Server : Apache/2.4.41 (Ubuntu) System : Linux absol.cf 5.4.0-198-generic #218-Ubuntu SMP Fri Sep 27 20:18:53 UTC 2024 x86_64 User : www-data ( 33) PHP Version : 7.4.33 Disable Function : pcntl_alarm,pcntl_fork,pcntl_waitpid,pcntl_wait,pcntl_wifexited,pcntl_wifstopped,pcntl_wifsignaled,pcntl_wifcontinued,pcntl_wexitstatus,pcntl_wtermsig,pcntl_wstopsig,pcntl_signal,pcntl_signal_get_handler,pcntl_signal_dispatch,pcntl_get_last_error,pcntl_strerror,pcntl_sigprocmask,pcntl_sigwaitinfo,pcntl_sigtimedwait,pcntl_exec,pcntl_getpriority,pcntl_setpriority,pcntl_async_signals,pcntl_unshare, Directory : /etc/alternatives/mpi/openmpi/orte/mca/snapc/
Upload File :
Current File : //etc/alternatives/mpi/openmpi/orte/mca/snapc/snapc.h
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2010 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file
*
* Snapshot Coordination (SNAPC) Interface
*
* Terminology:
* ------------
* Global Snapshot Coordinator:
* - HNP(s) coordination function.
* Local Snapshot Coordinator
* - VHNP(s) [e.g., orted] coordination function
* Application Snapshot Coordinator
* - Application level coordinaton function
* Local Snapshot
* - Snapshot generated by a single process in the parallel job
* Local Snapshot Reference
* - A generic reference to the physical Local Snapshot
* Global Snapshot
* - Snapshot generated for the entire parallel job
* Global Snapshot Reference
* - A generic reference to the physical Global Snapshot
*
* General Description:
* ---------------------
* This framework is tasked with:
* - Initiating the checkpoint in the system
* - Physically moving the local snapshot files to a location
* Initially this location, is the node on which the Head Node Process (HNP)
* is running, but later this will be a replicated checkpoint server or
* the like.
* - Generating a 'global snapshot handle' that the user can use to restart
* the parallel job.
*
* Each component will have 3 teirs of behavior that must behave in concert:
* - Global Snapshot Coordinator
* This is the HNPs tasks. Mostly distributing the notification of the
* checkpoint, and then compiling the physical and virtual nature of the
* global snapshot handle.
* - Local Snapshot Coordinator
* This is the VHNPs (or orted, if available) tasks. This will involve
* working with the Global Snapshot Coordinator to route the physical
* and virtual 'local snapshot's from the application to the desired
* location. This process must also notify the Global Snapshot Coordinator
* when it's set of processes have completed the checkpoint.
* - Application Snapshot Coordinator
* This is the application level coordinator. This is very light, just
* a subscription to be triggered when it needs to checkpoint, and then,
* once finished with the checkpoint, notify the Local Snapshot Coordinator
* that it is complete.
* If there is no orted (so no bootproxy), then the application assumes the
* responsibility of the Local Snapshot Coordinator as well.
*
*/
#ifndef MCA_SNAPC_H
#define MCA_SNAPC_H
#include "orte_config.h"
#include "orte/constants.h"
#include "orte/types.h"
#include "orte/mca/mca.h"
#include "opal/mca/base/base.h"
#include "opal/mca/crs/crs.h"
#include "opal/mca/crs/base/base.h"
#include "opal/class/opal_object.h"
#include "opal/class/opal_pointer_array.h"
#include "opal/util/output.h"
#include "orte/mca/sstore/sstore.h"
BEGIN_C_DECLS
/**
* States that a process can be in while checkpointing
*/
/* Reached an error */
#define ORTE_SNAPC_CKPT_STATE_ERROR 0
/* Doing no checkpoint -- Quiet state */
#define ORTE_SNAPC_CKPT_STATE_NONE 1
/* There has been a request for a checkpoint from one of the applications */
#define ORTE_SNAPC_CKPT_STATE_REQUEST 2
/* There is a Pending checkpoint for this process */
#define ORTE_SNAPC_CKPT_STATE_PENDING 3
/* Running the checkpoint */
#define ORTE_SNAPC_CKPT_STATE_RUNNING 4
/* INC Prep Finished */
#define ORTE_SNAPC_CKPT_STATE_INC_PREPED 5
/* All Processes have been stopped */
#define ORTE_SNAPC_CKPT_STATE_STOPPED 6
/* Finished the checkpoint locally */
#define ORTE_SNAPC_CKPT_STATE_FINISHED_LOCAL 7
/* Migrating */
#define ORTE_SNAPC_CKPT_STATE_MIGRATING 8
/* Finished establishing the checkpoint */
#define ORTE_SNAPC_CKPT_STATE_ESTABLISHED 9
/* Processes continuing or have been recovered (finished post-INC) */
#define ORTE_SNAPC_CKPT_STATE_RECOVERED 10
/* Unable to checkpoint this job */
#define ORTE_SNAPC_CKPT_STATE_NO_CKPT 11
/* Unable to restart this job */
#define ORTE_SNAPC_CKPT_STATE_NO_RESTART 12
#define ORTE_SNAPC_CKPT_MAX 13
/**
* Sufficiently high shift value to avoid colliding the process
* checkpointing states above with the ORTE process states
*/
#define ORTE_SNAPC_CKPT_SHIFT 131072
/* Uniquely encode the SNAPC state */
#define ORTE_SNAPC_CKPT_NOTIFY(state) (ORTE_SNAPC_CKPT_SHIFT + state)
/* Decode the SNAPC state */
#define ORTE_SNAPC_CKPT_STATE(state) (state - ORTE_SNAPC_CKPT_SHIFT)
/* Check whether a state is a SNAPC state or not. */
#define CHECK_ORTE_SNAPC_CKPT_STATE(state) (state >= ORTE_SNAPC_CKPT_SHIFT)
/**
* Definition of a orte local snapshot.
