diff --git a/src/salloc/opt.c b/src/salloc/opt.c
index b6a2e4e119419bb45221857fcaf3abb1d042dd24..a5a3058ab15476d78f56c2b17c1d2ca4d70e1bc3 100644
--- a/src/salloc/opt.c
+++ b/src/salloc/opt.c
@@ -160,6 +160,7 @@
/*---- global variables, defined in opt.h ----*/
opt_t opt;
+int error_exit = 1;
/*---- forward declarations of static functions ----*/
diff --git a/src/salloc/opt.h b/src/salloc/opt.h
index 16786e42e1fdf73ba2774fed49bcd622118eb31a..73217c9354cf35a0d1cd090b513664021f650bea 100644
--- a/src/salloc/opt.h
+++ b/src/salloc/opt.h
@@ -164,6 +164,7 @@ typedef struct salloc_options {
} opt_t;
extern opt_t opt;
+extern int error_exit;
/* process options:
* 1. set defaults
diff --git a/src/salloc/salloc.c b/src/salloc/salloc.c
index 09f6ec2f44cf372869f581ee9bacdb6a23680310..2a0a8af2fe11fa991e0abdb5b2b41a8e7acebb33 100644
--- a/src/salloc/salloc.c
+++ b/src/salloc/salloc.c
@@ -87,7 +87,6 @@ extern int spank_unset_job_env(const char *name);
char **command_argv;
int command_argc;
pid_t command_pid = -1;
-int error_exit = 1;
enum possible_allocation_states allocation_state = NOT_GRANTED;
pthread_mutex_t allocation_state_lock = PTHREAD_MUTEX_INITIALIZER;
diff --git a/src/salloc/salloc.h b/src/salloc/salloc.h
index 1e80a2c71fd6951616069990f9e2ed5be9e19849..fdd262aaf7b49575670d744e14e722759af9adcf 100644
--- a/src/salloc/salloc.h
+++ b/src/salloc/salloc.h
@@ -47,7 +47,6 @@
extern char **command_argv;
extern int command_argc;
extern pid_t command_pid;
-extern int error_exit;
enum possible_allocation_states {NOT_GRANTED, GRANTED, REVOKED};
extern enum possible_allocation_states allocation_state;
diff --git a/src/srun/allocate.c b/src/srun/allocate.c
index 18fbcaf0acebb4b5058eaf69b7c30327387238d2..784cbdb4f9c2dcc5085dc9199c26221584df9c11 100644
--- a/src/srun/allocate.c
+++ b/src/srun/allocate.c
@@ -437,7 +437,7 @@ relinquish:
slurm_free_resource_allocation_response_msg(resp);
if(!destroy_job)
slurm_complete_job(resp->job_id, 1);
- exit(1);
+ exit(error_exit);
return NULL;
}
@@ -475,7 +475,7 @@ existing_allocation(void)
old_job_id);
info ("Check SLURM_JOB_ID environment variable "
"for expired or invalid job.");
- exit(1);
+ exit(error_exit);
}
return resp;
@@ -495,10 +495,14 @@ slurmctld_msg_init(void)
slurmctld_fd = -1;
slurmctld_comm_addr.port = 0;
- if ((slurmctld_fd = slurm_init_msg_engine_port(0)) < 0)
- fatal("slurm_init_msg_engine_port error %m");
- if (slurm_get_stream_addr(slurmctld_fd, &slurm_address) < 0)
- fatal("slurm_get_stream_addr error %m");
+ if ((slurmctld_fd = slurm_init_msg_engine_port(0)) < 0) {
+ error("slurm_init_msg_engine_port error %m");
+ exit(error_exit);
+ }
+ if (slurm_get_stream_addr(slurmctld_fd, &slurm_address) < 0) {
+ error("slurm_get_stream_addr error %m");
+ exit(error_exit);
+ }
fd_set_nonblocking(slurmctld_fd);
/* hostname is not set, so slurm_get_addr fails
slurm_get_addr(&slurm_address, &port, hostname, sizeof(hostname)); */
diff --git a/src/srun/multi_prog.c b/src/srun/multi_prog.c
index 4c3e6fbcac54362883d1c1de2575446117016b2f..693d5d580006f9eedc626a486ba31f344d2371a2 100644
--- a/src/srun/multi_prog.c
