be/src/exprs/function/variant_inverted_index_search.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/function/variant_inverted_index_search.h" |
19 | | |
20 | | #include <CLucene/config/repl_wchar.h> |
21 | | #include <fmt/format.h> |
22 | | #include <glog/logging.h> |
23 | | |
24 | | #include <limits> |
25 | | #include <memory> |
26 | | #include <utility> |
27 | | |
28 | | #include "common/config.h" |
29 | | #include "common/exception.h" |
30 | | #include "common/logging.h" |
31 | | #include "exprs/function/function_search.h" |
32 | | #include "exprs/vexpr_context.h" |
33 | | #include "runtime/runtime_state.h" |
34 | | #include "storage/index/index_file_reader.h" |
35 | | #include "storage/index/inverted/analyzer/analyzer.h" |
36 | | #include "storage/index/inverted/inverted_index_compound_reader.h" |
37 | | #include "storage/index/inverted/inverted_index_parser.h" |
38 | | #include "storage/index/inverted/inverted_index_searcher.h" |
39 | | #include "storage/index/inverted/query_v2/bit_set_query/bit_set_scorer.h" |
40 | | #include "storage/index/inverted/query_v2/doc_set.h" |
41 | | #include "storage/index/inverted/query_v2/scorer.h" |
42 | | #include "storage/index/inverted/query_v2/term_query/term_query.h" |
43 | | #include "storage/index/inverted/query_v2/weight.h" |
44 | | #include "storage/index/inverted/util/string_helper.h" |
45 | | #include "storage/segment/segment.h" |
46 | | #include "storage/segment/variant/nested_group_path.h" |
47 | | #include "storage/segment/variant/nested_group_provider.h" |
48 | | #include "storage/segment/variant/variant_column_reader.h" |
49 | | #include "storage/utils.h" |
50 | | #include "util/debug_points.h" |
51 | | #include "util/time.h" |
52 | | |
53 | | namespace doris { |
54 | | |
55 | | namespace query_v2 = segment_v2::inverted_index::query_v2; |
56 | | |
57 | | namespace { |
58 | | |
59 | | void add_search_binding_diagnostic(const std::shared_ptr<IndexQueryContext>& context, |
60 | 5 | const std::string& diagnostic) { |
61 | 5 | VLOG_DEBUG << diagnostic; |
62 | 5 | if (context != nullptr && context->stats != nullptr) { |
63 | 0 | context->stats->inverted_index_stats.add_binding_diagnostic(diagnostic); |
64 | 0 | } |
65 | 5 | } |
66 | | |
67 | | } // namespace |
68 | | |
69 | | FieldReaderResolver::FieldReaderResolver( |
70 | | const std::unordered_map<std::string, IndexFieldNameAndTypePair>& data_type_with_names, |
71 | | const std::unordered_map<std::string, IndexIterator*>& iterators, |
72 | | std::shared_ptr<IndexQueryContext> context, |
73 | | const std::vector<TSearchFieldBinding>& field_bindings) |
74 | 43 | : _data_type_with_names(data_type_with_names), |
75 | 43 | _iterators(iterators), |
76 | 43 | _context(std::move(context)), |
77 | 43 | _field_bindings(field_bindings) { |
78 | 43 | for (const auto& binding : _field_bindings) { |
79 | 6 | if (binding.__isset.is_variant_subcolumn && binding.is_variant_subcolumn) { |
80 | 5 | _variant_subcolumn_fields.insert(binding.field_name); |
81 | 5 | } |
82 | 6 | _field_binding_map[binding.field_name] = &binding; |
83 | 6 | } |
84 | 43 | } |
85 | | |
86 | | Status FieldReaderResolver::resolve(const std::string& field_name, |
87 | | InvertedIndexQueryType query_type, |
88 | 33 | FieldReaderBinding* binding) { |
89 | 33 | DCHECK(binding != nullptr); |
90 | | |
91 | 33 | const bool is_variant_sub = is_variant_subcolumn(field_name); |
92 | | |
93 | 33 | auto data_it = _data_type_with_names.find(field_name); |
94 | 33 | if (data_it == _data_type_with_names.end()) { |
95 | 6 | if (is_variant_sub) { |
96 | 1 | add_search_binding_diagnostic( |
97 | 1 | _context, |
98 | 1 | fmt::format("[VariantSearchBinding] phase=field_resolve result=no_metadata " |
99 | 1 | "logical_field={} query_type={} reason=field_not_found", |
100 | 1 | field_name, query_type_to_string(query_type))); |
101 | 1 | *binding = FieldReaderBinding(); |
102 | 1 | return Status::OK(); |
103 | 1 | } |
104 | 5 | return Status::Error<ErrorCode::INVERTED_INDEX_FILE_NOT_FOUND>( |
105 | 5 | "field '{}' not found in inverted index metadata", field_name); |
106 | 6 | } |
107 | | |
108 | 27 | const auto& stored_field_name = data_it->second.first; |
109 | 27 | const auto binding_key = binding_key_for(stored_field_name, query_type); |
