github-actions[bot] commented on code in PR #66227: URL: https://github.com/apache/doris/pull/66227#discussion_r4061163320
########## be/src/format_v2/table/paimon_rust_predicate_converter.cpp: ########## @@ -0,0 +1,793 @@ +// 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 "format_v2/table/paimon_rust_predicate_converter.h" + +#include <algorithm> +#include <cctype> +#include <memory> +#include <utility> + +#include "common/logging.h" +#include "core/column/column_const.h" +#include "core/column/column_nullable.h" +#include "core/data_type/data_type.h" +#include "core/data_type/data_type_nullable.h" +#include "core/field.h" +#include "core/types.h" +#include "core/value/decimalv2_value.h" +#include "core/value/timestamptz_value.h" +#include "core/value/vdatetime_value.h" +#include "exprs/vcompound_pred.h" +#include "exprs/vdirect_in_predicate.h" +#include "exprs/vectorized_fn_call.h" +#include "exprs/vexpr.h" +#include "exprs/vin_predicate.h" +#include "exprs/vliteral.h" +#include "exprs/vslot_ref.h" + +namespace doris { + +namespace { +// paimon_datum tags (see paimon.h / bindings/c/src/table.rs::datum_from_c). +constexpr int32_t kTagBool = 0; +constexpr int32_t kTagTinyInt = 1; +constexpr int32_t kTagSmallInt = 2; +constexpr int32_t kTagInt = 3; +constexpr int32_t kTagLong = 4; +constexpr int32_t kTagDouble = 6; +constexpr int32_t kTagString = 7; +constexpr int32_t kTagDate = 8; +constexpr int32_t kTagTimestamp = 10; +constexpr int32_t kTagDecimal = 12; +constexpr int32_t kTagBytes = 13; + +// paimon decimal precision ceiling (paimon::Decimal::MAX_PRECISION). +constexpr int32_t kPaimonDecimalMaxPrecision = 38; + +// RAII for an owned paimon_predicate*. and/or/not consume their inputs, so we +// release() before handing pointers to them. +struct predicate_deleter { + void operator()(paimon_predicate* p) const { + if (p) { + paimon_predicate_free(p); + } + } +}; +using predicate_ptr = std::unique_ptr<paimon_predicate, predicate_deleter>; + +// RAII for an owned paimon_error*. +struct error_deleter { + void operator()(paimon_error* p) const { + if (p) { + paimon_error_free(p); + } + } +}; +using error_ptr = std::unique_ptr<paimon_error, error_deleter>; + +// Render a paimon_error into a string. Takes ownership of `err` via RAII so it +// is freed on every return path. Safe to call with nullptr. +std::string consume_predicate_error(paimon_error* err) { + error_ptr owned(err); + if (!owned) { + return "unknown error"; + } + std::string msg; + if (owned->message.data != nullptr && owned->message.len > 0) { + msg.assign(reinterpret_cast<const char*>(owned->message.data), owned->message.len); + } + return "code=" + std::to_string(owned->code) + ", msg=" + msg; +} +} // namespace + +PaimonRustPredicateConverter::PaimonRustPredicateConverter( + const std::vector<std::string>& column_names, const std::vector<DataTypePtr>& column_types, + const paimon_table* table) + : _table(table) { + DORIS_CHECK(column_names.size() == column_types.size()); + _columns_by_name.reserve(column_names.size()); + for (size_t i = 0; i < column_names.size(); ++i) { + _columns_by_name.emplace(_normalize_name(column_names[i]), + std::make_pair(column_names[i], column_types[i])); + } + // Paimon TIMESTAMP (wall clock) is stored as epoch-millis-of-the-wall-time + // and the DateTimeV2 serde decodes timezone-naive arrow values in UTC, so + // timestamp literals convert wall->epoch in UTC. utc_time_zone() needs no + // tzdata lookup, so the conversion cannot silently fall back to a + // machine-local zone. + _utc_tz = cctz::utc_time_zone(); +} + +paimon_predicate* PaimonRustPredicateConverter::build(const VExprContextSPtrs& conjuncts) { + if (_table == nullptr) { + return nullptr; + } + predicate_ptr result; + for (const auto& conjunct : conjuncts) { + if (!conjunct || !conjunct->root()) { + continue; + } + auto root = conjunct->root(); + if (root->is_rf_wrapper()) { + if (auto impl = root->get_impl()) { + root = impl; + } + } + // Preserve a safe prefix of the conjunct order: a later pushed + // predicate (e.g. an arrived IN runtime filter) could otherwise prune + // rows on which an earlier error-preserving conjunct — + // assert_true(...), a failing cast, ... — must still raise. The v1 + // partition-pruning path (FileScanner::_init_runtime_filter_partition_ + // prune_ctxs) stops at is_safe_to_execute_on_selected_rows() for the + // same reason, so a convertible predicate after an unsafe conjunct + // must not be pushed. Safe conjuncts that cannot be converted keep + // the old skip: they cannot raise, so pruning rows before they are + // evaluated as the residual never loses an error. + if (!root->is_safe_to_execute_on_selected_rows()) { + break; + } + predicate_ptr pred(_convert_expr(root)); + if (!pred) { + continue; + } + if (!result) { + result = std::move(pred); + } else { + // and consumes both inputs regardless of success. + result.reset(paimon_predicate_and(result.release(), pred.release())); + if (!result) { + return nullptr; + } + } + } + return result.release(); +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_expr(const VExprSPtr& expr) { + if (!expr) { + return nullptr; + } + + // Casts are not unwrapped anywhere (predicate root included): a cast node + // fails every dispatch below and the conjunct stays in the Doris residual, + // mirroring the FE converter, which keeps casted expressions unconverted. + if (auto* direct_in = dynamic_cast<VDirectInPredicate*>(expr.get())) { + VExprSPtr in_expr; + if (direct_in->get_slot_in_expr(in_expr)) { + return _convert_in(in_expr); + } + return nullptr; + } + + if (dynamic_cast<VInPredicate*>(expr.get()) != nullptr) { + return _convert_in(expr); + } + + switch (expr->op()) { + case TExprOpcode::COMPOUND_AND: + case TExprOpcode::COMPOUND_OR: + return _convert_compound(expr); + case TExprOpcode::COMPOUND_NOT: + return nullptr; + case TExprOpcode::EQ: + case TExprOpcode::EQ_FOR_NULL: + case TExprOpcode::NE: + case TExprOpcode::GE: + case TExprOpcode::GT: + case TExprOpcode::LE: + case TExprOpcode::LT: + return _convert_binary(expr); + default: + break; + } + + if (auto* fn = dynamic_cast<VectorizedFnCall*>(expr.get())) { + auto fn_name = _normalize_name(fn->function_name()); + if (fn_name == "is_null_pred" || fn_name == "is_not_null_pred") { + return _convert_is_null(expr, fn_name); + } + if (fn_name == "like") { + return _convert_like_prefix(expr); + } + } + + return nullptr; +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_compound(const VExprSPtr& expr) { + if (!expr || expr->get_num_children() != 2) { + return nullptr; + } + predicate_ptr left(_convert_expr(expr->get_child(0))); + if (!left) { + return nullptr; + } + predicate_ptr right(_convert_expr(expr->get_child(1))); + if (!right) { + return nullptr; + } + + if (expr->op() == TExprOpcode::COMPOUND_AND) { + return paimon_predicate_and(left.release(), right.release()); + } + if (expr->op() == TExprOpcode::COMPOUND_OR) { + return paimon_predicate_or(left.release(), right.release()); + } + return nullptr; +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_in(const VExprSPtr& expr) { + auto* in_pred = dynamic_cast<VInPredicate*>(expr.get()); + if (!in_pred || expr->get_num_children() < 2) { + return nullptr; + } + auto field_meta = _resolve_field(expr->get_child(0)); + if (!field_meta) { + return nullptr; + } + + const auto num_values = expr->get_num_children() - 1; + // Reserve up front so the backing strings never reallocate: each datum's + // str_data points into storages[i], which must stay stable. + std::vector<std::string> storages; + std::vector<paimon_datum> datums; + storages.reserve(num_values); + datums.reserve(num_values); + for (uint16_t i = 1; i < expr->get_num_children(); ++i) { + // Mirror FE's doInPredicate, which only accepts bare LiteralExpr + // children: a casted child would be unwrapped to its pre-cast value + // (debug_skip_fold_constant keeps such casts un-folded in the plan), + // so Doris would compare against the cast result while rust filters + // on the raw value — e.g. in `amount IN (CAST(1.24 AS DECIMAL(10,1)))` + // Doris keeps the 1.2 rows and the unwrapped 1.24 push would remove + // them permanently. Reject the whole predicate; the residual applies + // the cast correctly. + if (expr->get_child(i)->node_type() == TExprNodeType::CAST_EXPR) { + return nullptr; + } + auto holder = _convert_literal(expr->get_child(i), field_meta->type); + if (!holder) { + return nullptr; + } + storages.emplace_back(std::move(holder->storage)); + paimon_datum datum = holder->datum; + _bind_datum_storage(&datum, storages.back()); + datums.emplace_back(datum); + } + + if (datums.empty()) { + return nullptr; + } + if (in_pred->is_not_in()) { + return _take(paimon_predicate_is_not_in(_table, field_meta->column.c_str(), datums.data(), + datums.size())); + } + return _take(paimon_predicate_is_in(_table, field_meta->column.c_str(), datums.data(), + datums.size())); +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_binary(const VExprSPtr& expr) { + if (!expr || expr->get_num_children() != 2) { + return nullptr; + } + auto field_meta = _resolve_field(expr->get_child(0)); + if (!field_meta) { + return nullptr; + } + const char* column = field_meta->column.c_str(); + + // Convert the RHS first so EQ_FOR_NULL (<=>) only converts when the RHS is + // a convertible literal, mirroring the FE converter, which rejects a + // non-literal RHS. A column-to-column `a <=> b` must therefore stay in the + // Doris residual: it has no single-column rust predicate, and pushing + // `a IS NULL` would wrongly discard rows like (1, 1) — rows dropped by the + // pushed filter cannot be recovered by the residual conjunct. + auto holder = _convert_literal(expr->get_child(1), field_meta->type); + if (!holder) { + return nullptr; + } + + if (expr->op() == TExprOpcode::EQ_FOR_NULL) { + return _take(paimon_predicate_is_null(_table, column)); + } + + // `holder` is a local, so its storage stays put for the duration of the call. + _bind_datum_storage(&holder->datum, holder->storage); + const paimon_datum& datum = holder->datum; + + switch (expr->op()) { + case TExprOpcode::EQ: + return _take(paimon_predicate_equal(_table, column, datum)); + case TExprOpcode::NE: + return _take(paimon_predicate_not_equal(_table, column, datum)); + case TExprOpcode::GE: + return _take(paimon_predicate_greater_or_equal(_table, column, datum)); + case TExprOpcode::GT: + return _take(paimon_predicate_greater_than(_table, column, datum)); + case TExprOpcode::LE: + return _take(paimon_predicate_less_or_equal(_table, column, datum)); + case TExprOpcode::LT: + return _take(paimon_predicate_less_than(_table, column, datum)); + default: + break; + } + return nullptr; +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_is_null(const VExprSPtr& expr, + const std::string& fn_name) { + if (!expr || expr->get_num_children() != 1) { + return nullptr; + } + auto field_meta = _resolve_field(expr->get_child(0)); + if (!field_meta) { + return nullptr; + } + if (fn_name == "is_not_null_pred") { + return _take(paimon_predicate_is_not_null(_table, field_meta->column.c_str())); + } + return _take(paimon_predicate_is_null(_table, field_meta->column.c_str())); +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_like_prefix(const VExprSPtr& expr) { + if (!expr || expr->get_num_children() != 2) { + return nullptr; + } + auto field_meta = _resolve_field(expr->get_child(0)); + if (!field_meta || !