+++ b/src/srun/multi_prog.c
@@ -61,6 +61,7 @@
#include "src/common/xmalloc.h"
#include "src/common/xstring.h"
#include "src/srun/debugger.h"
+#include "src/srun/opt.h"
/* Given a program name, translate it to a fully qualified pathname
* as needed based upon the PATH environment variable */
@@ -226,12 +227,14 @@ mpir_init(int num_tasks)
{
MPIR_proctable_size = num_tasks;
MPIR_proctable = xmalloc(sizeof(MPIR_PROCDESC) * num_tasks);
- if (MPIR_proctable == NULL)
- fatal("Unable to initialize MPIR_proctable: %m");
+ if (MPIR_proctable == NULL) {
+ error("Unable to initialize MPIR_proctable: %m");
+ exit(error_exit);
+ }
}
extern void
-mpir_cleanup()
+mpir_cleanup(void)
{
int i;
@@ -249,9 +252,11 @@ mpir_set_executable_names(const char *executable_name)
for (i = 0; i < MPIR_proctable_size; i++) {
MPIR_proctable[i].executable_name = xstrdup(executable_name);
- if (MPIR_proctable[i].executable_name == NULL)
- fatal("Unable to set MPI_proctable executable_name:"
+ if (MPIR_proctable[i].executable_name == NULL) {
+ error("Unable to set MPI_proctable executable_name:"
" %m");
+ exit(error_exit);
+ }
}
}
diff --git a/src/srun/opt.c b/src/srun/opt.c
index 09854dc6154a14886cd2e440c9e89c6c8ff67a86..c9f42b61c29333585b6b8a98e006ef2a098ff45a 100644
--- a/src/srun/opt.c
+++ b/src/srun/opt.c
@@ -185,6 +185,7 @@
/*---- global variables, defined in opt.h ----*/
int _verbose;
opt_t opt;
+int error_exit = 1;
/*---- forward declarations of static functions ----*/
@@ -233,7 +234,7 @@ int initialize_and_process_args(int argc, char *argv[])
_opt_args(argc, argv);
if (!_opt_verify())
- exit(1);
+ exit(error_exit);
if (_verbose > 3)
_opt_list();
@@ -305,8 +306,10 @@ static void _opt_default()
opt.gid = getgid();
- if ((getcwd(buf, MAXPATHLEN)) == NULL)
- fatal("getcwd failed: %m");
+ if ((getcwd(buf, MAXPATHLEN)) == NULL) {
+ error("getcwd failed: %m");
+ exit(error_exit);
+ }
opt.cwd = xstrdup(buf);
opt.cwd_set = false;
@@ -587,13 +590,13 @@ _process_env_var(env_vars_t *e, const char *val)
case OPT_CPU_BIND:
if (slurm_verify_cpu_bind(val, &opt.cpu_bind,
&opt.cpu_bind_type))
- exit(1);
+ exit(error_exit);
break;
case OPT_MEM_BIND:
if (slurm_verify_mem_bind(val, &opt.mem_bind,
&opt.mem_bind_type))
- exit(1);
+ exit(error_exit);
break;
case OPT_NODES:
@@ -659,16 +662,18 @@ _process_env_var(env_vars_t *e, const char *val)
case OPT_MPI:
if (mpi_hook_client_init((char *)val) == SLURM_ERROR) {
- fatal("\"%s=%s\" -- invalid MPI type, "
+ error("\"%s=%s\" -- invalid MPI type, "
"--mpi=list for acceptable types.",
e->var, val);
+ exit(error_exit);
}
break;
case OPT_SIGNAL:
if (get_signal_opts(optarg, &opt.warn_signal,
&opt.warn_time)) {
- fatal("Invalid signal specification: %s", optarg);
+ error("Invalid signal specification: %s", optarg);
+ exit(error_exit);
}
break;
@@ -693,7 +698,7 @@ _get_int(const char *arg, const char *what, bool positive)
if ((*p != '\0') || (result < 0L)
|| (positive && (result <= 0L))) {
error ("Invalid numeric value \"%s\" for %s.", arg, what);
- exit(1);
+ exit(error_exit);
} else if (result > INT_MAX) {
