/*
* Current process is running as the user when this is called.
*/
void
exec_task(slurmd_job_t *job, int i)
{
uint32_t *gtids; /* pointer to arrary of ranks */
int fd, j;
slurmd_task_info_t *task = job->task[i];
char **tmp_env;
if (i == 0)
_make_tmpdir(job);
gtids = xmalloc(job->node_tasks * sizeof(uint32_t));
for (j = 0; j < job->node_tasks; j++)
gtids[j] = job->task[j]->gtid;
job->envtp->sgtids = _uint32_array_to_str(job->node_tasks, gtids);
xfree(gtids);
job->envtp->jobid = job->jobid;
job->envtp->stepid = job->stepid;
job->envtp->nodeid = job->nodeid;
job->envtp->cpus_on_node = job->cpus;
job->envtp->procid = task->gtid;
job->envtp->localid = task->id;
job->envtp->task_pid = getpid();
job->envtp->distribution = job->task_dist;
job->envtp->cpu_bind = xstrdup(job->cpu_bind);
job->envtp->cpu_bind_type = job->cpu_bind_type;
job->envtp->cpu_freq = job->cpu_freq;
job->envtp->mem_bind = xstrdup(job->mem_bind);
job->envtp->mem_bind_type = job->mem_bind_type;
job->envtp->distribution = -1;
job->envtp->ckpt_dir = xstrdup(job->ckpt_dir);
job->envtp->batch_flag = job->batch;
/* Modify copy of job's environment. Do not alter in place or
* concurrent searches of the environment can generate invalid memory
* references. */
job->envtp->env = env_array_copy((const char **) job->env);
setup_env(job->envtp, false);
setenvf(&job->envtp->env, "SLURMD_NODENAME", "%s", conf->node_name);
tmp_env = job->env;
job->env = job->envtp->env;
env_array_free(tmp_env);
job->envtp->env = NULL;
xfree(job->envtp->task_count);
if (task->argv[0] && *task->argv[0] != '/') {
/*
* Normally the client (srun) expands the command name
* to a fully qualified path, but in --multi-prog mode it
* is left up to the server to search the PATH for the
* executable.
*/
task->argv[0] = _build_path(task->argv[0], job->env);
}
if (!job->batch) {
if (interconnect_attach(job->switch_job, &job->env,
job->nodeid, (uint32_t) i, job->nnodes,
job->ntasks, task->gtid) < 0) {
error("Unable to attach to interconnect: %m");
log_fini();
exit(1);
}
if (_setup_mpi(job, i) != SLURM_SUCCESS) {
error("Unable to configure MPI plugin: %m");
log_fini();
exit(1);
}
}
/* task-specific pre-launch activities */
if (spank_user_task (job, i) < 0) {
error ("Failed to invoke task plugin stack");
exit (1);
}
/* task plugin hook */
if (pre_launch(job)) {
error ("Failed task affinity setup");
exit (1);
}
if (conf->task_prolog) {
char *my_prolog;
slurm_mutex_lock(&conf->config_mutex);
my_prolog = xstrdup(conf->task_prolog);
slurm_mutex_unlock(&conf->config_mutex);
_run_script_and_set_env("slurm task_prolog",
my_prolog, job);
xfree(my_prolog);
}
if (job->task_prolog) {
_run_script_and_set_env("user task_prolog",
job->task_prolog, job);
}
if (!job->batch)
pdebug_stop_current(job);
if (job->env == NULL) {
debug("job->env is NULL");
job->env = (char **)xmalloc(sizeof(char *));
job->env[0] = (char *)NULL;
}
if (job->restart_dir) {
info("restart from %s", job->restart_dir);
/* no return on success */
checkpoint_restart_task(job, job->restart_dir, task->gtid);
error("Restart task failed: %m");
exit(errno);
}
if (task->argv[0] == NULL) {
error("No executable program specified for this task");
exit(2);
}
/* Do this last so you don't worry too much about the users
limits including the slurmstepd in with it.