110 | | |
111 | 27 | auto cache_it = _cache.find(binding_key); |
112 | 27 | if (cache_it != _cache.end()) { |
113 | 4 | *binding = cache_it->second; |
114 | 4 | return Status::OK(); |
115 | 4 | } |
116 | | |
117 | 23 | auto iterator_it = _iterators.find(field_name); |
118 | 23 | if (iterator_it == _iterators.end() || iterator_it->second == nullptr) { |
119 | 18 | if (is_variant_sub) { |
120 | 1 | add_search_binding_diagnostic( |
121 | 1 | _context, |
122 | 1 | fmt::format("[VariantSearchBinding] phase=field_resolve result=no_iterator " |
123 | 1 | "logical_field={} stored_field={} query_type={} " |
124 | 1 | "reason=iterator_not_found", |
125 | 1 | field_name, stored_field_name, query_type_to_string(query_type))); |
126 | 1 | *binding = FieldReaderBinding(); |
127 | 1 | return Status::OK(); |
128 | 1 | } |
129 | 17 | return Status::Error<ErrorCode::INVERTED_INDEX_FILE_NOT_FOUND>( |
130 | 17 | "iterator not found for field '{}'", field_name); |
131 | 18 | } |
132 | | |
133 | 5 | auto* inverted_iterator = dynamic_cast<InvertedIndexIterator*>(iterator_it->second); |
134 | 5 | if (inverted_iterator == nullptr) { |
135 | 2 | return Status::Error<ErrorCode::INVERTED_INDEX_FILE_NOT_FOUND>( |
136 | 2 | "iterator for field '{}' is not InvertedIndexIterator", field_name); |
137 | 2 | } |
138 | | |
139 | 3 | InvertedIndexQueryType effective_query_type = query_type; |
140 | 3 | const auto& column_type = data_it->second.second; |
141 | 3 | const bool is_text_field = |
142 | 3 | column_type != nullptr && is_string_type(column_type->get_storage_field_type()); |
143 | 3 | auto fb_it = _field_binding_map.find(field_name); |
144 | 3 | std::string analyzer_key; |
145 | 3 | if (is_text_field && is_variant_sub && fb_it != _field_binding_map.end() && |
146 | 3 | fb_it->second->__isset.index_properties && !fb_it->second->index_properties.empty()) { |
147 | 1 | analyzer_key = normalize_analyzer_key( |
148 | 1 | build_analyzer_key_from_properties(fb_it->second->index_properties)); |
149 | 1 | if (inverted_index::InvertedIndexAnalyzer::should_analyzer( |
150 | 1 | fb_it->second->index_properties) && |
151 | 1 | (effective_query_type == InvertedIndexQueryType::EQUAL_QUERY || |
152 | 1 | effective_query_type == InvertedIndexQueryType::WILDCARD_QUERY)) { |
153 | 1 | effective_query_type = InvertedIndexQueryType::MATCH_ANY_QUERY; |
154 | 1 | } |
155 | 1 | } |
156 | | |
157 | 3 | Result<InvertedIndexReaderPtr> reader_result; |
158 | 3 | if (column_type) { |
159 | 3 | reader_result = inverted_iterator->select_best_reader(column_type, effective_query_type, |
160 | 3 | is_text_field ? analyzer_key : ""); |
161 | 3 | } else { |
162 | 0 | reader_result = inverted_iterator->select_best_reader(is_text_field ? analyzer_key : ""); |
163 | 0 | } |
164 | | |
165 | 3 | if (!reader_result.has_value()) { |
166 | 1 | if (is_variant_sub) { |
167 | 1 | add_search_binding_diagnostic( |
168 | 1 | _context, |
169 | 1 | fmt::format("[VariantSearchBinding] phase=field_resolve result=reject " |
170 | 1 | "logical_field={} stored_field={} query_type={} " |
171 | 1 | "effective_query_type={} analyzer_key={} reason={}", |
172 | 1 | field_name, stored_field_name, query_type_to_string(query_type), |
173 | 1 | query_type_to_string(effective_query_type), analyzer_key, |
174 | 1 | reader_result.error().to_string())); |
175 | 1 | } |
176 | 1 | return reader_result.error(); |
177 | 1 | } |
178 | | |
179 | 2 | auto inverted_reader = reader_result.value(); |
180 | 2 | if (inverted_reader == nullptr) { |
181 | 0 | if (is_variant_sub) { |
182 | 0 | add_search_binding_diagnostic( |
183 | 0 | _context, |
184 | 0 | fmt::format("[VariantSearchBinding] phase=field_resolve result=reject " |
185 | 0 | "logical_field={} stored_field={} query_type={} " |
186 | 0 | "effective_query_type={} reason=selected_reader_null", |
187 | 0 | field_name, stored_field_name, query_type_to_string(query_type), |
188 | 0 | query_type_to_string(effective_query_type))); |
189 | 0 | } |
190 | 0 | return Status::Error<ErrorCode::INVERTED_INDEX_FILE_NOT_FOUND>( |
191 | 0 | "selected reader is null for field '{}'", field_name); |
192 | 0 | } |
193 | | |
194 | 2 | FieldReaderBinding resolved; |
195 | 2 | resolved.logical_field_name = field_name; |
