SteNicholas commented on code in PR #485:
URL: https://github.com/apache/paimon-cpp/pull/485#discussion_r4496500542


##########
src/paimon/core/table/source/data_evolution_full_text_scan.cpp:
##########
@@ -156,6 +165,27 @@ bool DataEvolutionFullTextScan::IndexesField(const 
GlobalIndexMeta& meta, int32_
                meta.extra_field_ids->end();
 }
 
+bool DataEvolutionFullTextScan::IsScalarPreFilterFile(const GlobalIndexMeta& 
meta,
+                                                      const std::set<int32_t>& 
filter_field_ids,
+                                                      bool supports_full_text) 
{
+    bool has_extra_fields = meta.extra_field_ids && 
!meta.extra_field_ids->empty();
+    return IndexesAnyField(meta, filter_field_ids) && (!supports_full_text || 
has_extra_fields);

Review Comment:
   Good catch, done in 63a3f85a. Both consumers of these files drop multi-field 
indexes (`GlobalIndexScanImpl::CreateForScalarFilters()` here, and 
`createForScalarFilters` and `DataEvolutionGlobalIndexCoverage` in Java), so 
attaching them only cost validation and serialization. 
`IsScalarPreFilterFile()` now accepts only single-field indexes whose indexer 
does not support full-text search, and the manifest filter of the scan no 
longer loads files that reach a filter field only through their extra fields. 
`CreateForScalarFilters()` keeps its check, as in Java, and 
`TestScalarPreFilterFiles` covers the new rule.
   



