Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions docs/source/user_guide.rst
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,7 @@ User Guide
user_guide/data_types
user_guide/primary_key_table
user_guide/append_only_table
user_guide/system_tables
user_guide/write
user_guide/commit
user_guide/compaction
Expand Down
67 changes: 67 additions & 0 deletions docs/source/user_guide/system_tables.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1,67 @@
.. Copyright 2026-present Alibaba Inc.

.. Licensed 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.

System Tables
=============

Paimon C++ supports reading system tables by appending a system table suffix to
the data table path. For example, ``/warehouse/db.db/orders$snapshots`` reads
the snapshots system table for ``orders``.

Branch-qualified paths are also supported. For example,
``/warehouse/db.db/orders$branch_audit$files`` reads the files system table from
the ``audit`` branch.

Read-Optimized System Table
---------------------------

The read-optimized system table is addressed by the ``$ro`` suffix:

.. code-block:: text

/warehouse/db.db/orders$ro

For primary-key tables, ``$ro`` only plans data files from the highest LSM
level, which is the level produced by full compaction. This avoids merging data
from multiple LSM levels during query planning and reading. The tradeoff is that
the result may lag behind the latest committed data until a full compaction

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could we document that $ro does not necessarily represent one coherent older snapshot? Different buckets may complete full compaction at different times, so the top-level files selected from the latest snapshot can contain bucket states produced at different snapshots. The current “may lag” wording does not make this consistency caveat clear.

publishes the newest records into the highest level.

This stale view is not guaranteed to correspond to any single historical table
snapshot. Full compaction may finish independently for different buckets. When
``$ro`` scans the latest snapshot, the selected highest-level files can
therefore combine bucket states produced by full-compaction commits at different
snapshot IDs.

For primary-key tables, ``$ro`` also enables value-stats filtering, so file-level
pruning of the reader predicate can be more aggressive than the base table.

For append-only tables, ``$ro`` has the same read behavior as the base table,
including streaming reads.

Limitations
~~~~~~~~~~~

- Primary-key ``$ro`` scans are batch-only. Streaming scans are not supported.
- Freshness depends on full compaction frequency, and the result may not match
any single historical snapshot across buckets.
- Primary-key tables in bucket-unaware mode are not supported by the current C++
scan path.

Typical Usage
~~~~~~~~~~~~~

Use ``$ro`` for OLAP or batch workloads that can tolerate stale results and
prefer reading compacted files directly. Use the base table path when the query
must see the latest committed data.
1 change: 1 addition & 0 deletions src/paimon/CMakeLists.txt
Original file line number Diff line number Diff line change
Expand Up @@ -344,6 +344,7 @@ set(PAIMON_CORE_SRCS
core/table/system/binlog_system_table.cpp
core/table/system/in_memory_system_table.cpp
core/table/system/metadata_system_tables.cpp
core/table/system/read_optimized_system_table.cpp
core/table/system/system_table.cpp
core/table/system/system_table_scan.cpp
core/table/system/system_table_schema.cpp
Expand Down
11 changes: 8 additions & 3 deletions src/paimon/core/table/source/data_table_batch_scan.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -33,10 +33,15 @@ class DataSplit;

DataTableBatchScan::DataTableBatchScan(bool pk_table, const CoreOptions& core_options,
const std::shared_ptr<SnapshotReader>& snapshot_reader,
std::optional<int32_t> push_down_limit)
bool read_optimized, std::optional<int32_t> push_down_limit)
: AbstractTableScan(core_options, snapshot_reader), push_down_limit_(push_down_limit) {
if (pk_table && (core_options.DeletionVectorsEnabled() ||
core_options.GetMergeEngine() == MergeEngine::FIRST_ROW)) {
if (pk_table && read_optimized) {
int32_t top_level = core_options.GetNumLevels() - 1;
snapshot_reader_->WithLevelFilter(
[top_level](int32_t level) -> bool { return level == top_level; });
snapshot_reader_->EnableValueFilter();
} else if (pk_table && (core_options.DeletionVectorsEnabled() ||
core_options.GetMergeEngine() == MergeEngine::FIRST_ROW)) {
auto level_filter = [](int32_t level) -> bool { return level > 0; };
snapshot_reader_->WithLevelFilter(level_filter);
snapshot_reader_->EnableValueFilter();
Expand Down
2 changes: 1 addition & 1 deletion src/paimon/core/table/source/data_table_batch_scan.h
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ class SnapshotReader;
class DataTableBatchScan : public AbstractTableScan {
public:
DataTableBatchScan(bool pk_table, const CoreOptions& core_options,
const std::shared_ptr<SnapshotReader>& snapshot_reader,
const std::shared_ptr<SnapshotReader>& snapshot_reader, bool read_optimized,
std::optional<int32_t> push_down_limit);

