static void _fill_registration_msg(slurm_node_registration_status_msg_t *msg) { List steps; ListIterator i; step_loc_t *stepd; int n; char *arch, *os; struct utsname buf; static bool first_msg = true; static time_t slurmd_start_time = 0; Buf gres_info; msg->node_name = xstrdup (conf->node_name); msg->cpus = conf->cpus; msg->boards = conf->boards; msg->sockets = conf->sockets; msg->cores = conf->cores; msg->threads = conf->threads; msg->real_memory = conf->real_memory_size; msg->tmp_disk = conf->tmp_disk_space; msg->hash_val = slurm_get_hash_val(); get_cpu_load(&msg->cpu_load); gres_info = init_buf(1024); if (gres_plugin_node_config_pack(gres_info) != SLURM_SUCCESS) error("error packing gres configuration"); else msg->gres_info = gres_info; get_up_time(&conf->up_time); msg->up_time = conf->up_time; if (slurmd_start_time == 0) slurmd_start_time = time(NULL); msg->slurmd_start_time = slurmd_start_time; if (first_msg) { first_msg = false; info("CPUs=%u Boards=%u Sockets=%u Cores=%u Threads=%u " "Memory=%u TmpDisk=%u Uptime=%u", msg->cpus, msg->boards, msg->sockets, msg->cores, msg->threads, msg->real_memory, msg->tmp_disk, msg->up_time); } else { debug3("CPUs=%u Boards=%u Sockets=%u Cores=%u Threads=%u " "Memory=%u TmpDisk=%u Uptime=%u", msg->cpus, msg->boards, msg->sockets, msg->cores, msg->threads, msg->real_memory, msg->tmp_disk, msg->up_time); } uname(&buf); if ((arch = getenv("SLURM_ARCH"))) msg->arch = xstrdup(arch); else msg->arch = xstrdup(buf.machine); if ((os = getenv("SLURM_OS"))) msg->os = xstrdup(os); else msg->os = xstrdup(buf.sysname); if (msg->startup) { if (switch_g_alloc_node_info(&msg->switch_nodeinfo)) error("switch_g_alloc_node_info: %m"); if (switch_g_build_node_info(msg->switch_nodeinfo)) error("switch_g_build_node_info: %m"); } steps = stepd_available(conf->spooldir, conf->node_name); msg->job_count = list_count(steps); msg->job_id = xmalloc(msg->job_count * sizeof(*msg->job_id)); /* Note: Running batch jobs will have step_id == NO_VAL */ msg->step_id = xmalloc(msg->job_count * sizeof(*msg->step_id)); i = list_iterator_create(steps); n = 0; while ((stepd = list_next(i))) { int fd; fd = stepd_connect(stepd->directory, stepd->nodename, stepd->jobid, stepd->stepid); if (fd == -1) { --(msg->job_count); continue; } if (stepd_state(fd) == SLURMSTEPD_NOT_RUNNING) { debug("stale domain socket for stepd %u.%u ", stepd->jobid, stepd->stepid); --(msg->job_count); close(fd); continue; } close(fd); if (stepd->stepid == NO_VAL) debug("found apparently running job %u", stepd->jobid); else debug("found apparently running step %u.%u", stepd->jobid, stepd->stepid); msg->job_id[n] = stepd->jobid; msg->step_id[n] = stepd->stepid; n++; } list_iterator_destroy(i); list_destroy(steps); if (!msg->energy) msg->energy = acct_gather_energy_alloc(); acct_gather_energy_g_get_data(ENERGY_DATA_STRUCT, msg->energy); msg->timestamp = time(NULL); return; }