196 | 2 | resolved.stored_field_name = stored_field_name; |
197 | 2 | resolved.stored_field_wstr = StringHelper::to_wstring(resolved.stored_field_name); |
198 | 2 | resolved.column_type = column_type; |
199 | 2 | resolved.query_type = effective_query_type; |
200 | 2 | resolved.inverted_reader = inverted_reader; |
201 | 2 | resolved.binding_key = binding_key; |
202 | 2 | resolved.state = SearchFieldBindingState::BOUND; |
203 | 2 | if (fb_it != _field_binding_map.end() && fb_it->second->__isset.index_properties && |
204 | 2 | !fb_it->second->index_properties.empty()) { |
205 | 1 | resolved.index_properties = fb_it->second->index_properties; |
206 | 1 | } else { |
207 | 1 | resolved.index_properties = inverted_reader->get_index_properties(); |
208 | 1 | } |
209 | 2 | resolved.analyzer_key = |
210 | 2 | normalize_analyzer_key(build_analyzer_key_from_properties(resolved.index_properties)); |
211 | | |
212 | 2 | auto index_file_reader = inverted_reader->get_index_file_reader(); |
213 | 2 | if (index_file_reader == nullptr) { |
214 | 1 | if (is_variant_sub) { |
215 | 1 | add_search_binding_diagnostic( |
216 | 1 | _context, |
217 | 1 | fmt::format("[VariantSearchBinding] phase=field_resolve result=reject " |
218 | 1 | "logical_field={} stored_field={} index_id={} suffix={} " |
219 | 1 | "reason=index_file_reader_null", |
220 | 1 | field_name, stored_field_name, inverted_reader->get_index_id(), |
221 | 1 | inverted_reader->get_index_meta().get_index_suffix())); |
222 | 1 | } |
223 | 1 | return Status::Error<ErrorCode::INVERTED_INDEX_FILE_NOT_FOUND>( |
224 | 1 | "index file reader is null for field '{}'", field_name); |
225 | 1 | } |
226 | | |
227 | 1 | if (inverted_reader->type() == InvertedIndexReaderType::BKD) { |
228 | 1 | _cache.emplace(binding_key, resolved); |
229 | 1 | if (is_variant_sub) { |
230 | 1 | bool index_file_exists = false; |
231 | 1 | auto probe_status = index_file_reader->index_file_exist( |
232 | 1 | &inverted_reader->get_index_meta(), &index_file_exists); |
233 | 1 | add_search_binding_diagnostic( |
234 | 1 | _context, |
235 | 1 | fmt::format("[VariantSearchBinding] phase=field_resolve result=selected_direct " |
236 | 1 | "logical_field={} stored_field={} query_type={} " |
237 | 1 | "effective_query_type={} index_id={} suffix={} reader_type={} " |
238 | 1 | "index_file_exists={} probe_status={} index_file={}", |
239 | 1 | field_name, stored_field_name, query_type_to_string(query_type), |
240 | 1 | query_type_to_string(effective_query_type), |
241 | 1 | inverted_reader->get_index_id(), |
242 | 1 | inverted_reader->get_index_meta().get_index_suffix(), |
243 | 1 | reader_type_to_string(inverted_reader->type()), index_file_exists, |
244 | 1 | probe_status.ok() ? "OK" : probe_status.to_string(), |
245 | 1 | index_file_reader->get_index_file_path( |
246 | 1 | &inverted_reader->get_index_meta()))); |
247 | 1 | } |
248 | 1 | *binding = resolved; |
249 | 1 | return Status::OK(); |
250 | 1 | } |
251 | | |
252 | 0 | auto index_file_key = |
253 | 0 | index_file_reader->get_index_file_cache_key(&inverted_reader->get_index_meta()); |
254 | 0 | InvertedIndexSearcherCache::CacheKey searcher_cache_key(index_file_key); |
255 | 0 | InvertedIndexCacheHandle searcher_cache_handle; |
256 | |
|
257 | 0 | bool searcher_cache_enabled = |
258 | 0 | _context->runtime_state != nullptr && |
259 | 0 | _context->runtime_state->query_options().enable_inverted_index_searcher_cache; |
260 | |
|
261 | 0 | bool cache_hit = false; |
262 | 0 | if (searcher_cache_enabled) { |
263 | 0 | int64_t lookup_dummy = 0; |
264 | 0 | SCOPED_RAW_TIMER(_context->stats ? &_context->stats->inverted_index_lookup_timer |
265 | 0 | : &lookup_dummy); |
266 | 0 | cache_hit = InvertedIndexSearcherCache::instance()->lookup(searcher_cache_key, |
267 | 0 | &searcher_cache_handle); |
268 | 0 | } |
269 | |
|
270 | 0 | std::shared_ptr<lucene::index::IndexReader> reader_holder; |
271 | 0 | if (cache_hit) { |
272 | 0 | if (_context->stats) { |
273 | 0 | _context->stats->inverted_index_searcher_cache_hit++; |
274 | 0 | } |
275 | 0 | auto searcher_variant = searcher_cache_handle.get_index_searcher(); |
276 | 0 | auto* searcher_ptr = std::get_if<FulltextIndexSearcherPtr>(&searcher_variant); |
277 | 0 | if (searcher_ptr != nullptr && *searcher_ptr != nullptr) { |
