/* create a slurmd job structure from a launch tasks message */
extern stepd_step_rec_t *stepd_step_rec_create(launch_tasks_request_msg_t *msg,
uint16_t protocol_version)
{
stepd_step_rec_t *job = NULL;
srun_info_t *srun = NULL;
slurm_addr_t resp_addr;
slurm_addr_t io_addr;
int i, nodeid = NO_VAL;
xassert(msg != NULL);
xassert(msg->complete_nodelist != NULL);
debug3("entering stepd_step_rec_create");
if (acct_gather_check_acct_freq_task(msg->job_mem_lim, msg->acctg_freq))
return NULL;
job = xmalloc(sizeof(stepd_step_rec_t));
job->msg = msg;
#ifndef HAVE_FRONT_END
nodeid = nodelist_find(msg->complete_nodelist, conf->node_name);
job->node_name = xstrdup(conf->node_name);
#else
nodeid = 0;
job->node_name = xstrdup(msg->complete_nodelist);
#endif
if (nodeid < 0) {
error("couldn't find node %s in %s",
job->node_name, msg->complete_nodelist);
stepd_step_rec_destroy(job);
return NULL;
}
job->state = SLURMSTEPD_STEP_STARTING;
slurm_cond_init(&job->state_cond, NULL);
slurm_mutex_init(&job->state_mutex);
job->node_tasks = msg->tasks_to_launch[nodeid];
i = sizeof(uint16_t) * msg->nnodes;
job->task_cnts = xmalloc(i);
memcpy(job->task_cnts, msg->tasks_to_launch, i);
job->ntasks = msg->ntasks;
job->jobid = msg->job_id;
job->stepid = msg->job_step_id;
job->uid = (uid_t) msg->uid;
job->gid = (gid_t) msg->gid;
job->user_name = xstrdup(msg->user_name);
job->ngids = (int) msg->ngids;
job->gids = copy_gids(msg->ngids, msg->gids);
job->cwd = xstrdup(msg->cwd);
job->task_dist = msg->task_dist;
job->cpu_bind_type = msg->cpu_bind_type;
job->cpu_bind = xstrdup(msg->cpu_bind);
job->mem_bind_type = msg->mem_bind_type;
job->mem_bind = xstrdup(msg->mem_bind);
job->cpu_freq_min = msg->cpu_freq_min;
job->cpu_freq_max = msg->cpu_freq_max;
job->cpu_freq_gov = msg->cpu_freq_gov;
job->ckpt_dir = xstrdup(msg->ckpt_dir);
job->restart_dir = xstrdup(msg->restart_dir);
job->cpus_per_task = msg->cpus_per_task;
job->env = _array_copy(msg->envc, msg->env);
job->array_job_id = msg->job_id;
job->array_task_id = NO_VAL;
job->node_offset = msg->node_offset; /* Used for env vars */
job->pack_jobid = msg->pack_jobid; /* Used for env vars */
job->pack_nnodes = msg->pack_nnodes; /* Used for env vars */
if (msg->pack_nnodes && msg->pack_ntasks && msg->pack_task_cnts) {
job->pack_ntasks = msg->pack_ntasks; /* Used for env vars */
i = sizeof(uint16_t) * msg->pack_nnodes;
job->pack_task_cnts = xmalloc(i);
memcpy(job->pack_task_cnts, msg->pack_task_cnts, i);
}
job->pack_offset = msg->pack_offset; /* Used for env vars & labels */
job->pack_task_offset = msg->pack_task_offset; /* Used for env vars & labels */
job->pack_node_list = xstrdup(msg->pack_node_list);
for (i = 0; i < msg->envc; i++) {
/* 1234567890123456789 */
if (!xstrncmp(msg->env[i], "SLURM_ARRAY_JOB_ID=", 19))
job->array_job_id = atoi(msg->env[i] + 19);
/* 12345678901234567890 */
if (!xstrncmp(msg->env[i], "SLURM_ARRAY_TASK_ID=", 20))
job->array_task_id = atoi(msg->env[i] + 20);
}
job->eio = eio_handle_create(0);
job->sruns = list_create((ListDelF) _srun_info_destructor);
/*
* Based on my testing the next 3 lists here could use the
* eio_obj_destroy, but if you do you can get an invalid read. Since
* these stay until the end of the job it isn't that big of a deal.
