diff --git a/src/common/slurm_acct_gather_profile.c b/src/common/slurm_acct_gather_profile.c
index da3b5e04b4318d25ed1ac32bde64d5e58a3dac5e..f2e835dd0e55ff3b8ff4c784a1f9283c5ee71ec9 100644
--- a/src/common/slurm_acct_gather_profile.c
+++ b/src/common/slurm_acct_gather_profile.c
@@ -95,7 +95,11 @@ static bool init_run = false;
static int _get_int(const char *my_str)
{
char *end = NULL;
- int value = strtol(my_str, &end, 10);
+ int value;
+
+ if (!my_str)
+ return -1;
+ value = strtol(my_str, &end, 10);
//info("from %s I get %d and %s: %m", my_str, value, end);
/* means no numbers */
if (my_str == end)
diff --git a/src/plugins/acct_gather_energy/ipmi/acct_gather_energy_ipmi.c b/src/plugins/acct_gather_energy/ipmi/acct_gather_energy_ipmi.c
index 007f3a9d250c9f2eeac63563266165a5fc2fc404..e02db83e4d7244adce46db3c6d0c3998822f0368 100644
--- a/src/plugins/acct_gather_energy/ipmi/acct_gather_energy_ipmi.c
+++ b/src/plugins/acct_gather_energy/ipmi/acct_gather_energy_ipmi.c
@@ -544,12 +544,16 @@ static int _update_profile_message()
{
ipmi_message_profile_t *tmp;
int new_size;
- slurm_ctl_conf_t *conf_ptr;
if (profile_message_memory==0) {
- conf_ptr = slurm_conf_lock();
+ /* FIXME: now that job_acct_gather_freq is a string
+ this needs to be changed. This math looks wrong anyway.
+ */
+ /* new_size = 4 * */
+ /* (2+ conf_ptr->job_acct_gather_freq /slurm_ipmi_conf.freq); */
new_size = 4 *
- (2+ conf_ptr->job_acct_gather_freq /slurm_ipmi_conf.freq);
+ (2 + acct_gather_profile_timer[PROFILE_TASK].freq
+ / acct_gather_profile_timer[PROFILE_ENERGY].freq);
slurm_conf_unlock();
tmp = (ipmi_message_profile_t *)
xmalloc(sizeof(ipmi_message_profile_t)* new_size);
diff --git a/src/plugins/job_submit/lua/job_submit_lua.c b/src/plugins/job_submit/lua/job_submit_lua.c
index 4204cbb02e8160fcc534386780d61e30de269816..4a0ba6542d0358dd2b67a72317c20fd4a2e5fcf4 100644
--- a/src/plugins/job_submit/lua/job_submit_lua.c
+++ b/src/plugins/job_submit/lua/job_submit_lua.c
@@ -366,7 +366,7 @@ static int _get_job_req_field (lua_State *L)
} else if (!strcmp(name, "account")) {
lua_pushstring (L, job_desc->account);
} else if (!strcmp(name, "acctg_freq")) {
- lua_pushnumber (L, job_desc->acctg_freq);
+ lua_pushstring (L, job_desc->acctg_freq);
} else if (!strcmp(name, "begin_time")) {
lua_pushnumber (L, job_desc->begin_time);
} else if (!strcmp(name, "comment")) {
@@ -473,7 +473,10 @@ static int _set_job_req_field (lua_State *L)
if (strlen(value_str))
job_desc->account = xstrdup(value_str);
} else if (!strcmp(name, "acctg_freq")) {
- job_desc->acctg_freq = luaL_checknumber(L, 3);
+ value_str = luaL_checkstring(L, 3);
+ xfree(job_desc->acctg_freq);
+ if (strlen(value_str))
+ job_desc->acctg_freq = xstrdup(value_str);
} else if (!strcmp(name, "begin_time")) {
job_desc->begin_time = luaL_checknumber(L, 3);
} else if (!strcmp(name, "comment")) {
diff --git a/src/salloc/opt.c b/src/salloc/opt.c
index 35b6ebef445f1193f5cbfd29d716289526f4669b..9682da6fad26e31afb310ca6b03f9a419465a723 100644
--- a/src/salloc/opt.c
+++ b/src/salloc/opt.c
