Beispiel #1
0
int main(int argc, char **argv, const char **environ)
{

        pmix_status_t rc;

        pmix_info_t *info = NULL;

        bool flag;

        pmix_status_t retval;
        pmix_app_t *spawned_app = NULL;

        pmix_info_t *job_info = NULL;
        pmix_info_t *proc_info = NULL;
        int job_info_count = 0;
        int job_info_index = 0;
        int proc_info_count = 0;
        int proc_info_index = 0;

        char spawned_nsp[PMIX_MAX_NSLEN+1];
        char *path_to_app = NULL;
        char *host_to_use = NULL;
        int number_of_clients = 0;
        int temp_counter = 0;
        done_flag = false;
        gethostname(hostn, 500);
        int spawned_app_argc = 0;

        char **scr_environ = NULL;

        int proc_count = 1;
        int node_count = 0;
        bool blocking_mode = true;
        char *node_list = NULL;
        bool forward_all_scr_envs = false;
        bool pmix_mode = false;


        const char *optstring = "+n:N:L:x:bB:pPvhe";
        int temp_slen=0;
        /* todo: add arg parsing with ompi schizo */
        verbose_print = false;

        int sleep_max = 30;
        const int fixed_sleep = 5;
        int c;
        while((c = getopt(argc, argv, optstring)) != -1){
                switch(c){
                case 'h':
                        print_usage(argv[0]);
                        exit(0);
                        break;
                case 'n':
                        proc_count = atoi(optarg);
                        if(proc_count <= 0 || proc_count > 100){
                                printf("outside the range of allowable instances to spawn [1-100]\n");
                                exit(1);
                        }
                        if(verbose_print) {
                                printf("proc_count = %d\n", proc_count);
                        }
                        break;
                case 'N':
                        /* node_count = atoi(optarg); */
                        node_count = 1;
                        if(verbose_print) {
                                printf("node_count = %d\n", node_count);
                        }
                        break;
                case 'B':
                        blocking_mode = true;
                        sleep_max = atoi(optarg);
                        if(sleep_max < 0){
                                printf("can't sleep for less than 0 seconds\n");
                                exit(1);
                        }
                        if(verbose_print){
                                printf("blocking mode = %x\n", blocking_mode);
                        }
                        break;
                case 'b':
                        blocking_mode = false;
                        if(verbose_print){
                                printf("blocking mode = %x\n", blocking_mode);
                        }
                        break;
                case 'L':
                        node_list = optarg;
                        host_to_use = node_list;
                        if(verbose_print){
                                printf("node_list = '%s'\n", node_list);
                        }
                        break;
                case 'x':
                        temp_slen = strlen(optarg);
                        /*  check if the string is the same length as 'SCR', if so compare them */
                        if(temp_slen == strlen(SCR_STRING)){
                                if(strncmp(optarg, SCR_STRING, strlen(SCR_STRING)) == 0){
                                        /* if the string is SCR, then forward all SCR related env vars */
                                        if(verbose_print) printf("all scr envs will be forwarded\n");
                                        forward_all_scr_envs = true;
                                }
                                else{
                                        /* handled like a normal env var */
                                        handle_standard_env_var(optarg, &scr_environ);
                                }
                        }
                        else{
                                /*handled like a normal env var */
                                handle_standard_env_var(optarg, &scr_environ);
                        }
                        break;
                case 'v':
                        verbose_print = true;
                        break;
                case 'p':
                        pmix_mode = true;
                        if(verbose_print){
                                printf("pmix_mode = %x\n", pmix_mode);
                        }
                        break;
                case 'P':
                        pmix_mode = false;
                        if(verbose_print){
                                printf("pmix_mode = %x\n", pmix_mode);
                        }
                        break;
                case 'e':
                        experimental = true;
                        break;
                case '?':
                        printf("missing a required argument or invalid option: %x\n", optopt);
                        print_usage(argv[0]);
                        exit(1);
                        break;
                default:
                        printf("Unrecognized argument: %c\n", c);
                        print_usage(argv[0]);
                        exit(1);
                        break;
                }
        }

        /* number of instances to spawn */
        number_of_clients = proc_count;