error ("Numeric argument (%ld) to big for %s.", result, what);
} else if (result < INT_MIN) {
@@ -818,12 +823,13 @@ static void set_options(const int argc, char **argv)
};
char *opt_string = "+aA:bB:c:C:d:D:e:Eg:Hi:IjJ:kKlL:m:n:N:"
"o:Op:P:qQr:Rst:T:uU:vVw:W:x:XZ";
+ char *tmp_str;
struct option *optz = spank_option_table_create (long_options);
if (!optz) {
- error ("Unable to create option table");
- exit (1);
+ error("Unable to create option table");
+ exit(error_exit);
}
if(opt.progname == NULL)
@@ -838,7 +844,7 @@ static void set_options(const int argc, char **argv)
case (int)'?':
fprintf(stderr,
"Try \"srun --help\" for more information\n");
- exit(1);
+ exit(error_exit);
break;
case (int)'A':
case (int)'U': /* backwards compatibility */
@@ -848,11 +854,11 @@ static void set_options(const int argc, char **argv)
case (int)'a':
error("Please use the \"sattach\" command instead of "
"\"srun -a/--attach\".");
- exit(1);
+ exit(error_exit);
case (int)'b':
error("Please use the \"sbatch\" command instead of "
"\"srun -b/--batch\".");
- exit(1);
+ exit(error_exit);
case (int)'B':
opt.extra_set = verify_socket_core_thread_count(
optarg,
@@ -868,7 +874,7 @@ static void set_options(const int argc, char **argv)
if (opt.extra_set == false) {
error("invalid resource allocation -B `%s'",
optarg);
- exit(1);
+ exit(error_exit);
}
break;
case (int)'c':
@@ -891,8 +897,9 @@ static void set_options(const int argc, char **argv)
break;
case (int)'e':
if (opt.pty) {
- fatal("--error incompatable with --pty "
+ error("--error incompatable with --pty "
"option");
+ exit(error_exit);
}
xfree(opt.efname);
if (strncasecmp(optarg, "none", (size_t) 4) == 0)
@@ -905,15 +912,16 @@ static void set_options(const int argc, char **argv)
break;
case (int)'g':
if (verify_geometry(optarg, opt.geometry))
- exit(1);
+ exit(error_exit);
break;
case (int)'H':
opt.hold = true;
break;
case (int)'i':
if (opt.pty) {
- fatal("--input incompatable with "
+ error("--input incompatable with "
"--pty option");
+ exit(error_exit);
}
xfree(opt.ifname);
if (strncasecmp(optarg, "none", (size_t) 4) == 0)
@@ -954,7 +962,7 @@ static void set_options(const int argc, char **argv)
if (opt.distribution == SLURM_DIST_UNKNOWN) {
error("distribution type `%s' "
"is not recognized", optarg);
- exit(1);
+ exit(error_exit);
}
break;
case (int)'n':
@@ -972,12 +980,15 @@ static void set_options(const int argc, char **argv)
if (opt.nodes_set == false) {
error("invalid resource allocation -N `%s'",
optarg);
- exit(1);
+ exit(error_exit);
}
break;
case (int)'o':
- if (opt.pty)
- fatal("--output incompatable with --pty option");
+ if (opt.pty) {
+ error("--output incompatable with --pty "
+ "option");
+ exit(error_exit);
+ }
xfree(opt.ofname);
if (strncasecmp(optarg, "none", (size_t) 4) == 0)
opt.ofname = xstrdup("/dev/null");
@@ -1041,7 +1052,7 @@ static void set_options(const int argc, char **argv)
xfree(opt.exc_nodes);
opt.exc_nodes = xstrdup(optarg);
if (!_valid_node_list(&opt.exc_nodes))
- exit(1);
+ exit(error_exit);
break;
case (int)'X':
opt.disable_status = true;
@@ -1062,12 +1073,12 @@ static void set_options(const int argc, char **argv)