278 | 0 | reader_holder = std::shared_ptr<lucene::index::IndexReader>( |
279 | 0 | (*searcher_ptr)->getReader(), [](lucene::index::IndexReader*) {}); |
280 | 0 | } |
281 | 0 | } |
282 | |
|
283 | 0 | if (!reader_holder) { |
284 | 0 | if (_context->stats) { |
285 | 0 | _context->stats->inverted_index_searcher_cache_miss++; |
286 | 0 | } |
287 | 0 | int64_t dummy_timer = 0; |
288 | 0 | SCOPED_RAW_TIMER(_context->stats ? &_context->stats->inverted_index_searcher_open_timer |
289 | 0 | : &dummy_timer); |
290 | 0 | RETURN_IF_ERROR( |
291 | 0 | index_file_reader->init(config::inverted_index_read_buffer_size, _context->io_ctx)); |
292 | 0 | auto directory = DORIS_TRY( |
293 | 0 | index_file_reader->open(&inverted_reader->get_index_meta(), _context->io_ctx)); |
294 | |
|
295 | 0 | auto index_searcher_builder = DORIS_TRY( |
296 | 0 | IndexSearcherBuilder::create_index_searcher_builder(inverted_reader->type())); |
297 | 0 | auto searcher_result = |
298 | 0 | DORIS_TRY(index_searcher_builder->get_index_searcher(directory.get())); |
299 | 0 | auto reader_size = index_searcher_builder->get_reader_size(); |
300 | |
|
301 | 0 | auto* stream = static_cast<DorisCompoundReader*>(directory.get())->getDorisIndexInput(); |
302 | 0 | DBUG_EXECUTE_IF( |
303 | 0 | "FieldReaderResolver.resolve.io_ctx", ({ |
304 | 0 | const auto* cur_io_ctx = (const io::IOContext*)stream->getIoContext(); |
305 | 0 | if (cur_io_ctx->file_cache_stats) { |
306 | 0 | if (cur_io_ctx->file_cache_stats != &_context->stats->file_cache_stats) { |
307 | 0 | LOG(FATAL) << "search: io_ctx file_cache_stats mismatch: " |
308 | 0 | << cur_io_ctx->file_cache_stats << " vs " |
309 | 0 | << &_context->stats->file_cache_stats; |
310 | 0 | } |
311 | 0 | } |
312 | 0 | })); |
313 | 0 | stream->setIoContext(nullptr); |
314 | 0 | stream->setIndexFile(false); |
315 | |
|
316 | 0 | auto* cache_value = new InvertedIndexSearcherCache::CacheValue(std::move(searcher_result), |
317 | 0 | reader_size, UnixMillis()); |
318 | 0 | InvertedIndexSearcherCache::instance()->insert(searcher_cache_key, cache_value, |
319 | 0 | &searcher_cache_handle); |
320 | |
|
321 | 0 | auto new_variant = searcher_cache_handle.get_index_searcher(); |
322 | 0 | auto* new_ptr = std::get_if<FulltextIndexSearcherPtr>(&new_variant); |
323 | 0 | if (new_ptr != nullptr && *new_ptr != nullptr) { |
324 | 0 | reader_holder = std::shared_ptr<lucene::index::IndexReader>( |
325 | 0 | (*new_ptr)->getReader(), [](lucene::index::IndexReader*) {}); |
326 | 0 | } |
327 | |
|
328 | 0 | if (!reader_holder) { |
329 | 0 | return Status::Error<ErrorCode::INVERTED_INDEX_CLUCENE_ERROR>( |
330 | 0 | "failed to build IndexSearcher for field '{}'", field_name); |
331 | 0 | } |
332 | 0 | } |
333 | | |
334 | 0 | _searcher_cache_handles.push_back(std::move(searcher_cache_handle)); |
335 | |
|
336 | 0 | resolved.lucene_reader = reader_holder; |
337 | 0 | _binding_readers[binding_key] = reader_holder; |
338 | 0 | _field_readers[resolved.stored_field_wstr] = reader_holder; |
339 | 0 | _readers.emplace_back(reader_holder); |
340 | 0 | _cache.emplace(binding_key, resolved); |
341 | 0 | if (is_variant_sub) { |
342 | 0 | bool index_file_exists = false; |
343 | 0 | auto probe_status = index_file_reader->index_file_exist(&inverted_reader->get_index_meta(), |
344 | 0 | &index_file_exists); |
345 | 0 | add_search_binding_diagnostic( |
346 | 0 | _context, |
347 | 0 | fmt::format( |
348 | 0 | "[VariantSearchBinding] phase=field_resolve result=selected " |
349 | 0 | "logical_field={} stored_field={} query_type={} effective_query_type={} " |
350 | 0 | "index_id={} suffix={} reader_type={} analyzer_key={} " |
351 | 0 | "field_pattern={} index_file_exists={} probe_status={} " |
352 | 0 | "searcher_cache={} index_file={}", |
353 | 0 | field_name, stored_field_name, query_type_to_string(query_type), |
354 | 0 | query_type_to_string(effective_query_type), inverted_reader->get_index_id(), |
355 | 0 | inverted_reader->get_index_meta().get_index_suffix(), |
356 | 0 | reader_type_to_string(inverted_reader->type()), resolved.analyzer_key, |
357 | 0 | inverted_reader->get_index_meta().field_pattern(), index_file_exists, |
358 | 0 | probe_status.ok() ? "OK" : probe_status.to_string(), |