Result<std::shared_ptr<Plan>> CreatePlan() override;
Expand Down
24 changes: 24 additions & 0 deletions src/paimon/core/table/source/read_optimized_scan_options.h
Original file line number Diff line number Diff line change
@@ -0,0 +1,24 @@
/*
* Copyright 2026-present Alibaba Inc.
*
* Licensed 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.
*/

#pragma once

namespace paimon {

/// Internal scan option used by `T$ro` to request top-level-only planning.
inline constexpr char kReadOptimizedScanOption[] = "__paimon.internal.read-optimized";

} // namespace paimon
15 changes: 12 additions & 3 deletions src/paimon/core/table/source/table_scan.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,7 @@
#include "paimon/common/predicate/predicate_validator.h"
#include "paimon/common/types/data_field.h"
#include "paimon/common/utils/fields_comparator.h"
#include "paimon/common/utils/options_utils.h"
#include "paimon/core/core_options.h"
#include "paimon/core/index/index_file_handler.h"
#include "paimon/core/manifest/index_manifest_file.h"
Expand All @@ -47,6 +48,7 @@
#include "paimon/core/table/source/data_table_batch_scan.h"
#include "paimon/core/table/source/data_table_stream_scan.h"
#include "paimon/core/table/source/merge_tree_split_generator.h"
#include "paimon/core/table/source/read_optimized_scan_options.h"
#include "paimon/core/table/source/snapshot/snapshot_reader.h"
#include "paimon/core/table/source/split_generator.h"
#include "paimon/core/table/system/system_table.h"
Expand Down Expand Up @@ -225,6 +227,9 @@ Result<std::unique_ptr<TableScan>> NewDataTableScan(const std::shared_ptr<ScanCo
}
table_schema = latest_table_schema.value();
}
PAIMON_ASSIGN_OR_RAISE(bool read_optimized,
OptionsUtils::GetValueFromMap<bool>(context->GetOptions(),
kReadOptimizedScanOption, false));
// merge options
auto options = table_schema->Options();
for (const auto& [key, value] : context->GetOptions()) {
Expand Down Expand Up @@ -280,12 +285,16 @@ Result<std::unique_ptr<TableScan>> NewDataTableScan(const std::shared_ptr<ScanCo
context->GetMemoryPool()));
auto snapshot_reader = std::make_shared<SnapshotReader>(
file_store_scan, path_factory, std::move(split_generator), std::move(index_file_handler));
const bool pk_table = !table_schema->PrimaryKeys().empty();
if (read_optimized && pk_table && context->IsStreamingMode()) {
return Status::NotImplemented(
"read-optimized system table does not support streaming scan for primary key table");
}
if (context->IsStreamingMode()) {
return std::make_unique<DataTableStreamScan>(core_options, snapshot_reader);
}
auto batch_scan =
std::make_unique<DataTableBatchScan>(/*pk_table=*/!table_schema->PrimaryKeys().empty(),
core_options, snapshot_reader, context->GetLimit());
auto batch_scan = std::make_unique<DataTableBatchScan>(
/*pk_table=*/pk_table, core_options, snapshot_reader, read_optimized, context->GetLimit());
if (!core_options.DataEvolutionEnabled()) {
return batch_scan;
}
Expand Down
14 changes: 14 additions & 0 deletions src/paimon/core/table/source/table_scan_test.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@

