int main(int argc, char **argv) { pmix_status_t rc; pmix_proc_t myproc; pmix_info_t *info; size_t ninfo; /* init us */ if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, NULL, 0))) { fprintf(stderr, "PMIx_tool_init failed: %d\n", rc); exit(rc); } pmix_output(0, "Tool ns %s rank %d: Running", myproc.nspace, myproc.rank); /* query something */ ninfo = 2; PMIX_INFO_CREATE(info, ninfo); (void)strncpy(info[0].key, "foobar", PMIX_MAX_KEYLEN); (void)strncpy(info[1].key, "spastic", PMIX_MAX_KEYLEN); if (PMIX_SUCCESS != (rc = PMIx_Query_info(info, ninfo))) { pmix_output(0, "Client ns %s rank %d: PMIx_Query_info failed: %d", myproc.nspace, myproc.rank, rc); goto done; } if (0 != strncmp(info[0].key, "foobar", PMIX_MAX_KEYLEN)) { pmix_output(0, "Client ns %s rank %d: PMIx_Query_info key[0] wrong: %s vs foobar", myproc.nspace, myproc.rank, info[0].key); } if (0 != strncmp(info[1].key, "spastic", PMIX_MAX_KEYLEN)) { pmix_output(0, "Client ns %s rank %d: PMIx_Query_info key[0] wrong: %s vs spastic", myproc.nspace, myproc.rank, info[1].key); } if (PMIX_STRING != info[0].value.type) { pmix_output(0, "Client ns %s rank %d: PMIx_Query_info key[0] wrong type: %d vs %d", myproc.nspace, myproc.rank, info[0].value.type, PMIX_STRING); } if (PMIX_STRING != info[1].value.type) { pmix_output(0, "Client ns %s rank %d: PMIx_Query_info key[1] wrong type: %d vs %d", myproc.nspace, myproc.rank, info[1].value.type, PMIX_STRING); } if (0 != strcmp(info[0].value.data.string, "0")) { pmix_output(0, "Client ns %s rank %d: PMIx_Query_info key[0] wrong value: %s vs 0", myproc.nspace, myproc.rank, info[1].value.data.string); } if (0 != strcmp(info[1].value.data.string, "1")) { pmix_output(0, "Client ns %s rank %d: PMIx_Query_info key[1] wrong value: %s vs 1", myproc.nspace, myproc.rank, info[1].value.data.string); } PMIX_INFO_FREE(info, ninfo); done: /* finalize us */ pmix_output(0, "Client ns %s rank %d: Finalizing", myproc.nspace, myproc.rank); if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) { fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc); } else { fprintf(stderr, "Client ns %s rank %d:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank); } fflush(stderr); return(rc); }
int main(int argc, char **argv) { pmix_status_t rc; pmix_proc_t myproc; pmix_info_t *info; size_t ninfo; /* init us */ if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, NULL, 0))) { fprintf(stderr, "PMIx_tool_init failed: %s\n", PMIx_Error_string(rc)); exit(rc); } fprintf(stderr, "Tool ns %s rank %d: Running\n", myproc.nspace, myproc.rank); /* query something */ ninfo = 1; PMIX_INFO_CREATE(info, ninfo); (void)strncpy(info[0].key, PMIX_QUERY_NAMESPACES, PMIX_MAX_KEYLEN); if (PMIX_SUCCESS != (rc = PMIx_Query_info(info, ninfo))) { fprintf(stderr, "Tool ns %s rank %d: PMIx_Query_info failed: %d\n", myproc.nspace, myproc.rank, rc); goto done; } if (0 != strncmp(info[0].key, PMIX_QUERY_NAMESPACES, PMIX_MAX_KEYLEN)) { fprintf(stderr, "tool ns %s rank %d: PMIx_Query_info key[0] wrong: %s vs %s\n", myproc.nspace, myproc.rank, info[0].key, PMIX_QUERY_NAMESPACES); } if (PMIX_STRING != info[0].value.type) { fprintf(stderr, "Tool ns %s rank %d: PMIx_Query_info key[0] wrong type: %d vs %d\n", myproc.nspace, myproc.rank, info[0].value.type, PMIX_STRING); } fprintf(stderr, "Tool ns %s rank %d: PMIx_Query_info key[0] returned %s\n", myproc.nspace, myproc.rank, (NULL == info[0].value.data.string) ? "NULL" : info[0].value.data.string); PMIX_INFO_FREE(info, ninfo); done: /* finalize us */ fprintf(stderr, "Tool ns %s rank %d: Finalizing\n", myproc.nspace, myproc.rank); if (PMIX_SUCCESS != (rc = PMIx_tool_finalize())) { fprintf(stderr, "Tool ns %s rank %d:PMIx_tool_finalize failed: %d\n", myproc.nspace, myproc.rank, rc); } else { fprintf(stderr, "Tool ns %s rank %d:PMIx_tool_finalize successfully completed\n", myproc.nspace, myproc.rank); } fflush(stderr); return(rc); }
int main(int argc, char **argv) { pmix_status_t rc; pmix_value_t *val; pmix_proc_t proc; pmix_info_t *info; size_t ninfo; volatile int active; pmix_query_t *query; size_t nq, n; myquery_data_t myquery_data; fprintf(stderr, "I AM HERE\n"); fflush(stderr); sleep(10); exit(0); /* init us - since we were launched by the RM, our connection info * will have been provided at startup. */ if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, NULL, 0))) { fprintf(stderr, "Debugger daemon ns %s rank %d: PMIx_tool_init failed: %d\n", myproc.nspace, myproc.rank, rc); exit(0); } fprintf(stderr, "Debugger daemon ns %s rank %d: Running\n", myproc.nspace, myproc.rank); /* register our default event handler */ active = -1; PMIx_Register_event_handler(NULL, 0, NULL, 0, notification_fn, evhandler_reg_callbk, (void*)&active); while (-1 == active) { usleep(10); } if (0 != active) { exit(active); } /* get the nspace of the job we are to debug */ (void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_NSLEN); proc.rank = PMIX_RANK_WILDCARD; if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_DEBUG_JOB, NULL, 0, &val))) { fprintf(stderr, "[%s:%d] Failed to get job being debugged - error %d\n", myproc.nspace, myproc.rank, rc); goto done; } if (NULL == val) { fprintf(stderr, "Got NULL return\n"); goto done; } fprintf(stderr, "[%s:%d] Debugging %s\n", myproc.nspace, myproc.rank, val->data.string); /* get our local proctable - for scalability reasons, we don't want to * have our "root" debugger process get the proctable for everybody and * send it out to us. So ask the local PMIx server for the pid's of * our local target processes */ nq = 1; PMIX_QUERY_CREATE(query, nq); PMIX_ARGV_APPEND(rc, query[0].keys, PMIX_QUERY_LOCAL_PROC_TABLE); query[0].nqual = 1; PMIX_INFO_CREATE(query[0].qualifiers, 1); PMIX_INFO_LOAD(&query[0].qualifiers[0], PMIX_NSPACE, val->data.string, PMIX_STRING); // the nspace we are enquiring about /* setup the caddy to retrieve the data */ myquery_data.info = NULL; myquery_data.ninfo = 0; myquery_data.active = true; /* execute the query */ if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(query, nq, cbfunc, (void*)&myquery_data))) { fprintf(stderr, "PMIx_Query_info failed: %d\n", rc); goto done; } while (myquery_data.active) { usleep(10); } fprintf(stderr, "[%s:%d] Local proctable received\n", myproc.nspace, myproc.rank); /* now that we have the proctable for our local processes, we can do our * magic debugger stuff and attach to them. We then send a "release" event * to them - i.e., it's the equivalent to setting the MPIR breakpoint. We * do this with the event notification system */ (void)strncpy(proc.nspace, val->data.string, PMIX_MAX_NSLEN); proc.rank = PMIX_RANK_WILDCARD; /* we send the notification to just the local procs of the job being debugged */ ninfo = 1; PMIX_INFO_CREATE(info, ninfo); PMIX_INFO_LOAD(&info[0], PMIX_EVENT_CUSTOM_RANGE, &proc, PMIX_PROC); // deliver to the target nspace fprintf(stderr, "[%s:%u] Sending release\n", myproc.nspace, myproc.rank); PMIx_Notify_event(PMIX_ERR_DEBUGGER_RELEASE, NULL, PMIX_RANGE_LOCAL, info, ninfo, NULL, NULL); /* do some debugger magic */ n = 0; fprintf(stderr, "[%s:%u] Hanging around awhile, doing debugger magic\n", myproc.nspace, myproc.rank); while (n < 5) { usleep(1000); ++n; } done: /* finalize us */ fprintf(stderr, "Debugger daemon ns %s rank %d: Finalizing\n", myproc.nspace, myproc.rank); if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) { fprintf(stderr, "Debugger daemon ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc); } else { fprintf(stderr, "Debugger daemon ns %s rank %d:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank); } fflush(stderr); return(0); }
int main(int argc, char **argv) { pmix_status_t rc; pmix_info_t *info, *iptr; pmix_app_t *app; size_t ninfo, napps; char *nspace = NULL; int i; pmix_query_t *query; size_t nq, n; myquery_data_t myquery_data; bool cospawn = false, stop_on_exec = false, cospawn_reqd = false; char cwd[1024]; pmix_status_t code = PMIX_ERR_JOB_TERMINATED; mylock_t mylock; myrel_t myrel, launcher_ready, dbrel; pid_t pid; pmix_envar_t envar; char *launchers[] = { "prun", "mpirun", "mpiexec", "orterun", NULL }; pmix_proc_t proc; bool found; pmix_data_array_t darray; char *tmp; char clientspace[PMIX_MAX_NSLEN+1]; pid = getpid(); /* Process any arguments we were given */ for (i=1; i < argc; i++) { if (0 == strcmp(argv[i], "-h") || 0 == strcmp(argv[i], "--help")) { /* print the usage message and exit */ } if (0 == strcmp(argv[i], "-a") || 0 == strcmp(argv[i], "--attach")) { if (NULL != nspace) { /* can only support one */ fprintf(stderr, "Cannot attach to more than one nspace\n"); exit(1); } /* the next argument must be the nspace */ ++i; if (argc == i) { /* they goofed */ fprintf(stderr, "The %s option requires an <nspace> argument\n", argv[i]); exit(1); } nspace = strdup(argv[i]); } else if (0 == strcmp(argv[i], "-c") || 0 == strcmp(argv[i], "--cospawn")){ cospawn_reqd = true; break; } } info = NULL; ninfo = 0; /* use the system connection first, if available */ PMIX_INFO_CREATE(info, 1); PMIX_INFO_LOAD(&info[0], PMIX_CONNECT_SYSTEM_FIRST, NULL, PMIX_BOOL); /* init as a tool */ if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, info, ninfo))) { fprintf(stderr, "PMIx_tool_init failed: %s(%d)\n", PMIx_Error_string(rc), rc); exit(rc); } PMIX_INFO_FREE(info, ninfo); fprintf(stderr, "Debugger ns %s rank %d pid %lu: Running\n", myproc.nspace, myproc.rank, (unsigned long)pid); /* construct the debugger termination release */ DEBUG_CONSTRUCT_LOCK(&dbrel.lock); /* register a default event handler */ DEBUG_CONSTRUCT_LOCK(&mylock); PMIx_Register_event_handler(NULL, 0, NULL, 0, notification_fn, evhandler_reg_callbk, (void*)&mylock); DEBUG_WAIT_THREAD(&mylock); DEBUG_DESTRUCT_LOCK(&mylock); /* if we are attaching to a running job, then attach to it */ if (NULL != nspace) { if (PMIX_SUCCESS != (rc = attach_to_running_job(nspace))) { fprintf(stderr, "Failed to attach to nspace %s: error code %d\n", nspace, rc); goto done; } } done: DEBUG_DESTRUCT_LOCK(&myrel.lock); DEBUG_DESTRUCT_LOCK(&dbrel.lock); PMIx_tool_finalize(); return(rc); }