359 | 0 | cache_hit ? "hit" : "miss", |
360 | 0 | index_file_reader->get_index_file_path( |
361 | 0 | &inverted_reader->get_index_meta()))); |
362 | 0 | } |
363 | 0 | *binding = resolved; |
364 | 0 | return Status::OK(); |
365 | 0 | } |
366 | | |
367 | | segment_v2::IndexIterator* VariantSearchNullBitmapAdapter::iterator_for( |
368 | 0 | const query_v2::Scorer& /*scorer*/, const std::string& logical_field) const { |
369 | 0 | if (logical_field.empty()) { |
370 | 0 | return nullptr; |
371 | 0 | } |
372 | 0 | return _resolver.get_iterator(logical_field); |
373 | 0 | } |
374 | | |
375 | | void populate_variant_search_binding_context(const FieldReaderResolver& resolver, |
376 | 1 | query_v2::QueryExecutionContext* exec_ctx) { |
377 | 1 | DCHECK(exec_ctx != nullptr); |
378 | 1 | exec_ctx->readers = resolver.readers(); |
379 | 1 | exec_ctx->reader_bindings = resolver.reader_bindings(); |
380 | 1 | exec_ctx->field_reader_bindings = resolver.field_readers(); |
381 | 1 | for (const auto& [binding_key, binding] : resolver.binding_cache()) { |
382 | 0 | if (binding_key.empty()) { |
383 | 0 | continue; |
384 | 0 | } |
385 | 0 | query_v2::FieldBindingContext binding_ctx; |
386 | 0 | binding_ctx.logical_field_name = binding.logical_field_name; |
387 | 0 | binding_ctx.stored_field_name = binding.stored_field_name; |
388 | 0 | binding_ctx.stored_field_wstr = binding.stored_field_wstr; |
389 | 0 | exec_ctx->binding_fields.emplace(binding_key, std::move(binding_ctx)); |
390 | 0 | } |
391 | 1 | } |
392 | | |
393 | | query_v2::QueryExecutionContext build_variant_search_query_execution_context( |
394 | | uint32_t segment_num_rows, const FieldReaderResolver& resolver, |
395 | 1 | query_v2::NullBitmapResolver* null_resolver) { |
396 | 1 | query_v2::QueryExecutionContext exec_ctx; |
397 | 1 | exec_ctx.segment_num_rows = segment_num_rows; |
398 | 1 | populate_variant_search_binding_context(resolver, &exec_ctx); |
399 | 1 | exec_ctx.null_resolver = null_resolver; |
400 | 1 | return exec_ctx; |
401 | 1 | } |
402 | | |
403 | | namespace { |
404 | | |
405 | | class VariantNestedDocMappingWeight final : public query_v2::Weight { |
406 | | public: |
407 | | VariantNestedDocMappingWeight( |
408 | | query_v2::WeightPtr child_weight, |
409 | | std::vector<const segment_v2::NestedGroupReader*> child_to_parent_chain, |
410 | | const segment_v2::NestedGroupReadProvider* read_provider, |
411 | | segment_v2::ColumnIteratorOptions column_iter_opts) |
412 | 7 | : _child_weight(std::move(child_weight)), |
413 | 7 | _child_to_parent_chain(std::move(child_to_parent_chain)), |
414 | 7 | _read_provider(read_provider), |
415 | 7 | _column_iter_opts(std::move(column_iter_opts)) {} |
416 | | |
417 | | query_v2::ScorerPtr scorer(const query_v2::QueryExecutionContext& context, |
418 | 7 | const std::string& binding_key) override { |
419 | 7 | if (_child_weight == nullptr || _read_provider == nullptr || |
420 | 7 | _child_to_parent_chain.empty()) { |
421 | 1 | return std::make_shared<query_v2::EmptyScorer>(); |
422 | 1 | } |
423 | | |
424 | 6 | auto child_scorer = _child_weight->scorer(context, binding_key); |
425 | 6 | if (child_scorer == nullptr) { |
426 | 1 | return std::make_shared<query_v2::EmptyScorer>(); |
427 | 1 | } |
428 | | |
429 | 5 | roaring::Roaring child_true; |
430 | 5 | uint32_t doc = child_scorer->doc(); |
431 | 9 | while (doc != query_v2::TERMINATED) { |
432 | 4 | child_true.add(doc); |
433 | 4 | doc = child_scorer->advance(); |
434 | 4 | } |
435 | | |
436 | 5 | auto mapped_true = std::make_shared<roaring::Roaring>(); |
437 | 5 | if (!child_true.isEmpty()) { |
438 | 3 | auto status = _read_provider->map_elements_to_parent_ords( |
439 | 3 | _child_to_parent_chain, _column_iter_opts, child_true, mapped_true.get()); |
440 | 3 | if (!status.ok()) { |
441 | 1 | throw Exception(ErrorCode::INTERNAL_ERROR, |
442 | 1 | "failed to map nested search true bitmap: {}", status.to_string()); |
443 | 1 | } |
444 | 3 | } |
445 | | |
446 | 4 | std::shared_ptr<roaring::Roaring> mapped_null; |
447 | 4 | if (child_scorer->has_null_bitmap(context.null_resolver)) { |
448 | 3 | const auto* child_null = child_scorer->get_null_bitmap(context.null_resolver); |
449 | 3 | if (child_null != nullptr && !child_null->isEmpty()) { |