* Similar to the opal_crs_base_snapshot_t except that it
* contains process contact information.
*/
struct orte_snapc_base_local_snapshot_1_0_0_t {
/** List super object */
opal_list_item_t super;
/** ORTE Process name */
orte_process_name_t process_name;
/** State of the checkpoint */
int state;
/** Stable Storage Handle (must equal the global version) */
orte_sstore_base_handle_t ss_handle;
};
typedef struct orte_snapc_base_local_snapshot_1_0_0_t orte_snapc_base_local_snapshot_1_0_0_t;
typedef struct orte_snapc_base_local_snapshot_1_0_0_t orte_snapc_base_local_snapshot_t;
ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_snapc_base_local_snapshot_t);
/**
* Definition of the global snapshot.
* Each component is assumed to have extened this definition
* in the same way they extern the orte_snapc_base_compoinent_t below.
*/
struct orte_snapc_base_global_snapshot_1_0_0_t {
/** This is an object, so must have super */
opal_list_item_t super;
/** A list of orte_snapc_base_snapshot_t's */
opal_list_t local_snapshots;
/** Checkpoint Options */
opal_crs_base_ckpt_options_t *options;
/** Stable Storage Handle */
orte_sstore_base_handle_t ss_handle;
};
typedef struct orte_snapc_base_global_snapshot_1_0_0_t orte_snapc_base_global_snapshot_1_0_0_t;
typedef struct orte_snapc_base_global_snapshot_1_0_0_t orte_snapc_base_global_snapshot_t;
ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_snapc_base_global_snapshot_t);
struct orte_snapc_base_quiesce_1_0_0_t {
/** Parent is an object type */
opal_object_t super;
/** Current epoch */
int epoch;
/** Requested CRS */
char * crs_name;
/** Handle for reference */
char * handle;
/** snapshot list */
orte_snapc_base_global_snapshot_t *snapshot;
/** Stable Storage Handle */
orte_sstore_base_handle_t ss_handle;
/** Stable Storage Snapshot list */
orte_sstore_base_global_snapshot_info_t *ss_snapshot;
/** Target Directory */
char * target_dir;
/** Command Line */
char * cmdline;
/** State of operation if checkpointing */
opal_crs_state_type_t cr_state;
/** Checkpointing? */
bool checkpointing;
/** Restarting? */
bool restarting;
/** Migrating? */
bool migrating;
/** List of migrating processes */
int num_migrating;
opal_pointer_array_t migrating_procs;
};
typedef struct orte_snapc_base_quiesce_1_0_0_t orte_snapc_base_quiesce_1_0_0_t;
typedef struct orte_snapc_base_quiesce_1_0_0_t orte_snapc_base_quiesce_t;
ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_snapc_base_quiesce_t);
/**
* Application request for a global checkpoint related operation
*/
typedef enum {
ORTE_SNAPC_OP_NONE = 0,
ORTE_SNAPC_OP_INIT,
ORTE_SNAPC_OP_FIN,
ORTE_SNAPC_OP_FIN_ACK,
ORTE_SNAPC_OP_CHECKPOINT,
ORTE_SNAPC_OP_RESTART,
ORTE_SNAPC_OP_MIGRATE,
ORTE_SNAPC_OP_QUIESCE_START,
ORTE_SNAPC_OP_QUIESCE_CHECKPOINT,
ORTE_SNAPC_OP_QUIESCE_END
} orte_snapc_base_request_op_event_t;
struct orte_snapc_base_request_op_1_0_0_t {
/** Parent is an object type */
opal_object_t super;
/** Event to request */
orte_snapc_base_request_op_event_t event;
/** Is this request still active */
bool is_active;
/** Leader of the operation */
int leader;
/** Sequence Number */
int seq_num;
/** Global Handle */
char * global_handle;
/** Stable Storage Handle */
orte_sstore_base_handle_t ss_handle;
/** Migrating vpid list of participants */
int mig_num;
int *mig_vpids;
/** Migrating hostname preference list */
char (*mig_host_pref)[OPAL_MAX_PROCESSOR_NAME];
/** Migrating vpid preference list */
int *mig_vpid_pref;
/** Info key */
int *mig_off_node;
};
typedef struct orte_snapc_base_request_op_1_0_0_t orte_snapc_base_request_op_1_0_0_t;
typedef struct orte_snapc_base_request_op_1_0_0_t orte_snapc_base_request_op_t;
ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_snapc_base_request_op_t);
/**
* Module initialization function.