##########
src/paimon/core/table/source/filtered_row_id_reader.cpp:
##########
@@ -0,0 +1,248 @@
+/*
+ * 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/table/source/filtered_row_id_reader.h"
+
+#include <algorithm>
+#include <map>
+#include <memory>
+#include <optional>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "arrow/api.h"
+#include "arrow/c/bridge.h"
+#include "fmt/format.h"
+#include "paimon/common/table/special_fields.h"
+#include "paimon/common/types/data_field.h"
+#include "paimon/common/utils/arrow/status_utils.h"
+#include "paimon/common/utils/checked_cast.h"
+#include "paimon/common/utils/object_utils.h"
+#include "paimon/common/utils/scope_guard.h"
+#include "paimon/core/global_index/global_index_evaluator_impl.h"
+#include "paimon/core/operation/internal_read_context.h"
+#include "paimon/core/table/source/append_only_table_read.h"
+#include "paimon/core/utils/branch_manager.h"
+#include "paimon/core/utils/file_store_path_factory.h"
+#include "paimon/core/utils/snapshot_manager.h"
+#include "paimon/defs.h"
+#include "paimon/format/file_format.h"
+#include "paimon/global_index/bitmap_global_index_result.h"
+#include "paimon/predicate/compound_predicate.h"
+#include "paimon/predicate/leaf_predicate.h"
+#include "paimon/predicate/predicate_utils.h"
+#include "paimon/read_context.h"
+#include "paimon/reader/batch_reader.h"
+#include "paimon/scan_context.h"
+#include "paimon/table/source/plan.h"
+#include "paimon/table/source/startup_mode.h"
+#include "paimon/table/source/table_read.h"
+#include "paimon/table/source/table_scan.h"
+
+namespace paimon {
+namespace {
+/// Returns whether `predicate` has a leaf that a global index may answer with 
candidates only.
+bool HasCandidateOnlyLeaf(const std::shared_ptr<Predicate>& predicate) {
+    if (auto compound = 
std::dynamic_pointer_cast<CompoundPredicate>(predicate)) {
+        const auto& children = compound->Children();
+        return std::any_of(children.begin(), children.end(), 
HasCandidateOnlyLeaf);
+    }
+    if (auto leaf = std::dynamic_pointer_cast<LeafPredicate>(predicate)) {
+        Function::Type type = leaf->GetFunction().GetType();
+        return type == Function::Type::CONTAINS || type == 
Function::Type::ENDS_WITH ||
+               type == Function::Type::LIKE;
+    }
+    return false;
+}
+
+/// Returns `options` with the scan pinned to snapshot `snapshot_id`, as Java 
does. A tag may
+/// outlive its snapshot, so once the snapshot has expired a search of a tag 
reads the tag instead,
+/// which the caller must check still refers to the snapshot.
+Result<std::map<std::string, std::string>> PinToSnapshot(const std::string& 
root_path,
+                                                         const CoreOptions& 
options,
+                                                         int64_t snapshot_id) {
+    std::map<std::string, std::string> pinned = options.ToMap();
+    pinned[Options::SCAN_MODE] = StartupMode::FromSnapshot().ToString();
+    pinned.erase(Options::SCAN_TIMESTAMP);
+    pinned.erase(Options::SCAN_TIMESTAMP_MILLIS);
+    if (pinned.count(Options::SCAN_TAG_NAME) > 0) {
+        SnapshotManager snapshot_manager(options.GetFileSystem(), root_path,
+                                         
BranchManager::NormalizeBranch(options.GetBranch()));
+        PAIMON_ASSIGN_OR_RAISE(bool exists, 
snapshot_manager.SnapshotExists(snapshot_id));
+        if (!exists) {
+            pinned.erase(Options::SCAN_SNAPSHOT_ID);
+            return pinned;
+        }
+        pinned.erase(Options::SCAN_TAG_NAME);
+    }
+    pinned[Options::SCAN_SNAPSHOT_ID] = std::to_string(snapshot_id);
+    return pinned;
+}
+
+/// Creates a read of `read_context` with `table_schema` rather than the 
latest schema, as Java
+/// `FilteredRowIdReader` reads with the table that the search loaded, so that 
a column renamed or
+/// replaced since then cannot point the filter at another field.
+Result<std::unique_ptr<TableRead>> CreateReadWithSchema(
+    std::unique_ptr<ReadContext> read_context, const 
std::shared_ptr<TableSchema>& table_schema,
+    const std::map<std::string, std::string>& options, const 
std::shared_ptr<MemoryPool>& pool,
+    const std::shared_ptr<Executor>& executor) {
+    // Row ids come from row tracking, which a primary-key table does not 
support.
+    if (!table_schema->PrimaryKeys().empty()) {
+        return Status::NotImplemented("cannot evaluate a row filter on a 
primary-key table");
+    }
+    PAIMON_ASSIGN_OR_RAISE(
+        std::shared_ptr<InternalReadContext> internal_context,
+        InternalReadContext::Create(std::move(read_context), table_schema, 
options));
+    const CoreOptions& core_options = internal_context->GetCoreOptions();
+    auto arrow_schema = 
DataField::ConvertDataFieldsToArrowSchema(table_schema->Fields());
+    PAIMON_ASSIGN_OR_RAISE(std::vector<std::string> external_paths,