450 | 3 | mapped_null = std::make_shared<roaring::Roaring>(); |
451 | 3 | auto status = _read_provider->map_elements_to_parent_ords( |
452 | 3 | _child_to_parent_chain, _column_iter_opts, *child_null, mapped_null.get()); |
453 | 3 | if (!status.ok()) { |
454 | 1 | throw Exception(ErrorCode::INTERNAL_ERROR, |
455 | 1 | "failed to map nested search null bitmap: {}", |
456 | 1 | status.to_string()); |
457 | 1 | } |
458 | 2 | *mapped_null -= *mapped_true; |
459 | 2 | if (mapped_null->isEmpty()) { |
460 | 1 | mapped_null.reset(); |
461 | 1 | } |
462 | 2 | } |
463 | 3 | } |
464 | | |
465 | 3 | if (mapped_true->isEmpty() && (mapped_null == nullptr || mapped_null->isEmpty())) { |
466 | 1 | return std::make_shared<query_v2::EmptyScorer>(); |
467 | 1 | } |
468 | 2 | return std::make_shared<query_v2::BitSetScorer>(std::move(mapped_true), |
469 | 2 | std::move(mapped_null)); |
470 | 3 | } |
471 | | |
472 | | private: |
473 | | query_v2::WeightPtr _child_weight; |
474 | | std::vector<const segment_v2::NestedGroupReader*> _child_to_parent_chain; |
475 | | const segment_v2::NestedGroupReadProvider* _read_provider; |
476 | | segment_v2::ColumnIteratorOptions _column_iter_opts; |
477 | | }; |
478 | | |
479 | | class VariantNestedDocMappingQuery final : public query_v2::Query { |
480 | | public: |
481 | | VariantNestedDocMappingQuery( |
482 | | query_v2::QueryPtr child_query, |
483 | | std::vector<const segment_v2::NestedGroupReader*> child_to_parent_chain, |
484 | | const segment_v2::NestedGroupReadProvider* read_provider, |
485 | | segment_v2::ColumnIteratorOptions column_iter_opts) |
486 | 7 | : _child_query(std::move(child_query)), |
487 | 7 | _child_to_parent_chain(std::move(child_to_parent_chain)), |
488 | 7 | _read_provider(read_provider), |
489 | 7 | _column_iter_opts(std::move(column_iter_opts)) {} |
490 | | |
491 | 7 | query_v2::WeightPtr weight(bool enable_scoring) override { |
492 | 7 | if (_child_query == nullptr) { |
493 | 0 | return nullptr; |
494 | 0 | } |
495 | 7 | return std::make_shared<VariantNestedDocMappingWeight>(_child_query->weight(enable_scoring), |
496 | 7 | _child_to_parent_chain, |
497 | 7 | _read_provider, _column_iter_opts); |
498 | 7 | } |
499 | | |
500 | | private: |
501 | | query_v2::QueryPtr _child_query; |
502 | | std::vector<const segment_v2::NestedGroupReader*> _child_to_parent_chain; |
503 | | const segment_v2::NestedGroupReadProvider* _read_provider; |
504 | | segment_v2::ColumnIteratorOptions _column_iter_opts; |
505 | | }; |
506 | | |
507 | 2 | bool starts_with_root_field(const std::string& logical_field_name, const std::string& root_field) { |
508 | 2 | if (logical_field_name == root_field) { |
509 | 1 | return true; |
510 | 1 | } |
511 | 1 | return logical_field_name.size() > root_field.size() && |
512 | 1 | logical_field_name.compare(0, root_field.size(), root_field) == 0 && |
513 | 1 | logical_field_name[root_field.size()] == '.'; |
514 | 2 | } |
515 | | |
516 | | } // namespace |
517 | | |
518 | | query_v2::QueryPtr make_variant_nested_doc_mapping_query( |
519 | | query_v2::QueryPtr child_query, |
520 | | std::vector<const segment_v2::NestedGroupReader*> child_to_parent_chain, |
521 | | const segment_v2::NestedGroupReadProvider* read_provider, |
522 | 8 | segment_v2::ColumnIteratorOptions column_iter_opts) { |
523 | 8 | if (child_to_parent_chain.empty()) { |
524 | 1 | return child_query; |
525 | 1 | } |
526 | 7 | return std::make_shared<VariantNestedDocMappingQuery>( |
527 | 7 | std::move(child_query), std::move(child_to_parent_chain), read_provider, |
528 | 7 | std::move(column_iter_opts)); |
529 | 8 | } |
530 | | |
531 | | Status map_variant_nested_leaf_query_to_active_group(const VariantNestedDocMapperContext& context, |
532 | | const std::string& logical_field_name, |
533 | 3 | query_v2::QueryPtr* query) { |
534 | 3 | if (query == nullptr || *query == nullptr || context.variant_reader == nullptr || |
535 | 3 | context.read_provider == nullptr || context.active_group_chain.empty() || |
536 | 3 | context.root_field.empty()) { |
537 | 1 | return Status::OK(); |
538 | 1 | } |
539 | 2 | if (!starts_with_root_field(logical_field_name, context.root_field)) { |
540 | 1 | return Status::OK(); |
541 | 1 | } |
542 | | |
543 | 1 | std::string relative_path; |
