Abstract base class for merge-based snapshot write operations. More...

#include <merging_snapshot_update.h>

Inheritance diagram for iceberg::MergingSnapshotUpdate:

Public Member Functions
Result< std::vector< ManifestFile > >	Apply (const TableMetadata &metadata_to_update, const std::shared_ptr< Snapshot > &snapshot) override
	Apply the update's changes to the given metadata and snapshot.

Status	CleanUncommitted (const std::unordered_set< std::string > &committed) override
	Clean up any uncommitted manifests that were created.

std::unordered_map< std::string, std::string >	Summary () override
	Get the summary map for this operation.

Public Member Functions inherited from iceberg::SnapshotUpdate
Kind	kind () const override
	Return the kind of this pending update.

bool	IsRetryable () const override
	Whether this update can be retried after a commit conflict.

auto &	DeleteWith (this auto &self, std::function< Status(const std::string &)> delete_func)
	Set a callback to delete files instead of the table's default.

auto &	StageOnly (this auto &self)
	Stage a snapshot in table metadata, but do not make it current.

auto &	ScanManifestsWith (this auto &self, Executor &executor)
	Configure an executor for manifest planning work.

auto &	ToBranch (this auto &self, const std::string &branch)
	Perform operations on a particular branch.

auto &	Set (this auto &self, const std::string &property, const std::string &value)
	Set a summary property in the snapshot produced by this update.

auto &	WriteManifestsWith (this auto &self, Executor &executor, int32_t parallelism)
	Configure an executor and max writer count for writing new manifests.

Result< ApplyResult >	Apply ()
	Apply the update's changes to create a new snapshot.

Status	Finalize (Result< const TableMetadata * > commit_result) override
	Finalize the snapshot update, cleaning up any uncommitted files.

Public Member Functions inherited from iceberg::PendingUpdate
virtual Status	Commit ()
	Apply the pending changes and commit.

	PendingUpdate (const PendingUpdate &)=delete

PendingUpdate &	operator= (const PendingUpdate &)=delete

	PendingUpdate (PendingUpdate &&) noexcept=default

PendingUpdate &	operator= (PendingUpdate &&) noexcept=default

Public Member Functions inherited from iceberg::ErrorCollector
	ErrorCollector (ErrorCollector &&)=default

ErrorCollector &	operator= (ErrorCollector &&)=default

	ErrorCollector (const ErrorCollector &)=default

ErrorCollector &	operator= (const ErrorCollector &)=default

template<typename... Args>
auto &	AddError (this auto &self, ErrorKind kind, const std::format_string< Args... > fmt, Args &&... args)
	Add a specific error and return reference to derived class.

auto &	AddError (this auto &self, Error err)
	Add an existing error object and return reference to derived class.

auto &	AddError (this auto &self, std::unexpected< Error > err)
	Add an unexpected result's error and return reference to derived class.

bool	has_errors () const
	Check if any errors have been collected.

size_t	error_count () const
	Get the number of errors collected.

Status	CheckErrors () const
	Check for accumulated errors and return them if any exist.

void	ClearErrors ()
	Clear all accumulated errors.

const std::vector< Error > &	errors () const
	Get read-only access to all collected errors.

Protected Member Functions
	MergingSnapshotUpdate (std::string table_name, std::shared_ptr< TransactionContext > ctx)
	Constructor; reads merge configuration from table properties.

Status	AddDataFile (std::shared_ptr< DataFile > file)
	Stage a data file to be added to the table.

Status	AddDeleteFile (std::shared_ptr< DataFile > file)
	Stage a delete file to be added to the table.

Status	ValidateNewDeleteFile (const TableMetadata &metadata, const DataFile &file)
	Validate a delete file against the table format version rules.

Status	AddDeleteFile (std::shared_ptr< DataFile > file, int64_t data_sequence_number)
	Stage a delete file with an explicit data sequence number.

Status	AddManifest (ManifestFile manifest)
	Add all files in a pre-existing data manifest to the new snapshot.

Status	DeleteDataFile (std::shared_ptr< DataFile > file)
	Register a data file (by object) to be deleted from the table.

Status	DeleteDeleteFile (std::shared_ptr< DataFile > file)
	Register a delete file (by object) to be removed from the table.

Status	DeleteByPath (std::string_view path)
	Register a data file path to be deleted from the table.

Status	DeleteByRowFilter (std::shared_ptr< Expression > expr)
	Register an expression to delete matching rows.