* Returns ORTE_SUCCESS
*/
typedef int (*orte_snapc_base_module_init_fn_t)
(bool seed, bool app);
/**
* Module finalization function.
* Returns ORTE_SUCCESS
*/
typedef int (*orte_snapc_base_module_finalize_fn_t)
(void);
/**
* Setup the necessary structures for this job
* Returns ORTE_SUCCESS
*/
typedef int (*orte_snapc_base_setup_job_fn_t)
(orte_jobid_t jobid);
/**
* Setup the necessary structures for this job
* Returns ORTE_SUCCESS
*/
typedef int (*orte_snapc_base_release_job_fn_t)
(orte_jobid_t jobid);
/**
* Handle fault tolerance updates
*
* @param[in] state Fault tolerance state update
*
* @retval ORTE_SUCCESS The operation completed successfully
* @retval ORTE_ERROR An unspecifed error occurred
*/
typedef int (*orte_snapc_base_ft_event_fn_t)(int state);
/**
* Start a checkpoint originating from an internal source.
*
* This really only makes sense to call from an application, but in the future
* we may allow the checkpoint operation to use this function from the local
* coordinator.
*
* @param[out] epoch Epoch number to associate with this checkpoint operation
* Returns ORTE_SUCCESS
*/
typedef int (*orte_snapc_base_start_checkpoint_fn_t)
(orte_snapc_base_quiesce_t *datum);
/**
* Signal end of checkpoint epoch originating from an internal source.
*
* @param[in] epoch Epoch number to associate with this checkpoint operation
* Returns ORTE_SUCCESS
*/
typedef int (*orte_snapc_base_end_checkpoint_fn_t)
(orte_snapc_base_quiesce_t *datum);
/**
* Request a checkpoint related operation to take place
*/
typedef int (*orte_snapc_base_request_op_fn_t)
(orte_snapc_base_request_op_t *datum);
/**
* Structure for SNAPC components.
*/
struct orte_snapc_base_component_2_0_0_t {
/** MCA base component */
mca_base_component_t base_version;
/** MCA base data */
mca_base_component_data_t base_data;
/** Verbosity Level */
int verbose;
/** Output Handle for opal_output */
int output_handle;
/** Default Priority */
int priority;
};
typedef struct orte_snapc_base_component_2_0_0_t orte_snapc_base_component_2_0_0_t;
typedef struct orte_snapc_base_component_2_0_0_t orte_snapc_base_component_t;
/**
* Structure for SNAPC modules
*/
struct orte_snapc_base_module_1_0_0_t {
/** Initialization Function */
orte_snapc_base_module_init_fn_t snapc_init;
/** Finalization Function */
orte_snapc_base_module_finalize_fn_t snapc_finalize;
/** Setup structures for a job */
orte_snapc_base_setup_job_fn_t setup_job;
/** Release job */
orte_snapc_base_release_job_fn_t release_job;
/** Handle any FT Notifications */
orte_snapc_base_ft_event_fn_t ft_event;
/** Handle internal request for checkpoint */
orte_snapc_base_start_checkpoint_fn_t start_ckpt;
orte_snapc_base_end_checkpoint_fn_t end_ckpt;
/** Handle a checkpoint related request */
orte_snapc_base_request_op_fn_t request_op;
};
typedef struct orte_snapc_base_module_1_0_0_t orte_snapc_base_module_1_0_0_t;
typedef struct orte_snapc_base_module_1_0_0_t orte_snapc_base_module_t;
ORTE_DECLSPEC extern orte_snapc_base_module_t orte_snapc;
ORTE_DECLSPEC extern orte_snapc_base_component_t orte_snapc_base_selected_component;
/**
* Macro for use in components that are of type SNAPC
*/
#define ORTE_SNAPC_BASE_VERSION_2_0_0 \
ORTE_MCA_BASE_VERSION_2_1_0("snapc", 2, 0, 0)
END_C_DECLS
#endif /* ORTE_SNAPC_H */
Server : Apache/2.4.41 (Ubuntu)