#include "paimon/table/source/table_scan.h"

#include <memory>
#include <string>
#include <utility>
#include <vector>
Expand Down Expand Up @@ -59,4 +60,17 @@ TEST(TableScanTest, TestPkSchemaEvolutionScan) {
ASSERT_FALSE(plan->Splits().empty());
}

TEST(TableScanTest, TestReadOptimizedPrimaryKeyStreamingScanUnsupported) {
std::string path = paimon::test::GetDataDir() +
"/orc/pk_table_with_alter_table.db/pk_table_with_alter_table$ro";
ScanContextBuilder builder(path);
builder.AddOption(Options::FILE_FORMAT, "orc");
builder.WithStreamingMode(true);
ASSERT_OK_AND_ASSIGN(std::unique_ptr<ScanContext> context, builder.Finish());

ASSERT_NOK_WITH_MSG(TableScan::Create(std::move(context)),
"read-optimized system table does not support streaming scan for primary "
"key table");
}

} // namespace paimon::test
128 changes: 128 additions & 0 deletions src/paimon/core/table/system/read_optimized_system_table.cpp
Original file line number Diff line number Diff line change
@@ -0,0 +1,128 @@
/*
* Copyright 2026-present Alibaba Inc.
*
* Licensed 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/system/read_optimized_system_table.h"

#include <memory>
#include <string>
#include <utility>

#include "arrow/c/bridge.h"
#include "paimon/common/types/data_field.h"
#include "paimon/core/schema/table_schema.h"
#include "paimon/core/table/source/read_optimized_scan_options.h"
#include "paimon/defs.h"
#include "paimon/read_context.h"
#include "paimon/scan_context.h"
#include "paimon/table/source/table_read.h"
#include "paimon/table/source/table_scan.h"

namespace paimon {

ReadOptimizedSystemTable::ReadOptimizedSystemTable(std::string table_path,
std::shared_ptr<TableSchema> table_schema,
std::map<std::string, std::string> options)
: table_path_(std::move(table_path)),
table_schema_(std::move(table_schema)),
options_(std::move(options)) {}

std::string ReadOptimizedSystemTable::Name() const {
return kName;
}

Result<std::shared_ptr<arrow::Schema>> ReadOptimizedSystemTable::ArrowSchema() const {
return DataField::ConvertDataFieldsToArrowSchema(table_schema_->Fields());
}

std::map<std::string, std::string> ReadOptimizedSystemTable::ReadOptimizedOptions() const {
auto options = options_;
options[kReadOptimizedScanOption] = "true";
return options;
}

Result<std::unique_ptr<TableScan>> ReadOptimizedSystemTable::NewScan(
const std::shared_ptr<ScanContext>& context) const {
auto options = ReadOptimizedOptions();
ScanContextBuilder builder(table_path_);
builder.SetOptions(options)
.WithStreamingMode(context->IsStreamingMode())
.WithMemoryPool(context->GetMemoryPool())
.WithExecutor(context->GetExecutor())
.WithFileSystem(context->GetSpecificFileSystem())
.WithCache(context->GetCache());
if (context->GetLimit().has_value()) {
builder.SetLimit(context->GetLimit().value());
}
if (context->GetScanFilters()) {
if (context->GetScanFilters()->GetBucketFilter().has_value()) {
builder.SetBucketFilter(context->GetScanFilters()->GetBucketFilter().value());
}
builder.SetPartitionFilter(context->GetScanFilters()->GetPartitionFilters());
builder.SetPredicate(context->GetScanFilters()->GetPredicate());
}
if (context->GetGlobalIndexResult()) {
builder.SetGlobalIndexResult(context->GetGlobalIndexResult());
}
if (context->GetSpecificTableSchema().has_value()) {
builder.SetTableSchema(context->GetSpecificTableSchema().value());
}
PAIMON_ASSIGN_OR_RAISE(std::unique_ptr<ScanContext> base_context, builder.Finish());
return TableScan::Create(std::move(base_context));
}