@@ -356,7 +356,7 @@ static void _opt_default()
opt.egid = (gid_t) -1;
opt.bell = BELL_AFTER_DELAY;
- opt.acctg_freq = -1;
+ opt.acctg_freq = NULL;
opt.no_shell = false;
opt.get_user_env_time = -1;
opt.get_user_env_mode = -1;
@@ -387,7 +387,7 @@ struct env_vars {
env_vars_t env_vars[] = {
{"SALLOC_ACCOUNT", OPT_STRING, &opt.account, NULL },
- {"SALLOC_ACCTG_FREQ", OPT_INT, &opt.acctg_freq, NULL },
+ {"SALLOC_ACCTG_FREQ", OPT_STRING, &opt.acctg_freq, NULL },
{"SALLOC_BELL", OPT_BELL, NULL, NULL },
{"SALLOC_CONN_TYPE", OPT_CONN_TYPE, NULL, NULL },
{"SALLOC_CPU_BIND", OPT_CPU_BIND, NULL, NULL },
@@ -1104,7 +1104,8 @@ void set_options(const int argc, char **argv)
opt.ramdiskimage = xstrdup(optarg);
break;
case LONG_OPT_ACCTG_FREQ:
- opt.acctg_freq = _get_int(optarg, "acctg-freq");
+ xfree(opt.acctg_freq);
+ opt.acctg_freq = xstrdup(optarg);
break;
case LONG_OPT_NOSHELL:
opt.no_shell = true;
diff --git a/src/salloc/salloc.c b/src/salloc/salloc.c
index 84f999745c8c25a05489384cefd29c260e52cd7b..85c7cd94d287d2c3d3af7a6f61cfc9f59d11d5f9 100644
--- a/src/salloc/salloc.c
+++ b/src/salloc/salloc.c
@@ -407,8 +407,8 @@ int main(int argc, char *argv[])
env_array_append_fmt(&env, "SLURM_OVERCOMMIT", "%d",
opt.overcommit);
}
- if (opt.acctg_freq >= 0) {
- env_array_append_fmt(&env, "SLURM_ACCTG_FREQ", "%d",
+ if (opt.acctg_freq) {
+ env_array_append_fmt(&env, "SLURM_ACCTG_FREQ", "%s",
opt.acctg_freq);
}
if (opt.network)
diff --git a/src/sbatch/opt.c b/src/sbatch/opt.c
index cb36cd04fe481a02647b05b144e5cd165ea955cd..a5c9828dfae27587a752dbf04035e9239be32909 100644
--- a/src/sbatch/opt.c
+++ b/src/sbatch/opt.c
@@ -450,7 +450,7 @@ struct env_vars {
env_vars_t env_vars[] = {
{"SBATCH_ACCOUNT", OPT_STRING, &opt.account, NULL },
{"SBATCH_ARRAY_INX", OPT_STRING, &opt.array_inx, NULL },
- {"SBATCH_ACCTG_FREQ", OPT_INT, &opt.acctg_freq, NULL },
+ {"SBATCH_ACCTG_FREQ", OPT_STRING, &opt.acctg_freq, NULL },
{"SBATCH_BLRTS_IMAGE", OPT_STRING, &opt.blrtsimage, NULL },
{"SBATCH_CHECKPOINT", OPT_STRING, &opt.ckpt_interval_str, NULL },
{"SBATCH_CHECKPOINT_DIR",OPT_STRING, &opt.ckpt_dir, NULL },
@@ -2454,8 +2454,8 @@ static bool _opt_verify(void)
acct_gather_profile_to_string(opt.profile));
- if (opt.acctg_freq >= 0)
- setenvf(NULL, "SLURM_ACCTG_FREQ", "%d", opt.acctg_freq);
+ if (opt.acctg_freq)
+ setenvf(NULL, "SLURM_ACCTG_FREQ", "%s", opt.acctg_freq);
#ifdef HAVE_AIX
if (opt.network == NULL)
diff --git a/src/slurmd/common/slurmstepd_init.c b/src/slurmd/common/slurmstepd_init.c
index d3def654f8d72e8f09803644c9860841e2059d60..88b3e2f631e36f183b111ddaa1cc722e3c0f1a97 100644
--- a/src/slurmd/common/slurmstepd_init.c
+++ b/src/slurmd/common/slurmstepd_init.c
@@ -112,6 +112,8 @@ extern int unpack_slurmd_conf_lite_no_alloc(slurmd_conf_t *conf, Buf buffer)
unpack_error:
error("unpack_error in unpack_slurmd_conf_lite_no_alloc: %m");
+ xfree(conf->job_acct_gather_freq);
+ xfree(conf->job_acct_gather_type);
xfree(conf->hostname);
xfree(conf->spooldir);
xfree(conf->node_name);
diff --git a/src/slurmd/slurmstepd/slurmstepd.c b/src/slurmd/slurmstepd/slurmstepd.c