        /* check to make sure an application was specified to launch */
        if( optind < argc ){
                /* if optind is < argc, it means there is at least one more arg
                 * beyond the args for this program */
                path_to_app = argv[optind];
                spawned_app_argc = argc - optind;
                if(verbose_print) {
                        printf("app to launch: %s @ %s:%d\n",
                               path_to_app, __FILE__, __LINE__);
                }
        }
        else{
                printf("program_to_spawn option was not provded\n");
                print_usage(argv[0]);
                exit(1);
        }

        if(verbose_print){
                printf("master process will spawn %d instances; app to run: %s\n\n",
                       number_of_clients, path_to_app);
                printf("pmix version: %s (host: %s)\n", PMIx_Get_version(), hostn);
        }
        /* init pmix */
        retval = PMIx_Init(&main_proc, NULL, 0);
        if(retval != PMIX_SUCCESS){
                error_helper(retval, hostn, "error initializing pmix");
                exit(0);
        }

        if(verbose_print){
                printf("rank %d, host '%s', nspace: '%s' init'd pmix succesfully\n\n",
                       main_proc.rank, hostn, main_proc.nspace);
        }



        /* we need to attach to a "system" PMIx server so we
         * can ask it to spawn applications for us. There can
         * only be one such connection on a node, so we will
         * instruct the tool library to only look for it */
        int ninfo = 1;
        PMIX_INFO_CREATE(info, ninfo);
        flag = true;
        PMIX_INFO_LOAD(&info[0], PMIX_CONNECT_TO_SYSTEM, &flag, PMIX_BOOL);

        /* initialize the library and make the connection */
        if (PMIX_SUCCESS != (rc = PMIx_tool_init(&tool_proc, NULL, 0 ))) {
                fprintf(stderr, "PMIx_tool_init failed: %d\n", rc );
                exit(rc);
        }
        if (0 < ninfo) {
                PMIX_INFO_FREE(info, ninfo);
        }




        /* first call fence to sync all processes */
        retval = fence_helper();
        if(retval != PMIX_SUCCESS)
        {
                error_helper(retval, hostn, "error fencing");
                exit(retval);
        }

        /* Process SCR env vars if needed */

        if(forward_all_scr_envs){
                parse_all_scr_envs(&scr_environ, environ);
        }

        /* finalize the env array so a NULL is in place */
        finalize_array(scr_environ);

        /* Setup info structs to pass to this: */
        /* pmix_info_t *error_info = NULL; */
        /*  PMIX_INFO_CREATE(error_info, 1); */
        /*
          strncpy(error_info[0].key, PMIX_ERROR_GROUP_ABORT, PMIX_MAX_KEYLEN);
          error_info[0].value.type = PMIX_BOOL;
          error_info[0].value.data.flag = true;
        */

        /*  strncpy(error_info[0].key, PMIX_ERROR_GROUP_SPAWN, PMIX_MAX_KEYLEN);
            int t_val = 1;
            pmix_value_load(&error_info[1].value, &t_val, PMIX_BOOL);
        */

        /*error_info[1].value.type = PMIX_BOOL;
        error_info[1].value.data.flag = true; */

        /*  strncpy(error_info[2].key, PMIX_ERROR_GROUP_GENERAL, PMIX_MAX_KEYLEN);
            error_info[2].value.type = PMIX_BOOL;
            error_info[2].value.data.flag = true;
        */


        /* TODO: setup error handling when implemented in pmix with the
         * following error codes: */
        /*
        pmix_status_t registered_codes[5];
        registered_codes[0] = PMIX_ERR_JOB_TERMINATED;
        registered_codes[1] = PMIX_ERR_PROC_ABORTED;
        registered_codes[2] = PMIX_ERR_PROC_ABORTING;
        */
        PMIx_Register_event_handler(NULL, 0,
                                    NULL, 0,
                                    errhandler_cb,
                                    errhandler_reg_callbk,
                                    (void *) NULL);

        /*  PMIX_INFO_DESTRUCT(error_info); */

        /* allocate memory to hold the spawend app struct */
        PMIX_APP_CREATE(spawned_app, 1);

        /* maxprocs isn't documented very well, but it appears to control
         * how many instances of the spanwed app are created */
        spawned_app->maxprocs = number_of_clients;

        /* set the app to run */
        (void)asprintf(&spawned_app->cmd, "%s", path_to_app);

        /* set argv for spawned app starting with remaining argv  */
        spawned_app->argv = &argv[optind];

        /* set the environment pointer */
        spawned_app->env = scr_environ;

        /*--START: add all proc level infos */

        /* add things to the proc level info */
        if(!pmix_mode){
                job_info_count++;
        }

        if(host_to_use != NULL){
                proc_info_count++;
        }

        if(verbose_print){
                printf("enabling debug feature for forwarding stdout/stderr\n");
                proc_info_count+=2;
                /* add PMIX_FWD_STDOUT and PMIX_FWD_STDERR later*/

        }

        if(experimental){
                job_info_count++;
        }
        if(node_count == 1){
                job_info_count++;
        }
        /*--END: add all proc level infos */