_is_string_type(field_meta->type->get_primitive_type())) { + return nullptr; + } + + auto pattern_opt = _extract_string_literal(expr->get_child(1)); + if (!pattern_opt) { + return nullptr; + } + const std::string& pattern = *pattern_opt; + // Only prefix matches (`abc%`) are convertible to a range scan. + if (!pattern.empty() && pattern.front() == '%') { + return nullptr; + } + if (pattern.empty() || pattern.back() != '%') { + return nullptr; + } + + const char* column = field_meta->column.c_str(); + std::string prefix = pattern.substr(0, pattern.size() - 1); + + // lower bound: column >= prefix + paimon_datum lower {}; + lower.tag = kTagString; + _bind_datum_storage(&lower, prefix); + predicate_ptr lower_pred(_take(paimon_predicate_greater_or_equal(_table, column, lower))); + if (!lower_pred) { + return nullptr; + } + + auto upper_prefix = _next_prefix(prefix); + if (!upper_prefix) { + return lower_pred.release(); + } + + // upper bound: column < next_prefix + paimon_datum upper {}; + upper.tag = kTagString; + _bind_datum_storage(&upper, *upper_prefix); + predicate_ptr upper_pred(_take(paimon_predicate_less_than(_table, column, upper))); + if (!upper_pred) { + // No usable upper bound: fall back to the (still correct) lower bound. + return lower_pred.release(); + } + return paimon_predicate_and(lower_pred.release(), upper_pred.release()); +} + +std::optional<PaimonRustPredicateConverter::FieldMeta> PaimonRustPredicateConverter::_resolve_field( + const VExprSPtr& expr) const { + if (!expr) { + return std::nullopt; + } + // Mirror the FE converter's convertDorisExprToSlotRef: a casted column is + // rejected, never unwrapped. Stripping a lossy cast changes which rows match + // — for a DECIMAL(10,2) column, CAST(amount AS DECIMAL(10,1)) = 1.2 keeps + // the row 1.24 while the unwrapped `amount = 1.2` prunes it — and rows + // pruned by the pushed filter cannot be recovered by the Doris residual. + // The conjunct stays in the residual instead. + auto* slot_ref = dynamic_cast<VSlotRef*>(expr.get()); + if (!slot_ref) { + return std::nullopt; + } + // FileScannerV2 rewrites conjunct VSlotRefs to table global indices, so slot_id + // is a position, not a slot id; resolve by the carried column name against the + // projected-column registry instead of the desc table. + auto it = _columns_by_name.find(_normalize_name(slot_ref->column_name())); + if (it == _columns_by_name.end()) { + return std::nullopt; + } + const auto& [column, type] = it->second; + if (!_is_supported_slot_type(type->get_primitive_type(), type->get_precision())) { + return std::nullopt; + } + return FieldMeta {column, type}; +} + +std::optional<PaimonRustPredicateConverter::DatumHolder> +PaimonRustPredicateConverter::_convert_literal(const VExprSPtr& expr, + const DataTypePtr& column_type) const { + // Mirror the FE converter's convertDorisExprToLiteralExpr: a bare literal + // or a single cast wrapping a direct literal converts; anything deeper is + // rejected. Unwrapping recursively would silently apply the inner casts' + // lossy semantics (e.g. a DECIMAL scale reduction), which FE also rejects + // (its instanceof check only unwraps one CastExpr around a LiteralExpr). + VExprSPtr literal_expr = expr; + if (expr->node_type() == TExprNodeType::CAST_EXPR) { Review Comment: [P1] Reject or evaluate casted RHS literals in binary predicates. This shared literal helper strips a one-level CAST without executing it; with constant folding disabled, a `DECIMAL(10,2)` predicate `amount = CAST(1.24 AS DECIMAL(10,1))` is evaluated by Doris against `1.20`, while Rust receives the raw `1.24` and can discard the matching `1.20` row before the residual runs. The IN-list path now rejects this exact unsafe shape, but `_convert_binary` still reaches it, and `SingleCastLiteralIsStillPushed` locks the behavior in. Please keep casted RHS values residual unless the cast result is evaluated exactly, and cover a scale-reducing binary differential case. ########## fe/fe-core/src/main/java/org/apache/doris/datasource/paimon/source/PaimonScanNode.java: ########## @@ -411,10 +413,138 @@ private void setPaimonParams(TFileRangeDesc rangeDesc, PaimonSplit paimonSplit) String fileFormat = getFileFormat(paimonSplit.getPathString()); Review Comment: [P1] Gate ORC `TIMESTAMP_LTZ` off the Rust reader. The new differential suite explicitly says the pinned paimon-rust ORC decoder shifts LTZ values by the writer timezone and therefore tests only the Parquet twin, but this already-computed file format is not used by `canUseRust`. A logical ORC `DataSplit` (for example with `force_jni_scanner=true` or when raw conversion is unavailable) therefore selects Rust and returns a different instant from JNI; applying the session timezone in BE cannot repair an epoch shifted during decode. Keep ORC schemas containing `TIMESTAMP_WITH_LOCAL_TIME_ZONE` on JNI until the decoder is fixed, and add the existing ORC fixture to the differential coverage. ########## be/src/format_v2/table/paimon_rust_table_reader.cpp: ########## @@ -0,0 +1,830 @@ +// 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 "format_v2/table/paimon_rust_table_reader.h" + +#include <algorithm> +#include <utility> + +#include "arrow/c/abi.h" +#include "arrow/c/bridge.h" +#include "arrow/record_batch.h" +#include "arrow/result.h" +#include "common/logging.h" +#include "core/block/block.h" +#include "core/block/column_with_type_and_name.h" +#include "core/column/column_const.h" +#include "exprs/vexpr_context.h" +#include "exprs/vliteral.h" +#include "format_v2/column_mapper.h" +#include "format_v2/table/paimon_rust_predicate_converter.h" +#include "runtime/descriptors.h" +#include "runtime/file_scan_profile.h" +#include "runtime/runtime_state.h" +#include "util/string_util.h" +#include "util/timezone_utils.h" +#include "util/url_coding.h" + +extern "C" { +#include "paimon_rust/paimon.h" +} + +namespace doris::format::paimon { + +namespace { +constexpr const char* VALUE_KIND_FIELD = "_VALUE_KIND"; + +// --------------------------------------------------------------------------- +// RAII wrappers over the paimon-rust C handles. Each handle is an opaque +// pointer owned by Rust and released by a matching paimon_*_free function. +// --------------------------------------------------------------------------- +#define PAIMON_OWNED(type, freefn) \ + struct type##_deleter { \ + void operator()(paimon_##type* p) const { \ + if (p) { \ + freefn(p); \ + } \ + } \ + }; \ + using type##_ptr = std::unique_ptr<paimon_##type, type##_deleter> + +PAIMON_OWNED(table, paimon_table_free); +PAIMON_OWNED(read_builder, paimon_read_builder_free); +PAIMON_OWNED(plan, paimon_plan_free); +PAIMON_OWNED(table_read, paimon_table_read_free); +PAIMON_OWNED(record_batch_reader, paimon_record_batch_reader_free); +PAIMON_OWNED(error, paimon_error_free); + +#undef PAIMON_OWNED + +// One Arrow batch (schema + array containers). Owning it requires a two-step +// teardown that the unique_ptr deleters above can't express: first invoke the +// Arrow C Data Interface `release` callback on each struct (hands buffers back +// to the producer), then free the container structs via paimon_arrow_batch_free. +class ArrowBatch { +public: + explicit ArrowBatch(paimon_arrow_batch batch) : batch_(batch) {} + ~ArrowBatch() { + auto* schema = static_cast<ArrowSchema*>(batch_.schema); + auto* array = static_cast<ArrowArray*>(batch_.array); + if (array && array->release) { + array->release(array); + } + if (schema && schema->release) { + schema->release(schema); + } + paimon_arrow_batch_free(batch_); + } + + ArrowBatch(const ArrowBatch&) = delete; + ArrowBatch& operator=(const ArrowBatch&) = delete; + + ArrowSchema* schema() const { return static_cast<ArrowSchema*>(batch_.schema); } + ArrowArray* array() const { return static_cast<ArrowArray*>(batch_.array); } + +private: + paimon_arrow_batch batch_; +}; + +// Render a paimon_error into a string. Takes ownership of `err` via RAII so it +// is freed on every return path. Safe to call with nullptr. +std::string consume_error(paimon_error* err) { + error_ptr owned(err); + if (!owned) { + return "unknown error"; + } + std::string msg; + if (owned->message.data != nullptr && owned->message.len > 0) { + msg.assign(reinterpret_cast<const char*>(owned->message.data), owned->message.len); + } + return "code=" + std::to_string(owned->code) + ", msg=" + msg; +} + +// Render storage option KEYS for diagnostics. Values are never rendered: +// credential keys arrive under many spellings and cases (AWS_SECRET_KEY, +// AWS_TOKEN, fs.oss.accessKeySecret, s3.secret-key, ...), and a key-name +// blocklist that misses one alias leaks the value into the INFO log, so +// only the key names are printed at all. +std::string format_options(const std::map<std::string, std::string>& options) { + std::string out; + for (const auto& kv : options) { + if (!out.empty()) { + out += ", "; + } + out += kv.first; + } + return out; +} + +} // namespace + +// Paimon-rust handles. Order of members matters: destruction runs in reverse +// declaration order, and the read_builder depends on the table while the arrow +// reader depends on the whole pipeline above it. So the table MUST be declared +// first (destroyed last) and the record batch reader last. +struct PaimonRustTableReader::PaimonHandles { + table_ptr table; + read_builder_ptr read_builder; + plan_ptr plan; + table_read_ptr table_read; + record_batch_reader_ptr reader; +}; + +PaimonRustTableReader::PaimonRustTableReader() = default; + +PaimonRustTableReader::~PaimonRustTableReader() = default; + +Status PaimonRustTableReader::init(format::TableReadOptions&& options) { + RETURN_IF_ERROR(format::TableReader::init(std::move(options))); + { + // Base and derived scopes must not overlap on the same counter: RuntimeProfile timers + // add deltas, so nested use would double-count instead of extending lifecycle coverage. + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.init_timer); + // Materialize TIMESTAMP_LTZ in the session timezone — the same + // convention as the JNI reader (PaimonJniScanner