C++ ORTE_ACTIVATE_JOB_STATE函数代码示例

/* When working in this function, ALWAYS jump to "cleanup" if
 * you encounter an error so that orterun will be woken up and
 * the job can cleanly terminate
 */
static int plm_alps_launch_job(orte_job_t *jdata)
{
    orte_app_context_t *app;

    for (int i = 0 ; i < jdata->apps->size ; ++i) {
        int env_count;

        if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, i))) {
            continue;
        }

        for (env_count = 0 ; app->env && app->env[env_count] ; ++env_count);
        /* disable PMI for the application. this will prevent the pmi library from printing useless warnings */
        opal_argv_append (&env_count, &app->env, "PMI_NO_FORK=1");
        opal_argv_append (&env_count, &app->env, "PMI_NO_PREINITIALIZE=1");
    }

    if (ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_RESTART)) {
        /* this is a restart situation - skip to the mapping stage */
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_MAP);
    } else {
        /* new job - set it up */
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_INIT);
    }
    return ORTE_SUCCESS;
}

static void alps_wait_cb(orte_proc_t *proc, void* cbdata){
    orte_job_t *jdata;

    /* According to the ALPS folks, alps always returns the highest exit
       code of our remote processes. Thus, a non-zero exit status doesn't
       necessarily mean that alps failed - it could be that an orted returned
       a non-zero exit status. Of course, that means the orted failed(!), so
       the end result is the same - the job didn't start.
    
       As a result, we really can't do much with the exit status itself - it
       could be something in errno (if alps itself failed), or it could be
       something returned by an orted, or it could be something returned by
       the OS (e.g., couldn't find the orted binary). Somebody is welcome
       to sort out all the options and pretty-print a better error message. For
       now, though, the only thing that really matters is that
       alps failed. Report the error and make sure that orterun
       wakes up - otherwise, do nothing!
    */
    jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
    
    if (0 != proc->exit_code) {
        if (failed_launch) {
            /* report that the daemon has failed so we break out of the daemon
             * callback receive and exit
             */
            ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_FAILED_TO_START);            
        } else {
            /* an orted must have died unexpectedly after launch - report
             * that the daemon has failed so we exit
             */
            ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_ABORTED);
        }
    }
}

/* When working in this function, ALWAYS jump to "cleanup" if
 * you encounter an error so that orterun will be woken up and
 * the job can cleanly terminate
 */
static int plm_alps_launch_job(orte_job_t *jdata)
{
    if (ORTE_JOB_CONTROL_RESTART & jdata->controls) {
        /* this is a restart situation - skip to the mapping stage */
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_MAP);
    } else {
        /* new job - set it up */
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_INIT);
    }
    return ORTE_SUCCESS;
}

/* When working in this function, ALWAYS jump to "cleanup" if
 * you encounter an error so that orterun will be woken up and
 * the job can cleanly terminate
 */
static int plm_slurm_launch_job(orte_job_t *jdata)
{
    if (ORTE_FLAG_TEST(jdata, ORTE_JOB_FLAG_RESTART)) {
        /* this is a restart situation - skip to the mapping stage */
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_MAP);
    } else {
        /* new job - set it up */
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_INIT);
    }
    return ORTE_SUCCESS;
}

int
orte_rml_oob_ft_event(int state) {
    int exit_status = ORTE_SUCCESS;
    int ret;

    if(OPAL_CRS_CHECKPOINT == state) {
        ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_FT_CHECKPOINT);
    }
    else if(OPAL_CRS_CONTINUE == state) {
        ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_FT_CONTINUE);
    }
    else if(OPAL_CRS_RESTART == state) {
        ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_FT_RESTART);
    }
    else if(OPAL_CRS_TERM == state ) {
        ;
    }
    else {
        ;
    }


    if(OPAL_CRS_CHECKPOINT == state) {
        ;
    }
    else if(OPAL_CRS_CONTINUE == state) {
        ;
    }
    else if(OPAL_CRS_RESTART == state) {
        (void) mca_base_framework_close(&orte_oob_base_framework);