Status	DropPartition (int32_t spec_id, PartitionValues partition)
	Register a partition to be dropped.

void	FailMissingDeletePaths ()
	Fail if any registered delete path is not found in any manifest.

void	FailAnyDelete ()
	Fail if any manifest entry matches a delete condition.

void	SetNewDataFilesDataSequenceNumber (int64_t sequence_number)
	Override the data sequence number assigned to all newly-added data files.

bool	HasDataSequenceNumber () const
	Returns true if SetNewDataFilesDataSequenceNumber was called.

void	CaseSensitive (bool case_sensitive)
	Set case sensitivity for row filter and expression evaluation.

bool	IsCaseSensitive () const
	Returns true if case-sensitive matching is enabled (default: true).

bool	AddsDataFiles () const
	Returns true if any data files have been staged for addition.

bool	AddsDeleteFiles () const
	Returns true if any delete files have been staged for addition.

bool	DeletesDataFiles () const
	Returns true if any data files have been registered for deletion.

bool	DeletesDeleteFiles () const
	Returns true if any delete files have been registered for removal.

const std::shared_ptr< Expression > &	RowFilter () const
	Returns the row-filter expression set via DeleteByRowFilter, or nullptr.

Result< std::shared_ptr< PartitionSpec > >	DataSpec () const
	Returns the single partition spec for all staged data files.

std::vector< std::shared_ptr< DataFile > >	AddedDataFiles () const
	Returns all data files staged for addition.

Status	ValidateNoNewDeletesForDataFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, const DataFileSet &replaced_files, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io) const
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a delete file that covers a file in replaced_files.

Status	ValidateAddedDVs (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, std::shared_ptr< Expression > conflict_filter, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io) const
	Return an error if a staged deletion vector conflicts with a deletion vector added since starting_snapshot_id.

Protected Member Functions inherited from iceberg::SnapshotUpdate
	SnapshotUpdate (std::shared_ptr< TransactionContext > ctx)

Result< std::vector< ManifestFile > >	WriteDataManifests (std::span< const std::shared_ptr< DataFile > > files, const std::shared_ptr< PartitionSpec > &spec, std::optional< int64_t > data_sequence_number=std::nullopt)
	Write data manifests for the given data files.

Result< std::vector< ManifestFile > >	WriteDeleteManifests (std::span< const ContentFileWithSequenceNumber > files, const std::shared_ptr< PartitionSpec > &spec)

const std::string &	target_branch () const

bool	can_inherit_snapshot_id () const

const std::string &	commit_uuid () const

int32_t	manifest_count () const

int32_t	attempt () const

int64_t	target_manifest_size_bytes () const

virtual std::string	operation ()=0
	A string that describes the action that produced the new snapshot.

virtual Status	Validate (const TableMetadata &current_metadata, const std::shared_ptr< Snapshot > &snapshot)
	Validate the current metadata.

virtual bool	CleanupAfterCommit () const
	Check if cleanup should happen after commit.

int64_t	SnapshotId ()
	Get or generate the snapshot ID for the new snapshot.

Status	DeleteFile (const std::string &path)
	Delete a file at the given path.

std::string	ManifestPath ()

std::string	ManifestListPath ()

SnapshotSummaryBuilder &	summary_builder ()

SnapshotSummaryBuilder	BuildManifestCountSummary (std::span< const ManifestFile > manifests, int32_t replaced_manifests_count)

Protected Member Functions inherited from iceberg::PendingUpdate
	PendingUpdate (std::shared_ptr< TransactionContext > ctx)

const TableMetadata &	base () const

Static Protected Member Functions
static Status	ValidateAddedDataFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, std::shared_ptr< Expression > filter, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io, bool case_sensitive=true)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a data file matching the given filter expression.

static Status	ValidateAddedDataFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, const PartitionSet &partition_set, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a data file in any partition of the given partition set.

static Status	ValidateDataFilesExist (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, const std::unordered_set< std::string > &file_paths, bool skip_deletes, std::shared_ptr< Expression > filter, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io, bool case_sensitive=true)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, removed a file whose path is in file_paths (and skip_deletes is false).

static Status	ValidateNoNewDeletesForDataFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, const DataFileSet &replaced_files, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io, bool ignore_equality_deletes=false)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a delete file that covers a file in replaced_files.

static Status	ValidateNoNewDeletesForDataFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, std::shared_ptr< Expression > data_filter, const DataFileSet &replaced_files, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io, bool case_sensitive=true)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a delete file matching the data filter that covers a file in replaced_files.

static Status	ValidateNoNewDeleteFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, std::shared_ptr< Expression > data_filter, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io, bool case_sensitive=true)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a delete file matching the given row filter.

static Status	ValidateNoNewDeleteFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, const PartitionSet &partition_set, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a delete file matching any partition in the given partition set.

static Status	ValidateDeletedDataFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, std::shared_ptr< Expression > data_filter, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io, bool case_sensitive=true)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, deleted a data file matching the given row filter.

static Status	ValidateDeletedDataFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, const PartitionSet &partition_set, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, deleted a data file in any partition of the given partition set.

static Result< std::unique_ptr< DeleteFileIndex > >	AddedDeleteFiles (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, std::shared_ptr< Expression > data_filter, std::shared_ptr< PartitionSet > partition_set, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io, bool case_sensitive=true)
	Build a DeleteFileIndex of delete files added since starting_snapshot_id.

static Status	ValidateAddedDVs (const TableMetadata &metadata, std::optional< int64_t > starting_snapshot_id, std::shared_ptr< Expression > conflict_filter, const std::unordered_set< std::string > &referenced_data_files, const std::shared_ptr< Snapshot > &parent, std::shared_ptr< FileIO > io, bool case_sensitive=true)
	Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a deletion vector that conflicts with DVs being written.

Additional Inherited Members
Public Types inherited from iceberg::PendingUpdate
enum class	Kind : uint8_t { kExpireSnapshots , kSetSnapshot , kUpdateLocation , kUpdatePartitionSpec , kUpdatePartitionStatistics , kUpdateProperties , kUpdateSchema , kUpdateSnapshot , kUpdateSnapshotReference , kUpdateSortOrder , kUpdateStatistics }

Protected Attributes inherited from iceberg::SnapshotUpdate
SnapshotSummaryBuilder	summary_

Protected Attributes inherited from iceberg::PendingUpdate
std::shared_ptr< TransactionContext >	ctx_

Protected Attributes inherited from iceberg::ErrorCollector
std::vector< Error >	errors_

Detailed Description

Abstract base class for merge-based snapshot write operations.

This class is the C++ counterpart of Java's MergingSnapshotProducer. It provides the shared filter, write, and merge pipeline used by operations that accumulate file additions and deletions, produce a new Snapshot, and commit that snapshot as current. Commit conflicts are resolved by applying the same staged changes to the new latest snapshot and reattempting the commit.

Subclasses describe their changes by calling the protected primitive API and implement operation() to set the snapshot operation summary.

The Apply() pipeline:

Filter data manifests (via data_filter_manager_)
Compute min data sequence number and set up delete filter cleanup
Filter delete manifests (via delete_filter_manager_)
Write new data manifests (cached for commit retry)
Write new delete manifests (cached for commit retry)
Merge data manifests (via data_merge_manager_)
Merge delete manifests (via delete_merge_manager_)

TODO(Guotao): Java MergingSnapshotProducer overrides updateEvent() to return a CreateSnapshotEvent(tableName, operation, snapshotId, sequenceNumber, summary) for commit listeners. The C++ update framework does not yet have an event notification mechanism, so this is intentionally not implemented here. Add it once an equivalent CreateSnapshotEvent / listener facility exists.

Member Function Documentation

◆ AddManifest()

Status iceberg::MergingSnapshotUpdate::AddManifest ( ManifestFile manifest )

protected

Add all files in a pre-existing data manifest to the new snapshot.

The manifest must contain only appended data files. If snapshot ID inheritance is enabled and the manifest has no snapshot ID assigned, it is used directly; otherwise it is copied with the current snapshot ID.

◆ Apply()

Result< std::vector< ManifestFile > > iceberg::MergingSnapshotUpdate::Apply	(	const TableMetadata &	metadata_to_update,
		const std::shared_ptr< Snapshot > &	snapshot
	)

overridevirtual

Apply the update's changes to the given metadata and snapshot.

Parameters

metadata_to_update	The base table metadata to apply changes to
snapshot	Snapshot to apply the changes to

Returns: A vector of manifest files for the new snapshot

Implements iceberg::SnapshotUpdate.

◆ CleanUncommitted()

Status iceberg::MergingSnapshotUpdate::CleanUncommitted ( const std::unordered_set< std::string > & committed )

overridevirtual

Clean up any uncommitted manifests that were created.

Manifests may not be committed if Apply is called multiple times because a commit conflict has occurred. Implementations may keep around manifests because the same changes will be made by both Apply calls. This method instructs the implementation to clean up those manifests and passes the paths of the manifests that were actually committed.

Parameters

committed A set of manifest paths that were actually committed

Implements iceberg::SnapshotUpdate.

◆ DataSpec()

Result< std::shared_ptr< PartitionSpec > > iceberg::MergingSnapshotUpdate::DataSpec ( ) const

protected

Returns the single partition spec for all staged data files.

Precondition: exactly one partition spec ID must be represented among staged data files.

◆ DeleteByPath()

Status iceberg::MergingSnapshotUpdate::DeleteByPath ( std::string_view path )

protected

Register a data file path to be deleted from the table.

Note: Only applies to data files. To remove delete files, use DeleteDeleteFile().

◆ DeleteByRowFilter()

Status iceberg::MergingSnapshotUpdate::DeleteByRowFilter ( std::shared_ptr< Expression > expr )

protected

Register an expression to delete matching rows.

Both data and delete filter managers receive the expression: delete files that match the row filter can also be removed because those rows will be deleted.

◆ DropPartition()

Status iceberg::MergingSnapshotUpdate::DropPartition	(	int32_t	spec_id,
		PartitionValues	partition
	)

protected

Register a partition to be dropped.

Both data and delete filter managers receive the partition drop, since dropping data in a partition also drops all delete files in that partition.

◆ Summary()

std::unordered_map< std::string, std::string > iceberg::MergingSnapshotUpdate::Summary ( )

overridevirtual

Get the summary map for this operation.

Returns: A map of summary properties

Implements iceberg::SnapshotUpdate.

◆ ValidateNewDeleteFile()

Status iceberg::MergingSnapshotUpdate::ValidateNewDeleteFile	(	const TableMetadata &	metadata,
		const DataFile &	file
	)

protected

Validate a delete file against the table format version rules.

Format v1: deletes are not supported.
Format v2: position deletes must NOT be deletion vectors (DVs).
Format v3+: position deletes MUST be deletion vectors (DVs).

◆ ValidateNoNewDeletesForDataFiles() [1/2]

Status iceberg::MergingSnapshotUpdate::ValidateNoNewDeletesForDataFiles	(	const TableMetadata &	metadata,
		std::optional< int64_t >	starting_snapshot_id,
		const DataFileSet &	replaced_files,
		const std::shared_ptr< Snapshot > &	parent,
		std::shared_ptr< FileIO >	io
	)		const

inlineprotected

Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a delete file that covers a file in replaced_files.

Whether equality deletes are checked is derived automatically from whether a custom data sequence number was set via SetNewDataFilesDataSequenceNumber(): if set, equality deletes are ignored because they still apply to the rewritten files and are not a conflict.

Subclasses should prefer this overload over the static one.

◆ ValidateNoNewDeletesForDataFiles() [2/2]

Status iceberg::MergingSnapshotUpdate::ValidateNoNewDeletesForDataFiles	(	const TableMetadata &	metadata,
		std::optional< int64_t >	starting_snapshot_id,
		const DataFileSet &	replaced_files,
		const std::shared_ptr< Snapshot > &	parent,
		std::shared_ptr< FileIO >	io,
		bool	ignore_equality_deletes = `false`
	)

staticprotected

Return an error if any snapshot after starting_snapshot_id, or from the beginning if unset, added a delete file that covers a file in replaced_files.

Parameters

ignore_equality_deletes If true, only position deletes are checked. Set to true when replaced data files have the same sequence number as the new files (e.g. RewriteFiles), so equality deletes at higher sequence numbers still apply and are not a conflict.

The documentation for this class was generated from the following files:

iceberg/update/merging_snapshot_update.h
iceberg/update/merging_snapshot_update.cc

Public Member Functions

Protected Member Functions

Static Protected Member Functions

Additional Inherited Members

Detailed Description

Member Function Documentation

◆ AddManifest()

◆ Apply()

◆ CleanUncommitted()

◆ DataSpec()

◆ DeleteByPath()

◆ DeleteByRowFilter()

◆ DropPartition()

◆ Summary()

◆ ValidateNewDeleteFile()

◆ ValidateNoNewDeletesForDataFiles() [1/2]

◆ ValidateNoNewDeletesForDataFiles() [2/2]