*/
job->clients = list_create(NULL); /* FIXME! Needs destructor */
job->stdout_eio_objs = list_create(NULL); /* FIXME! Needs destructor */
job->stderr_eio_objs = list_create(NULL); /* FIXME! Needs destructor */
job->free_incoming = list_create(NULL); /* FIXME! Needs destructor */
job->incoming_count = 0;
job->free_outgoing = list_create(NULL); /* FIXME! Needs destructor */
job->outgoing_count = 0;
job->outgoing_cache = list_create(NULL); /* FIXME! Needs destructor */
job->envtp = xmalloc(sizeof(env_t));
job->envtp->jobid = -1;
job->envtp->stepid = -1;
job->envtp->procid = -1;
job->envtp->localid = -1;
job->envtp->nodeid = -1;
job->envtp->distribution = 0;
job->envtp->cpu_bind_type = 0;
job->envtp->cpu_bind = NULL;
job->envtp->mem_bind_type = 0;
job->envtp->mem_bind = NULL;
job->envtp->ckpt_dir = NULL;
if (!msg->resp_port)
msg->num_resp_port = 0;
if (msg->num_resp_port) {
job->envtp->comm_port =
msg->resp_port[nodeid % msg->num_resp_port];
memcpy(&resp_addr, &msg->orig_addr, sizeof(slurm_addr_t));
slurm_set_addr(&resp_addr,
msg->resp_port[nodeid % msg->num_resp_port],
NULL);
} else {
memset(&resp_addr, 0, sizeof(slurm_addr_t));
}
if (!msg->io_port)
msg->flags |= LAUNCH_USER_MANAGED_IO;
if ((msg->flags & LAUNCH_USER_MANAGED_IO) == 0) {
memcpy(&io_addr, &msg->orig_addr, sizeof(slurm_addr_t));
slurm_set_addr(&io_addr,
msg->io_port[nodeid % msg->num_io_port],
NULL);
} else {
memset(&io_addr, 0, sizeof(slurm_addr_t));
}
srun = srun_info_create(msg->cred, &resp_addr, &io_addr,
protocol_version);
job->profile = msg->profile;
job->task_prolog = xstrdup(msg->task_prolog);
job->task_epilog = xstrdup(msg->task_epilog);
job->argc = msg->argc;
job->argv = _array_copy(job->argc, msg->argv);
job->nnodes = msg->nnodes;
job->nodeid = nodeid;
job->debug = msg->slurmd_debug;
job->cpus = msg->node_cpus;
job->job_core_spec = msg->job_core_spec;
/* This needs to happen before acct_gather_profile_startpoll
and only really looks at the profile in the job.
*/
acct_gather_profile_g_node_step_start(job);
acct_gather_profile_startpoll(msg->acctg_freq,
conf->job_acct_gather_freq);
job->timelimit = (time_t) -1;
job->flags = msg->flags;
job->switch_job = msg->switch_job;
job->open_mode = msg->open_mode;
job->options = msg->options;
format_core_allocs(msg->cred, conf->node_name, conf->cpus,
&job->job_alloc_cores, &job->step_alloc_cores,
&job->job_mem, &job->step_mem);
/* If users have configured MemLimitEnforce=no
* in their slurm.conf keep going.
*/
if (job->step_mem
&& conf->mem_limit_enforce) {
jobacct_gather_set_mem_limit(job->jobid, job->stepid,
job->step_mem);
} else if (job->job_mem
&& conf->mem_limit_enforce) {
jobacct_gather_set_mem_limit(job->jobid, job->stepid,
job->job_mem);
}
#ifdef HAVE_ALPS_CRAY
/* This is only used for Cray emulation mode where slurmd is used to
* launch job steps. On a real Cray system, ALPS is used to launch
* the tasks instead of SLURM. SLURM's task launch RPC does NOT
* contain the reservation ID, so just use some non-zero value here
* for testing purposes. */
job->resv_id = 1;
select_g_select_jobinfo_set(msg->select_jobinfo, SELECT_JOBDATA_RESV_ID,
&job->resv_id);
#endif
/* only need these values on the extern step, don't copy otherwise */
if ((msg->job_step_id == SLURM_EXTERN_CONT) && msg->x11) {
job->x11 = msg->x11;
job->x11_magic_cookie = xstrdup(msg->x11_magic_cookie);
job->x11_target_host = xstrdup(msg->x11_target_host);
job->x11_target_port = msg->x11_target_port;
}
get_cred_gres(msg->cred, conf->node_name,
&job->job_gres_list, &job->step_gres_list);
list_append(job->sruns, (void *) srun);
_job_init_task_info(job, msg->global_task_ids,
msg->ifname, msg->ofname, msg->efname);
return job;
}
/* create a slurmd job structure from a launch tasks message */
extern stepd_step_rec_t *
