Checkpoint/Restart infrastructure. More...

#include <checkpoint_control.hpp>

Public Member Functions
	CheckpointControl (const Dist::Comm &comm, const LAFEM::SerialConfig &config=LAFEM::SerialConfig())
	Constructor. More...

	~CheckpointControl ()
	Destructor. More...

template<typename OT_ >
void	add_object (String identifier, OT_ &object)
	Register checkpointable object. More...

void	clear_input ()
	Delete all read input. More...

String	get_identifier_list ()
	Retrieve a list of all items stored in the checkpoint. More...

void	load (BinaryStream &bs)
	Load checkpoint from stream. More...

void	load (const String filename)
	Load checkpoint from disk. More...

void	remove_object (String identifier)
	Remove checkpointable object. More...

template<typename OT_ >
void	restore_object (String identifier, OT_ &object, bool add_to_checkpoint_control=true)
	Restore checkpointable object. More...

void	save (BinaryStream &bs)
	Save checkpoint to a stream. More...

void	save (const String filename)
	Write checkpoint file to disk. More...

void	set_config (FEAT::LAFEM::SerialConfig &conf)
	Set a new serialize configuration. More...

Private Member Functions
std::uint64_t	_collect_checkpoint_data (std::vector< char > &buffer)
	Build checkpoint buffer. More...

void	_load (const String name)
	Load checkpoint from disk. More...

void	_restore_checkpoint_data ()
	Extract input buffer. More...

void	_save (const String name)
	Write checkpoint files to disk. More...

Private Attributes
std::map< String, std::shared_ptr< Checkpointable > >	_checkpointable_by_identifier
	Mapping of identifier string to pointer to the checkpointable object. More...

const Dist::Comm &	_comm
	the mpi communicator identifying our mpi context More...

FEAT::LAFEM::SerialConfig	_config
	The config class that controls the compression modes. More...

std::vector< char >	_input_array
	Vector holding the array read from the input file during restore. More...

std::map< String, std::uint64_t >	_offset_by_identifier
	Mapping of identifier string to the offset in the input file. More...

Detailed Description

Checkpoint/Restart infrastructure.

This class provides a general interface for checkpoint/restart usecases. Compatible objects can be summed up to a single checkpoint. This checkpoint can then be stored and restored via mpi io mechanics. Afterwards, any object can be restored from the recreated checkpoint.

Note: The checkpoint control class object is used on a per rank basis. The contents of all ranks objects will be written into one single file. Later on, the loaded objects will, again, be local objects in every rank with distinct contents.

Todo:: do we need extra handling for machines with no global acessible file servers?

Definition at line 140 of file checkpoint_control.hpp.

Constructor & Destructor Documentation

◆ CheckpointControl()

FEAT::Control::CheckpointControl::CheckpointControl	(	const Dist::Comm &	comm,
		const LAFEM::SerialConfig &	config = `LAFEM::SerialConfig()`
	)

inline

Constructor.

Initalize the input array as NULL pointer.

Parameters

[in]	comm	The communicator common to all stored objects
[in]	config	A config class, controlling the compression of the written out data

Definition at line 280 of file checkpoint_control.hpp.

References _input_array.

◆ ~CheckpointControl()

FEAT::Control::CheckpointControl::~CheckpointControl ( )

inline

Destructor.

Destroy the checkpoint control instance and delete the input array.

Definition at line 292 of file checkpoint_control.hpp.

Member Function Documentation

◆ _collect_checkpoint_data()

std::uint64_t FEAT::Control::CheckpointControl::_collect_checkpoint_data ( std::vector< char > & buffer )

inlineprivate

Build checkpoint buffer.

Collect all data needed for a checkpoint from every object registered at the checkpoint control.

Parameters

[in] buffer buffer containing the collected data

Returns: the size of the buffer

Definition at line 163 of file checkpoint_control.hpp.

References _checkpointable_by_identifier, and _config.

Referenced by _save(), and save().

◆ _load()

void FEAT::Control::CheckpointControl::_load ( const String name )

inlineprivate

Load checkpoint from disk.

Read binary file [name].cp holding a safed state of all checkpointable objects and store it to _input_array.