*/
if (set_user_limits(job) < 0) {
debug("Unable to set user limits");
log_fini();
exit(5);
}
execve(task->argv[0], task->argv, job->env);
/*
* print error message and clean up if execve() returns:
*/
if ((errno == ENOENT) &&
((fd = open(task->argv[0], O_RDONLY)) >= 0)) {
char buf[256], *eol;
int sz;
sz = read(fd, buf, sizeof(buf));
if ((sz >= 3) && (strncmp(buf, "#!", 2) == 0)) {
eol = strchr(buf, '\n');
if (eol)
eol[0] = '\0';
else
buf[sizeof(buf)-1] = '\0';
error("execve(): bad interpreter(%s): %m", buf+2);
exit(errno);
}
}
error("execve(): %s: %m", task->argv[0]);
exit(errno);
}
/* execute archive script */
extern int archive_run_script(slurmdb_archive_cond_t *arch_cond,
char *cluster_name, time_t last_submit)
{
char * args[] = {arch_cond->archive_script, NULL};
struct stat st;
char **env = NULL;
time_t curr_end;
if (stat(arch_cond->archive_script, &st) < 0) {
error("archive_run_script: failed to stat %s: %m",
arch_cond->archive_script);
return SLURM_ERROR;
}
if (!(st.st_mode & S_IFREG)) {
errno = EACCES;
error("archive_run_script: %s isn't a regular file",
arch_cond->archive_script);
return SLURM_ERROR;
}
if (access(arch_cond->archive_script, X_OK) < 0) {
errno = EACCES;
error("archive_run_script: %s is not executable",
arch_cond->archive_script);
return SLURM_ERROR;
}
env = env_array_create();
env_array_append_fmt(&env, "SLURM_ARCHIVE_CLUSTER", "%s",
cluster_name);
if (arch_cond->purge_event != NO_VAL) {
if (!(curr_end = archive_setup_end_time(
last_submit, arch_cond->purge_event))) {
error("Parsing purge events failed");
return SLURM_ERROR;
}
env_array_append_fmt(&env, "SLURM_ARCHIVE_EVENTS", "%u",
SLURMDB_PURGE_ARCHIVE_SET(
arch_cond->purge_event));
env_array_append_fmt(&env, "SLURM_ARCHIVE_LAST_EVENT", "%ld",
(long)curr_end);
}
if (arch_cond->purge_job != NO_VAL) {
if (!(curr_end = archive_setup_end_time(
last_submit, arch_cond->purge_job))) {
error("Parsing purge job failed");
return SLURM_ERROR;
}
env_array_append_fmt(&env, "SLURM_ARCHIVE_JOBS", "%u",
SLURMDB_PURGE_ARCHIVE_SET(
arch_cond->purge_job));
env_array_append_fmt(&env, "SLURM_ARCHIVE_LAST_JOB", "%ld",
(long)curr_end);
}
if (arch_cond->purge_resv != NO_VAL) {
if (!(curr_end = archive_setup_end_time(
last_submit, arch_cond->purge_job))) {
error("Parsing purge job failed");
return SLURM_ERROR;
}
env_array_append_fmt(&env, "SLURM_ARCHIVE_RESV", "%u",
SLURMDB_PURGE_ARCHIVE_SET(
arch_cond->purge_job));
env_array_append_fmt(&env, "SLURM_ARCHIVE_LAST_RESV", "%ld",
(long)curr_end);
}
if (arch_cond->purge_step != NO_VAL) {
if (!(curr_end = archive_setup_end_time(
last_submit, arch_cond->purge_step))) {
error("Parsing purge step");
return SLURM_ERROR;
}
env_array_append_fmt(&env, "SLURM_ARCHIVE_STEPS", "%u",
SLURMDB_PURGE_ARCHIVE_SET(
arch_cond->purge_step));
env_array_append_fmt(&env, "SLURM_ARCHIVE_LAST_STEP", "%ld",
(long)curr_end);
}
if (arch_cond->purge_suspend != NO_VAL) {
if (!(curr_end = archive_setup_end_time(