stepd_step_rec_create(launch_tasks_request_msg_t *msg)
{
stepd_step_rec_t *job = NULL;
srun_info_t *srun = NULL;
slurm_addr_t resp_addr;
slurm_addr_t io_addr;
int i, nodeid = NO_VAL;
xassert(msg != NULL);
xassert(msg->complete_nodelist != NULL);
debug3("entering stepd_step_rec_create");
if (!_valid_uid_gid((uid_t)msg->uid, &(msg->gid), &(msg->user_name)))
return NULL;
if (_check_acct_freq_task(msg->job_mem_lim, msg->acctg_freq))
return NULL;
job = xmalloc(sizeof(stepd_step_rec_t));
#ifndef HAVE_FRONT_END
nodeid = nodelist_find(msg->complete_nodelist, conf->node_name);
job->node_name = xstrdup(conf->node_name);
#else
nodeid = 0;
job->node_name = xstrdup(msg->complete_nodelist);
#endif
if (nodeid < 0) {
error("couldn't find node %s in %s",
job->node_name, msg->complete_nodelist);
stepd_step_rec_destroy(job);
return NULL;
}
job->state = SLURMSTEPD_STEP_STARTING;
job->node_tasks = msg->tasks_to_launch[nodeid];
job->ntasks = msg->ntasks;
job->jobid = msg->job_id;
job->stepid = msg->job_step_id;
job->uid = (uid_t) msg->uid;
job->user_name = xstrdup(msg->user_name);
job->gid = (gid_t) msg->gid;
job->cwd = xstrdup(msg->cwd);
job->task_dist = msg->task_dist;
job->cpu_bind_type = msg->cpu_bind_type;
job->cpu_bind = xstrdup(msg->cpu_bind);
job->mem_bind_type = msg->mem_bind_type;
job->mem_bind = xstrdup(msg->mem_bind);
job->cpu_freq = msg->cpu_freq;
job->ckpt_dir = xstrdup(msg->ckpt_dir);
job->restart_dir = xstrdup(msg->restart_dir);
job->cpus_per_task = msg->cpus_per_task;
job->env = _array_copy(msg->envc, msg->env);
job->array_job_id = msg->job_id;
job->array_task_id = (uint16_t) NO_VAL;
for (i = 0; i < msg->envc; i++) {
/* 1234567890123456789 */
if (!strncmp(msg->env[i], "SLURM_ARRAY_JOB_ID=", 19))
job->array_job_id = atoi(msg->env[i] + 19);
/* 12345678901234567890 */
if (!strncmp(msg->env[i], "SLURM_ARRAY_TASK_ID=", 20))
job->array_task_id = atoi(msg->env[i] + 20);
}
job->eio = eio_handle_create();
job->sruns = list_create((ListDelF) _srun_info_destructor);
job->clients = list_create(NULL); /* FIXME! Needs destructor */
job->stdout_eio_objs = list_create(NULL); /* FIXME! Needs destructor */
job->stderr_eio_objs = list_create(NULL); /* FIXME! Needs destructor */
job->free_incoming = list_create(NULL); /* FIXME! Needs destructor */
job->incoming_count = 0;
job->free_outgoing = list_create(NULL); /* FIXME! Needs destructor */
job->outgoing_count = 0;
job->outgoing_cache = list_create(NULL); /* FIXME! Needs destructor */
job->envtp = xmalloc(sizeof(env_t));
job->envtp->jobid = -1;
job->envtp->stepid = -1;
job->envtp->procid = -1;
job->envtp->localid = -1;
job->envtp->nodeid = -1;
job->envtp->distribution = 0;
job->envtp->cpu_bind_type = 0;
job->envtp->cpu_bind = NULL;
job->envtp->mem_bind_type = 0;
job->envtp->mem_bind = NULL;
job->envtp->ckpt_dir = NULL;
job->envtp->comm_port = msg->resp_port[nodeid % msg->num_resp_port];
memcpy(&resp_addr, &msg->orig_addr, sizeof(slurm_addr_t));
slurm_set_addr(&resp_addr,
msg->resp_port[nodeid % msg->num_resp_port],
NULL);
job->user_managed_io = msg->user_managed_io;
if (!msg->user_managed_io) {
memcpy(&io_addr, &msg->orig_addr, sizeof(slurm_addr_t));
slurm_set_addr(&io_addr,
msg->io_port[nodeid % msg->num_io_port],
NULL);
}
srun = srun_info_create(msg->cred, &resp_addr, &io_addr);
job->buffered_stdio = msg->buffered_stdio;
job->labelio = msg->labelio;
job->profile = msg->profile;
job->task_prolog = xstrdup(msg->task_prolog);
job->task_epilog = xstrdup(msg->task_epilog);
job->argc = msg->argc;
job->argv = _array_copy(job->argc, msg->argv);
job->nnodes = msg->nnodes;
job->nodeid = nodeid;
job->debug = msg->slurmd_debug;
job->cpus = msg->cpus_allocated[nodeid];
/* This needs to happen before acct_gather_profile_startpoll
and only really looks at the profile in the job.
