hci.c File Reference

#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include "utils.h"
#include "hci.h"

Include dependency graph for hci.c:

Go to the source code of this file.

Functions
static double	hci_now (HCIManager *manager)
void	hci_init (HCIManager *manager, char *prefix, double WallClockTimeLimit, double AutoCheckPointTime, int FOFEnabled)
void	hci_action_init (HCIAction *action)
void	hci_override_now (HCIManager *manager, double now)
static double	hci_get_elapsed_time (HCIManager *manager)
static void	hci_update_query_timer (HCIManager *manager)
int	hci_query_filesystem (HCIManager *manager, const char *filename, char **request)
static int	hci_query_timeout (HCIManager *manager, char **request)
static int	hci_query_auto_checkpoint (HCIManager *manager, char **request)
int	hci_query (HCIManager *manager, HCIAction *action)

Function Documentation

hci_action_init()

void hci_action_init

(

HCIAction *

action

)

Definition at line 38 of file hci.c.

{
    action->type = HCI_NO_ACTION;
    action->write_snapshot = 0;
    action->write_fof = 0;
}

References HCI_NO_ACTION, HCIAction::type, HCIAction::write_fof, and HCIAction::write_snapshot.

Referenced by hci_query(), and run().

Here is the caller graph for this function:

hci_get_elapsed_time()

static double hci_get_elapsed_time ( HCIManager *  manager )

static

Definition at line 55 of file hci.c.

{
    double e = hci_now(manager) - manager->timer_begin;
    return e;
}

References hci_now(), manager, and HCIManager::timer_begin.

Referenced by hci_query(), hci_query_auto_checkpoint(), and hci_query_timeout().

Here is the call graph for this function:

Here is the caller graph for this function:

hci_init()

void hci_init	(	HCIManager *	manager,
		char *	prefix,
		double	WallClockTimeLimit,
		double	AutoCheckPointTime,
		int	FOFEnabled
	)

Definition at line 25 of file hci.c.

{
    manager->prefix = prefix;
    manager->timer_begin = hci_now(manager);
    manager->timer_query_begin = manager->timer_begin;
 
    manager->WallClockTimeLimit = WallClockTimeLimit;
    manager->AutoCheckPointTime = AutoCheckPointTime;
    manager->FOFEnabled = FOFEnabled;
    manager->LongestTimeBetweenQueries = 0;
}

References HCIManager::AutoCheckPointTime, HCIManager::FOFEnabled, hci_now(), HCIManager::LongestTimeBetweenQueries, manager, HCIManager::prefix, prefix, HCIManager::timer_begin, HCIManager::timer_query_begin, and HCIManager::WallClockTimeLimit.

Referenced by run(), test_hci_auto_checkpoint(), test_hci_auto_checkpoint2(), test_hci_checkpoint(), test_hci_no_action(), test_hci_stop(), test_hci_terminate(), and test_hci_timeout().

Here is the call graph for this function:

Here is the caller graph for this function:

hci_now()

static double hci_now ( HCIManager *  manager )

static

Definition at line 11 of file hci.c.

{
    double e;
    if(manager->OVERRIDE_NOW) {
        e = manager->_now;
    } else {
        e = MPI_Wtime();
    }
    /* must be consistent between all ranks. */
    MPI_Bcast(&e, 1, MPI_DOUBLE, 0, MPI_COMM_WORLD);
    return e;
}

References HCIManager::_now, manager, and HCIManager::OVERRIDE_NOW.

Referenced by hci_get_elapsed_time(), hci_init(), and hci_update_query_timer().

Here is the caller graph for this function:

hci_override_now()

void hci_override_now	(	HCIManager *	manager,
		double	now
	)

Definition at line 48 of file hci.c.

{
    manager->_now = now;
    manager->OVERRIDE_NOW = 1;
}

References HCIManager::_now, manager, and HCIManager::OVERRIDE_NOW.

Referenced by test_hci_auto_checkpoint(), test_hci_auto_checkpoint2(), test_hci_checkpoint(), test_hci_no_action(), test_hci_stop(), test_hci_terminate(), and test_hci_timeout().

Here is the caller graph for this function:

hci_query()

int hci_query	(	HCIManager *	manager,
		HCIAction *	action
	)

Definition at line 152 of file hci.c.

{
    hci_action_init(action);
 
    /* measure time since last query */
    hci_update_query_timer(manager);
 
    /* Check whether we need to interrupt the run */
 
    char * request;
 
    /* Will we run out of time by the query ? highest priority.
     */
    if(hci_query_timeout(manager, &request)) {
        message(0, "HCI: Stopping due to TimeLimitCPU, dumping a CheckPoint.\n");
        action->type = HCI_TIMEOUT;
        action->write_snapshot = 1;
        if(manager->FOFEnabled)
            action->write_fof = 1;
        return 1;
    }
 
    if(hci_query_filesystem(manager, "reconfigure", &request))
    {
        /* FIXME: This is not implemented
         * it shall reread the configuration file and update the parameters of
         * the module listed in the request.
         * see the comment about update_IO_params
         * */
        message(0, "HCI: updating io parameters, this is not supported yet.\n");
        myfree(request);
        return 0;
    }
 
    if(hci_query_filesystem(manager, "checkpoint", &request))
    {
        message(0, "HCI: human controlled stop with checkpoint at next PM.\n");
        action->type = HCI_CHECKPOINT;
        /* will write checkpoint in this PM timestep */
        action->write_snapshot = 1;
        /* Write fof as well*/
        if(manager->FOFEnabled)
            action->write_fof = 1;
        myfree(request);
        manager->TimeLastCheckPoint = hci_get_elapsed_time(manager);
        return 0;
    }
 