last_submit, arch_cond->purge_suspend))) {
error("Parsing purge suspend");
return SLURM_ERROR;
}
env_array_append_fmt(&env, "SLURM_ARCHIVE_SUSPEND", "%u",
SLURMDB_PURGE_ARCHIVE_SET(
arch_cond->purge_suspend));
env_array_append_fmt(&env, "SLURM_ARCHIVE_LAST_SUSPEND", "%ld",
(long)curr_end);
}
if (arch_cond->purge_txn != NO_VAL) {
if (!(curr_end = archive_setup_end_time(
last_submit, arch_cond->purge_txn))) {
error("Parsing purge txn");
return SLURM_ERROR;
}
env_array_append_fmt(&env, "SLURM_ARCHIVE_TXN", "%u",
SLURMDB_PURGE_ARCHIVE_SET(
arch_cond->purge_txn));
env_array_append_fmt(&env, "SLURM_ARCHIVE_LAST_TXN", "%ld",
(long)curr_end);
}
if (arch_cond->purge_usage != NO_VAL) {
if (!(curr_end = archive_setup_end_time(
last_submit, arch_cond->purge_usage))) {
error("Parsing purge usage");
return SLURM_ERROR;
}
env_array_append_fmt(&env, "SLURM_ARCHIVE_USAGE", "%u",
SLURMDB_PURGE_ARCHIVE_SET(
arch_cond->purge_usage));
env_array_append_fmt(&env, "SLURM_ARCHIVE_LAST_USAGE", "%ld",
(long)curr_end);
}
#ifdef _PATH_STDPATH
env_array_append (&env, "PATH", _PATH_STDPATH);
#else
env_array_append (&env, "PATH", "/bin:/usr/bin");
#endif
execve(arch_cond->archive_script, args, env);
env_array_free(env);
return SLURM_SUCCESS;
}
/*
* Current process is running as the user when this is called.
*/
extern void exec_task(stepd_step_rec_t *job, int local_proc_id)
{
uint32_t *gtids; /* pointer to array of ranks */
int fd, j;
stepd_step_task_info_t *task = job->task[local_proc_id];
char **tmp_env;
int saved_errno;
uint32_t node_offset = 0, task_offset = 0;
if (job->node_offset != NO_VAL)
node_offset = job->node_offset;
if (job->pack_task_offset != NO_VAL)
task_offset = job->pack_task_offset;
gtids = xmalloc(job->node_tasks * sizeof(uint32_t));
for (j = 0; j < job->node_tasks; j++)
gtids[j] = job->task[j]->gtid + task_offset;
job->envtp->sgtids = _uint32_array_to_str(job->node_tasks, gtids);
xfree(gtids);
if (job->pack_jobid != NO_VAL)
job->envtp->jobid = job->pack_jobid;
else
job->envtp->jobid = job->jobid;
job->envtp->stepid = job->stepid;
job->envtp->nodeid = job->nodeid + node_offset;
job->envtp->cpus_on_node = job->cpus;
job->envtp->procid = task->gtid + task_offset;
job->envtp->localid = task->id;
job->envtp->task_pid = getpid();
job->envtp->distribution = job->task_dist;
job->envtp->cpu_bind = xstrdup(job->cpu_bind);
job->envtp->cpu_bind_type = job->cpu_bind_type;
job->envtp->cpu_freq_min = job->cpu_freq_min;
job->envtp->cpu_freq_max = job->cpu_freq_max;
job->envtp->cpu_freq_gov = job->cpu_freq_gov;
job->envtp->mem_bind = xstrdup(job->mem_bind);
job->envtp->mem_bind_type = job->mem_bind_type;
job->envtp->distribution = -1;
job->envtp->ckpt_dir = xstrdup(job->ckpt_dir);
job->envtp->batch_flag = job->batch;
job->envtp->uid = job->uid;
job->envtp->user_name = xstrdup(job->user_name);
/*
* Modify copy of job's environment. Do not alter in place or
* concurrent searches of the environment can generate invalid memory
* references.