Fill the _offset_by_identifier map with the data from the array.

Parameters

[in] name String holding the name of the file (without the extension .zcp)

Warning: Reading another input file / stream will overwrite values in _input_array and _offset_by_identifier

Definition at line 260 of file checkpoint_control.hpp.

References _comm, _input_array, _restore_checkpoint_data(), and FEAT::DistFileIO::read_combined().

Referenced by load().

◆ _restore_checkpoint_data()

void FEAT::Control::CheckpointControl::_restore_checkpoint_data ( )

inlineprivate

Extract input buffer.

Extract the identifier string and offset for its data in the input array for every object in the read checkpoint, and store this information for restoration in this checkpoint control object.

Definition at line 208 of file checkpoint_control.hpp.

References _input_array, and _offset_by_identifier.

Referenced by _load(), and load().

◆ _save()

void FEAT::Control::CheckpointControl::_save ( const String name )

inlineprivate

Write checkpoint files to disk.

Write one binary file: [name].cp: a uncompressed binary file holding the current status of all checkpointable objects added to the checkpoint control

Parameters

[in] name String holding the name of the file (without the extension .cp)

Definition at line 240 of file checkpoint_control.hpp.

References _collect_checkpoint_data(), _comm, and FEAT::DistFileIO::write_combined().

Referenced by save().

◆ add_object()

template<typename OT_ >

void FEAT::Control::CheckpointControl::add_object	(	String	identifier,
		OT_ &	object
	)

inline

Register checkpointable object.

Adds a pointer to a checkpointable object to the checkpoint control.

The object will be internally identified by its indentifier string (especially during restoration).

Parameters

[in]	identifier	String used as identifier
[in]	object	Pointer to the checkpointable object

Definition at line 346 of file checkpoint_control.hpp.

References _checkpointable_by_identifier, and XASSERTM.

Referenced by restore_object().

◆ clear_input()

void FEAT::Control::CheckpointControl::clear_input ( )

inline

Delete all read input.

Clear the input read before from this checkpoint control object. Does not touch the map holding the pointers to the checkpointable objects included into the following written checkpoints.

Definition at line 313 of file checkpoint_control.hpp.

References _input_array, and _offset_by_identifier.

◆ get_identifier_list()

String FEAT::Control::CheckpointControl::get_identifier_list ( )

inline

Retrieve a list of all items stored in the checkpoint.

Definition at line 320 of file checkpoint_control.hpp.

References _checkpointable_by_identifier.

◆ load() [1/2]

void FEAT::Control::CheckpointControl::load ( BinaryStream & bs )

inline

Load checkpoint from stream.

Read the BinaryStream, to restore the safed state of all objects registered to the checkpoint control and store it to _input_array.

Fill the _offset_by_identifier map with the data from the array.

Parameters

[in] bs BinaryStream that shall be read from

Warning: Reading another input file / stream will overwrite values in _input_array and _offset_by_identifier

Definition at line 488 of file checkpoint_control.hpp.

References _input_array, _restore_checkpoint_data(), FEAT::BinaryStream::data(), and XASSERTM.

◆ load() [2/2]

void FEAT::Control::CheckpointControl::load ( const String filename )

inline

Load checkpoint from disk.

Read the [filename.cp] file, to restore the size of the checkpoint data per rank.

Read binary file [filename] holding a safed state of all checkpointable objects and store it to _input_array.

Fill the _offset_by_identifier map with the data from the array.

Parameters

[in] filename String holding the name of the file

Warning: Reading another input file / stream will overwrite values in _input_array and _offset_by_identifier

Definition at line 440 of file checkpoint_control.hpp.

References _input_array, _load(), XABORTM, and XASSERTM.

◆ remove_object()

void FEAT::Control::CheckpointControl::remove_object ( String identifier )

inline

Remove checkpointable object.

Remove pointer to a checkpointable object from the checkpoint control.

Parameters

[in] identifier Identifier of the object to remove

Definition at line 361 of file checkpoint_control.hpp.

References _checkpointable_by_identifier, and XASSERTM.