case LONG_OPT_CPU_BIND:
if (slurm_verify_cpu_bind(optarg, &opt.cpu_bind,
&opt.cpu_bind_type))
- exit(1);
+ exit(error_exit);
break;
case LONG_OPT_MEM_BIND:
if (slurm_verify_mem_bind(optarg, &opt.mem_bind,
&opt.mem_bind_type))
- exit(1);
+ exit(error_exit);
break;
case LONG_OPT_CORE:
opt.core_type = core_format_type (optarg);
@@ -1094,7 +1105,7 @@ static void set_options(const int argc, char **argv)
if (opt.job_min_memory < 0) {
error("invalid memory constraint %s",
optarg);
- exit(1);
+ exit(error_exit);
}
break;
case LONG_OPT_MEM_PER_CPU:
@@ -1102,15 +1113,16 @@ static void set_options(const int argc, char **argv)
if (opt.mem_per_cpu < 0) {
error("invalid memory constraint %s",
optarg);
- exit(1);
+ exit(error_exit);
}
break;
case LONG_OPT_MPI:
if (mpi_hook_client_init((char *)optarg)
== SLURM_ERROR) {
- fatal("\"--mpi=%s\" -- long invalid MPI type, "
+ error("\"--mpi=%s\" -- long invalid MPI type, "
"--mpi=list for acceptable types.",
optarg);
+ exit(error_exit);
}
break;
case LONG_OPT_RESV_PORTS:
@@ -1123,7 +1135,7 @@ static void set_options(const int argc, char **argv)
opt.job_min_tmp_disk = str_to_bytes(optarg);
if (opt.job_min_tmp_disk < 0) {
error("invalid tmp value %s", optarg);
- exit(1);
+ exit(error_exit);
}
break;
case LONG_OPT_JOBID:
@@ -1143,16 +1155,24 @@ static void set_options(const int argc, char **argv)
_get_int(optarg, "max-exit-timeout", true);
break;
case LONG_OPT_UID:
- if (opt.euid != (uid_t) -1)
- fatal ("duplicate --uid option");
- if (uid_from_string (optarg, &opt.euid) < 0)
- fatal ("--uid=\"%s\" invalid", optarg);
+ if (opt.euid != (uid_t) -1) {
+ error("duplicate --uid option");
+ exit(error_exit);
+ }
+ if (uid_from_string (optarg, &opt.euid) < 0) {
+ error("--uid=\"%s\" invalid", optarg);
+ exit(error_exit);
+ }
break;
case LONG_OPT_GID:
- if (opt.egid != (gid_t) -1)
- fatal ("duplicate --gid option");
- if (gid_from_string (optarg, &opt.egid) < 0)
- fatal ("--gid=\"%s\" invalid", optarg);
+ if (opt.egid != (gid_t) -1) {
+ error("duplicate --gid option");
+ exit(error_exit);
+ }
+ if (gid_from_string (optarg, &opt.egid) < 0) {
+ error("--gid=\"%s\" invalid", optarg);
+ exit(error_exit);
+ }
break;
case LONG_OPT_DEBUG_TS:
opt.debugger_test = true;
@@ -1200,14 +1220,17 @@ static void set_options(const int argc, char **argv)
case LONG_OPT_BEGIN:
opt.begin = parse_time(optarg, 0);
if (opt.begin == 0) {
- fatal("Invalid time specification %s",
+ error("Invalid time specification %s",
optarg);
+ exit(error_exit);
}
break;
case LONG_OPT_MAIL_TYPE:
opt.mail_type |= parse_mail_type(optarg);
- if (opt.mail_type == 0)
- fatal("--mail-type=%s invalid", optarg);
+ if (opt.mail_type == 0) {
+ error("--mail-type=%s invalid", optarg);
+ exit(error_exit);
+ }
break;
case LONG_OPT_MAIL_USER:
xfree(opt.mail_user);
@@ -1229,7 +1252,7 @@ static void set_options(const int argc, char **argv)
if (abs(opt.nice) > NICE_OFFSET) {
error("Invalid nice value, must be between "
"-%d and %d", NICE_OFFSET, NICE_OFFSET);
- exit(1);
+ exit(error_exit);
}
if (opt.nice < 0) {
uid_t my_uid = getuid();