*/
job->envtp->env = env_array_copy((const char **) job->env);
setup_env(job->envtp, false);
setenvf(&job->envtp->env, "SLURM_JOB_GID", "%d", job->gid);
setenvf(&job->envtp->env, "SLURMD_NODENAME", "%s", conf->node_name);
if (job->tres_bind) {
setenvf(&job->envtp->env, "SLURMD_TRES_BIND", "%s",
job->tres_bind);
}
if (job->tres_freq) {
setenvf(&job->envtp->env, "SLURMD_TRES_FREQ", "%s",
job->tres_freq);
}
tmp_env = job->env;
job->env = job->envtp->env;
env_array_free(tmp_env);
job->envtp->env = NULL;
xfree(job->envtp->task_count);
if (task->argv[0] && *task->argv[0] != '/') {
/*
* Normally the client (srun) expands the command name
* to a fully qualified path, but in --multi-prog mode it
* is left up to the server to search the PATH for the
* executable.
*/
task->argv[0] = _build_path(task->argv[0], job->env, NULL);
}
if (!job->batch && (job->stepid != SLURM_EXTERN_CONT)) {
if (switch_g_job_attach(job->switch_job, &job->env,
job->nodeid, (uint32_t) local_proc_id,
job->nnodes, job->ntasks,
task->gtid) < 0) {
error("Unable to attach to interconnect: %m");
log_fini();
exit(1);
}
if (_setup_mpi(job, local_proc_id) != SLURM_SUCCESS) {
error("Unable to configure MPI plugin: %m");
log_fini();
exit(1);
}
}
/* task-specific pre-launch activities */
/* task plugin hook */
if (task_g_pre_launch(job)) {
error("Failed to invoke task plugins: task_p_pre_launch error");
exit(1);
}
if (!job->batch &&
(job->accel_bind_type || job->tres_bind || job->tres_freq)) {
/*
* Modify copy of job's environment. Do not alter in place or
* concurrent searches of the environment can generate invalid
* memory references.
*
* Also sets GRES frequency as needed.
*/
job->envtp->env = env_array_copy((const char **) job->env);
gres_plugin_step_set_env(&job->envtp->env, job->step_gres_list,
job->accel_bind_type, job->tres_bind,
job->tres_freq, local_proc_id);
tmp_env = job->env;
job->env = job->envtp->env;
env_array_free(tmp_env);
}
if (spank_user_task(job, local_proc_id) < 0) {
error("Failed to invoke spank plugin stack");
exit(1);
}
if (conf->task_prolog) {
char *my_prolog;
slurm_mutex_lock(&conf->config_mutex);
my_prolog = xstrdup(conf->task_prolog);
slurm_mutex_unlock(&conf->config_mutex);
_run_script_and_set_env("slurm task_prolog",
my_prolog, job);
xfree(my_prolog);
}
if (job->task_prolog) {
_run_script_and_set_env("user task_prolog",
job->task_prolog, job);
}
/*
* Set TMPDIR after running prolog scripts, since TMPDIR
* might be set or changed in one of the prolog scripts.
*/
if (local_proc_id == 0)
_make_tmpdir(job);
if (!job->batch)
pdebug_stop_current(job);
if (job->env == NULL) {
debug("job->env is NULL");
job->env = (char **)xmalloc(sizeof(char *));
job->env[0] = (char *)NULL;
}
if (job->restart_dir) {
info("restart from %s", job->restart_dir);
/* no return on success */
checkpoint_restart_task(job, job->restart_dir, task->gtid);
error("Restart task failed: %m");
exit(errno);
}
if (task->argv[0] == NULL) {
error("No executable program specified for this task");
exit(2);
}
/* Do this last so you don't worry too much about the users
limits including the slurmstepd in with it.