544 | 1 | if (logical_field_name.size() > context.root_field.size()) { |
545 | 0 | relative_path = logical_field_name.substr(context.root_field.size() + 1); |
546 | 0 | } |
547 | 1 | if (relative_path.empty()) { |
548 | 1 | return Status::OK(); |
549 | 1 | } |
550 | | |
551 | 0 | auto [found, leaf_group_chain, _] = |
552 | 0 | context.variant_reader->collect_nested_group_chain(relative_path); |
553 | 0 | if (!found) { |
554 | 0 | return Status::OK(); |
555 | 0 | } |
556 | 0 | if (leaf_group_chain.size() < context.active_group_chain.size()) { |
557 | 0 | return Status::InvalidArgument( |
558 | 0 | "nested search leaf field '{}' is outside active nested path", logical_field_name); |
559 | 0 | } |
560 | 0 | for (size_t i = 0; i < context.active_group_chain.size(); ++i) { |
561 | 0 | if (leaf_group_chain[i] != context.active_group_chain[i]) { |
562 | 0 | return Status::InvalidArgument( |
563 | 0 | "nested search leaf field '{}' is outside active nested path", |
564 | 0 | logical_field_name); |
565 | 0 | } |
566 | 0 | } |
567 | 0 | if (leaf_group_chain.size() == context.active_group_chain.size()) { |
568 | 0 | return Status::OK(); |
569 | 0 | } |
570 | | |
571 | 0 | std::vector<const segment_v2::NestedGroupReader*> child_to_parent_chain( |
572 | 0 | leaf_group_chain.begin() + context.active_group_chain.size(), leaf_group_chain.end()); |
573 | 0 | *query = make_variant_nested_doc_mapping_query(std::move(*query), |
574 | 0 | std::move(child_to_parent_chain), |
575 | 0 | context.read_provider, context.column_iter_opts); |
576 | 0 | return Status::OK(); |
577 | 0 | } |
578 | | |
579 | | Status VariantNestedSearchEvaluator::evaluate( |
580 | | const TSearchParam& search_param, const TSearchClause& nested_clause, |
581 | | const std::shared_ptr<segment_v2::IndexQueryContext>& context, |
582 | | FieldReaderResolver& resolver, uint32_t num_rows, const IndexExecContext* index_exec_ctx, |
583 | | const std::unordered_map<std::string, int>& field_name_to_column_id, |
584 | 7 | std::shared_ptr<roaring::Roaring>& result_bitmap) const { |
585 | 7 | (void)num_rows; |
586 | 7 | (void)field_name_to_column_id; |
587 | 7 | if (!(nested_clause.__isset.nested_path)) { |
588 | 2 | return Status::InvalidArgument("NESTED clause missing nested_path"); |
589 | 2 | } |
590 | 5 | if (!(nested_clause.__isset.children) || nested_clause.children.empty()) { |
591 | 2 | return Status::InvalidArgument("NESTED clause missing inner query"); |
592 | 2 | } |
593 | 3 | if (result_bitmap == nullptr) { |
594 | 2 | result_bitmap = std::make_shared<roaring::Roaring>(); |
595 | 2 | } else { |
596 | 1 | *result_bitmap = roaring::Roaring(); |
597 | 1 | } |
598 | | |
599 | 3 | std::string root_field = nested_clause.nested_path; |
600 | 3 | auto dot_pos = nested_clause.nested_path.find('.'); |
601 | 3 | if (dot_pos != std::string::npos) { |
602 | 1 | root_field = nested_clause.nested_path.substr(0, dot_pos); |
603 | 1 | } |
604 | 3 | if (index_exec_ctx == nullptr || index_exec_ctx->segment() == nullptr) { |
605 | 3 | return Status::InvalidArgument("NESTED query requires IndexExecContext with valid segment"); |
606 | 3 | } |
607 | 0 | auto* segment = index_exec_ctx->segment(); |
608 | 0 | const int32_t ordinal = segment->tablet_schema()->field_index(root_field); |
609 | 0 | if (ordinal < 0) { |
610 | 0 | return Status::InvalidArgument("Column '{}' not found in tablet schema for nested query", |
611 | 0 | root_field); |
612 | 0 | } |
613 | 0 | const ColumnId column_id = static_cast<ColumnId>(ordinal); |
614 | |
|
615 | 0 | std::shared_ptr<segment_v2::ColumnReader> column_reader; |
616 | 0 | RETURN_IF_ERROR(segment->get_column_reader(segment->tablet_schema()->column(column_id), |
617 | 0 | &column_reader, |
618 | 0 | index_exec_ctx->column_iter_opts().stats)); |
619 | 0 | auto* variant_reader = dynamic_cast<segment_v2::VariantColumnReader*>(column_reader.get()); |
620 | 0 | if (variant_reader == nullptr) { |
621 | 0 | return Status::InvalidArgument("Column '{}' is not VARIANT for nested query", root_field); |
622 | 0 | } |
623 | | |
624 | 0 | std::string array_path; |
625 | 0 | if (dot_pos == std::string::npos) { |
626 | 0 | array_path = std::string(segment_v2::kRootNestedGroupPath); |
627 | 0 | } else { |