@@ -1279,7 +1302,7 @@ static void set_options(const int argc, char **argv)
&opt.min_threads_per_core,
&opt.max_threads_per_core,
&opt.cpu_bind_type)) {
- exit(1);
+ exit(error_exit);
}
break;
case LONG_OPT_NTASKSPERNODE:
@@ -1322,13 +1345,21 @@ static void set_options(const int argc, char **argv)
opt.pty = true;
opt.unbuffered = true; /* implicit */
if (opt.ifname)
- fatal("--input incompatable with --pty option");
- if (opt.ofname)
- fatal("--output incompatable with --pty option");
- if (opt.efname)
- fatal("--error incompatable with --pty option");
+ tmp_str = "--input";
+ else if (opt.ofname)
+ tmp_str = "--output";
+ else if (opt.efname)
+ tmp_str = "--error";
+ else
+ tmp_str = NULL;
+ if (tmp_str) {
+ error("%s incompatable with --pty option",
+ tmp_str);
+ exit(error_exit);
+ }
#else
- error("--pty not currently supported on this system type");
+ error("--pty not currently supported on this system "
+ "type");
#endif
break;
case LONG_OPT_CHECKPOINT:
@@ -1368,13 +1399,14 @@ static void set_options(const int argc, char **argv)
case LONG_OPT_SIGNAL:
if (get_signal_opts(optarg, &opt.warn_signal,
&opt.warn_time)) {
- fatal("Invalid signal specification: %s",
+ error("Invalid signal specification: %s",
optarg);
+ exit(error_exit);
}
break;
default:
if (spank_process_option (opt_char, optarg) < 0) {
- exit (1);
+ exit(error_exit);
}
}
}
@@ -1394,17 +1426,17 @@ static void _load_multi(int *argc, char **argv)
if ((config_fd = open(argv[0], O_RDONLY)) == -1) {
error("Could not open multi_prog config file %s",
argv[0]);
- exit(1);
+ exit(error_exit);
}
if (fstat(config_fd, &stat_buf) == -1) {
error("Could not stat multi_prog config file %s",
argv[0]);
- exit(1);
+ exit(error_exit);
}
if (stat_buf.st_size > 60000) {
error("Multi_prog config file %s is too large",
argv[0]);
- exit(1);
+ exit(error_exit);
}
data_buf = xmalloc(stat_buf.st_size + 1);
while ((i = read(config_fd, &data_buf[data_read], stat_buf.st_size
@@ -1412,7 +1444,7 @@ static void _load_multi(int *argc, char **argv)
if (i < 0) {
error("Error reading multi_prog config file %s",
argv[0]);
- exit(1);
+ exit(error_exit);
} else
data_read += i;
}
@@ -1461,7 +1493,7 @@ static void _opt_args(int argc, char **argv)
#endif
error("Too few processes for the requested "
"{plane,node} distribution");
- exit(1);
+ exit(error_exit);
}
}
}
@@ -1500,7 +1532,7 @@ static void _opt_args(int argc, char **argv)
if (opt.multi_prog) {
if (opt.argc < 1) {
error("configuration file not specified");
- exit(1);
+ exit(error_exit);
}
_load_multi(&opt.argc, opt.argv);
}
@@ -1514,7 +1546,7 @@ static void _opt_args(int argc, char **argv)
}
if (opt.multi_prog && verify_multi_name(opt.argv[0], opt.nprocs))
- exit(1);
+ exit(error_exit);
}
/*
@@ -1586,7 +1618,7 @@ static bool _opt_verify(void)
if (!_valid_node_list(&opt.nodelist)) {
error("Failure getting NodeNames from "
"hostfile");
- exit(1);
+ exit(error_exit);
} else {
debug("loaded nodes (%s) from hostfile",
opt.nodelist);
@@ -1594,7 +1626,7 @@ static bool _opt_verify(void)
}
} else {
if (!_valid_node_list(&opt.nodelist))
- exit(1);
+ exit(error_exit);