*/
if (set_user_limits(job) < 0) {
debug("Unable to set user limits");
log_fini();
exit(5);
}
execve(task->argv[0], task->argv, job->env);
saved_errno = errno;
/*
* print error message and clean up if execve() returns:
*/
if ((errno == ENOENT) &&
((fd = open(task->argv[0], O_RDONLY)) >= 0)) {
char buf[256], *eol;
int sz;
sz = read(fd, buf, sizeof(buf));
if ((sz >= 3) && (xstrncmp(buf, "#!", 2) == 0)) {
buf[sizeof(buf)-1] = '\0';
eol = strchr(buf, '\n');
if (eol)
eol[0] = '\0';
slurm_seterrno(saved_errno);
error("execve(): bad interpreter(%s): %m", buf+2);
exit(errno);
}
}
slurm_seterrno(saved_errno);
error("execve(): %s: %m", task->argv[0]);
exit(errno);
}
int main(int argc, char *argv[])
{
log_options_t logopt = LOG_OPTS_STDERR_ONLY;
job_desc_msg_t desc;
resource_allocation_response_msg_t *alloc;
time_t before, after;
allocation_msg_thread_t *msg_thr;
char **env = NULL;
int status = 0;
int retries = 0;
pid_t pid = getpid();
pid_t tpgid = 0;
pid_t rc_pid = 0;
int i, rc = 0;
static char *msg = "Slurm job queue full, sleeping and retrying.";
slurm_allocation_callbacks_t callbacks;
log_init(xbasename(argv[0]), logopt, 0, NULL);
_set_exit_code();
if (spank_init_allocator() < 0) {
error("Failed to initialize plugin stack");
exit(error_exit);
}
/* Be sure to call spank_fini when salloc exits
*/
if (atexit((void (*) (void)) spank_fini) < 0)
error("Failed to register atexit handler for plugins: %m");
if (initialize_and_process_args(argc, argv) < 0) {
error("salloc parameter parsing");
exit(error_exit);
}
/* reinit log with new verbosity (if changed by command line) */
if (opt.verbose || opt.quiet) {
logopt.stderr_level += opt.verbose;
logopt.stderr_level -= opt.quiet;
logopt.prefix_level = 1;
log_alter(logopt, 0, NULL);
}
if (spank_init_post_opt() < 0) {
error("Plugin stack post-option processing failed");
exit(error_exit);
}
_set_spank_env();
_set_submit_dir_env();
if (opt.cwd && chdir(opt.cwd)) {
error("chdir(%s): %m", opt.cwd);
exit(error_exit);
}
if (opt.get_user_env_time >= 0) {
char *user = uid_to_string(opt.uid);
if (strcmp(user, "nobody") == 0) {
error("Invalid user id %u: %m", (uint32_t)opt.uid);
exit(error_exit);
}
env = env_array_user_default(user,
opt.get_user_env_time,
opt.get_user_env_mode);
xfree(user);
if (env == NULL)
exit(error_exit); /* error already logged */
_set_rlimits(env);
}
/*
* Job control for interactive salloc sessions: only if ...
*
* a) input is from a terminal (stdin has valid termios attributes),
* b) controlling terminal exists (non-negative tpgid),
* c) salloc is not run in allocation-only (--no-shell) mode,
* d) salloc runs in its own process group (true in interactive
* shells that support job control),
* e) salloc has been configured at compile-time to support background
* execution and is not currently in the background process group.
*/
if (tcgetattr(STDIN_FILENO, &saved_tty_attributes) < 0) {
/*
* Test existence of controlling terminal (tpgid > 0)
* after first making sure stdin is not redirected.
*/
} else if ((tpgid = tcgetpgrp(STDIN_FILENO)) < 0) {
if (!opt.no_shell) {
error("no controlling terminal: please set --no-shell");
exit(error_exit);
}
} else if ((!opt.no_shell) && (pid == getpgrp())) {
if (tpgid == pid)
is_interactive = true;
#ifdef SALLOC_RUN_FOREGROUND
while (tcgetpgrp(STDIN_FILENO) != pid) {
if (!is_interactive) {
error("Waiting for program to be placed in "
"the foreground");
is_interactive = true;
}
killpg(pid, SIGTTIN);
}
#endif
}
/*
* Reset saved tty attributes at exit, in case a child
* process died before properly resetting terminal.