◆ restore_object()

template<typename OT_ >

void FEAT::Control::CheckpointControl::restore_object	(	String	identifier,
		OT_ &	object,
		bool	add_to_checkpoint_control = `true`
	)

inline

Restore checkpointable object.

Restore a checkpointable object from the checkpoint control. The objects contents will be found by its identifier string and restored.

Parameters

[in]	identifier	Identifier of the object to restore
[out]	object	Reference to object
[in]	add_to_checkpoint_control	If true the restored object is added to the checkpoint control, default true

Warning: The user has to load an input file before and has to ensure proper size/data/whatsover type layout.

Definition at line 380 of file checkpoint_control.hpp.

References _input_array, _offset_by_identifier, add_object(), and XASSERTM.

◆ save() [1/2]

void FEAT::Control::CheckpointControl::save ( BinaryStream & bs )

inline

Save checkpoint to a stream.

Save a checkpoint, holding the current status of all registered objects, in a BinaryStream

Parameters

[in] bs BinaryStream that shall be written to

Definition at line 467 of file checkpoint_control.hpp.

References _collect_checkpoint_data().

◆ save() [2/2]

void FEAT::Control::CheckpointControl::save ( const String filename )

inline

Write checkpoint file to disk.

Write one uncompressed binary file: filename: [name.cp]: holding the current status of all checkpointable objects added to the checkpoint control

Parameters

[in] filename String holding the complete name of the file with extension .cp

Definition at line 407 of file checkpoint_control.hpp.

References _save(), and XASSERTM.

◆ set_config()

void FEAT::Control::CheckpointControl::set_config ( FEAT::LAFEM::SerialConfig & conf )

inline

Set a new serialize configuration.

Parameters

[in] conf LAFEM::SerialConfig, a config class holding information about the compression parameters.

Set a new SerialConfig object.

Definition at line 303 of file checkpoint_control.hpp.

References _config.

Member Data Documentation

◆ _checkpointable_by_identifier

std::map<String, std::shared_ptr<Checkpointable> > FEAT::Control::CheckpointControl::_checkpointable_by_identifier

private

Mapping of identifier string to pointer to the checkpointable object.

Definition at line 144 of file checkpoint_control.hpp.

Referenced by _collect_checkpoint_data(), add_object(), get_identifier_list(), and remove_object().

◆ _comm

const Dist::Comm& FEAT::Control::CheckpointControl::_comm

private

the mpi communicator identifying our mpi context

Definition at line 150 of file checkpoint_control.hpp.

Referenced by _load(), and _save().

◆ _config

FEAT::LAFEM::SerialConfig FEAT::Control::CheckpointControl::_config

private

The config class that controls the compression modes.

Definition at line 152 of file checkpoint_control.hpp.

Referenced by _collect_checkpoint_data(), and set_config().

◆ _input_array

std::vector<char> FEAT::Control::CheckpointControl::_input_array

private

Vector holding the array read from the input file during restore.

Definition at line 146 of file checkpoint_control.hpp.

Referenced by CheckpointControl(), _load(), _restore_checkpoint_data(), clear_input(), load(), and restore_object().

◆ _offset_by_identifier

std::map<String, std::uint64_t> FEAT::Control::CheckpointControl::_offset_by_identifier

private

Mapping of identifier string to the offset in the input file.

Definition at line 148 of file checkpoint_control.hpp.

Referenced by _restore_checkpoint_data(), clear_input(), and restore_object().

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

control/checkpoint_control.hpp

Public Member Functions

Private Member Functions

Private Attributes

Detailed Description

Constructor & Destructor Documentation

◆ CheckpointControl()

◆ ~CheckpointControl()

Member Function Documentation

◆ _collect_checkpoint_data()

◆ _load()

◆ _restore_checkpoint_data()

◆ _save()

◆ add_object()

◆ clear_input()

◆ get_identifier_list()

◆ load() [1/2]

◆ load() [2/2]

◆ remove_object()

◆ restore_object()

◆ save() [1/2]

◆ save() [2/2]

◆ set_config()

Member Data Documentation

◆ _checkpointable_by_identifier

◆ _comm

◆ _config

◆ _input_array

◆ _offset_by_identifier