selected[row_idx]) { |
217 | 0 | sel[new_size++] = row_idx; |
218 | 0 | } |
219 | 0 | } |
220 | 0 | } |
221 | 0 | return new_size; |
222 | 0 | } |
223 | 0 | return _nested->evaluate(column, sel, size); |
224 | 0 | } |
225 | | |
226 | | std::shared_ptr<ColumnPredicate> _nested; |
227 | | }; |
228 | | |
229 | | } //namespace doris |