*/
if (is_interactive)
atexit (_reset_input_mode);
/*
* Request a job allocation
*/
slurm_init_job_desc_msg(&desc);
if (_fill_job_desc_from_opts(&desc) == -1) {
exit(error_exit);
}
if (opt.gid != (gid_t) -1) {
if (setgid(opt.gid) < 0) {
error("setgid: %m");
exit(error_exit);
}
}
callbacks.ping = _ping_handler;
callbacks.timeout = _timeout_handler;
callbacks.job_complete = _job_complete_handler;
callbacks.user_msg = _user_msg_handler;
callbacks.node_fail = _node_fail_handler;
/* create message thread to handle pings and such from slurmctld */
msg_thr = slurm_allocation_msg_thr_create(&desc.other_port,
&callbacks);
/* NOTE: Do not process signals in separate pthread. The signal will
* cause slurm_allocate_resources_blocking() to exit immediately. */
for (i = 0; sig_array[i]; i++)
xsignal(sig_array[i], _signal_while_allocating);
before = time(NULL);
while ((alloc = slurm_allocate_resources_blocking(&desc, opt.immediate,
_pending_callback)) == NULL) {
if ((errno != ESLURM_ERROR_ON_DESC_TO_RECORD_COPY) ||
(retries >= MAX_RETRIES))
break;
if (retries == 0)
error("%s", msg);
else
debug("%s", msg);
sleep (++retries);
}
/* become the user after the allocation has been requested. */
if (opt.uid != (uid_t) -1) {
if (setuid(opt.uid) < 0) {
error("setuid: %m");
exit(error_exit);
}
}
if (alloc == NULL) {
if (allocation_interrupted) {
/* cancelled by signal */
info("Job aborted due to signal");
} else if (errno == EINTR) {
error("Interrupted by signal."
" Allocation request rescinded.");
} else if (opt.immediate &&
((errno == ETIMEDOUT) ||
(errno == ESLURM_NOT_TOP_PRIORITY) ||
(errno == ESLURM_NODES_BUSY))) {
error("Unable to allocate resources: %m");
error_exit = immediate_exit;
} else {
error("Failed to allocate resources: %m");
}
slurm_allocation_msg_thr_destroy(msg_thr);
exit(error_exit);
} else if (!allocation_interrupted) {
/*
* Allocation granted!
*/
info("Granted job allocation %u", alloc->job_id);
pending_job_id = alloc->job_id;
#ifdef HAVE_BG
if (!_wait_bluegene_block_ready(alloc)) {
if(!allocation_interrupted)
error("Something is wrong with the "
"boot of the block.");
goto relinquish;
}
#else
if (!_wait_nodes_ready(alloc)) {
if(!allocation_interrupted)
error("Something is wrong with the "
"boot of the nodes.");
goto relinquish;
}
#endif
}
after = time(NULL);
if (opt.bell == BELL_ALWAYS
|| (opt.bell == BELL_AFTER_DELAY
&& ((after - before) > DEFAULT_BELL_DELAY))) {
_ring_terminal_bell();
}
if (opt.no_shell)
exit(0);
if (allocation_interrupted) {
/* salloc process received a signal after
* slurm_allocate_resources_blocking returned with the
* allocation, but before the new signal handlers were
* registered.
*/
goto relinquish;
}
/*
* Run the user's command.