reads time_zone from + // its scan params) and lance_reader. Timezone-naive (paimon TIMESTAMP) + // arrow values are decoded in UTC by the DateTimeV2 serde regardless + // of _ctz, so NTZ wall-clock semantics are preserved. + DORIS_CHECK(_runtime_state != nullptr); + _ctz = _runtime_state->timezone_obj(); + if (_scanner_profile != nullptr) { + file_scan_profile::ensure_hierarchy(_scanner_profile); + _rust_total_time = ADD_CHILD_TIMER(_scanner_profile, "PaimonRustReader", + file_scan_profile::TABLE_READER); + _rust_open_split_time = + ADD_CHILD_TIMER(_scanner_profile, "OpenSplitTime", "PaimonRustReader"); + _rust_read_batch_time = + ADD_CHILD_TIMER(_scanner_profile, "ReadBatchTime", "PaimonRustReader"); + _rust_arrow_to_block_time = + ADD_CHILD_TIMER(_scanner_profile, "ArrowToBlockTime", "PaimonRustReader"); + } + // Projected column name -> fixed output position, registered with both the exact and + // the lower-case spelling so mixed-case Rust schema output still resolves (v1 + // semantics: exact match first, lower-case fallback on lookup). + _output_name_to_idx.reserve(_projected_columns.size() * 2); + for (size_t idx = 0; idx < _projected_columns.size(); ++idx) { + _output_name_to_idx.emplace(_projected_columns[idx].name, idx); + _output_name_to_idx.emplace(to_lower(_projected_columns[idx].name), idx); + } + } + return Status::OK(); +} + +Status PaimonRustTableReader::prepare_split(const format::SplitReadOptions& options) { + // EOF belongs to the previous split. Keep it set after closing that split so repeated reads + // are idempotent, and clear it only when a new split is explicitly prepared. + _close_split_reader(); + _split_eof = false; + _current_range = options.current_range; + RETURN_IF_ERROR(format::TableReader::prepare_split(options)); + if (current_split_pruned()) { + return Status::OK(); + } + if (_is_table_level_count_active()) { + // No rust pipeline is opened; get_block emits the synthetic count rows. + return Status::OK(); + } + RETURN_IF_ERROR(_validate_rust_split(options.current_range)); + { + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.prepare_split_timer); + SCOPED_TIMER(_rust_open_split_time); + RETURN_IF_ERROR(_open_split_reader(options.current_range)); + } + return Status::OK(); +} + +Status PaimonRustTableReader::get_block(Block* block, bool* eos) { + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.exec_timer); + SCOPED_TIMER(_rust_total_time); + DORIS_CHECK(block != nullptr); + DORIS_CHECK(eos != nullptr); + DORIS_CHECK(block->columns() == _projected_columns.size()); + block->clear_column_data(_projected_columns.size()); + *eos = false; + + if (_is_table_level_count_active()) { + return _read_table_level_count(block, eos); + } + + // num_splits == 0 yields an empty (but valid) stream: report EOF. + if (_split_eof) { + *eos = true; + return Status::OK(); + } + if (!_handles || !_handles->reader) { + return Status::InternalError("paimon-rust reader is not initialized"); + } + + while (true) { + // Mirror the base TableReader cancellation contract so a cancelled query does not + // drain the whole split. + if (_io_ctx != nullptr && _io_ctx->should_stop) { + _split_eof = true; + _close_split_reader(); + *eos = true; + return Status::OK(); + } + + paimon_result_next_batch next; + { + SCOPED_TIMER(_rust_read_batch_time); + next = paimon_record_batch_reader_next(_handles->reader.get()); + } + if (next.error != nullptr) { + return Status::InternalError("paimon-rust read batch failed: {}", + consume_error(next.error)); + } + // End of stream: both pointers are null. + if (next.batch.array == nullptr && next.batch.schema == nullptr) { + _split_eof = true; + _close_split_reader(); + *eos = true; + return Status::OK(); + } + + // RAII: the batch's Arrow release callbacks + container free run when + // `batch` leaves this scope, including on any early return. + ArrowBatch batch(next.batch); + + auto* c_array = batch.array(); + auto* c_schema = batch.schema(); + arrow::Result<std::shared_ptr<arrow::RecordBatch>> import_result = + arrow::ImportRecordBatch(c_array, c_schema); + if (!import_result.ok()) { + return Status::InternalError("failed to import paimon-rust arrow batch: {}", + import_result.status().message()); + } + + auto record_batch = std::move(import_result).ValueUnsafe(); + const auto rows = static_cast<size_t>(record_batch->num_rows()); + if (rows == 0) { + // Skip empty batches and keep draining the stream. + continue; + } + RETURN_IF_ERROR(_fill_block_from_record_batch(record_batch, block, rows)); + _record_scan_rows(rows); + *eos = false; + return Status::OK(); + } +} + +Status PaimonRustTableReader::abort_split() { + { + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.close_timer); + _close_split_reader(); + _split_eof = false; + } + return format::TableReader::abort_split(); +} + +#ifdef BE_TEST +std::string PaimonRustTableReader::TEST_format_options( + const std::map<std::string, std::string>& options) { + return format_options(options); +} + +std::map<std::string, std::string> PaimonRustTableReader::TEST_build_options( + TFileScanRangeParams* scan_params, const TFileRangeDesc& range) { + TFileScanRangeParams* previous_params = _scan_params; + TFileRangeDesc previous_range = _current_range; + _scan_params = scan_params; + _current_range = range; + std::map<std::string, std::string> options = _build_options(); + _scan_params = previous_params; + _current_range = std::move(previous_range); + return options; +} +#endif + +Status PaimonRustTableReader::close() { + { + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.close_timer); + _close_split_reader(); + _close_table(); + } + return format::TableReader::close(); +} + +Status PaimonRustTableReader::_validate_rust_split(const TFileRangeDesc& range) const { + if (!range.__isset.table_format_params || !range.table_format_params.__isset.paimon_params) { + return Status::InternalError( + "missing paimon_params for paimon rust reader, possibly caused by FE/BE protocol " + "mismatch"); + } + const auto& params = range.table_format_params.paimon_params; + if (!params.__isset.paimon_split || params.paimon_split.empty()) { + return Status::InternalError( + "missing paimon_split for paimon rust reader, possibly caused by FE/BE protocol " + "mismatch"); + } + if (params.__isset.reader_type && params.reader_type != TPaimonReaderType::PAIMON_RUST) { + return Status::InternalError( + "invalid reader_type for paimon rust reader, possibly caused by FE/BE protocol " + "mismatch"); + } + if (!_resolve_table_path(range).has_value()) { + return Status::InternalError( + "paimon-rust missing paimon_table; cannot resolve paimon table location"); + } + if (!_resolve_db_name(range).has_value()) { + return Status::InternalError( + "paimon-rust missing db_name; cannot open paimon table via schema json"); + } + if (!_resolve_table_name(range).has_value()) { + return Status::InternalError( + "paimon-rust missing table_name; cannot open paimon table via schema json"); + } + if (!_resolve_table_schema_json(range).has_value()) { + return Status::InternalError( + "paimon-rust missing paimon_table_schema_json; cannot open paimon table via " + "schema json"); + } + return Status::OK(); +} + +Status PaimonRustTableReader::_open_split_reader(const TFileRangeDesc& range) { + // 1. Decode the FE-planned split first so we fail fast (and without any + // filesystem IO) when it is missing or malformed. + std::string split_bytes; + RETURN_IF_ERROR(_decode_split_bytes(&split_bytes)); + + // 2. Resolve identifier + table_path + FE-supplied TableSchema JSON. + auto table_path = _resolve_table_path(range).value(); + auto db_name = _resolve_db_name(range).value(); + auto table_name = _resolve_table_name(range).value(); + auto schema_json = _resolve_table_schema_json(range).value(); + auto branch_opt = _resolve_branch(range); + + // 3. Assemble storage options: FE-supplied paimon options + hadoop_conf + + // OSS/S3 → AWS_* translations. These feed FileIO only (per + // paimon_table_from_schema_json contract); they are NOT merged into the + // supplied table schema. + auto options = _build_options(); + + auto opened_table_key = + std::make_tuple(table_path, schema_json, db_name, table_name, branch_opt, options); + if (!_handles || !_handles->table || _opened_table_key != opened_table_key) { + // A paimon scan reads one table, so the handle is opened at most once per + // distinct identity (e.g. re-created after a close); splits of the same + // table reuse it and only rebuild the read pipeline below. + _close_table(); + _handles = std::make_unique<PaimonHandles>(); + + std::vector<paimon_option> c_options; + c_options.reserve(options.size()); + for (const auto& kv : options) { + c_options.push_back(paimon_option {kv.first.c_str(), kv.second.c_str()}); + } + + LOG(INFO) << "paimon-rust opening table via schema json: db=" << db_name + << " table=" << table_name << " path=" << table_path + << " branch=" << (branch_opt.has_value() ? branch_opt.value() : "main") + << " storage_options=[" << format_options(options) << "]"; + + // Build the table directly from the FE-supplied schema JSON. The Rust + // side rejects null / empty branch, so we default to paimon's canonical + // "main" sentinel when FE did not set paimon_branch (i.e. the table is + // on the main branch — matches upstream Identifier.DEFAULT_MAIN_BRANCH). + const std::string& branch_str = branch_opt.has_value() ? branch_opt.value() : "main"; + paimon_result_get_table tbl_res = paimon_table_from_schema_json( + table_path.c_str(), schema_json.c_str(), db_name.c_str(), table_name.c_str(), + branch_str.c_str(), c_options.empty() ? nullptr : c_options.data(), + c_options.size()); + if (tbl_res.error != nullptr) { + return Status::InternalError( + "paimon-rust table_from_schema_json failed: db={} table={} err={}", db_name, + table_name, consume_error(tbl_res.error)); + } + _handles->table.reset(tbl_res.table); + _opened_table_key = std::move(opened_table_key); + } + + // 4. Build the read pipeline: read_builder -> case-insensitive -> projection. + paimon_result_read_builder rb_res = paimon_table_new_read_builder(_handles->table.get()); + if (rb_res.error != nullptr) { + return Status::InternalError("paimon-rust new read builder failed: {}", + consume_error(rb_res.error)); + } + _handles->read_builder.reset(rb_res.read_builder); + + // Fold column casing on the Rust side so FE-normalized lowercase names + // resolve against tables with mixed-case column definitions. + if (paimon_error* case_err = + paimon_read_builder_with_case_sensitive(_handles->read_builder.get(), false)) { + return Status::InternalError("paimon-rust set case_sensitive failed: {}", + consume_error(case_err)); + } + + // Partition keys are excluded: they are materialized from