    /* Is the stop-file present? If yes, interrupt the run with a snapshot. */
    if(hci_query_filesystem(manager, "stop", &request))
    {
        /* will write checkpoint in this PM timestep, then stop */
        action->type = HCI_STOP;
        action->write_snapshot = 1;
        myfree(request);
        return 1;
    }
 
    /* Is the terminate-file present? If yes, interrupt the run immediately. */
    if(hci_query_filesystem(manager, "terminate", &request))
    {
        message(0, "HCI: human triggered termination.\n");
        /* the caller shall take care of immediate termination.
         * This action is better than KILL as it avoids corrupt/incomplete snapshot files.*/
        action->type = HCI_TERMINATE;
        action->write_snapshot = 0;
        myfree(request);
        return 1;
    }
 
    /* lower priority */
    if(hci_query_auto_checkpoint(manager, &request))
    {
        message(0, "HCI: Auto checkpoint due to AutoCheckPointTime.\n");
        action->type = HCI_AUTO_CHECKPOINT;
        /* Write when the PM timestep completes*/
        action->write_snapshot = 1;
        if(manager->FOFEnabled)
            action->write_fof = 1;
        manager->TimeLastCheckPoint = hci_get_elapsed_time(manager);
        return 0;
    }
 
    message(0, "HCI: Nothing happened. \n");
    return 0;
}

References HCIManager::FOFEnabled, hci_action_init(), HCI_AUTO_CHECKPOINT, HCI_CHECKPOINT, hci_get_elapsed_time(), hci_query_auto_checkpoint(), hci_query_filesystem(), hci_query_timeout(), HCI_STOP, HCI_TERMINATE, HCI_TIMEOUT, hci_update_query_timer(), manager, message(), myfree, HCIManager::TimeLastCheckPoint, HCIAction::type, HCIAction::write_fof, and HCIAction::write_snapshot.

Referenced by run(), test_hci_auto_checkpoint(), test_hci_auto_checkpoint2(), test_hci_checkpoint(), test_hci_no_action(), test_hci_stop(), test_hci_terminate(), and test_hci_timeout().

Here is the call graph for this function:

Here is the caller graph for this function:

hci_query_auto_checkpoint()

static int hci_query_auto_checkpoint ( HCIManager *  manager, char **  request  )

static

Definition at line 134 of file hci.c.

{
    /* this function is collective because we take care to ensure manager is
     * collective */
    if(manager->AutoCheckPointTime <= 0) return 0;
 
    /* How long since the last checkpoint? */
    double now = hci_get_elapsed_time(manager);
    if(now - manager->TimeLastCheckPoint >= manager->AutoCheckPointTime) {
        return 1;
    }
    return 0;
}

References HCIManager::AutoCheckPointTime, hci_get_elapsed_time(), manager, and HCIManager::TimeLastCheckPoint.

Referenced by hci_query().

Here is the call graph for this function:

Here is the caller graph for this function:

hci_query_filesystem()

int hci_query_filesystem	(	HCIManager *	manager,
		const char *	filename,
		char **	request
	)

Definition at line 77 of file hci.c.

{
    int ThisTask;
    int NTask;
    MPI_Comm_size(MPI_COMM_WORLD, &NTask);
    MPI_Comm_rank(MPI_COMM_WORLD, &ThisTask);
    int size = 0;
    char * content = NULL;
    if(ThisTask == 0) {
        char * fullname = fastpm_strdup_printf("%s/%s", manager->prefix, filename);
        content = fastpm_file_get_content(fullname);
        if(content) {
            size = strlen(content);
            remove(fullname);
        } else {
            size = -1;
        }
        myfree(fullname);
    }
    MPI_Bcast(&size, 1, MPI_INT, 0, MPI_COMM_WORLD);
 
    if(size != -1) {
        if(ThisTask != 0) {
            content = ta_malloc("hcicontent", char, size + 1);
        }
        MPI_Bcast(content, size+1, MPI_BYTE, 0, MPI_COMM_WORLD);
    } else {
        content = NULL;
    }
 
    *request = content;
    return *request != NULL;
}

References fastpm_file_get_content(), fastpm_strdup_printf(), manager, myfree, NTask, HCIManager::prefix, ta_malloc, and ThisTask.

Referenced by hci_query().

Here is the call graph for this function:

Here is the caller graph for this function:

hci_query_timeout()

static int hci_query_timeout ( HCIManager *  manager, char **  request  )

static

Definition at line 112 of file hci.c.

{
    /* this function is collective because we take care to ensure manager is
     * collective */
    double now = hci_get_elapsed_time(manager);
    /*
     * factor 0.9 is a safety tolerance
     * for possible inconsistency between measured time and the true wallclock
     *
     * If there likely isn't time for a new query, then we shall timeout as well.
     * */
 
    *request = NULL;
    if (now + manager->LongestTimeBetweenQueries < manager->WallClockTimeLimit * 0.95) {
        return 0;
    }
 
    /* any freeable string would work. */
    return 1;
}

References hci_get_elapsed_time(), HCIManager::LongestTimeBetweenQueries, manager, and HCIManager::WallClockTimeLimit.

Referenced by hci_query().

Here is the call graph for this function:

Here is the caller graph for this function:

hci_update_query_timer()

static void hci_update_query_timer ( HCIManager *  manager )

static

Definition at line 62 of file hci.c.

{
    double e = hci_now(manager);
    double g = e - manager->timer_query_begin;
    if(g > manager->LongestTimeBetweenQueries)
        manager->LongestTimeBetweenQueries = g;
 
    manager->timer_query_begin = e;
}

References hci_now(), HCIManager::LongestTimeBetweenQueries, manager, and HCIManager::timer_query_begin.

Referenced by hci_query().

Here is the call graph for this function:

Here is the caller graph for this function: