zjw1111 commented on code in PR #448: URL: https://github.com/apache/paimon-cpp/pull/448#discussion_r4344385916
########## src/paimon/core/global_index/live_row_global_index_reader.cpp: ########## @@ -0,0 +1,232 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +#include "paimon/core/global_index/live_row_global_index_reader.h" + +#include <algorithm> +#include <cassert> +#include <mutex> +#include <utility> + +#include "paimon/common/global_index/union_global_index_reader.h" + +namespace paimon { +namespace { +/// Returns whether `full_text_search` may rank a row of `range`: it has no include row ids, or one +/// of them is in `range`. +bool MayRankRowsOf(const FullTextSearch& full_text_search, const Range& range) { + if (!full_text_search.include_row_ids) { + return true; + } + const RoaringBitmap64& include_row_ids = full_text_search.include_row_ids.value(); + auto first = include_row_ids.EqualOrLarger(range.from); + return first != include_row_ids.End() && *first <= range.to; +} + +/// A union of LiveRowGlobalIndexReaders that loads the live rows on the visiting thread before a +/// full-text search fans out to them. +class LiveRowUnionGlobalIndexReader : public UnionGlobalIndexReader { + public: + /// @param ranked_ranges The global row ids the shards rank in a full-text search, leaving out + /// the shards that rank none. + LiveRowUnionGlobalIndexReader(std::vector<std::shared_ptr<GlobalIndexReader>>&& readers, + std::vector<Range>&& row_ranges, + std::vector<std::optional<Range>>&& search_ranges, + std::vector<Range>&& ranked_ranges, + const std::shared_ptr<Executor>& executor, + LiveRowGlobalIndexReader::LiveRowsSupplier live_rows_supplier) + : UnionGlobalIndexReader(std::move(readers), std::move(row_ranges), + std::move(search_ranges), executor), + ranked_ranges_(std::move(ranked_ranges)), + live_rows_supplier_(std::move(live_rows_supplier)) {} + + Result<std::shared_ptr<ScoredGlobalIndexResult>> VisitFullTextSearch( + const std::shared_ptr<FullTextSearch>& full_text_search) override { + // Only a shard search needs live rows: the union rejects a search without a positive limit + // before searching any shard, and skips the shards whose rows the include row ids miss. + if (full_text_search && full_text_search->limit > 0 && + std::any_of(ranked_ranges_.begin(), ranked_ranges_.end(), [&](const Range& range) { + return MayRankRowsOf(*full_text_search, range); + })) { + PAIMON_RETURN_NOT_OK(live_rows_supplier_()); + } + return UnionGlobalIndexReader::VisitFullTextSearch(full_text_search); + } + + private: + std::vector<Range> ranked_ranges_; + LiveRowGlobalIndexReader::LiveRowsSupplier live_rows_supplier_; +}; +} // namespace + +LiveRowGlobalIndexReader::LiveRowsSupplier LiveRowGlobalIndexReader::CacheLiveRows( + std::function<Result<std::optional<RoaringBitmap64>>()> load) { + struct State { + std::mutex mutex; + bool loaded = false; + std::shared_ptr<const RoaringBitmap64> live_rows; + }; + auto state = std::make_shared<State>(); + return [state, load = std::move(load)]() -> Result<std::shared_ptr<const RoaringBitmap64>> { + std::lock_guard<std::mutex> lock(state->mutex); + if (!state->loaded) { + PAIMON_ASSIGN_OR_RAISE(std::optional<RoaringBitmap64> live_rows, load()); + if (live_rows) { + state->live_rows = + std::make_shared<const RoaringBitmap64>(std::move(live_rows).value()); + } + state->loaded = true; + } + return state->live_rows; + }; +} + +std::shared_ptr<GlobalIndexReader> LiveRowGlobalIndexReader::CreateUnion( + std::vector<std::shared_ptr<GlobalIndexReader>>&& shard_readers, + std::vector<Range>&& shard_ranges, std::vector<std::optional<Range>>&& search_ranges, + const std::shared_ptr<Executor>& executor, const LiveRowsSupplier& live_rows_supplier) { + assert(shard_readers.size() == shard_ranges.size()); + std::vector<std::shared_ptr<GlobalIndexReader>> readers; + std::vector<Range> ranked_ranges; + readers.reserve(shard_readers.size()); + ranked_ranges.reserve(shard_readers.size()); + for (size_t i = 0; i < shard_readers.size(); i++) { + readers.push_back(std::make_shared<LiveRowGlobalIndexReader>( + std::move(shard_readers[i]), shard_ranges[i], live_rows_supplier)); + if (search_ranges.empty()) { + ranked_ranges.push_back(shard_ranges[i]); + } else if (search_ranges[i]) { + ranked_ranges.push_back(search_ranges[i].value()); + } + } + return std::make_shared<LiveRowUnionGlobalIndexReader>( + std::move(readers), std::move(shard_ranges), std::move(search_ranges), + std::move(ranked_ranges), executor, live_rows_supplier); +} + +LiveRowGlobalIndexReader::LiveRowGlobalIndexReader(std::shared_ptr<GlobalIndexReader>&& wrapped, + const Range& shard_range, + LiveRowsSupplier live_rows_supplier) + : wrapped_(std::move(wrapped)), + shard_range_(shard_range), + live_rows_supplier_(std::move(live_rows_supplier)) {} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitIsNotNull() { + return wrapped_->VisitIsNotNull(); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitIsNull() { + return wrapped_->VisitIsNull(); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitEqual( + const Literal& literal) { + return wrapped_->VisitEqual(literal); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitNotEqual( + const Literal& literal) { + return wrapped_->VisitNotEqual(literal); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitLessThan( + const Literal& literal) { + return wrapped_->VisitLessThan(literal); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitLessOrEqual( + const Literal& literal) { + return wrapped_->VisitLessOrEqual(literal); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitGreaterThan( + const Literal& literal) { + return wrapped_->VisitGreaterThan(literal); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitGreaterOrEqual( + const Literal& literal) { + return wrapped_->VisitGreaterOrEqual(literal); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitIn( + const std::vector<Literal>& literals) { + return wrapped_->VisitIn(literals); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitNotIn( + const std::vector<Literal>& literals) { + return wrapped_->VisitNotIn(literals); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitStartsWith( + const Literal& prefix) { + return wrapped_->VisitStartsWith(prefix); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitEndsWith( + const Literal& suffix) { + return wrapped_->VisitEndsWith(suffix); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitContains( + const Literal& literal) { + return wrapped_->VisitContains(literal); +} + +Result<std::shared_ptr<GlobalIndexResult>> LiveRowGlobalIndexReader::VisitLike( + const Literal& literal) { + return wrapped_->VisitLike(literal); +} + +Result<std::shared_ptr<ScoredGlobalIndexResult>> LiveRowGlobalIndexReader::VisitVectorSearch( + const std::shared_ptr<VectorSearch>& vector_search) { + return wrapped_->VisitVectorSearch(vector_search); Review Comment: Thanks for fixing the full-text case. As noted in the description, vector search on data-evolution tables with deletion vectors enabled still has the same top-k issue: deleted rows can occupy result slots and then be dropped during the data read. Do you have plans to address this in a follow-up? If not, we'll schedule support on our side. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