split metadata + // (see _fill_non_arrow_columns), and paimon-rust does not emit them. + auto read_columns = _build_read_columns(); + std::vector<const char*> projection; + projection.reserve(read_columns.size() + 1); + for (const auto& col : read_columns) { + projection.push_back(col.c_str()); + } + projection.push_back(nullptr); + if (paimon_error* proj_err = paimon_read_builder_with_projection(_handles->read_builder.get(), + projection.data())) { + return Status::InternalError("paimon-rust set projection failed: {}", + consume_error(proj_err)); + } + + // Convert the scanner conjuncts into a paimon-rust filter and apply it. + RETURN_IF_ERROR(_apply_predicate()); + + // 5. Deserialize the FE-planned split into a one-split plan, so this + // scanner reads exactly the split it was assigned rather than replanning + // the whole table. The wire form is identical to what paimon-cpp consumes + // (`paimon::table::DataSplit::serialize`). + paimon_result_plan plan_res = paimon_plan_from_split_bytes( + reinterpret_cast<const uint8_t*>(split_bytes.data()), split_bytes.size()); + if (plan_res.error != nullptr) { + return Status::InternalError("paimon-rust build plan failed: {}", + consume_error(plan_res.error)); + } + _handles->plan.reset(plan_res.plan); + + size_t num_splits = paimon_plan_num_splits(_handles->plan.get()); + if (num_splits == 0) { + _split_eof = true; + return Status::OK(); + } + + // 6. Open the arrow stream over the plan. + paimon_result_new_read read_res = paimon_read_builder_new_read(_handles->read_builder.get()); + if (read_res.error != nullptr) { + return Status::InternalError("paimon-rust new read failed: {}", + consume_error(read_res.error)); + } + _handles->table_read.reset(read_res.read); + + paimon_result_record_batch_reader rdr_res = paimon_table_read_to_arrow( + _handles->table_read.get(), _handles->plan.get(), /*offset=*/0, /*length=*/num_splits); + if (rdr_res.error != nullptr) { + return Status::InternalError("paimon-rust open arrow reader failed: {}", + consume_error(rdr_res.error)); + } + _handles->reader.reset(rdr_res.reader); + return Status::OK(); +} + +void PaimonRustTableReader::_close_split_reader() { + if (!_handles) { + return; + } + // Reverse of the declaration order in PaimonHandles. + _handles->reader.reset(); + _handles->table_read.reset(); + _handles->plan.reset(); + _handles->read_builder.reset(); +} + +void PaimonRustTableReader::_close_table() { + if (!_handles) { + return; + } + _close_split_reader(); + _handles->table.reset(); + _opened_table_key.reset(); +} + +Status PaimonRustTableReader::_apply_predicate() { + if (_conjuncts.empty() || !_handles || !_handles->table || !_handles->read_builder) { + return Status::OK(); + } + LOG(INFO) << "paimon-rust predicate pushdown: " << _conjuncts.size() << " conjunct(s) input"; + // The conjunct VSlotRefs carry table global indices (positions), so the v2 + // converter mode resolves fields by the projected column names; partition + // keys are excluded because the rust reader does not read them. + std::vector<std::string> names; + std::vector<DataTypePtr> types; + names.reserve(_projected_columns.size()); + types.reserve(_projected_columns.size()); + for (const auto& col : _projected_columns) { + if (col.is_partition_key) { + continue; + } + names.push_back(col.name); + types.push_back(col.type); + } + PaimonRustPredicateConverter converter(names, types, _handles->table.get()); + paimon_predicate* predicate = converter.build(_conjuncts); + if (predicate == nullptr) { + LOG(INFO) << "paimon-rust predicate pushdown: nothing convertible, no filter applied"; + return Status::OK(); + } + // paimon_read_builder_with_filter consumes the predicate (ownership moves to + // the builder) on every path, so we must not free it here. + if (paimon_error* err = + paimon_read_builder_with_filter(_handles->read_builder.get(), predicate)) { + return Status::InternalError("paimon-rust apply filter failed: {}", consume_error(err)); + } + LOG(INFO) << "paimon-rust predicate pushdown: applied"; + return Status::OK(); +} + +Status PaimonRustTableReader::_fill_block_from_record_batch( + const std::shared_ptr<arrow::RecordBatch>& batch, Block* block, size_t rows) { + SCOPED_TIMER(_rust_arrow_to_block_time); + DORIS_CHECK(batch != nullptr); + DORIS_CHECK(block != nullptr); + std::unordered_set<size_t> materialized_indices; + materialized_indices.reserve(_projected_columns.size()); + { + auto columns_guard = block->mutate_columns_scoped(); + auto& columns = columns_guard.mutable_columns(); + for (int c = 0; c < batch->num_columns(); ++c) { + const auto& field = batch->schema()->field(c); + if (field->name() == VALUE_KIND_FIELD) { + continue; + } + // Projected column names are FE-normalized to lowercase. + // paimon-rust's case_sensitive=false setting also case-folds column + // names in the schema output, so exact match works — but tolerate + // mixed-case Rust output by folding here as well. + auto it = _output_name_to_idx.find(field->name()); + if (it == _output_name_to_idx.end()) { + it = _output_name_to_idx.find(to_lower(field->name())); + } + if (it == _output_name_to_idx.end()) { + // Skip columns that are not in the block (e.g. columns dropped by + // slot pruning). + continue; + } + const auto output_idx = it->second; + if (!materialized_indices.emplace(output_idx).second) { + return Status::InternalError("paimon-rust returned duplicate column '{}'", + field->name()); + } + try { + RETURN_IF_ERROR(columns_guard.get_datatype_by_position(output_idx) + ->get_serde() + ->read_column_from_arrow(*columns[output_idx], + batch->column(c).get(), 0, rows, + _ctz)); + } catch (Exception& e) { + return Status::InternalError("Failed to convert from arrow to block: {}", e.what()); + } + } + } + // Partition columns and other projected columns absent from the arrow batch + // are back-filled from split metadata / defaults. + return _fill_non_arrow_columns(block, rows, materialized_indices); +} + +Status PaimonRustTableReader::_fill_non_arrow_columns( + Block* block, size_t rows, const std::unordered_set<size_t>& materialized_indices) { + for (size_t idx = 0; idx < _projected_columns.size(); ++idx) { + if (materialized_indices.count(idx) != 0) { + continue; + } + const auto& column = _projected_columns[idx]; + VExprContextSPtr constant_expr; + if (const Field* value = find_partition_value(column, _partition_values); + column.is_partition_key && value != nullptr) { + // Partition values are split constants (same materialization the + // TableColumnMapper builds for native readers). + constant_expr = + VExprContext::create_shared(VLiteral::create_shared(column.type, *value)); + } else if (column.default_expr != nullptr) { + constant_expr = column.default_expr; + } else { + // The column is genuinely absent from the arrow batch. Schema + // evolution is handled by paimon-rust itself, so reaching here means + // an unexpected schema drift: fill defaults so the scan remains + // well-defined instead of failing the query. + LOG(WARNING) << "paimon-rust did not return projected column '" << column.name + << "'; filling with defaults"; + auto data = column.type->create_column(); + data->insert_many_defaults(rows); + block->replace_by_position(idx, std::move(data)); + continue; + } + ColumnPtr constant_column; + RETURN_IF_ERROR(_materialize_constant_column(constant_expr, column.type, column.name, rows, + &constant_column)); + block->replace_by_position(idx, std::move(constant_column)); + } + return Status::OK(); +} + +Status PaimonRustTableReader::_materialize_constant_column(const VExprContextSPtr& expr, + const DataTypePtr& type, + const std::string& name, size_t rows, + ColumnPtr* column) { + DORIS_CHECK(expr != nullptr); + DORIS_CHECK(column != nullptr); + RowDescriptor row_desc; + RETURN_IF_ERROR(expr->prepare(_runtime_state, row_desc)); + RETURN_IF_ERROR(expr->open(_runtime_state)); + // Constants evaluate per input row, so a rows-sized synthetic block yields a + // rows-sized result for both plain literals and default expressions. + Block eval_block; + eval_block.insert({type->create_column_const_with_default_value(rows), type, name}); + int result_column_id = -1; + RETURN_IF_ERROR(expr->execute(&eval_block, &result_column_id)); + DORIS_CHECK(result_column_id >= 0); + ColumnPtr result_column = eval_block.get_by_position(result_column_id).column; + if (result_column->size() == 1 && rows > 1) { + result_column = ColumnConst::create(std::move(result_column), rows); + } + *column = std::move(result_column); + return Status::OK(); +} + +Status PaimonRustTableReader::_decode_split_bytes(std::string* out) const { + if (!_current_range.__isset.table_format_params || + !_current_range.table_format_params.__isset.paimon_params || + !_current_range.table_format_params.paimon_params.__isset.paimon_split) { + return Status::InternalError("paimon-rust missing paimon_split in scan range"); + } + const auto& encoded_split = _current_range.table_format_params.paimon_params.paimon_split; + if (!base64_decode(encoded_split, out)) { + return Status::InternalError("paimon-rust base64 decode paimon_split failed"); + } + if (out->empty()) { + return Status::InternalError("paimon-rust decoded paimon_split is empty"); + } + return Status::OK(); +} + +std::optional<std::string> PaimonRustTableReader::_resolve_table_path( + const TFileRangeDesc& range) const { + if (range.__isset.table_format_params && range.table_format_params.__isset.paimon_params && + range.table_format_params.paimon_params.__isset.paimon_table && + !range.table_format_params.paimon_params.paimon_table.empty()) { + return range.table_format_params.paimon_params.paimon_table; + } + return std::nullopt; +} + +std::optional<std::string> PaimonRustTableReader::_resolve_db_name( + const TFileRangeDesc& range) const { + if (range.__isset.table_format_params && range.table_format_params.__isset.paimon_params && + range.table_format_params.paimon_params.__isset.db_name && + !range.table_format_params.paimon_params.db_name.empty()) { + return range.table_format_params.paimon_params.db_name; + } + return std::nullopt; +} + +std::optional<std::string> PaimonRustTableReader::_resolve_table_name( + const TFileRangeDesc& range) const { + if (range.__isset.table_format_params && range.table_format_params.__isset.paimon_params && + range.table_format_params.paimon_params.__isset.table_name && + !range.table_format_params.paimon_params.table_name.empty()) { + return range.table_format_params.paimon_params.table_name; + } + return std::nullopt; +} + +std::optional<std::string> PaimonRustTableReader::_resolve_table_schema_json( + const TFileRangeDesc& range) const { + if (range.__isset.table_format_params && range.table_format_params.__isset.paimon_params && + range.table_format_params.paimon_params.