*/
acct_gather_profile_g_node_step_start(job);
acct_gather_profile_startpoll(msg->acctg_freq,
conf->job_acct_gather_freq);
job->multi_prog = msg->multi_prog;
job->timelimit = (time_t) -1;
job->task_flags = msg->task_flags;
job->switch_job = msg->switch_job;
job->pty = msg->pty;
job->open_mode = msg->open_mode;
job->options = msg->options;
format_core_allocs(msg->cred, conf->node_name, conf->cpus,
&job->job_alloc_cores, &job->step_alloc_cores,
&job->job_mem, &job->step_mem);
if (job->step_mem) {
jobacct_gather_set_mem_limit(job->jobid, job->stepid,
job->step_mem);
} else if (job->job_mem) {
jobacct_gather_set_mem_limit(job->jobid, job->stepid,
job->job_mem);
}
#ifdef HAVE_ALPS_CRAY
/* This is only used for Cray emulation mode where slurmd is used to
* launch job steps. On a real Cray system, ALPS is used to launch
* the tasks instead of SLURM. SLURM's task launch RPC does NOT
* contain the reservation ID, so just use some non-zero value here
* for testing purposes. */
job->resv_id = 1;
select_g_select_jobinfo_set(msg->select_jobinfo, SELECT_JOBDATA_RESV_ID,
&job->resv_id);
#endif
get_cred_gres(msg->cred, conf->node_name,
&job->job_gres_list, &job->step_gres_list);
list_append(job->sruns, (void *) srun);
_job_init_task_info(job, msg->global_task_ids[nodeid],
msg->ifname, msg->ofname, msg->efname);
return job;
}
/* create a slurmd job structure from a launch tasks message */
slurmd_job_t *
job_create(launch_tasks_request_msg_t *msg)
{
struct passwd *pwd = NULL;
slurmd_job_t *job = NULL;
srun_info_t *srun = NULL;
slurm_addr_t resp_addr;
slurm_addr_t io_addr;
int nodeid = NO_VAL;
xassert(msg != NULL);
xassert(msg->complete_nodelist != NULL);
debug3("entering job_create");
if ((pwd = _pwd_create((uid_t)msg->uid)) == NULL) {
error("uid %ld not found on system", (long) msg->uid);
slurm_seterrno (ESLURMD_UID_NOT_FOUND);
return NULL;
}
if (!_valid_gid(pwd, &(msg->gid))) {
slurm_seterrno (ESLURMD_GID_NOT_FOUND);
_pwd_destroy(pwd);
return NULL;
}
if (msg->job_mem_lim && (msg->acctg_freq != (uint16_t) NO_VAL)
&& (msg->acctg_freq > conf->job_acct_gather_freq)) {
error("Can't set frequency to %u, it is higher than %u. "
"We need it to be at least at this level to "
"monitor memory usage.",
msg->acctg_freq, conf->job_acct_gather_freq);
slurm_seterrno (ESLURMD_INVALID_ACCT_FREQ);
_pwd_destroy(pwd);
return NULL;
}
job = xmalloc(sizeof(slurmd_job_t));
#ifndef HAVE_FRONT_END
nodeid = nodelist_find(msg->complete_nodelist, conf->node_name);
job->node_name = xstrdup(conf->node_name);
#else
nodeid = 0;
job->node_name = xstrdup(msg->complete_nodelist);
#endif
if(nodeid < 0) {
error("couldn't find node %s in %s",
job->node_name, msg->complete_nodelist);
job_destroy(job);
return NULL;
}
job->state = SLURMSTEPD_STEP_STARTING;
job->pwd = pwd;
job->node_tasks = msg->tasks_to_launch[nodeid];
job->ntasks = msg->ntasks;
job->jobid = msg->job_id;
job->stepid = msg->job_step_id;
job->uid = (uid_t) msg->uid;
job->gid = (gid_t) msg->gid;
job->cwd = xstrdup(msg->cwd);
job->task_dist = msg->task_dist;
job->cpu_bind_type = msg->cpu_bind_type;
job->cpu_bind = xstrdup(msg->cpu_bind);
job->mem_bind_type = msg->mem_bind_type;
job->mem_bind = xstrdup(msg->mem_bind);
job->cpu_freq = msg->cpu_freq;