        if (ORTE_SUCCESS != (ret = mca_base_framework_open(&orte_oob_base_framework, 0))) {
            ORTE_ERROR_LOG(ret);
            exit_status = ret;
            goto cleanup;
        }

        if( ORTE_SUCCESS != (ret = orte_oob_base_select())) {
            ORTE_ERROR_LOG(ret);
            exit_status = ret;
            goto cleanup;
        }
    }
    else if(OPAL_CRS_TERM == state ) {
        ;
    }
    else {
        ;
    }

 cleanup:
    return exit_status;
}

/**
* Terminate the orteds for a given job
 */
static int plm_alps_terminate_orteds(void)
{
    int rc;
    orte_job_t *jdata;

    OPAL_OUTPUT_VERBOSE((10, orte_plm_base_framework.framework_output,
                            "%s plm:alps: terminating orteds",
                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));

    /* deregister the waitpid callback to ensure we don't make it look like
     * alps failed when it didn't. Since the alps may have already completed,
     * do NOT ERROR_LOG any return code to avoid confusing, duplicate error
     * messages
     */
    if (NULL != alpsrun) {
        orte_wait_cb_cancel(alpsrun);
    }

    /* now tell them to die */
    if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
        ORTE_ERROR_LOG(rc);
    }

    jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
    ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_TERMINATED);

    OPAL_OUTPUT_VERBOSE((10, orte_plm_base_framework.framework_output,
                            "%s plm:alps: terminated orteds",
                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
    return rc;
}

static void poll_spawns(int fd, short args, void *cbdata)
{
    orte_state_caddy_t *state = (orte_state_caddy_t*)cbdata;
    int i, rc;
    bool failed_launch = true;
    int local_err;
    tm_event_t event;

    /* TM poll for all the spawns */
    for (i = 0; i < launched; ++i) {
        rc = tm_poll(TM_NULL_EVENT, &event, 1, &local_err);
        if (TM_SUCCESS != rc) {
            opal_output(0, "plm:tm: failed to poll for a spawned daemon, return status = %d", rc);
            goto cleanup;
        }
        if (TM_SUCCESS != local_err) {
            opal_output(0, "plm:tm: failed to spawn daemon, error code = %d", local_err );
            goto cleanup;
        }
    }
    failed_launch = false;

 cleanup:
    /* cleanup */
    OBJ_RELEASE(state);

    /* check for failed launch - if so, force terminate */
    if (failed_launch) {
        ORTE_ACTIVATE_JOB_STATE(state->jdata, ORTE_JOB_STATE_FAILED_TO_START);
    }
}

/**
* Terminate the orteds for a given job
 */
static int plm_slurm_terminate_orteds(void)
{
    int rc=ORTE_SUCCESS;
    orte_job_t *jdata;
        
    /* check to see if the primary pid is set. If not, this indicates
     * that we never launched any additional daemons, so we cannot
     * not wait for a waitpid to fire and tell us it's okay to
     * exit. Instead, we simply trigger an exit for ourselves
     */
    if (primary_pid_set) {
        if (ORTE_SUCCESS != (rc = orte_plm_base_orted_exit(ORTE_DAEMON_EXIT_CMD))) {
            ORTE_ERROR_LOG(rc);
        }
    } else {
        OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                             "%s plm:slurm: primary daemons complete!",
                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
        jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
        /* need to set the #terminated value to avoid an incorrect error msg */
        jdata->num_terminated = jdata->num_procs;
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_TERMINATED);
    }
    
    return rc;
}

static void hnp_notify(int sd, short args, void *cbdata)
{
    orte_state_caddy_t *caddy = (orte_state_caddy_t*)cbdata;
    orte_job_t *jdata = caddy->jdata;
    orte_process_name_t parent, target, *npptr;