__isset.paimon_table_schema_json && + !range.table_format_params.paimon_params.paimon_table_schema_json.empty()) { + return range.table_format_params.paimon_params.paimon_table_schema_json; + } + return std::nullopt; +} + +std::optional<std::string> PaimonRustTableReader::_resolve_branch( + const TFileRangeDesc& range) const { + // FE only sets paimon_branch when the branch is not `main` (matches + // upstream paimon commit 742da63: null-if-DEFAULT_MAIN_BRANCH). Unset here + // means main-branch semantics. + if (range.__isset.table_format_params && range.table_format_params.__isset.paimon_params && + range.table_format_params.paimon_params.__isset.paimon_branch && + !range.table_format_params.paimon_params.paimon_branch.empty()) { + return range.table_format_params.paimon_params.paimon_branch; + } + return std::nullopt; +} + +std::vector<std::string> PaimonRustTableReader::_build_read_columns() const { + std::vector<std::string> columns; + columns.reserve(_projected_columns.size()); + for (const auto& column : _projected_columns) { + if (column.is_partition_key) { + continue; + } + columns.emplace_back(column.name); + } + return columns; +} + +std::map<std::string, std::string> PaimonRustTableReader::_build_options() const { + std::map<std::string, std::string> options; + if (_scan_params && _scan_params->__isset.paimon_options && + !_scan_params->paimon_options.empty()) { + options.insert(_scan_params->paimon_options.begin(), _scan_params->paimon_options.end()); + } else if (_current_range.__isset.table_format_params && + _current_range.table_format_params.__isset.paimon_params && + _current_range.table_format_params.paimon_params.__isset.paimon_options) { + options.insert(_current_range.table_format_params.paimon_params.paimon_options.begin(), + _current_range.table_format_params.paimon_params.paimon_options.end()); + } + + if (_scan_params && _scan_params->__isset.properties && !_scan_params->properties.empty()) { + for (const auto& kv : _scan_params->properties) { + options[kv.first] = kv.second; + } + } else if (_current_range.__isset.table_format_params && + _current_range.table_format_params.__isset.paimon_params && + _current_range.table_format_params.paimon_params.__isset.hadoop_conf) { + for (const auto& kv : _current_range.table_format_params.paimon_params.hadoop_conf) { + options[kv.first] = kv.second; + } + } + + auto copy_if_missing = [&](const char* from_key, const char* to_key) { + if (options.find(to_key) != options.end()) { + return; + } + auto it = options.find(from_key); + if (it != options.end() && !it->second.empty()) { + options[to_key] = it->second; + } + }; + + // The pinned paimon-rust storage dispatcher selects the parser from the + // table path's URI scheme (io/storage.rs): `oss://` tables read the OSS + // parser, which requires fs.oss.endpoint / fs.oss.accessKeyId / + // fs.oss.accessKeySecret (plus optional fs.oss.securityToken for STS); + // `s3://` tables read the S3 parser, whose family is paimon-java's + // s3.* keys (s3.access-key, s3.secret-key, s3.session.token, s3.endpoint, + // s3.region, s3.path-style-access, normalized from the fs.s3a. / s3a. / + // s3. prefixes). The FE's storage-properties channel delivers both + // protocols' credentials under the AWS_* / use_path_style aliases, so map + // them to the key family the table's scheme actually dispatches to — + // mapping everything to s3.* would leave OSS catalogs failing to open + // ("Missing required OSS config: fs.oss.endpoint"). + const std::string table_path = _resolve_table_path(_current_range).value_or(""); + const bool is_oss = table_path.rfind("oss://", 0) == 0; Review Comment: [P1] Gate or translate every enabled warehouse scheme. The pinned C crate enables separate COS, OBS, GCS, and Azure parsers, but this branch maps every non-`oss://` property map only to `s3.*`. Doris sends ordinary COS/OBS/GCS/shared-key Azure credentials as `AWS_*` aliases, so a `cosn://` table, for example, reaches the Rust COS parser without `fs.cosn.userinfo.secretId`/`secretKey` and fails instead of using JNI. Add a scheme capability gate (falling back for unverified schemes) or translate each scheme's exact key/auth family, with production-property-map open tests. ########## thirdparty/build-thirdparty.sh: ########## @@ -2094,6 +2094,60 @@ build_pugixml() { } # lance-c +# liblance_c.a and libpaimon_c.a are both Rust staticlibs linked into the +# same BE binary and must be built with the SAME rustc toolchain (see the +# in-function NOTE for the rust_eh_personality collision). LANCE_C_CARGO and +# PAIMON_RUST_CARGO are selected independently and each version check accepts +# any toolchain at least the minimum, so supported overrides could build the +# two libraries with different std hashes and defer the collision to the +# final BE link, where it surfaces as an opaque duplicate-symbol error. +# Compare the exact rustc identity (-vV: version, commit-hash, host) across +# both builds and fail early in the second one, stamping the identity so the +# invariant also holds across separate build-thirdparty.sh invocations +# (--continue / package lists). +ensure_same_rust_toolchain() { + # Portable array passing (bash 3.2 / macOS safe): the caller spreads its + # cargo_env entries as trailing arguments; at the call sites they are all + # space-free KEY=VALUE pairs (CFLAGS is appended only afterwards). + local pkg="$1" + local cargo_bin="$2" + shift 2 + + # Locate the rustc this cargo dispatches to: an explicit cargo path + # usually has rustc beside it; otherwise rustc resolves via PATH and the + # RUSTUP_TOOLCHAIN entry of env_arr dispatches the rustup shim. + local rustc_bin="rustc" + if [[ "${cargo_bin}" == */* && -x "${cargo_bin%/*}/rustc" ]]; then + rustc_bin="${cargo_bin%/*}/rustc" + fi + local identity + if ! identity="$(env "$@" "${rustc_bin}" -vV 2>&1)"; then + echo "failed to resolve the rustc identity for ${pkg} ('${rustc_bin}' -vV):" + echo "${identity}" + exit 1 + fi + + local stamp="${TP_INSTALL_DIR}/.doris-rust-toolchain-id" + if [[ -f "${stamp}" ]]; then + if ! diff -q <(printf '%s\n' "${identity}") "${stamp}" >/dev/null; then + echo "${pkg} would use a different rustc than the one recorded for" + echo "the other Rust static library:" + echo "-- recorded (${stamp}):" + cat "${stamp}" + echo "-- this build (${pkg}, '${rustc_bin}' -vV):" + printf '%s\n' "${identity}" + echo "liblance_c.a and libpaimon_c.a must be built with the SAME rustc" + echo "(different std hashes pull two std copies into the BE link and collide" + echo "on the unmangled rust_eh_personality symbol). Point LANCE_C_CARGO and" + echo "PAIMON_RUST_CARGO at one toolchain, or rebuild all Rust packages." + exit 1 + fi + else + printf '%s\n' "${identity}" > "${stamp}" Review Comment: [P2] Commit this toolchain identity only with an installed archive. The first Rust package writes the shared stamp before `cargo build` and archive copy, so a failed build leaves a compiler identity recorded even though no matching library exists; a retry with the corrected compiler then fails here. The advertised paired rebuild after changing toolchains also cannot start, because `--clean` removes only extracted sources and the first package is rejected against the old install-prefix stamp. Please track identity per successfully installed archive (atomically after copy), or make an explicit all-Rust rebuild remove both archives and the shared stamp before rebuilding. ########## be/src/format_v2/table/paimon_rust_predicate_converter.cpp: ########## @@ -0,0 +1,793 @@ +// 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 "format_v2/table/paimon_rust_predicate_converter.h" + +#include <algorithm> +#include <cctype> +#include <memory> +#include <utility> + +#include "common/logging.h" +#include "core/column/column_const.h" +#include "core/column/column_nullable.h" +#include "core/data_type/data_type.h" +#include "core/data_type/data_type_nullable.h" +#include "core/field.h" +#include "core/types.h" +#include "core/value/decimalv2_value.h" +#include "core/value/timestamptz_value.h" +#include "core/value/vdatetime_value.h" +#include "exprs/vcompound_pred.h" +#include "exprs/vdirect_in_predicate.h" +#include "exprs/vectorized_fn_call.h" +#include "exprs/vexpr.h" +#include "exprs/vin_predicate.h" +#include "exprs/vliteral.h" +#include "exprs/vslot_ref.h" + +namespace doris { + +namespace { +// paimon_datum tags (see paimon.h / bindings/c/src/table.rs::datum_from_c). +constexpr int32_t kTagBool = 0; +constexpr int32_t kTagTinyInt = 1; +constexpr int32_t kTagSmallInt = 2; +constexpr int32_t kTagInt = 3; +constexpr int32_t kTagLong = 4; +constexpr int32_t kTagDouble = 6; +constexpr int32_t kTagString = 7; +constexpr int32_t kTagDate = 8; +constexpr int32_t kTagTimestamp = 10; +constexpr int32_t kTagDecimal = 12; +constexpr int32_t kTagBytes = 13; + +// paimon decimal precision ceiling (paimon::Decimal::MAX_PRECISION). +constexpr int32_t kPaimonDecimalMaxPrecision = 38; + +// RAII for an owned paimon_predicate*. and/or/not consume their inputs, so we +// release() before handing pointers to them. +struct predicate_deleter { + void operator()(paimon_predicate* p) const { + if (p) { + paimon_predicate_free(p); + } + } +}; +using predicate_ptr = std::unique_ptr<paimon_predicate, predicate_deleter>; + +// RAII for an owned paimon_error*. +struct error_deleter { + void operator()(paimon_error* p) const { + if (p) { + paimon_error_free(p); + } + } +}; +using error_ptr = std::unique_ptr<paimon_error, error_deleter>; + +// Render a paimon_error into a string. Takes ownership of `err` via RAII so it +// is freed on every return path. Safe to call with nullptr. +std::string consume_predicate_error(paimon_error* err) { + error_ptr owned(err); + if (!owned) { + return "unknown error"; + } + std::string msg; + if (owned->message.data != nullptr && owned->message.len > 0) { + msg.assign(reinterpret_cast<const char*>(owned->message.data), owned->message.len); + } + return "code=" + std::to_string(owned->code) + ", msg=" + msg; +} +} // namespace + +PaimonRustPredicateConverter::PaimonRustPredicateConverter( + const std::vector<std::string>& column_names, const std::vector<DataTypePtr>& column_types, + const paimon_table* table) + : _table(table) { + DORIS_CHECK(column_names.size() == column_types.size()); + _columns_by_name.reserve(column_names.size()); + for (size_t i = 0; i < column_names.size(); ++i) { + _columns_by_name.emplace(_normalize_name(column_names[i]), + std::make_pair(column_names[i], column_types[i])); + } + // Paimon TIMESTAMP (wall clock) is stored as epoch-millis-of-the-wall-time + // and the DateTimeV2 serde decodes timezone-naive arrow values in UTC, so + // timestamp literals convert wall->epoch in