628 | 0 | array_path = nested_clause.nested_path.substr(dot_pos + 1); |
629 | 0 | } |
630 | |
|
631 | 0 | auto [found, group_chain, _] = variant_reader->collect_nested_group_chain(array_path); |
632 | 0 | if (!found || group_chain.empty()) { |
633 | 0 | return Status::OK(); |
634 | 0 | } |
635 | | |
636 | 0 | auto read_provider = segment_v2::create_nested_group_read_provider(); |
637 | 0 | if (!read_provider || !read_provider->should_enable_nested_group_read_path()) { |
638 | 0 | return Status::NotSupported( |
639 | 0 | "NestedGroup search is an enterprise capability, not available in this build"); |
640 | 0 | } |
641 | | |
642 | 0 | auto& leaf_group = group_chain.back(); |
643 | 0 | uint64_t total_elements = 0; |
644 | 0 | RETURN_IF_ERROR(read_provider->get_total_elements(index_exec_ctx->column_iter_opts(), |
645 | 0 | leaf_group, &total_elements)); |
646 | 0 | if (total_elements == 0) { |
647 | 0 | return Status::OK(); |
648 | 0 | } |
649 | 0 | if (total_elements > std::numeric_limits<uint32_t>::max()) { |
650 | 0 | return Status::InvalidArgument("nested element_count exceeds uint32_t max"); |
651 | 0 | } |
652 | | |
653 | 0 | std::string default_operator = "or"; |
654 | 0 | if (search_param.__isset.default_operator && !search_param.default_operator.empty()) { |
655 | 0 | default_operator = search_param.default_operator; |
656 | 0 | } |
657 | 0 | int32_t minimum_should_match = -1; |
658 | 0 | if (search_param.__isset.minimum_should_match) { |
659 | 0 | minimum_should_match = search_param.minimum_should_match; |
660 | 0 | } |
661 | |
|
662 | 0 | query_v2::QueryPtr inner_query; |
663 | 0 | std::string inner_binding_key; |
664 | 0 | VariantNestedDocMapperContext mapper_context; |
665 | 0 | mapper_context.root_field = root_field; |
666 | 0 | mapper_context.active_group_chain = group_chain; |
667 | 0 | mapper_context.variant_reader = variant_reader; |
668 | 0 | mapper_context.read_provider = read_provider.get(); |
669 | 0 | mapper_context.column_iter_opts = index_exec_ctx->column_iter_opts(); |
670 | 0 | resolver.set_leaf_query_mapper( |
671 | 0 | [mapper_context](const std::string& logical_field_name, query_v2::QueryPtr* query) { |
672 | 0 | return map_variant_nested_leaf_query_to_active_group(mapper_context, |
673 | 0 | logical_field_name, query); |
674 | 0 | }); |
675 | 0 | struct ScopedLeafMapperReset { |
676 | 0 | FieldReaderResolver& resolver; |
677 | 0 | ~ScopedLeafMapperReset() { resolver.set_leaf_query_mapper(nullptr); } |
678 | 0 | } mapper_reset {resolver}; |
679 | 0 | RETURN_IF_ERROR(_function_search.build_query_recursive( |
680 | 0 | nested_clause.children[0], context, resolver, &inner_query, &inner_binding_key, |
681 | 0 | default_operator, minimum_should_match, static_cast<uint32_t>(total_elements))); |
682 | 0 | if (inner_query == nullptr) { |
683 | 0 | return Status::OK(); |
684 | 0 | } |
685 | | |
686 | 0 | VariantSearchNullBitmapAdapter null_resolver(resolver); |
687 | 0 | query_v2::QueryExecutionContext exec_ctx = build_variant_search_query_execution_context( |
688 | 0 | static_cast<uint32_t>(total_elements), resolver, &null_resolver); |
689 | |
|
690 | 0 | auto weight = inner_query->weight(false); |
691 | 0 | if (!weight) { |
692 | 0 | return Status::OK(); |
693 | 0 | } |
694 | 0 | auto scorer = weight->scorer(exec_ctx, inner_binding_key); |
695 | 0 | if (!scorer) { |
696 | 0 | return Status::OK(); |
697 | 0 | } |
698 | | |
699 | 0 | roaring::Roaring element_bitmap; |
700 | 0 | uint32_t doc = scorer->doc(); |
701 | 0 | while (doc != query_v2::TERMINATED) { |
702 | 0 | element_bitmap.add(doc); |
703 | 0 | doc = scorer->advance(); |
704 | 0 | } |
705 | |
|
706 | 0 | if (scorer->has_null_bitmap(exec_ctx.null_resolver)) { |
707 | 0 | const auto* bitmap = scorer->get_null_bitmap(exec_ctx.null_resolver); |
708 | 0 | if (bitmap != nullptr && !bitmap->isEmpty()) { |
709 | 0 | element_bitmap -= *bitmap; |
710 | 0 | } |
711 | 0 | } |
712 | |
|
713 | 0 | roaring::Roaring parent_bitmap; |
714 | 0 | RETURN_IF_ERROR(read_provider->map_elements_to_parent_ords( |
715 | 0 | group_chain, index_exec_ctx->column_iter_opts(), element_bitmap, &parent_bitmap)); |
716 | 0 | *result_bitmap = std::move(parent_bitmap); |
717 | 0 | return Status::OK(); |
718 | 0 | } |
719 | | |
720 | | } // namespace doris |