+                           core_options.CreateExternalPaths());
+    PAIMON_ASSIGN_OR_RAISE(std::optional<std::string> 
global_index_external_path,
+                           core_options.CreateGlobalIndexExternalPath());
+    PAIMON_ASSIGN_OR_RAISE(
+        std::shared_ptr<FileStorePathFactory> path_factory,
+        FileStorePathFactory::Create(
+            internal_context->GetPath(), arrow_schema, 
table_schema->PartitionKeys(),
+            core_options.GetPartitionDefaultName(), 
core_options.GetFileFormat()->Identifier(),
+            core_options.DataFilePrefix(), 
core_options.LegacyPartitionNameEnabled(),
+            external_paths, global_index_external_path, 
core_options.IndexFileInDataFileDir(),
+            pool));
+    return std::unique_ptr<TableRead>(
+        std::make_unique<AppendOnlyTableRead>(path_factory, internal_context, 
pool, executor));
+}
+}  // namespace
+
+FilteredRowIdReader::FilteredRowIdReader(
+    const std::string& root_path, const std::shared_ptr<TableSchema>& 
table_schema,
+    const CoreOptions& options, const std::shared_ptr<PredicateFilter>& 
partition_filter,
+    const std::shared_ptr<Predicate>& filter, int64_t snapshot_id,
+    const std::shared_ptr<Executor>& executor, const 
std::shared_ptr<MemoryPool>& pool)
+    : root_path_(root_path),
+      table_schema_(table_schema),
+      options_(options),
+      partition_filter_(partition_filter),
+      filter_(filter),
+      snapshot_id_(snapshot_id),
+      executor_(executor),
+      pool_(pool) {}
+
+Result<bool> FilteredRowIdReader::IsExact(const TableSchema& table_schema,
+                                          const std::shared_ptr<Predicate>& 
filter,
+                                          const std::set<int32_t>& 
contributing_field_ids) {
+    PAIMON_ASSIGN_OR_RAISE(std::set<int32_t> filter_field_ids,
+                           
GlobalIndexEvaluatorImpl::CollectFieldIds(table_schema, filter));
+    if (!std::includes(contributing_field_ids.begin(), 
contributing_field_ids.end(),
+                       filter_field_ids.begin(), filter_field_ids.end())) {
+        return false;
+    }
+    return !HasCandidateOnlyLeaf(filter);
+}
+
+Result<RoaringBitmap64> FilteredRowIdReader::MatchingRowIds(const 
RoaringBitmap64& rows) const {
+    RoaringBitmap64 matching;
+    if (rows.IsEmpty()) {
+        return matching;
+    }
+    std::map<std::string, std::string> pinned_options;
+    PAIMON_ASSIGN_OR_RAISE(pinned_options, PinToSnapshot(root_path_, options_, 
snapshot_id_));
+
+    // Unlike the read, the scan may load the latest schema, as it selects 
files by partition and
+    // row id alone.
+    ScanContextBuilder scan_builder(root_path_);
+    scan_builder.SetOptions(pinned_options)
+        .SetPredicate(partition_filter_)
+        .SetGlobalIndexResult(std::make_shared<BitmapGlobalIndexResult>(
+            [rows]() -> Result<RoaringBitmap64> { return rows; }))
+        .WithFileSystem(options_.GetFileSystem())
+        .WithMemoryPool(pool_)
+        .WithExecutor(executor_);
+    PAIMON_ASSIGN_OR_RAISE(std::unique_ptr<ScanContext> scan_context, 
scan_builder.Finish());
+    PAIMON_ASSIGN_OR_RAISE(std::unique_ptr<TableScan> table_scan,
+                           TableScan::Create(std::move(scan_context)));
+    PAIMON_ASSIGN_OR_RAISE(std::shared_ptr<Plan> plan, 
table_scan->CreatePlan());
+    // An empty plan of another snapshot would drop every candidate, so it is 
checked too.
+    if (plan->SnapshotId() != snapshot_id_) {
+        return Status::Invalid(fmt::format(
+            "cannot evaluate a row filter on snapshot {}, as the data scan 
resolved another one",
+            snapshot_id_));
+    }
+    if (plan->Splits().empty()) {
+        return matching;
+    }
+
+    // The read schema decides the field indexes that a predicate of the read 
must refer to.
+    std::set<std::string> filter_fields;
+    PAIMON_RETURN_NOT_OK(PredicateUtils::GetAllNames(filter_, &filter_fields));
+    std::vector<std::string> read_fields;
+    for (const DataField& field : table_schema_->Fields()) {
+        if (filter_fields.count(field.Name()) > 0) {
+            read_fields.push_back(field.Name());
+        }
+    }
+    read_fields.push_back(SpecialFields::RowId().Name());
+    PAIMON_ASSIGN_OR_RAISE(std::shared_ptr<Predicate> read_filter,
+                           PredicateUtils::CreatePickedFieldFilter(
+                               filter_, 
ObjectUtils::CreateIdentifierToIndexMap(read_fields)));
+
+    ReadContextBuilder read_builder(root_path_);
+    read_builder.SetOptions(pinned_options)

Review Comment:
   Done in 63a3f85a, as the other internal readers do (global index build, blob 
view lookup, compaction). `MatchedRows()` runs on the caller's thread before 
the searches fan out, so the prefetch tasks on the executor cannot deadlock it.
   



-- 
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]

Reply via email to