/* * main - slurmctld main function, start various threads and process RPCs * test7.17.prog <TRES_PER_NODE> <CONFIG_DIR_HEAD> <CONFIG_SUB_DIR> <CPU_COUNT> * */ int main(int argc, char *argv[]) { log_options_t opts = LOG_OPTS_STDERR_ONLY; int rc; uint32_t cpu_count, cpu_alloc, job_id = 12345; char *node_name, *reason_down = NULL; char *orig_config, *new_config = NULL, *tres_per_node = NULL; Buf buffer; List job_gres_list = NULL, node_gres_list = NULL; bitstr_t *cpu_bitmap; char config_dir[10000], test[1000]; char slurm_conf[1000]; uint32_t num_tasks = 1; uint32_t min_nodes = 1; uint32_t max_nodes = 1; uint16_t ntasks_per_node = NO_VAL16; uint16_t ntasks_per_socket = NO_VAL16; uint16_t sockets_per_node = NO_VAL16; uint16_t cpus_per_task = NO_VAL16; int core_count, sock_count; /* Setup slurm.conf and gres.conf test paths */ strcpy(config_dir, argv[2]); strcpy(config_dir,strcat(config_dir, "/test7.17_configs")); strcpy(test, strcat(config_dir, argv[3])); strcpy(slurm_conf, strcat(test, "/slurm.conf")); /* Enable detailed logging for now */ opts.stderr_level = LOG_LEVEL_DEBUG; log_init(argv[0], opts, SYSLOG_FACILITY_USER, NULL); /* * Logic normally executed by slurmd daemon */ setenv("SLURM_CONF", slurm_conf, 1); rc = gres_plugin_init(); if (rc != SLURM_SUCCESS) { slurm_perror("failure: gres_plugin_init"); exit(1); } setenv("SLURM_CONFIG_DIR", config_dir, 1); cpu_count = strtol(argv[4], NULL, 10); node_name = "test_node"; rc = gres_plugin_node_config_load(cpu_count, node_name, NULL, NULL, NULL); if (rc != SLURM_SUCCESS) { slurm_perror("failure: gres_plugin_node_config_load"); exit(1); } buffer = init_buf(1024); rc = gres_plugin_node_config_pack(buffer); if (rc != SLURM_SUCCESS) { slurm_perror("failure: gres_plugin_node_config_pack"); exit(1); } /* * Logic normally executed by slurmctld daemon */ orig_config = "gpu:8"; rc = gres_plugin_init_node_config(node_name, orig_config, &node_gres_list); if (rc != SLURM_SUCCESS) { slurm_perror("failure: gres_plugin_init_node_config"); exit(1); } set_buf_offset(buffer, 0); rc = gres_plugin_node_config_unpack(buffer, node_name); if (rc != SLURM_SUCCESS) { slurm_perror("failure: gres_plugin_node_config_unpack"); exit(1); } core_count = cpu_count; sock_count = 1; rc = gres_plugin_node_config_validate(node_name, orig_config, &new_config, &node_gres_list, cpu_count, core_count, sock_count, 0, &reason_down); if (rc != SLURM_SUCCESS) { slurm_perror("failure: gres_plugin_node_config_validate"); exit(1); } if (argc > 2) tres_per_node = xstrdup(argv[1]); rc = gres_plugin_job_state_validate(NULL, /* cpus_per_tres */ NULL, /* tres_freq */ NULL, /* tres_per_job */ tres_per_node, NULL, /* tres_per_socket */ NULL, /* tres_per_task */ NULL, /* mem_per_tres */ &num_tasks, &min_nodes, &max_nodes, &ntasks_per_node, &ntasks_per_socket, &sockets_per_node, &cpus_per_task, &job_gres_list); if (rc != SLURM_SUCCESS) { slurm_seterrno(rc); slurm_perror("failure: gres_plugin_job_state_validate"); exit(1); } gres_plugin_node_state_log(node_gres_list, node_name); gres_plugin_job_state_log(job_gres_list, job_id); cpu_bitmap = bit_alloc(cpu_count); bit_nset(cpu_bitmap, 0, cpu_count - 1); cpu_alloc = gres_plugin_job_test(job_gres_list, node_gres_list, true, cpu_bitmap, 0, cpu_count - 1, job_id, node_name); if (cpu_alloc == NO_VAL) printf("cpu_alloc=ALL\n"); else printf("cpu_alloc=%u\n", cpu_alloc); rc = gres_plugin_fini(); if (rc != SLURM_SUCCESS) { slurm_perror("failure: gres_plugin_fini"); exit(1); } printf("Test %s ran to completion\n\n", argv[3]); exit(0); }