        /*--START: append actual proc level info */
        PMIX_INFO_CREATE(job_info, job_info_count);
        PMIX_INFO_CREATE(proc_info, proc_info_count);
        /* PMIX_VAL_set_assign(_v, _field, _val )  */
        /* PMIX_VAL_set_strdup(_v, _field, _val )  */

        if(host_to_use != NULL){
                /* add info struct to the spawned app itself for the host */

                /* old way */
                strncpy(proc_info[proc_info_index].key, PMIX_HOST, PMIX_MAX_KEYLEN);
                //proc_info[proc_info_index].value.type = PMIX_STRING;
                /* set the data for host list to use */
                //proc_info[proc_info_index].value.data.string = host_to_use;
                /* end old way */
                if(verbose_print) printf("about to set host val\n");
                PMIX_VAL_SET(&(proc_info[proc_info_index].value), string,
                                    host_to_use );
                proc_info_index++;
        }

        if(!pmix_mode){
                strncpy(job_info[job_info_index].key, PMIX_NON_PMI,
                        PMIX_MAX_KEYLEN);
                if(verbose_print) printf("about to set non pmix flag\n");
                PMIX_VAL_SET(&(job_info[job_info_index].value), flag, true);
                job_info_index++;
        }
        if(verbose_print){
                strncpy(proc_info[proc_info_index].key, PMIX_FWD_STDOUT,
                        PMIX_MAX_KEYLEN);
                if(verbose_print) printf("about to set stdout flag\n");
                PMIX_VAL_SET(&(proc_info[proc_info_index].value), flag, true );
                proc_info_index++;

                strncpy(proc_info[proc_info_index].key, PMIX_FWD_STDERR,
                        PMIX_MAX_KEYLEN);
                if(verbose_print) printf("about to set stderr flag\n");
                PMIX_VAL_SET(&(proc_info[proc_info_index].value), flag, true );
                proc_info_index++;
        }
        if(experimental){
                printf("attempting to perform experiment\n");
                bool local_flag = true;
                PMIX_INFO_LOAD(&job_info[job_info_index], PMIX_NOTIFY_COMPLETION, &local_flag, PMIX_BOOL);
                job_info_index++;
        }
        if(node_count == 1){
                strncpy(job_info[job_info_index].key, PMIX_PPR,
                        PMIX_MAX_KEYLEN);
                PMIX_VAL_SET(&(job_info[job_info_index].value), string,
                             "1:n");
                job_info_index++;
        }
        /*--END: append actual proc level info */

        /* sanity check to make sure we covered all the info structs */
        if(proc_info_index != proc_info_count ){
                printf("bug: mismatch with appending proc info\n");
                exit(1);
        }
        if(job_info_index != job_info_count){
                printf("bug: mismatch with appending job info\n");
                exit(1);
        }

        /* TODO: TEST PMIX_NOTIFY_COMPLETION WHEN IT'S IMPLEMENTED IN PMIX */

        /* fill in job_info */
        /*
        strncpy(job_info[0].key, PMIX_TIMEOUT, PMIX_MAX_KEYLEN);
        job_info[0].value.type = PMIX_INT;
        job_info[0].value.data.integer = 10; */

        /* strncpy(job_info[0].key, PMIX_NOTIFY_COMPLETION, PMIX_MAX_KEYLEN);
           job_info[0].value.type = PMIX_BOOL;
           job_info[0].value.data.flag = true; */

        /*strncpy(spawned_app->info[0].key, PMIX_DISPLAY_MAP, PMIX_MAX_KEYLEN);
          job_info[0].value.type = PMIX_BOOL;
          job_info[0].value.data.flag = true;*/


        /* TODO: TEST PMIX_NOTIFY_COMPLETION WHEN IT'S IMPLEMENTED IN PMIX */
        spawned_app->info = proc_info;
        spawned_app->ninfo = proc_info_count;

        if(verbose_print){
                printf("proc level info count: %d\n", proc_info_count);
        }
        /* call spawn */
        retval = PMIx_Spawn(job_info, job_info_count,
                            spawned_app, 1, spawned_nsp);

        if(verbose_print) {
                printf("rank %d (host %s) just called spawn; spawned nspace: %s, retval:%d\n",
                       main_proc.rank,
                       hostn,
                       spawned_nsp,
                       retval);
        }
        if(retval != PMIX_SUCCESS){
                error_helper(retval,  hostn, "error with spawn");
                goto done;
        }

        /* TODO: TEMPORARY WORKAROUND TO WAIT FOR A SPAWNED PROCESS */
        if(blocking_mode){

                sleep(fixed_sleep);

                /* wait until app completes: */
                while(!done_flag){
                        sleep(fixed_sleep);
                        temp_counter++;
                        if(temp_counter*fixed_sleep >= sleep_max) {
                                if(verbose_print) printf("broke out early\n");
                                break;
                        }
                }
                if(verbose_print){
                        if(done_flag == true) {
                                printf("done_flag was set to true!\n");
                        }
                }