    /* if they requested notification upon completion, provide it */
    if (orte_get_attribute(&jdata->attributes, ORTE_JOB_NOTIFY_COMPLETION, NULL, OPAL_BOOL)) {
        /* notify_completion => notify the parent of the termination
         * of this child job. So get the parent jobid info */
        npptr = &parent;
        if (!orte_get_attribute(&jdata->attributes, ORTE_JOB_LAUNCH_PROXY, (void**)&npptr, OPAL_NAME)) {
            /* notify everyone who asked for it */
            target.jobid = jdata->jobid;
            target.vpid = ORTE_VPID_WILDCARD;
            _send_notification(OPAL_ERR_JOB_TERMINATED, caddy->proc_state, &target, ORTE_NAME_WILDCARD);
        } else {
            target.jobid = jdata->jobid;
            target.vpid = ORTE_VPID_WILDCARD;
            _send_notification(OPAL_ERR_JOB_TERMINATED, caddy->proc_state, &target, &parent);
        }
    }
    ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_NOTIFIED);

    OBJ_RELEASE(caddy);
}

void orte_routed_base_xcast_routing(orte_grpcomm_collective_t *coll,
                                    opal_list_t *my_children)
{
    opal_list_item_t *item;
    orte_routed_tree_t *child;
    orte_namelist_t *nm;
    int i;
    orte_proc_t *proc;
    orte_job_t *daemons;

    /* if we are the HNP and an abnormal termination is underway,
     * then send it directly to everyone
     */
    if (ORTE_PROC_IS_HNP) {
        if (orte_abnormal_term_ordered) {
            daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
            for (i=1; i < daemons->procs->size; i++) {
                if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(daemons->procs, i))) {
                    continue;
                }
                /* exclude anyone known not alive */
                if (proc->alive) {
                    nm = OBJ_NEW(orte_namelist_t);
                    nm->name.jobid = ORTE_PROC_MY_NAME->jobid;
                    nm->name.vpid = proc->name.vpid;
                    opal_list_append(&coll->targets, &nm->super);
                }
            }
            /* if nobody is known alive, then we need to die */
            if (0 == opal_list_get_size(&coll->targets)) {
                ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_DAEMONS_TERMINATED);
            }
        } else {
            /* the xcast always goes to our children */
            for (item = opal_list_get_first(my_children);
                 item != opal_list_get_end(my_children);
                 item = opal_list_get_next(item)) {
                child = (orte_routed_tree_t*)item;
                nm = OBJ_NEW(orte_namelist_t);
                nm->name.jobid = ORTE_PROC_MY_NAME->jobid;
                nm->name.vpid = child->vpid;
                opal_list_append(&coll->targets, &nm->super);
            }
        }
    } else {
        /* I am a daemon - route to my children */
        for (item = opal_list_get_first(my_children);
             item != opal_list_get_end(my_children);
             item = opal_list_get_next(item)) {
            child = (orte_routed_tree_t*)item;
            nm = OBJ_NEW(orte_namelist_t);
            nm->name.jobid = ORTE_PROC_MY_NAME->jobid;
            nm->name.vpid = child->vpid;
            opal_list_append(&coll->targets, &nm->super);
        }
    }
}

static void shutdown_signal(int fd, short flags, void *arg)
{
    /* trigger the call to shutdown callback to protect
     * against race conditions - the trigger event will
     * check the one-time lock
     */
    ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
    ORTE_ACTIVATE_JOB_STATE(NULL, ORTE_JOB_STATE_FORCED_EXIT);
}

/* after we allocate, we need to map the processes
 * so we know what nodes will be used
 */
static void allocation_complete(int fd, short args, void *cbdata)
{
    orte_state_caddy_t *state = (orte_state_caddy_t*)cbdata;
    orte_job_t *jdata;
    orte_job_t *daemons;
    orte_topology_t *t;
    orte_node_t *node;
    int i;

    ORTE_ACQUIRE_OBJECT(caddy);
    jdata = state->jdata;

    jdata->state = ORTE_JOB_STATE_ALLOCATION_COMPLETE;