index 3e0a104e2f2ddbc26fcb48ddbcb2c0aa91bfc168..1be7975dbcb630c73bdf3e4cc557249bd05788b8 100644
--- a/src/slurmd/slurmstepd/slurmstepd.c
+++ b/src/slurmd/slurmstepd/slurmstepd.c
@@ -390,11 +390,6 @@ _init_from_slurmd(int sock, char **argv,
log_alter(conf->log_opts, 0, conf->logfile);
debug2("debug level is %d.", conf->debug_level);
- /* FIXME: comment this out when we get this to work like a
- * string. It will be handled in
- * acct_gather_profile_startpoll when that happens
- */
- jobacct_gather_startpoll(conf->job_acct_gather_freq);
switch_g_slurmd_step_init();
diff --git a/src/slurmd/slurmstepd/slurmstepd_job.c b/src/slurmd/slurmstepd/slurmstepd_job.c
index 4c06372d31330d4a5015b03f6016dcbd4c66a339..87e3c92adcceaf795512941845f0ec33cfb1a384 100644
--- a/src/slurmd/slurmstepd/slurmstepd_job.c
+++ b/src/slurmd/slurmstepd/slurmstepd_job.c
@@ -180,16 +180,17 @@ job_create(launch_tasks_request_msg_t *msg)
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;
- }
+ /* FIXME: handle this now that acctg_freq is a string */
+ /* 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
@@ -293,17 +294,14 @@ job_create(launch_tasks_request_msg_t *msg)
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);
/* 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);
- /* FIXME: uncomment this when we get acctg-freq working like a
- string */
- //acct_gather_profile_startpoll("Network=10","5");
+ acct_gather_profile_startpoll(msg->acctg_freq,
+ conf->job_acct_gather_freq);
job->multi_prog = msg->multi_prog;
job->timelimit = (time_t) -1;
@@ -382,16 +380,18 @@ job_batch_job_create(batch_job_launch_msg_t *msg)
_pwd_destroy(pwd);
return NULL;
}
- if (msg->job_mem && (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;
- }
+
+ /* FIXME: handle this now that acctg_freq is a string */
+ /* 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));
@@ -407,8 +407,14 @@ job_batch_job_create(batch_job_launch_msg_t *msg)
job->array_task_id = msg->array_task_id;
job->batch = true;
- if (msg->acctg_freq != (uint16_t) NO_VAL)
- jobacct_gather_change_poll(msg->acctg_freq);
+ /* 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);
+ /* needed for the jobacct_gather plugin to start */
+ acct_gather_profile_startpoll(msg->acctg_freq,
+ conf->job_acct_gather_freq);
+
job->multi_prog = 0;
job->open_mode = msg->open_mode;
job->overcommit = (bool) msg->overcommit;
diff --git a/src/srun/libsrun/opt.c b/src/srun/libsrun/opt.c
index d8d6a908c5e810db3c6bf18e567265b3a0085754..d1ab97712de4814864f81a2e7a0976e529b7ed8d 100644
--- a/src/srun/libsrun/opt.c
+++ b/src/srun/libsrun/opt.c
@@ -542,7 +542,7 @@ struct env_vars {
env_vars_t env_vars[] = {
{"SLURMD_DEBUG", OPT_INT, &opt.slurmd_debug, NULL },
{"SLURM_ACCOUNT", OPT_STRING, &opt.account, NULL },
-{"SLURM_ACCTG_FREQ", OPT_INT, &opt.acctg_freq, NULL },
+{"SLURM_ACCTG_FREQ", OPT_STRING, &opt.acctg_freq, NULL },
{"SLURM_BLRTS_IMAGE", OPT_STRING, &opt.blrtsimage, NULL },
{"SLURM_CHECKPOINT", OPT_STRING, &opt.ckpt_interval_str, NULL },
{"SLURM_CHECKPOINT_DIR",OPT_STRING, &opt.ckpt_dir, NULL },