UTC. utc_time_zone() needs no + // tzdata lookup, so the conversion cannot silently fall back to a + // machine-local zone. + _utc_tz = cctz::utc_time_zone(); +} + +paimon_predicate* PaimonRustPredicateConverter::build(const VExprContextSPtrs& conjuncts) { + if (_table == nullptr) { + return nullptr; + } + predicate_ptr result; + for (const auto& conjunct : conjuncts) { + if (!conjunct || !conjunct->root()) { + continue; + } + auto root = conjunct->root(); + if (root->is_rf_wrapper()) { + if (auto impl = root->get_impl()) { + root = impl; + } + } + // Preserve a safe prefix of the conjunct order: a later pushed + // predicate (e.g. an arrived IN runtime filter) could otherwise prune + // rows on which an earlier error-preserving conjunct — + // assert_true(...), a failing cast, ... — must still raise. The v1 + // partition-pruning path (FileScanner::_init_runtime_filter_partition_ + // prune_ctxs) stops at is_safe_to_execute_on_selected_rows() for the + // same reason, so a convertible predicate after an unsafe conjunct + // must not be pushed. Safe conjuncts that cannot be converted keep + // the old skip: they cannot raise, so pruning rows before they are + // evaluated as the residual never loses an error. + if (!root->is_safe_to_execute_on_selected_rows()) { + break; + } + predicate_ptr pred(_convert_expr(root)); + if (!pred) { + continue; + } + if (!result) { + result = std::move(pred); + } else { + // and consumes both inputs regardless of success. + result.reset(paimon_predicate_and(result.release(), pred.release())); + if (!result) { + return nullptr; + } + } + } + return result.release(); +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_expr(const VExprSPtr& expr) { + if (!expr) { + return nullptr; + } + + // Casts are not unwrapped anywhere (predicate root included): a cast node + // fails every dispatch below and the conjunct stays in the Doris residual, + // mirroring the FE converter, which keeps casted expressions unconverted. + if (auto* direct_in = dynamic_cast<VDirectInPredicate*>(expr.get())) { + VExprSPtr in_expr; + if (direct_in->get_slot_in_expr(in_expr)) { + return _convert_in(in_expr); + } + return nullptr; + } + + if (dynamic_cast<VInPredicate*>(expr.get()) != nullptr) { + return _convert_in(expr); + } + + switch (expr->op()) { + case TExprOpcode::COMPOUND_AND: + case TExprOpcode::COMPOUND_OR: + return _convert_compound(expr); + case TExprOpcode::COMPOUND_NOT: + return nullptr; + case TExprOpcode::EQ: + case TExprOpcode::EQ_FOR_NULL: + case TExprOpcode::NE: + case TExprOpcode::GE: + case TExprOpcode::GT: + case TExprOpcode::LE: + case TExprOpcode::LT: + return _convert_binary(expr); + default: + break; + } + + if (auto* fn = dynamic_cast<VectorizedFnCall*>(expr.get())) { + auto fn_name = _normalize_name(fn->function_name()); + if (fn_name == "is_null_pred" || fn_name == "is_not_null_pred") { + return _convert_is_null(expr, fn_name); + } + if (fn_name == "like") { + return _convert_like_prefix(expr); + } + } + + return nullptr; +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_compound(const VExprSPtr& expr) { + if (!expr || expr->get_num_children() != 2) { + return nullptr; + } + predicate_ptr left(_convert_expr(expr->get_child(0))); + if (!left) { + return nullptr; + } + predicate_ptr right(_convert_expr(expr->get_child(1))); + if (!right) { + return nullptr; + } + + if (expr->op() == TExprOpcode::COMPOUND_AND) { + return paimon_predicate_and(left.release(), right.release()); + } + if (expr->op() == TExprOpcode::COMPOUND_OR) { + return paimon_predicate_or(left.release(), right.release()); + } + return nullptr; +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_in(const VExprSPtr& expr) { + auto* in_pred = dynamic_cast<VInPredicate*>(expr.get()); + if (!in_pred || expr->get_num_children() < 2) { + return nullptr; + } + auto field_meta = _resolve_field(expr->get_child(0)); + if (!field_meta) { + return nullptr; + } + + const auto num_values = expr->get_num_children() - 1; + // Reserve up front so the backing strings never reallocate: each datum's + // str_data points into storages[i], which must stay stable. + std::vector<std::string> storages; + std::vector<paimon_datum> datums; + storages.reserve(num_values); + datums.reserve(num_values); + for (uint16_t i = 1; i < expr->get_num_children(); ++i) { + // Mirror FE's doInPredicate, which only accepts bare LiteralExpr + // children: a casted child would be unwrapped to its pre-cast value + // (debug_skip_fold_constant keeps such casts un-folded in the plan), + // so Doris would compare against the cast result while rust filters + // on the raw value — e.g. in `amount IN (CAST(1.24 AS DECIMAL(10,1)))` + // Doris keeps the 1.2 rows and the unwrapped 1.24 push would remove + // them permanently. Reject the whole predicate; the residual applies + // the cast correctly. + if (expr->get_child(i)->node_type() == TExprNodeType::CAST_EXPR) { + return nullptr; + } + auto holder = _convert_literal(expr->get_child(i), field_meta->type); + if (!holder) { + return nullptr; + } + storages.emplace_back(std::move(holder->storage)); + paimon_datum datum = holder->datum; + _bind_datum_storage(&datum, storages.back()); + datums.emplace_back(datum); + } + + if (datums.empty()) { + return nullptr; + } + if (in_pred->is_not_in()) { + return _take(paimon_predicate_is_not_in(_table, field_meta->column.c_str(), datums.data(), + datums.size())); + } + return _take(paimon_predicate_is_in(_table, field_meta->column.c_str(), datums.data(), + datums.size())); +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_binary(const VExprSPtr& expr) { + if (!expr || expr->get_num_children() != 2) { + return nullptr; + } + auto field_meta = _resolve_field(expr->get_child(0)); + if (!field_meta) { + return nullptr; + } + const char* column = field_meta->column.c_str(); + + // Convert the RHS first so EQ_FOR_NULL (<=>) only converts when the RHS is + // a convertible literal, mirroring the FE converter, which rejects a + // non-literal RHS. A column-to-column `a <=> b` must therefore stay in the + // Doris residual: it has no single-column rust predicate, and pushing + // `a IS NULL` would wrongly discard rows like (1, 1) — rows dropped by the + // pushed filter cannot be recovered by the residual conjunct. + auto holder = _convert_literal(expr->get_child(1), field_meta->type); + if (!holder) { + return nullptr; + } + + if (expr->op() == TExprOpcode::EQ_FOR_NULL) { + return _take(paimon_predicate_is_null(_table, column)); + } + + // `holder` is a local, so its storage stays put for the duration of the call. + _bind_datum_storage(&holder->datum, holder->storage); + const paimon_datum& datum = holder->datum; + + switch (expr->op()) { + case TExprOpcode::EQ: + return _take(paimon_predicate_equal(_table, column, datum)); + case TExprOpcode::NE: + return _take(paimon_predicate_not_equal(_table, column, datum)); + case TExprOpcode::GE: + return _take(paimon_predicate_greater_or_equal(_table, column, datum)); + case TExprOpcode::GT: + return _take(paimon_predicate_greater_than(_table, column, datum)); + case TExprOpcode::LE: + return _take(paimon_predicate_less_or_equal(_table, column, datum)); + case TExprOpcode::LT: + return _take(paimon_predicate_less_than(_table, column, datum)); + default: + break; + } + return nullptr; +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_is_null(const VExprSPtr& expr, + const std::string& fn_name) { + if (!expr || expr->get_num_children() != 1) { + return nullptr; + } + auto field_meta = _resolve_field(expr->get_child(0)); + if (!field_meta) { + return nullptr; + } + if (fn_name == "is_not_null_pred") { + return _take(paimon_predicate_is_not_null(_table, field_meta->column.c_str())); + } + return _take(paimon_predicate_is_null(_table, field_meta->column.c_str())); +} + +paimon_predicate* PaimonRustPredicateConverter::_convert_like_prefix(const VExprSPtr& expr) { + if (!expr || expr->get_num_children() != 2) { + return nullptr; + } + auto field_meta = _resolve_field(expr->get_child(0)); + if (!field_meta || !_is_string_type(field_meta->type->get_primitive_type())) { + return nullptr; + } + + auto pattern_opt = _extract_string_literal(expr->get_child(1)); + if (!pattern_opt) { + return nullptr; + } + const std::string& pattern = *pattern_opt; + // Only prefix matches (`abc%`) are convertible to a range scan. + if (!pattern.empty() && pattern.front() == '%') { + return nullptr; + } + if (pattern.empty() || pattern.back() != '%') { Review Comment: [P1] Push only a provably literal LIKE prefix. These checks accept any pattern ending in %, including internal wildcards and escapes. For example, `s LIKE 'a_b%'` matches `axb`, but this code pushes the range `[a_b, a_c)`, so Rust discards `axb` before Doris can apply the residual. Escaped `_`/`%` and backslashes are likewise treated as literal range bytes rather than LIKE syntax. Please parse Doris escape semantics and accept only one literal prefix plus one unescaped trailing `%`, or leave LIKE entirely residual; add internal-wildcard and escaped-pattern differential cases. ########## fe/fe-core/src/main/java/org/apache/doris/datasource/paimon/source/PaimonScanNode.java: ########## @@ -411,10 +413,138 @@ private void setPaimonParams(TFileRangeDesc rangeDesc, PaimonSplit paimonSplit) String fileFormat = getFileFormat(paimonSplit.getPathString()); if (split != null) { + // use jni reader / paimon-cpp reader / paimon-rust reader rangeDesc.setFormatType(TFileFormatType.FORMAT_JNI); - // A logical DataSplit may span multiple files, so keep it intact for the JNI reader. - fileDesc.setReaderType(TPaimonReaderType.PAIMON_JNI); - fileDesc.setPaimonSplit(PaimonUtil.encodeObjectToString(split)); + // paimon-cpp and paimon-rust both consume Paimon native binary serialization, + // which only supports DataSplit. Any other split type falls back to JNI. + boolean nativeSplit = split instanceof DataSplit; + // Fallback-read splits stay on JNI: FallbackDataSplit extends + // DataSplit, so the instanceof above passes, but its serializer + // appends an isFallback byte after the ordinary split that the + // pinned rust decoder rejects outright ("trailing bytes after + // DataSplit" — it requires full-buffer consumption), and even a + // permissive decode would still lack the second table identity + // needed to honor the fallback-side discriminator. Both sides of a + // FallbackReadFileStoreTable wrap their splits, so the table + // wrapper is gated as a whole (any split from it routes to JNI) + // until the rust ABI represents both sides; the FallbackSplit + // interface also catches a wrapper split regardless of how the + // table was resolved here. + boolean fallbackRead = split instanceof FallbackReadFileStoreTable.FallbackSplit + || processedTable instanceof