job->ckpt_dir = xstrdup(msg->ckpt_dir);
job->restart_dir = xstrdup(msg->restart_dir);
job->cpus_per_task = msg->cpus_per_task;
job->env = _array_copy(msg->envc, msg->env);
job->eio = eio_handle_create();
job->sruns = list_create((ListDelF) _srun_info_destructor);
job->clients = list_create(NULL); /* FIXME! Needs destructor */
job->stdout_eio_objs = list_create(NULL); /* FIXME! Needs destructor */
job->stderr_eio_objs = list_create(NULL); /* FIXME! Needs destructor */
job->free_incoming = list_create(NULL); /* FIXME! Needs destructor */
job->incoming_count = 0;
job->free_outgoing = list_create(NULL); /* FIXME! Needs destructor */
job->outgoing_count = 0;
job->outgoing_cache = list_create(NULL); /* FIXME! Needs destructor */
job->envtp = xmalloc(sizeof(env_t));
job->envtp->jobid = -1;
job->envtp->stepid = -1;
job->envtp->procid = -1;
job->envtp->localid = -1;
job->envtp->nodeid = -1;
job->envtp->distribution = 0;
job->envtp->cpu_bind_type = 0;
job->envtp->cpu_bind = NULL;
job->envtp->mem_bind_type = 0;
job->envtp->mem_bind = NULL;
job->envtp->ckpt_dir = NULL;
//job->envtp->comm_port = msg->resp_port[nodeid % msg->num_resp_port];
/*memcpy(&resp_addr, &msg->orig_addr, sizeof(slurm_addr_t));
slurm_set_addr(&resp_addr,
msg->resp_port[nodeid % msg->num_resp_port],
NULL);
job->user_managed_io = msg->user_managed_io;
if (!msg->user_managed_io) {
memcpy(&io_addr, &msg->orig_addr, sizeof(slurm_addr_t));
slurm_set_addr(&io_addr,
msg->io_port[nodeid % msg->num_io_port],
NULL);
}*/
//srun = srun_info_create(msg->cred, &resp_addr, &io_addr);
srun = srun_info_create(NULL, NULL, NULL);
job->buffered_stdio = msg->buffered_stdio;
job->labelio = msg->labelio;
job->task_prolog = xstrdup(msg->task_prolog);
job->task_epilog = xstrdup(msg->task_epilog);
job->argc = msg->argc;
job->argv = _array_copy(job->argc, msg->argv);
job->nnodes = msg->nnodes;
job->nodeid = nodeid;
job->debug = msg->slurmd_debug;
job->cpus = msg->cpus_allocated[nodeid];
if (msg->acctg_freq != (uint16_t) NO_VAL)
jobacct_gather_change_poll(msg->acctg_freq);
job->multi_prog = msg->multi_prog;
job->timelimit = (time_t) -1;
job->task_flags = msg->task_flags;
job->switch_job = msg->switch_job;
job->pty = msg->pty;
job->open_mode = msg->open_mode;
job->options = msg->options;
format_core_allocs(msg->cred, conf->node_name,
&job->job_alloc_cores, &job->step_alloc_cores,
&job->job_mem, &job->step_mem);
if (job->step_mem) {
jobacct_gather_set_mem_limit(job->jobid, job->stepid,
job->step_mem);
} else if (job->job_mem) {
jobacct_gather_set_mem_limit(job->jobid, job->stepid,
job->job_mem);
}
#ifdef HAVE_CRAY
/* This is only used for Cray emulation mode where slurmd is used to
* launch job steps. On a real Cray system, ALPS is used to launch
* the tasks instead of SLURM. SLURM's task launch RPC does NOT
* contain the reservation ID, so just use some non-zero value here
* for testing purposes. */
job->resv_id = 1;
select_g_select_jobinfo_set(msg->select_jobinfo, SELECT_JOBDATA_RESV_ID,
&job->resv_id);
#endif
get_cred_gres(msg->cred, conf->node_name,
&job->job_gres_list, &job->step_gres_list);
list_append(job->sruns, (void *) srun);
_job_init_task_info(job, msg->global_task_ids[nodeid],
msg->ifname, msg->ofname, msg->efname);
return job;
}