Result<std::unique_ptr<TableRead>> ReadOptimizedSystemTable::NewRead(
const std::shared_ptr<ReadContext>& context) const {
auto options = options_;
std::string branch = context->GetBranch();
// SystemTableLoader injects Options::BRANCH when parsing paths such as `T$branch_dev$ro`.
auto branch_iter = options.find(Options::BRANCH);
if (branch_iter != options.end()) {
branch = branch_iter->second;
}
ReadContextBuilder builder(table_path_);
builder.SetOptions(options)
.WithBranch(branch)
.SetPredicate(context->GetPredicate())
.EnablePredicateFilter(context->EnablePredicateFilter())
.EnablePrefetch(context->EnablePrefetch())
.SetPrefetchBatchCount(context->GetPrefetchBatchCount())
.SetPrefetchMaxParallelNum(context->GetPrefetchMaxParallelNum())
.EnableMultiThreadRowToBatch(context->EnableMultiThreadRowToBatch())
.SetRowToBatchThreadNumber(context->GetRowToBatchThreadNumber())
.WithMemoryPool(context->GetMemoryPool())
.WithExecutor(context->GetExecutor())
.WithFileSystem(context->GetSpecificFileSystem())
.WithFileSystemSchemeToIdentifierMap(context->GetFileSystemSchemeToIdentifierMap())
.SetPrefetchCacheMode(context->GetPrefetchCacheMode())
.WithCacheConfig(context->GetCacheConfig())
.WithCache(context->GetCache())
.SetReadFieldNames(context->GetReadFieldNames())
.SetReadFieldIds(context->GetReadFieldIds());

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could we preserve the projected Arrow schema from ReadContext::SetReadSchema() here? The wrapper currently copies field names and IDs only. For nested-column pruning those vectors are normally empty, so $ro silently falls back to the full table schema and may return a different shape from the requested projection. This is separate from the existing top-level projection test; could we transfer or re-export the read schema and add a nested projection test for $ro?

if (context->HasReadSchema()) {
PAIMON_ASSIGN_OR_RAISE_FROM_ARROW(std::shared_ptr<arrow::Schema> read_schema,
arrow::ImportSchema(context->GetReadSchema()));
auto c_read_schema = std::make_unique<::ArrowSchema>();
PAIMON_RETURN_NOT_OK_FROM_ARROW(arrow::ExportSchema(*read_schema, c_read_schema.get()));
builder.SetReadSchema(std::move(c_read_schema));
}
if (context->GetSpecificTableSchema().has_value()) {
builder.SetTableSchema(context->GetSpecificTableSchema().value());
}
PAIMON_ASSIGN_OR_RAISE(std::unique_ptr<ReadContext> base_context, builder.Finish());
return TableRead::Create(std::move(base_context));
}

} // namespace paimon
51 changes: 51 additions & 0 deletions src/paimon/core/table/system/read_optimized_system_table.h
Original file line number Diff line number Diff line change
@@ -0,0 +1,51 @@
/*
* Copyright 2026-present Alibaba Inc.
*
* Licensed 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.
*/

#pragma once

#include <map>
#include <memory>
#include <string>

#include "paimon/core/table/system/system_table.h"

namespace paimon {
class TableSchema;

/// System table for `T$ro`, exposing read-optimized data.
class ReadOptimizedSystemTable : public SystemTable {
public:
static constexpr const char* kName = "ro";

ReadOptimizedSystemTable(std::string table_path, std::shared_ptr<TableSchema> table_schema,
std::map<std::string, std::string> options);

std::string Name() const override;
Result<std::shared_ptr<arrow::Schema>> ArrowSchema() const override;
Result<std::unique_ptr<TableScan>> NewScan(
const std::shared_ptr<ScanContext>& context) const override;
Result<std::unique_ptr<TableRead>> NewRead(
const std::shared_ptr<ReadContext>& context) const override;

private:
std::map<std::string, std::string> ReadOptimizedOptions() const;

std::string table_path_;
std::shared_ptr<TableSchema> table_schema_;
std::map<std::string, std::string> options_;
};

} // namespace paimon
Loading
Loading