kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType > Class Template Reference

A plugin implementing a wrapper around the User-Level Failure-Mitigation (ULFM) feature of the upcoming MPI 4 standard. This plugin and the accompanying example is tested with OpenMPI 5.0.2. More...

#include <ulfm.hpp>

Inheritance diagram for kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >:

Collaboration diagram for kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >:

Public Member Functions
	UserLevelFailureMitigation ()
	Default constructor; sets the error handler of MPI_COMM_WORLD (!) to MPI_ERRORS_RETURN. Although the standard allows setting the error handler for only a specific communicator; neither MPICH nor OpenMPI currently (March 2024) support this.
void	revoke ()
	Revokes the current communicator.
uint32_t	ack_failed (uint32_t const num_to_ack)
	Acknowledges that the application intends to ignore the effect of currently known failures on wildcard receive completions and agreement return values.
uint32_t	num_ack_failed ()
	Gets the number of acknowledged failures.
uint32_t	ack_all_failed ()
	Acknowledge all failures.
Comm	shrink ()
	Creates a new communicator from this communicator, excluding the failed processes.
int	agree (int flag)
	Agrees on a flag from all live processes and distributes the result back to all live processes, even after process failures.
bool	agree (bool flag)
	Agrees on a boolean flag from all live processes and distributes the result back to all live processes, even after process failures.
Group	get_failed ()
	Obtains the group of currently failed processes.
bool	is_revoked ()
	Checks if this communicator has been revoked.
void	mpi_error_handler (int const ret, std::string const &callee) const
	Overwrite the on-MPI-error handler to throw appropriate exceptions for then hardware faults happened.

Detailed Description

template<typename Comm, template< typename... > typename DefaultContainerType>
class kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >

A plugin implementing a wrapper around the User-Level Failure-Mitigation (ULFM) feature of the upcoming MPI 4 standard. This plugin and the accompanying example is tested with OpenMPI 5.0.2.

Member Function Documentation

ack_all_failed()

template<typename Comm , template< typename... > typename DefaultContainerType>

uint32_t kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >::ack_all_failed ( )

inline

Acknowledge all failures.

Returns

The overall number of failures acknowledged.

ack_failed()

template<typename Comm , template< typename... > typename DefaultContainerType>

uint32_t kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >::ack_failed ( uint32_t const num_to_ack )

inline

Acknowledges that the application intends to ignore the effect of currently known failures on wildcard receive completions and agreement return values.

Parameters

num_to_ack

The number of failures to acknowledge.

Returns

The overall number of failures acknowledged.

agree() [1/2]

template<typename Comm , template< typename... > typename DefaultContainerType>

bool kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >::agree ( bool flag )

inline

Agrees on a boolean flag from all live processes and distributes the result back to all live processes, even after process failures.

Parameters

flag	The flag to agree on.

Returns

The bitwise AND over the contributed input values of flag.

agree() [2/2]

template<typename Comm , template< typename... > typename DefaultContainerType>

int kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >::agree ( int flag )

inline

Agrees on a flag from all live processes and distributes the result back to all live processes, even after process failures.

Parameters

flag	The flag to agree on.

Returns

The bitwise AND over the contributed input values of flag.

get_failed()

template<typename Comm , template< typename... > typename DefaultContainerType>

Group kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >::get_failed ( )

inline

Obtains the group of currently failed processes.

Returns

The group of currently failed processes.

is_revoked()

template<typename Comm , template< typename... > typename DefaultContainerType>

bool kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >::is_revoked ( )

inline

Checks if this communicator has been revoked.

Returns

True if the communicator has been revoked, false otherwise.

num_ack_failed()

template<typename Comm , template< typename... > typename DefaultContainerType>

uint32_t kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >::num_ack_failed ( )

inline

Gets the number of acknowledged failures.

Returns

The number of acknowledged failures.

shrink()

template<typename Comm , template< typename... > typename DefaultContainerType>

Comm kamping::plugin::UserLevelFailureMitigation< Comm, DefaultContainerType >::shrink ( )

inline

Creates a new communicator from this communicator, excluding the failed processes.

Returns

The new communicator.

---

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

• include/kamping/plugin/ulfm.hpp