*/
if (env_array_for_job(&env, alloc, &desc) != SLURM_SUCCESS)
goto relinquish;
/* Add default task count for srun, if not already set */
if (opt.ntasks_set) {
env_array_append_fmt(&env, "SLURM_NTASKS", "%d", opt.ntasks);
/* keep around for old scripts */
env_array_append_fmt(&env, "SLURM_NPROCS", "%d", opt.ntasks);
}
if (opt.cpus_per_task > 1) {
env_array_append_fmt(&env, "SLURM_CPUS_PER_TASK", "%d",
opt.cpus_per_task);
}
if (opt.overcommit) {
env_array_append_fmt(&env, "SLURM_OVERCOMMIT", "%d",
opt.overcommit);
}
if (opt.acctg_freq >= 0) {
env_array_append_fmt(&env, "SLURM_ACCTG_FREQ", "%d",
opt.acctg_freq);
}
if (opt.network)
env_array_append_fmt(&env, "SLURM_NETWORK", "%s", opt.network);
env_array_set_environment(env);
env_array_free(env);
pthread_mutex_lock(&allocation_state_lock);
if (allocation_state == REVOKED) {
error("Allocation was revoked for job %u before command could "
"be run", alloc->job_id);
pthread_mutex_unlock(&allocation_state_lock);
if (slurm_complete_job(alloc->job_id, status) != 0) {
error("Unable to clean up allocation for job %u: %m",
alloc->job_id);
}
return 1;
}
allocation_state = GRANTED;
pthread_mutex_unlock(&allocation_state_lock);
/* Ensure that salloc has initial terminal foreground control. */
if (is_interactive) {
/*
* Ignore remaining job-control signals (other than those in
* sig_array, which at this state act like SIG_IGN).
*/
xsignal(SIGTSTP, SIG_IGN);
xsignal(SIGTTIN, SIG_IGN);
xsignal(SIGTTOU, SIG_IGN);
pid = getpid();
setpgid(pid, pid);
tcsetpgrp(STDIN_FILENO, pid);
}
command_pid = _fork_command(command_argv);
/*
* Wait for command to exit, OR for waitpid to be interrupted by a
* signal. Either way, we are going to release the allocation next.
*/
if (command_pid > 0) {
setpgid(command_pid, command_pid);
if (is_interactive)
tcsetpgrp(STDIN_FILENO, command_pid);
/* NOTE: Do not process signals in separate pthread.
* The signal will cause waitpid() to exit immediately. */
xsignal(SIGHUP, _exit_on_signal);
/* Use WUNTRACED to treat stopped children like terminated ones */
do {
rc_pid = waitpid(command_pid, &status, WUNTRACED);
} while ((rc_pid == -1) && (!exit_flag));
if ((rc_pid == -1) && (errno != EINTR))
error("waitpid for %s failed: %m", command_argv[0]);
}
if (is_interactive)
tcsetpgrp(STDIN_FILENO, pid);
/*
* Relinquish the job allocation (if not already revoked).
*/
relinquish:
pthread_mutex_lock(&allocation_state_lock);
if (allocation_state != REVOKED) {
pthread_mutex_unlock(&allocation_state_lock);
info("Relinquishing job allocation %d", alloc->job_id);
if ((slurm_complete_job(alloc->job_id, status) != 0) &&
(slurm_get_errno() != ESLURM_ALREADY_DONE))
error("Unable to clean up job allocation %d: %m",
alloc->job_id);
pthread_mutex_lock(&allocation_state_lock);
allocation_state = REVOKED;
}
pthread_mutex_unlock(&allocation_state_lock);
slurm_free_resource_allocation_response_msg(alloc);
slurm_allocation_msg_thr_destroy(msg_thr);
/*
* Figure out what return code we should use. If the user's command
* exited normally, return the user's return code.
*/
rc = 1;
if (rc_pid != -1) {
if (WIFEXITED(status)) {
rc = WEXITSTATUS(status);
} else if (WIFSIGNALED(status)) {
verbose("Command \"%s\" was terminated by signal %d",
command_argv[0], WTERMSIG(status));
/* if we get these signals we return a normal
* exit since this was most likely sent from the
* user */
switch(WTERMSIG(status)) {
case SIGHUP:
case SIGINT:
case SIGQUIT:
case SIGKILL:
rc = 0;
break;
default:
break;
}
}
}
return rc;
}