FallbackReadFileStoreTable; + // Serialize the same effective table that planning and the JNI reader use. + // Relation options such as t@options('read.batch-size'='1') are applied by + // getProcessedTable() (doInitialize caches it in processedTable), and the + // rust reader derives its read batch size from the schema options — the raw + // cached table would silently drop the override. Copies, delegates and + // fallback wrappers of getProcessedTable() are still FileStoreTable, so the + // instanceof gate keeps its semantics. + Table paimonTable = processedTable; + FileStoreTable paimonFileStoreTable = + paimonTable instanceof FileStoreTable ? (FileStoreTable) paimonTable : null; + // query-auth.enabled tables stay on JNI: when catalog authorization + // succeeds with no row filter or column mask, Paimon still leaves an + // ordinary DataSplit (restricted results use QueryAuthSplit and are + // already handled by the nativeSplit gate above), so this table shape + // passes the compound gate — but the shipped schema keeps + // query-auth.enabled=true and the pinned rust ReadBuilder rejects + // every such table (its CoreOptions::ensure_read_authorized fails + // closed because the client cannot enforce the row filter / column + // masking), turning a valid authorized scan into a BE-open failure. + // Until the authorization result can be transported and enforced by + // the rust ABI, these tables route to JNI. + boolean queryAuthTable = false; + if (paimonFileStoreTable != null) { + CoreOptions queryAuthOptions = paimonFileStoreTable.coreOptions(); + // Null-safe: a table handle whose CoreOptions is not resolved + // (e.g. some wrapper shapes) stays rust-eligible rather than + // failing the scan here — the rust open itself rejects such a + // table if the option is really set. + queryAuthTable = queryAuthOptions != null && queryAuthOptions.queryAuthEnabled(); + } + // paimon-rust additionally requires (a) FileScannerV2: the V1 FileScanner + // explicitly rejects PAIMON_RUST, so with enable_file_scanner_v2 disabled + // the split falls back to JNI instead of encoding a rust request that the + // selected scanner cannot consume, and (b) a FileStoreTable: BE opens the + // table via paimon_table_from_schema_json, which needs the resolved + // TableSchema that only FileStoreTable exposes via schema(). If the table + // is not a FileStoreTable (e.g. a sys table backed by DataSplit), we cannot + // ship a schema JSON, so fall back to CPP / JNI rather than sending an + // incomplete PAIMON_RUST request that BE would reject. + // + // The paimon-rust S3 bridge maps static credentials, anonymous + // access (AWS_CREDENTIALS_PROVIDER_TYPE=ANONYMOUS -> s3.anonymous) + // and assume-role (AWS_ROLE_ARN / AWS_EXTERNAL_ID -> + // s3.assumed.role.*), but the remaining credential-provider modes + // are ambient JVM provider chains (ENV, SYSTEM_PROPERTIES, + // WEB_IDENTITY, CONTAINER, INSTANCE_PROFILE) with no paimon-rust + // equivalent — rust would silently sign with whatever the ambient + // chain resolves to. Gate those modes away from the rust reader + // here so the configured provider is honored via the JNI path. + boolean providerModeTranslatable = true; + String providerType = backendStorageProperties == null + ? null : backendStorageProperties.get("AWS_CREDENTIALS_PROVIDER_TYPE"); + if (providerType != null) { + String mode = providerType.trim().toUpperCase(Locale.ROOT); + providerModeTranslatable = mode.equals("DEFAULT") + || mode.equals("ANONYMOUS"); + // The rust OSS FileIO parser (oss:// warehouses) has no + // skip-signature switch, so an anonymous OSS catalog cannot be + // served by the rust reader either — fall back to JNI. + if (mode.equals("ANONYMOUS")) { + String location = source.getTableLocation(); + if (location != null && location.startsWith("oss://")) { + providerModeTranslatable = false; + } + } + } + // Incremental scans (binlog / changelog / delta / diff) must stay + // on the JNI path: this wire format carries only an ordinary + // DataSplit and the rust reader invokes TableRead::to_arrow, but + // paimon 1.4 marks incremental splits as streaming (which the + // pinned rust deserializer rejects), diff requires a separate + // IncrementalPlan instead of an ordinary plan, and ordinary + // primary-key reads can merge versions rather than return the + // changes — until the C ABI transports the mode and plan, the + // rust reader cannot express any of these. + TableScanParams incrementalParams = getScanParams(); + boolean isIncremental = incrementalParams != null && incrementalParams.incrementalRead(); + boolean canUseRust = sessionVariable.isEnablePaimonRustReader() + && sessionVariable.enableFileScannerV2 && nativeSplit && !fallbackRead + && !isIncremental && providerModeTranslatable && !queryAuthTable + && paimonFileStoreTable != null; Review Comment: [P1] Keep unsupported primary-key DV modes on JNI. This gate admits ordinary `DataSplit`s from partial-update or aggregation tables without checking their deletion-vector mode, but the pinned Rust `read_pk` path explicitly returns Unsupported for `deletion-vectors.merge-on-read=true` and for DV splits that are not fully materialized. Those are valid Java/JNI scans, so enabling Rust turns them into BE-open failures. Gate on the effective `CoreOptions` and split capability (or otherwise fall back) and cover both merge engines plus uncompacted DV splits. ########## be/src/format_v2/table/paimon_rust_table_reader.cpp: ########## @@ -0,0 +1,830 @@ +// 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 "format_v2/table/paimon_rust_table_reader.h" + +#include <algorithm> +#include <utility> + +#include "arrow/c/abi.h" +#include "arrow/c/bridge.h" +#include "arrow/record_batch.h" +#include "arrow/result.h" +#include "common/logging.h" +#include "core/block/block.h" +#include "core/block/column_with_type_and_name.h" +#include "core/column/column_const.h" +#include "exprs/vexpr_context.h" +#include "exprs/vliteral.h" +#include "format_v2/column_mapper.h" +#include "format_v2/table/paimon_rust_predicate_converter.h" +#include "runtime/descriptors.h" +#include "runtime/file_scan_profile.h" +#include "runtime/runtime_state.h" +#include "util/string_util.h" +#include "util/timezone_utils.h" +#include "util/url_coding.h" + +extern "C" { +#include "paimon_rust/paimon.h" +} + +namespace doris::format::paimon { + +namespace { +constexpr const char* VALUE_KIND_FIELD = "_VALUE_KIND"; + +// --------------------------------------------------------------------------- +// RAII wrappers over the paimon-rust C handles. Each handle is an opaque +// pointer owned by Rust and released by a matching paimon_*_free function. +// --------------------------------------------------------------------------- +#define PAIMON_OWNED(type, freefn) \ + struct type##_deleter { \ + void operator()(paimon_##type* p) const { \ + if (p) { \ + freefn(p); \ + } \ + } \ + }; \ + using type##_ptr = std::unique_ptr<paimon_##type, type##_deleter> + +PAIMON_OWNED(table, paimon_table_free); +PAIMON_OWNED(read_builder, paimon_read_builder_free); +PAIMON_OWNED(plan, paimon_plan_free); +PAIMON_OWNED(table_read, paimon_table_read_free); +PAIMON_OWNED(record_batch_reader, paimon_record_batch_reader_free); +PAIMON_OWNED(error, paimon_error_free); + +#undef PAIMON_OWNED + +// One Arrow batch (schema + array containers). Owning it requires a two-step +// teardown that the unique_ptr deleters above can't express: first invoke the +// Arrow C Data Interface `release` callback on each struct (hands buffers back +// to the producer), then free the container structs via paimon_arrow_batch_free. +class ArrowBatch { +public: + explicit ArrowBatch(paimon_arrow_batch batch) : batch_(batch) {} + ~ArrowBatch() { + auto* schema = static_cast<ArrowSchema*>(batch_.schema); + auto* array = static_cast<ArrowArray*>(batch_.array); + if (array && array->release) { + array->release(array); + } + if (schema && schema->release) { + schema->release(schema); + } + paimon_arrow_batch_free(batch_); + } + + ArrowBatch(const ArrowBatch&) = delete; + ArrowBatch& operator=(const ArrowBatch&) = delete; + + ArrowSchema* schema() const { return static_cast<ArrowSchema*>(batch_.schema); } + ArrowArray* array() const { return static_cast<ArrowArray*>(batch_.array); } + +private: + paimon_arrow_batch batch_; +}; + +// Render a paimon_error into a string. Takes ownership of `err` via RAII so it +// is freed on every return path. Safe to call with nullptr. +std::string consume_error(paimon_error* err) { + error_ptr owned(err); + if (!owned) { + return "unknown error"; + } + std::string msg; + if (owned->message.data != nullptr && owned->message.len > 0) { + msg.assign(reinterpret_cast<const char*>(owned->message.data), owned->message.len); + } + return "code=" + std::to_string(owned->code) + ", msg=" + msg; +} + +// Render storage option KEYS for diagnostics. Values are never rendered: +// credential keys arrive under many spellings and cases (AWS_SECRET_KEY, +// AWS_TOKEN, fs.oss.accessKeySecret, s3.secret-key, ...), and a key-name +// blocklist that misses one alias leaks the value into the INFO log, so +// only the key names are printed at all. +std::string format_options(const std::map<std::string, std::string>& options) { + std::string out; + for (const auto& kv : options) { + if (!out.empty()) { + out += ", "; + } + out += kv.first; + } + return out; +} + +} // namespace + +// Paimon-rust handles. Order of members matters: destruction runs in reverse +// declaration order, and the read_builder depends on the table while the arrow +// reader depends on the whole pipeline above it. So the table MUST be declared +// first (destroyed last) and the record batch reader last. +struct PaimonRustTableReader::PaimonHandles { + table_ptr table; + read_builder_ptr read_builder; + plan_ptr plan; + table_read_ptr table_read; + record_batch_reader_ptr reader; +}; + +PaimonRustTableReader::PaimonRustTableReader() = default; + +PaimonRustTableReader::~PaimonRustTableReader() = default; + +Status PaimonRustTableReader::init(format::TableReadOptions&& options) { + RETURN_IF_ERROR(format::TableReader::init(std::move(options))); + { + // Base and derived scopes must not overlap on the same counter: RuntimeProfile timers + // add deltas, so nested use would double-count instead of extending lifecycle coverage. + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.init_timer); + // Materialize TIMESTAMP_LTZ in the session timezone — the same + // convention as the JNI reader (PaimonJniScanner reads time_zone from + // its scan params) and lance_reader. Timezone-naive (paimon TIMESTAMP) + // arrow values are decoded in UTC by the DateTimeV2 serde regardless + // of _ctz, so NTZ wall-clock semantics are