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; }
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); }
int main(int argc, char **argv) { pmix_status_t rc; pmix_value_t *val; pmix_proc_t proc; pmix_info_t *info; size_t ninfo; pmix_query_t *query; size_t nq, n; myquery_data_t myquery_data; pid_t pid; pmix_status_t code = PMIX_ERR_JOB_TERMINATED; mylock_t mylock; myrel_t myrel; uint16_t localrank; char *target = NULL; pid = getpid(); /* init us - since we were launched by the RM, our connection info * will have been provided at startup. */ if (PMIX_SUCCESS != (rc = PMIx_tool_init(&myproc, NULL, 0))) { fprintf(stderr, "Debugger daemon: PMIx_tool_init failed: %d\n", rc); exit(0); } fprintf(stderr, "Debugger daemon ns %s rank %d pid %lu: Running\n", myproc.nspace, myproc.rank, (unsigned long)pid); /* register our default event handler */ DEBUG_CONSTRUCT_LOCK(&mylock); PMIx_Register_event_handler(NULL, 0, NULL, 0, notification_fn, evhandler_reg_callbk, (void*)&mylock); DEBUG_WAIT_THREAD(&mylock); if (PMIX_SUCCESS != mylock.status) { rc = mylock.status; DEBUG_DESTRUCT_LOCK(&mylock); goto done; } DEBUG_DESTRUCT_LOCK(&mylock); /* get the nspace of the job we are to debug - it will be in our JOB info */ #ifdef PMIX_LOAD_PROCID PMIX_LOAD_PROCID(&proc, myproc.nspace, PMIX_RANK_WILDCARD); #else PMIX_PROC_CONSTRUCT(&proc); (void)strncpy(proc.nspace, myproc.nspace, PMIX_MAX_KEYLEN); proc.rank = PMIX_RANK_WILDCARD; #endif if (PMIX_SUCCESS != (rc = PMIx_Get(&proc, PMIX_DEBUG_JOB, NULL, 0, &val))) { fprintf(stderr, "[%s:%d:%lu] Failed to get job being debugged - error %s\n", myproc.nspace, myproc.rank, (unsigned long)pid, PMIx_Error_string(rc)); goto done; } if (NULL == val || PMIX_STRING != val->type || NULL == val->data.string) { fprintf(stderr, "[%s:%d:%lu] Failed to get job being debugged - NULL data returned\n", myproc.nspace, myproc.rank, (unsigned long)pid); goto done; } /* save it for later */ target = strdup(val->data.string); PMIX_VALUE_RELEASE(val); fprintf(stderr, "[%s:%d:%lu] Debugging %s\n", myproc.nspace, myproc.rank, (unsigned long)pid, target); /* get my local rank so I can determine which local proc is "mine" * to debug */ val = NULL; if (PMIX_SUCCESS != (rc = PMIx_Get(&myproc, PMIX_LOCAL_RANK, NULL, 0, &val))) { fprintf(stderr, "[%s:%d:%lu] Failed to get my local rank - error %s\n", myproc.nspace, myproc.rank, (unsigned long)pid, PMIx_Error_string(rc)); goto done; } if (NULL == val) { fprintf(stderr, "[%s:%d:%lu] Failed to get my local rank - NULL data returned\n", myproc.nspace, myproc.rank, (unsigned long)pid); goto done; } if (PMIX_UINT16 != val->type) { fprintf(stderr, "[%s:%d:%lu] Failed to get my local rank - returned wrong type %s\n", myproc.nspace, myproc.rank, (unsigned long)pid, PMIx_Data_type_string(val->type)); goto done; } /* save the data */ localrank = val->data.uint16; PMIX_VALUE_RELEASE(val); fprintf(stderr, "[%s:%d:%lu] my local rank %d\n", myproc.nspace, myproc.rank, (unsigned long)pid, (int)localrank); /* register another handler specifically for when the target * job completes */ DEBUG_CONSTRUCT_LOCK(&myrel.lock); myrel.nspace = strdup(proc.nspace); PMIX_INFO_CREATE(info, 2); PMIX_INFO_LOAD(&info[0], PMIX_EVENT_RETURN_OBJECT, &myrel, PMIX_POINTER); /* only call me back when this specific job terminates */ PMIX_LOAD_PROCID(&proc, target, PMIX_RANK_WILDCARD); PMIX_INFO_LOAD(&info[1], PMIX_EVENT_AFFECTED_PROC, &proc, PMIX_PROC); fprintf(stderr, "[%s:%d:%lu] registering for termination of %s\n", myproc.nspace, myproc.rank, (unsigned long)pid, proc.nspace); DEBUG_CONSTRUCT_LOCK(&mylock); PMIx_Register_event_handler(&code, 1, info, 2, release_fn, evhandler_reg_callbk, (void*)&mylock); DEBUG_WAIT_THREAD(&mylock); if (PMIX_SUCCESS != mylock.status) { rc = mylock.status; DEBUG_DESTRUCT_LOCK(&mylock); PMIX_INFO_FREE(info, 2); goto done; } DEBUG_DESTRUCT_LOCK(&mylock); PMIX_INFO_FREE(info, 2); /* get our local proctable - for scalability reasons, we don't want to * have our "root" debugger process get the proctable for everybody and * send it out to us. So ask the local PMIx server for the pid's of * our local target processes */ nq = 1; PMIX_QUERY_CREATE(query, nq); PMIX_ARGV_APPEND(rc, query[0].keys, PMIX_QUERY_LOCAL_PROC_TABLE); query[0].nqual = 1; PMIX_INFO_CREATE(query[0].qualifiers, 1); PMIX_INFO_LOAD(&query[0].qualifiers[0], PMIX_NSPACE, target, PMIX_STRING); // the nspace we are enquiring about /* setup the caddy to retrieve the data */ DEBUG_CONSTRUCT_LOCK(&myquery_data.lock); myquery_data.info = NULL; myquery_data.ninfo = 0; /* execute the query */ if (PMIX_SUCCESS != (rc = PMIx_Query_info_nb(query, nq, cbfunc, (void*)&myquery_data))) { fprintf(stderr, "PMIx_Query_info failed: %d\n", rc); goto done; } DEBUG_WAIT_THREAD(&myquery_data.lock); DEBUG_DESTRUCT_LOCK(&myquery_data.lock); PMIX_QUERY_FREE(query, nq); if (PMIX_SUCCESS != myquery_data.status) { rc = myquery_data.status; goto done; } fprintf(stderr, "[%s:%d:%lu] Local proctable received\n", myproc.nspace, myproc.rank, (unsigned long)pid); /* now that we have the proctable for our local processes, we can do our * magic debugger stuff and attach to them. We then send a "release" event * to them - i.e., it's the equivalent to setting the MPIR breakpoint. We * do this with the event notification system. For this example, we just * send it to all local procs of the job being debugged */ (void)strncpy(proc.nspace, target, PMIX_MAX_NSLEN); proc.rank = PMIX_RANK_WILDCARD; ninfo = 2; PMIX_INFO_CREATE(info, ninfo); PMIX_INFO_LOAD(&info[0], PMIX_EVENT_CUSTOM_RANGE, &proc, PMIX_PROC); // deliver to the target nspace PMIX_INFO_LOAD(&info[1], PMIX_EVENT_NON_DEFAULT, NULL, PMIX_BOOL); // deliver to the target nspace fprintf(stderr, "[%s:%u:%lu] Sending release\n", myproc.nspace, myproc.rank, (unsigned long)pid); rc = PMIx_Notify_event(PMIX_ERR_DEBUGGER_RELEASE, NULL, PMIX_RANGE_CUSTOM, info, ninfo, NULL, NULL); if (PMIX_SUCCESS != rc) { fprintf(stderr, "%s[%s:%u:%lu] Sending release failed with error %s(%d)\n", myproc.nspace, myproc.rank, (unsigned long)pid, PMIx_Error_string(rc), rc); goto done; } /* do some debugger magic while waiting for the job to terminate */ DEBUG_WAIT_THREAD(&myrel.lock); done: if (NULL != target) { free(target); } /* finalize us */ fprintf(stderr, "Debugger daemon ns %s rank %d pid %lu: Finalizing\n", myproc.nspace, myproc.rank, (unsigned long)pid); if (PMIX_SUCCESS != (rc = PMIx_Finalize(NULL, 0))) { fprintf(stderr, "Debugger daemon ns %s rank %d:PMIx_Finalize failed: %d\n", myproc.nspace, myproc.rank, rc); } else { fprintf(stderr, "Debugger daemon ns %s rank %d pid %lu:PMIx_Finalize successfully completed\n", myproc.nspace, myproc.rank, (unsigned long)pid); } fflush(stderr); return(0); }