}
/* set up the proc and node counts based on the arbitrary list
@@ -1729,8 +1761,10 @@ static bool _opt_verify(void)
/* massage the numbers */
if (opt.nodelist) {
hl = hostlist_create(opt.nodelist);
- if (!hl)
- fatal("memory allocation failure");
+ if (!hl) {
+ error("memory allocation failure");
+ exit(error_exit);
+ }
hostlist_uniq(hl);
hl_cnt = hostlist_count(hl);
if (opt.nodes_set)
@@ -1764,8 +1798,10 @@ static bool _opt_verify(void)
/* massage the numbers */
if (opt.nodelist) {
hl = hostlist_create(opt.nodelist);
- if (!hl)
- fatal("memory allocation failure");
+ if (!hl) {
+ error("memory allocation failure");
+ exit(error_exit);
+ }
if(opt.distribution == SLURM_DIST_ARBITRARY
&& !opt.nprocs_set) {
opt.nprocs = hostlist_count(hl);
@@ -1836,7 +1872,7 @@ static bool _opt_verify(void)
if (opt.labelio && opt.unbuffered) {
error("Do not specify both -l (--label) and "
"-u (--unbuffered)");
- exit(1);
+ exit(error_exit);
}
/*
@@ -1849,7 +1885,7 @@ static bool _opt_verify(void)
opt.time_limit = time_str2mins(opt.time_limit_str);
if ((opt.time_limit < 0) && (opt.time_limit != INFINITE)) {
error("Invalid time limit specification");
- exit(1);
+ exit(error_exit);
}
if (opt.time_limit == 0)
opt.time_limit = INFINITE;
@@ -1860,7 +1896,7 @@ static bool _opt_verify(void)
if ((opt.ckpt_interval < 0) &&
(opt.ckpt_interval != INFINITE)) {
error("Invalid checkpoint interval specification");
- exit(1);
+ exit(error_exit);
}
}
@@ -1875,7 +1911,7 @@ static bool _opt_verify(void)
if (slurm_verify_cpu_bind(NULL, &opt.cpu_bind,
&opt.cpu_bind_type))
- exit(1);
+ exit(error_exit);
return verified;
}
diff --git a/src/srun/opt.h b/src/srun/opt.h
index e4827c882cd7e52e8ff65f16ddffcf94cda627be..55fddb45666335567813ed3b99e413068d9c90e6 100644
--- a/src/srun/opt.h
+++ b/src/srun/opt.h
@@ -230,6 +230,8 @@ typedef struct srun_options {
extern opt_t opt;
+extern int error_exit;
+
/* return whether any constraints were specified by the user
* (if new constraints are added above, might want to add them to this
* macro or move this to a function if it gets a little complicated)
diff --git a/src/srun/srun.c b/src/srun/srun.c
index d9211358d318dbb075c0b40018c530d7f4553c28..b0719a46a6d50f6e60643e25da2ab9e4b275af37 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -143,6 +143,7 @@ static void _run_srun_prolog (srun_job_t *job);
static void _run_srun_epilog (srun_job_t *job);
static int _run_srun_script (srun_job_t *job, char *script);
static void _set_cpu_env_var(resource_allocation_response_msg_t *resp);
+static void _set_exit_code(void);
static int _setup_signals();
static void _step_opt_exclusive(void);
static void _set_stdio_fds(srun_job_t *job, slurm_step_io_fds_t *cio_fds);
@@ -196,6 +197,7 @@ int srun(int ac, char **av)
debug_level = _slurm_debug_env_val();
logopt.stderr_level += debug_level;
log_init(xbasename(av[0]), logopt, 0, NULL);
+ _set_exit_code();
/* xsignal(SIGQUIT, _ignore_signal); */
/* xsignal(SIGPIPE, _ignore_signal); */
@@ -205,7 +207,8 @@ int srun(int ac, char **av)
/* Initialize plugin stack, read options from plugins, etc.