preserved. + DORIS_CHECK(_runtime_state != nullptr); + _ctz = _runtime_state->timezone_obj(); + if (_scanner_profile != nullptr) { + file_scan_profile::ensure_hierarchy(_scanner_profile); + _rust_total_time = ADD_CHILD_TIMER(_scanner_profile, "PaimonRustReader", + file_scan_profile::TABLE_READER); + _rust_open_split_time = + ADD_CHILD_TIMER(_scanner_profile, "OpenSplitTime", "PaimonRustReader"); + _rust_read_batch_time = + ADD_CHILD_TIMER(_scanner_profile, "ReadBatchTime", "PaimonRustReader"); + _rust_arrow_to_block_time = + ADD_CHILD_TIMER(_scanner_profile, "ArrowToBlockTime", "PaimonRustReader"); + } + // Projected column name -> fixed output position, registered with both the exact and + // the lower-case spelling so mixed-case Rust schema output still resolves (v1 + // semantics: exact match first, lower-case fallback on lookup). + _output_name_to_idx.reserve(_projected_columns.size() * 2); + for (size_t idx = 0; idx < _projected_columns.size(); ++idx) { + _output_name_to_idx.emplace(_projected_columns[idx].name, idx); + _output_name_to_idx.emplace(to_lower(_projected_columns[idx].name), idx); + } + } + return Status::OK(); +} + +Status PaimonRustTableReader::prepare_split(const format::SplitReadOptions& options) { + // EOF belongs to the previous split. Keep it set after closing that split so repeated reads + // are idempotent, and clear it only when a new split is explicitly prepared. + _close_split_reader(); + _split_eof = false; + _current_range = options.current_range; + RETURN_IF_ERROR(format::TableReader::prepare_split(options)); + if (current_split_pruned()) { + return Status::OK(); + } + if (_is_table_level_count_active()) { + // No rust pipeline is opened; get_block emits the synthetic count rows. + return Status::OK(); + } + RETURN_IF_ERROR(_validate_rust_split(options.current_range)); + { + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.prepare_split_timer); + SCOPED_TIMER(_rust_open_split_time); + RETURN_IF_ERROR(_open_split_reader(options.current_range)); + } + return Status::OK(); +} + +Status PaimonRustTableReader::get_block(Block* block, bool* eos) { + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.exec_timer); + SCOPED_TIMER(_rust_total_time); + DORIS_CHECK(block != nullptr); + DORIS_CHECK(eos != nullptr); + DORIS_CHECK(block->columns() == _projected_columns.size()); + block->clear_column_data(_projected_columns.size()); + *eos = false; + + if (_is_table_level_count_active()) { + return _read_table_level_count(block, eos); + } + + // num_splits == 0 yields an empty (but valid) stream: report EOF. + if (_split_eof) { + *eos = true; + return Status::OK(); + } + if (!_handles || !_handles->reader) { + return Status::InternalError("paimon-rust reader is not initialized"); + } + + while (true) { + // Mirror the base TableReader cancellation contract so a cancelled query does not + // drain the whole split. + if (_io_ctx != nullptr && _io_ctx->should_stop) { + _split_eof = true; + _close_split_reader(); + *eos = true; + return Status::OK(); + } + + paimon_result_next_batch next; + { + SCOPED_TIMER(_rust_read_batch_time); + next = paimon_record_batch_reader_next(_handles->reader.get()); + } + if (next.error != nullptr) { + return Status::InternalError("paimon-rust read batch failed: {}", + consume_error(next.error)); + } + // End of stream: both pointers are null. + if (next.batch.array == nullptr && next.batch.schema == nullptr) { + _split_eof = true; + _close_split_reader(); + *eos = true; + return Status::OK(); + } + + // RAII: the batch's Arrow release callbacks + container free run when + // `batch` leaves this scope, including on any early return. + ArrowBatch batch(next.batch); + + auto* c_array = batch.array(); + auto* c_schema = batch.schema(); + arrow::Result<std::shared_ptr<arrow::RecordBatch>> import_result = + arrow::ImportRecordBatch(c_array, c_schema); + if (!import_result.ok()) { + return Status::InternalError("failed to import paimon-rust arrow batch: {}", + import_result.status().message()); + } + + auto record_batch = std::move(import_result).ValueUnsafe(); + const auto rows = static_cast<size_t>(record_batch->num_rows()); + if (rows == 0) { + // Skip empty batches and keep draining the stream. + continue; + } + RETURN_IF_ERROR(_fill_block_from_record_batch(record_batch, block, rows)); + _record_scan_rows(rows); + *eos = false; + return Status::OK(); + } +} + +Status PaimonRustTableReader::abort_split() { + { + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.close_timer); + _close_split_reader(); + _split_eof = false; + } + return format::TableReader::abort_split(); +} + +#ifdef BE_TEST +std::string PaimonRustTableReader::TEST_format_options( + const std::map<std::string, std::string>& options) { + return format_options(options); +} + +std::map<std::string, std::string> PaimonRustTableReader::TEST_build_options( + TFileScanRangeParams* scan_params, const TFileRangeDesc& range) { + TFileScanRangeParams* previous_params = _scan_params; + TFileRangeDesc previous_range = _current_range; + _scan_params = scan_params; + _current_range = range; + std::map<std::string, std::string> options = _build_options(); + _scan_params = previous_params; + _current_range = std::move(previous_range); + return options; +} +#endif + +Status PaimonRustTableReader::close() { + { + SCOPED_TIMER(_profile.total_timer); + SCOPED_TIMER(_profile.close_timer); + _close_split_reader(); + _close_table(); + } + return format::TableReader::close(); +} + +Status PaimonRustTableReader::_validate_rust_split(const TFileRangeDesc& range) const { + if (!range.__isset.table_format_params || !range.table_format_params.__isset.paimon_params) { + return Status::InternalError( + "missing paimon_params for paimon rust reader, possibly caused by FE/BE protocol " + "mismatch"); + } + const auto& params = range.table_format_params.paimon_params; + if (!params.__isset.paimon_split || params.paimon_split.empty()) { + return Status::InternalError( + "missing paimon_split for paimon rust reader, possibly caused by FE/BE protocol " + "mismatch"); + } + if (params.__isset.reader_type && params.reader_type != TPaimonReaderType::PAIMON_RUST) { + return Status::InternalError( + "invalid reader_type for paimon rust reader, possibly caused by FE/BE protocol " + "mismatch"); + } + if (!_resolve_table_path(range).has_value()) { + return Status::InternalError( + "paimon-rust missing paimon_table; cannot resolve paimon table location"); + } + if (!_resolve_db_name(range).has_value()) { + return Status::InternalError( + "paimon-rust missing db_name; cannot open paimon table via schema json"); + } + if (!_resolve_table_name(range).has_value()) { + return Status::InternalError( + "paimon-rust missing table_name; cannot open paimon table via schema json"); + } + if (!_resolve_table_schema_json(range).has_value()) { + return Status::InternalError( + "paimon-rust missing paimon_table_schema_json; cannot open paimon table via " + "schema json"); + } + return Status::OK(); +} + +Status PaimonRustTableReader::_open_split_reader(const TFileRangeDesc& range) { + // 1. Decode the FE-planned split first so we fail fast (and without any + // filesystem IO) when it is missing or malformed. + std::string split_bytes; + RETURN_IF_ERROR(_decode_split_bytes(&split_bytes)); + + // 2. Resolve identifier + table_path + FE-supplied TableSchema JSON. + auto table_path = _resolve_table_path(range).value(); + auto db_name = _resolve_db_name(range).value(); + auto table_name = _resolve_table_name(range).value(); + auto schema_json = _resolve_table_schema_json(range).value(); + auto branch_opt = _resolve_branch(range); + + // 3. Assemble storage options: FE-supplied paimon options + hadoop_conf + + // OSS/S3 → AWS_* translations. These feed FileIO only (per + // paimon_table_from_schema_json contract); they are NOT merged into the + // supplied table schema. + auto options = _build_options(); + + auto opened_table_key = + std::make_tuple(table_path, schema_json, db_name, table_name, branch_opt, options); + if (!_handles || !_handles->table || _opened_table_key != opened_table_key) { + // A paimon scan reads one table, so the handle is opened at most once per + // distinct identity (e.g. re-created after a close); splits of the same + // table reuse it and only rebuild the read pipeline below. + _close_table(); + _handles = std::make_unique<PaimonHandles>(); + + std::vector<paimon_option> c_options; + c_options.reserve(options.size()); + for (const auto& kv : options) { + c_options.push_back(paimon_option {kv.first.c_str(), kv.second.c_str()}); + } + + LOG(INFO) << "paimon-rust opening table via schema json: db=" << db_name + << " table=" << table_name << " path=" << table_path + << " branch=" << (branch_opt.has_value() ? branch_opt.value() : "main") + << " storage_options=[" << format_options(options) << "]"; + + // Build the table directly from the FE-supplied schema JSON. The Rust + // side rejects null / empty branch, so we default to paimon's canonical + // "main" sentinel when FE did not set paimon_branch (i.e. the table is + // on the main branch — matches upstream Identifier.DEFAULT_MAIN_BRANCH). + const std::string& branch_str = branch_opt.has_value() ? branch_opt.value() : "main"; + paimon_result_get_table tbl_res = paimon_table_from_schema_json( + table_path.c_str(), schema_json.c_str(), db_name.c_str(), table_name.c_str(), + branch_str.c_str(), c_options.empty() ? nullptr : c_options.data(), + c_options.size()); + if (tbl_res.error != nullptr) { + return Status::InternalError( + "paimon-rust table_from_schema_json failed: db={} table={} err={}", db_name, + table_name, consume_error(tbl_res.error)); + } + _handles->table.reset(tbl_res.table); + _opened_table_key = std::move(opened_table_key); + } + + // 4. Build the read pipeline: read_builder -> case-insensitive -> projection. + paimon_result_read_builder rb_res = paimon_table_new_read_builder(_handles->table.get()); Review Comment: [P1] Do not time-travel the FE-resolved schema again here. Doris deliberately uses `copyWithLatestSchema` plus `copyWithoutTimeTravel(scan.snapshot-id)` so a schema-only ALTER after the last data commit retains the new schema while reading the pinned data snapshot, and FE serializes that resolved `TableSchema`. The pinned C entry point called here invokes `copy_with_time_travel(empty)`, which reuses the embedded selector and replaces the supplied schema with the older snapshot schema; projecting the newly added column then fails before schema evolution can fill it, while JNI succeeds. Add/use a builder path that trusts the transported schema (or strip only the planning selector without changing its fields), and cover `ADD COLUMN` with no later data commit. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