        }

done:
        /* fence first */
        retval = fence_helper();
        if(retval != PMIX_SUCCESS){
                if(verbose_print) printf("error fencing, finalize may fail ! \n");
        }
        /* finalize */

        PMIx_Deregister_event_handler(_g_errhandler_ref, NULL, NULL);

        if(verbose_print){
                fprintf(stdout,
                        "spawn master process (rank %d) (host %s) finalizing\n",
                        main_proc.rank,
                        hostn);
        }

        /* clean up pmix */

        retval = PMIx_tool_finalize();

        if(retval == PMIX_SUCCESS)
        {
                if(verbose_print){
                        printf("spawn master process %d finalize success\n\n",
                               main_proc.rank);
                }
        }
        else
        {
                printf("spawn master process %d pmix_finalize FAILURE: %d\n\n",
                       main_proc.rank,
                       retval);
        }

        retval = PMIx_Finalize(NULL, 0);
        fflush(stdout);

        /*  cleanup before returning */
        PMIX_INFO_FREE(job_info, job_info_count);
        spawned_app->argv = NULL;
        PMIX_APP_FREE(spawned_app, 1);
        if(verbose_print) printf("%s exiting cleanly :)\n", argv[0]);
        return 0;

}
Beispiel #2
0
int main(int argc, char **argv)
{
    pmix_status_t rc;
    pmix_proc_t myproc;
    pmix_info_t *info;
    pmix_app_t *app;
    size_t ninfo, napps;

    /* check for user directives - this would include:
     * - a flag indicating we want to attach to a specified application
     * - application info if we are launching a new app
     */

    /* init us - if a PMIx server pid was provided, then pass it along */
    if (0 < server_pid) {
        ninfo = 1;
        PMIX_INFO_CREATE(info, ninfo);
        PMIX_INFO_LOAD(&info[0], PMIX_SERVER_PIDINFO, server_pid, PMIX_UINT32);
    } else {
        info = NULL;
        ninfo = 0;
    }
    if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, info, ninfo))) {
        fprintf(stderr, "PMIx_tool_init failed: %d\n", rc);
        exit(rc);
    }
    if (0 < ninfo) {
        PMIX_INFO_FREE(info, ninfo);
    }
    fprintf(stderr, "Tool ns %s rank %d: Running\n", myproc.nspace, myproc.rank);

    /* if we are attaching to a running job, then attach to it */
    if (attach) {
        ret = attach_to_running_job(argv[1]);
    } else {
        /* this is an initial launch - we need to launch the application
         * plus the debugger daemons, letting the RM know we are debugging
         * so that it will "pause" the app procs until we are ready */
        napps = 2;
        PMIX_APP_CREATE(app, napps);
        /* setup the executable */
        app[0].cmd = strdup("app");
        app[0].argc = 1;
        app[0].argv = (char**)malloc(2*sizeof(char*));
        app[0].argv[0] = strdup("app");
        app[0].argv[1] = NULL;
        /* provide directives so the apps do what the user requested */
        ninfo = 2;
        PMIX_INFO_CREATE(app[0].info, ninfo);
        PMIX_INFO_LOAD(&app[0].info[0], PMIX_NP, 128, PMIX_UINT64);
        PMIX_INFO_LOAD(&app[0].info[0], PMIX_MAPBY, "slot", PMIX_STRING);

        /* setup the name of the daemon executable to launch */
        app[1].cmd = strdup("debuggerdaemon");
        app[1].argc = 1;
        app[1].argv = (char**)malloc(2*sizeof(char*));
        app[1].argv[0] = strdup("debuggerdaemon");
        app[1].argv[1] = NULL;
        /* provide directives so the daemons go where we want, and
         * let the RM know these are debugger daemons */
        ninfo = 2;
        PMIX_INFO_CREATE(app[1].info, ninfo);
        PMIX_INFO_LOAD(&app[1].info[0], PMIX_MAPBY, "ppr:1:node", PMIX_STRING);  // instruct the RM to launch one copy of the executable on each node
        PMIX_INFO_LOAD(&app[1].info[1], PMIX_DEBUGGER_DAEMONS, true, PMIX_BOOL); // these are debugger daemons
        /* spawn the daemons */
        PMIx_Spawn(NULL, 0, app, napps, dspace);
        /* cleanup */
        PMIX_APP_FREE(app, napps);

        /* this is where a debugger tool would wait until the debug operation is complete */
    }


 done:
    PMIx_tool_finalize(NULL, 0);

    return(ret);
}