*/
if (spank_init(NULL) < 0) {
- fatal("Plug-in initialization failed");
+ error("Plug-in initialization failed");
+ exit(error_exit);
_define_symbols();
}
@@ -223,9 +226,11 @@ int srun(int ac, char **av)
}
record_ppid();
- if (spank_init_post_opt() < 0)
- fatal("Plugin stack post-option processing failed.");
-
+ if (spank_init_post_opt() < 0) {
+ error("Plugin stack post-option processing failed.");
+ exit(error_exit);
+ }
+
/* reinit log with new verbosity (if changed by command line)
*/
if (_verbose || opt.quiet) {
@@ -262,7 +267,7 @@ int srun(int ac, char **av)
info("do not allocate resources");
job = job_create_noalloc();
if (create_job_step(job, false) < 0) {
- exit(1);
+ exit(error_exit);
}
} else if ((resp = existing_allocation())) {
@@ -280,7 +285,7 @@ int srun(int ac, char **av)
" are already allocated.");
if (!job || create_job_step(job, false) < 0)
- exit(1);
+ exit(error_exit);
} else {
/* Combined job allocation and job step launch */
#ifdef HAVE_FRONT_END
@@ -288,12 +293,12 @@ int srun(int ac, char **av)
if ((my_uid != 0)
&& (my_uid != slurm_get_slurm_user_id())) {
error("srun task launch not supported on this system");
- exit(1);
+ exit(error_exit);
}
#endif
if ( !(resp = allocate_nodes()) )
- exit(1);
+ exit(error_exit);
got_alloc = 1;
_print_job_information(resp);
_set_cpu_env_var(resp);
@@ -306,7 +311,7 @@ int srun(int ac, char **av)
}
if (!job || create_job_step(job, true) < 0) {
slurm_complete_job(resp->job_id, 1);
- exit(1);
+ exit(error_exit);
}
slurm_free_resource_allocation_response_msg(resp);
@@ -430,7 +435,7 @@ int srun(int ac, char **av)
if (_call_spank_local_user (job) < 0) {
error("Failure in local plugin stack");
slurm_step_launch_abort(job->step_ctx);
- exit(1);
+ exit(error_exit);
}
update_job_state(job, SRUN_JOB_LAUNCHING);
@@ -467,10 +472,10 @@ int srun(int ac, char **av)
slurm_step_ctx_destroy(job->step_ctx);
if (got_alloc) {
if (create_job_step(job, true) < 0)
- exit(1);
+ exit(error_exit);
} else {
if (create_job_step(job, false) < 0)
- exit(1);
+ exit(error_exit);
}
task_state_destroy(task_state);
goto re_launch;
@@ -659,6 +664,20 @@ static void _set_prio_process_env(void)
debug ("propagating SLURM_PRIO_PROCESS=%d", retval);
}
+static void _set_exit_code(void)
+{
+ int i;
+ char *val = getenv("SLURM_ERROR_EXIT");
+
+ if (val) {
+ i = atoi(val);
+ if (i == 0)
+ error("SLURM_ERROR_EXIT has zero value");
+ else
+ error_exit = i;
+ }
+}
+
static void _set_cpu_env_var(resource_allocation_response_msg_t *resp)
{
char *tmp;
@@ -686,8 +705,10 @@ static int _set_rlimit_env(void)
slurm_rlimits_info_t *rli;
/* Modify limits with any command-line options */
- if (opt.propagate && parse_rlimits( opt.propagate, PROPAGATE_RLIMITS))
- fatal( "--propagate=%s is not valid.", opt.propagate );
+ if (opt.propagate && parse_rlimits( opt.propagate, PROPAGATE_RLIMITS)){
+ error( "--propagate=%s is not valid.", opt.propagate );
+ exit(error_exit);
+ }