    /* get the daemon job object */
    if (NULL == (daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid))) {
        ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
        ORTE_FORCED_TERMINATE(ORTE_ERROR_DEFAULT_EXIT_CODE);
        goto done;
    }
    /* mark that we are not using a VM */
    orte_set_attribute(&daemons->attributes, ORTE_JOB_NO_VM, ORTE_ATTR_GLOBAL, NULL, OPAL_BOOL);

    /* ensure that all nodes point to our topology - we
     * cannot support hetero nodes with this state machine
     */
    t = (orte_topology_t*)opal_pointer_array_get_item(orte_node_topologies, 0);
    for (i=1; i < orte_node_pool->size; i++) {
        if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, i))) {
            continue;
        }
        node->topology = t;
    }
    if (!orte_managed_allocation) {
        if (NULL != orte_set_slots &&
            0 != strncmp(orte_set_slots, "none", strlen(orte_set_slots))) {
            for (i=0; i < orte_node_pool->size; i++) {
                if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(orte_node_pool, i))) {
                    continue;
                }
                if (!ORTE_FLAG_TEST(node, ORTE_NODE_FLAG_SLOTS_GIVEN)) {
                    OPAL_OUTPUT_VERBOSE((5, orte_plm_base_framework.framework_output,
                                         "%s plm:base:setting slots for node %s by %s",
                                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), node->name, orte_set_slots));
                    orte_plm_base_set_slots(node);
                }
            }
        }
    }

    /* move to the map stage */
    ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_MAP);

 done:
    /* cleanup */
    OBJ_RELEASE(state);
}

/**
* Terminate the orteds for a given job
 */
static int plm_yarn_terminate_orteds(void) {
	finish_app_master(0 == orte_exit_status);
    orte_job_t* jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
    if (ORTE_JOB_STATE_DAEMONS_TERMINATED != jdata->state) {
		/* need to set the #terminated value to avoid an incorrect error msg */
		jdata->num_terminated = jdata->num_procs;
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_TERMINATED);
    }
    return ORTE_SUCCESS;
}

static void launch_daemons(int fd, short args, void *cbdata)
{
    orte_state_caddy_t *state = (orte_state_caddy_t*)cbdata;

    /* there are no daemons to launch, so just trigger the
     * daemon-launch-complete state
     */
    ORTE_ACTIVATE_JOB_STATE(state->jdata, ORTE_JOB_STATE_DAEMONS_REPORTED);
    OBJ_RELEASE(state);
}

static void alps_wait_cb(pid_t pid, int status, void* cbdata){
    orte_job_t *jdata;

    /* According to the ALPS folks, alps always returns the highest exit
       code of our remote processes. Thus, a non-zero exit status doesn't
       necessarily mean that alps failed - it could be that an orted returned
       a non-zero exit status. Of course, that means the orted failed(!), so
       the end result is the same - the job didn't start.
    
       As a result, we really can't do much with the exit status itself - it
       could be something in errno (if alps itself failed), or it could be
       something returned by an orted, or it could be something returned by
       the OS (e.g., couldn't find the orted binary). Somebody is welcome
       to sort out all the options and pretty-print a better error message. For
       now, though, the only thing that really matters is that
       alps failed. Report the error and make sure that orterun
       wakes up - otherwise, do nothing!
    */
    jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
    
    if (0 != status) {
        if (failed_launch) {
            /* we have a problem during launch */
            opal_output(0, "ERROR: alps failed to start the required daemons.");
            opal_output(0, "ERROR: This could be due to an inability to find the orted binary (--prefix)");
            opal_output(0, "ERROR: on one or more remote nodes, compilation of the orted with dynamic libraries,");
            opal_output(0, "ERROR: lack of authority to execute on one or more specified nodes,");
            opal_output(0, "ERROR: or the inability to write startup files into /tmp (--tmpdir/orte_tmpdir_base).");
            
            /* report that the daemon has failed so we break out of the daemon
             * callback receive and exit
             */
            ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_FAILED_TO_START);            
        } else {
            /* an orted must have died unexpectedly after launch - report
             * that the daemon has failed so we exit
             */
            ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_ABORTED);
        }
    }
    
}