for (rli = get_slurm_rlimits_info(); rli->name != NULL; rli++ ) {
@@ -856,12 +877,15 @@ _set_stdio_fds(srun_job_t *job, slurm_step_io_fds_t *cio_fds)
* create stdin file descriptor
*/
if (_is_local_file(job->ifname)) {
- if ((job->ifname->name == NULL) || (job->ifname->taskid != -1)) {
+ if ((job->ifname->name == NULL) ||
+ (job->ifname->taskid != -1)) {
cio_fds->in.fd = STDIN_FILENO;
} else {
cio_fds->in.fd = open(job->ifname->name, O_RDONLY);
- if (cio_fds->in.fd == -1)
- fatal("Could not open stdin file: %m");
+ if (cio_fds->in.fd == -1) {
+ error("Could not open stdin file: %m");
+ exit(error_exit);
+ }
}
if (job->ifname->type == IO_ONE) {
cio_fds->in.taskid = job->ifname->taskid;
@@ -875,13 +899,16 @@ _set_stdio_fds(srun_job_t *job, slurm_step_io_fds_t *cio_fds)
* create stdout file descriptor
*/
if (_is_local_file(job->ofname)) {
- if ((job->ofname->name == NULL) || (job->ofname->taskid != -1)) {
+ if ((job->ofname->name == NULL) ||
+ (job->ofname->taskid != -1)) {
cio_fds->out.fd = STDOUT_FILENO;
} else {
cio_fds->out.fd = open(job->ofname->name,
file_flags, 0644);
- if (cio_fds->out.fd == -1)
- fatal("Could not open stdout file: %m");
+ if (cio_fds->out.fd == -1) {
+ error("Could not open stdout file: %m");
+ exit(error_exit);
+ }
}
if (job->ofname->name != NULL
&& job->efname->name != NULL
@@ -899,13 +926,16 @@ _set_stdio_fds(srun_job_t *job, slurm_step_io_fds_t *cio_fds)
cio_fds->err.fd = cio_fds->out.fd;
cio_fds->err.taskid = cio_fds->out.taskid;
} else if (_is_local_file(job->efname)) {
- if ((job->efname->name == NULL) || (job->efname->taskid != -1)) {
+ if ((job->efname->name == NULL) ||
+ (job->efname->taskid != -1)) {
cio_fds->err.fd = STDERR_FILENO;
} else {
cio_fds->err.fd = open(job->efname->name,
file_flags, 0644);
- if (cio_fds->err.fd == -1)
- fatal("Could not open stderr file: %m");
+ if (cio_fds->err.fd == -1) {
+ error("Could not open stderr file: %m");
+ exit(error_exit);
+ }
}
}
}
@@ -938,12 +968,18 @@ static void _step_opt_exclusive(void)
opt.min_nodes = 1;
opt.max_nodes = 0;
}
- if (!opt.nprocs_set)
- fatal("--ntasks must be set with --exclusive");
- if (opt.relative_set)
- fatal("--relative disabled, incompatible with --exclusive");
- if (opt.exc_nodes)
- fatal("--exclude is incompatible with --exclusive");
+ if (!opt.nprocs_set) {
+ error("--ntasks must be set with --exclusive");
+ exit(error_exit);
+ }
+ if (opt.relative_set) {
+ error("--relative disabled, incompatible with --exclusive");
+ exit(error_exit);
+ }
+ if (opt.exc_nodes) {
+ error("--exclude is incompatible with --exclusive");
+ exit(error_exit);
+ }
}
static void
@@ -1055,8 +1091,10 @@ _task_array_to_string(int ntasks, uint32_t taskids[])
int i;
tasks_bitmap = bit_alloc(job->ntasks);
- if (!tasks_bitmap)
- fatal("bit_alloc: memory allocation failure");
+ if (!tasks_bitmap) {
+ error("bit_alloc: memory allocation failure");
+ exit(error_exit);
+ }
for (i=0; i<ntasks; i++)
bit_set(tasks_bitmap, taskids[i]);
str = xmalloc(2048);