diff --git a/doc/html/elastic_computing.shtml b/doc/html/elastic_computing.shtml
index 6e90de7ed4747cf4ff260c99e58c58b603949a41..2fa6789fce0a4179673704f2ea277b48ea83485e 100644
--- a/doc/html/elastic_computing.shtml
+++ b/doc/html/elastic_computing.shtml
@@ -126,10 +126,18 @@ option is used.</li>
<p>The <i>SuspendProgram</i> only needs to relinquish the node back to the
cloud.</p>
+<p>An environment variable SLURM_NODE_ALIASES is set with the node name and
+communication address pairs. The variable is set by salloc, sbatch, and srun.
+It is then used by srun to determine the destination for job launch
+communication messages.</p>
+
<h2>Remaining Work</h2>
<ul>
-<li>srun commands are not currently aware of the addresses of nodes in the cloud.</li>
+<li>The sbatch logicneeds modifcation to set the SLURM_NODE_ALIASES
+environment variable.</li>
+<li>The SLURM_NODE_ALIASES environment varilable needs to change if jobs
+size changes.</li>
<li>We need scripts to provision resources from EC2.</li>
</ul>
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index 5ca6e5d4c57c7162ec03e6b6416a4edc9a35cf0b..73f95817a997d49bda3f4e41cb9f1904612727e2 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -1501,6 +1501,7 @@ typedef struct delete_partition_msg {
typedef struct resource_allocation_response_msg {
uint32_t job_id; /* assigned job id */
char *node_list; /* assigned list of nodes */
+ char *alias_list; /* node name/address aliases */
uint32_t num_cpu_groups;/* elements in below cpu arrays */
uint16_t *cpus_per_node;/* cpus per node */
uint32_t *cpu_count_reps;/* how many nodes have same cpu count */
diff --git a/src/common/env.c b/src/common/env.c
index 55872f98542cdfca6d732850e68b9f0e5f69e134..51f56a17e0858f489077740ed5be21f25ca1b3e5 100644
--- a/src/common/env.c
+++ b/src/common/env.c
@@ -943,17 +943,19 @@ env_array_for_job(char ***dest, const resource_allocation_response_msg_t *alloc,
env_array_overwrite_fmt(dest, "SLURM_JOB_NUM_NODES", "%u", node_cnt);
env_array_overwrite_fmt(dest, "SLURM_JOB_NODELIST", "%s",
alloc->node_list);
+ env_array_overwrite_fmt(dest, "SLURM_NODE_ALIASES", "%s",
+ alloc->alias_list);
_set_distribution(desc->task_dist, &dist, &lllp_dist);
- if(dist)
+ if (dist)
env_array_overwrite_fmt(dest, "SLURM_DISTRIBUTION", "%s",
dist);
- if(desc->task_dist == SLURM_DIST_PLANE)
+ if (desc->task_dist == SLURM_DIST_PLANE)
env_array_overwrite_fmt(dest, "SLURM_DIST_PLANESIZE",
"%u", desc->plane_size);
- if(lllp_dist)
+ if (lllp_dist)
env_array_overwrite_fmt(dest, "SLURM_DIST_LLLP", "%s",
lllp_dist);
diff --git a/src/common/slurm_protocol_defs.c b/src/common/slurm_protocol_defs.c
index 84294549ca05d2ffee31e6200532affcc4828184..9304a531e6429f405239768afff058b555507220 100644
--- a/src/common/slurm_protocol_defs.c
+++ b/src/common/slurm_protocol_defs.c
@@ -1794,6 +1794,7 @@ extern void slurm_free_resource_allocation_response_msg (
if (msg) {
select_g_select_jobinfo_free(msg->select_jobinfo);
msg->select_jobinfo = NULL;
+ xfree(msg->alias_list);
xfree(msg->node_list);
xfree(msg->cpus_per_node);
xfree(msg->cpu_count_reps);
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 872e7c82c3b8e70f04d95b4cecb019e9a25fb22f..25dea356911df687dbd7e6fbad07f371bbf6ef3c 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -2465,6 +2465,7 @@ _pack_resource_allocation_response_msg(resource_allocation_response_msg_t *msg,
pack32(msg->error_code, buffer);
pack32(msg->job_id, buffer);
pack32(msg->pn_min_memory, buffer);
+ packstr(msg->alias_list, buffer);
packstr(msg->node_list, buffer);
pack32(msg->num_cpu_groups, buffer);
@@ -2493,31 +2494,61 @@ _unpack_resource_allocation_response_msg(
*msg = tmp_ptr;
/* load the data values */
- safe_unpack32(&tmp_ptr->error_code, buffer);
- safe_unpack32(&tmp_ptr->job_id, buffer);
- safe_unpack32(&tmp_ptr->pn_min_memory, buffer);
- safe_unpackstr_xmalloc(&tmp_ptr->node_list, &uint32_tmp, buffer);
+ if (protocol_version >= SLURM_2_4_PROTOCOL_VERSION) {
+ safe_unpack32(&tmp_ptr->error_code, buffer);
+ safe_unpack32(&tmp_ptr->job_id, buffer);
+ safe_unpack32(&tmp_ptr->pn_min_memory, buffer);
+ safe_unpackstr_xmalloc(&tmp_ptr->alias_list, &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&tmp_ptr->node_list, &uint32_tmp,
+ buffer);
- safe_unpack32(&tmp_ptr->num_cpu_groups, buffer);
- if (tmp_ptr->num_cpu_groups > 0) {
- safe_unpack16_array(&tmp_ptr->cpus_per_node, &uint32_tmp,
- buffer);
- if (tmp_ptr->num_cpu_groups != uint32_tmp)
- goto unpack_error;
- safe_unpack32_array(&tmp_ptr->cpu_count_reps, &uint32_tmp,
- buffer);
- if (tmp_ptr->num_cpu_groups != uint32_tmp)
+ safe_unpack32(&tmp_ptr->num_cpu_groups, buffer);
+ if (tmp_ptr->num_cpu_groups > 0) {
+ safe_unpack16_array(&tmp_ptr->cpus_per_node,
+ &uint32_tmp, buffer);
+ if (tmp_ptr->num_cpu_groups != uint32_tmp)
+ goto unpack_error;
+ safe_unpack32_array(&tmp_ptr->cpu_count_reps,
+ &uint32_tmp, buffer);
+ if (tmp_ptr->num_cpu_groups != uint32_tmp)
+ goto unpack_error;
+ } else {
+ tmp_ptr->cpus_per_node = NULL;
+ tmp_ptr->cpu_count_reps = NULL;
+ }
+
+ safe_unpack32(&tmp_ptr->node_cnt, buffer);
+ if (select_g_select_jobinfo_unpack(&tmp_ptr->select_jobinfo,
+ buffer, protocol_version))
goto unpack_error;
} else {
- tmp_ptr->cpus_per_node = NULL;
- tmp_ptr->cpu_count_reps = NULL;
- }
+ safe_unpack32(&tmp_ptr->error_code, buffer);
+ safe_unpack32(&tmp_ptr->job_id, buffer);
+ safe_unpack32(&tmp_ptr->pn_min_memory, buffer);
+ safe_unpackstr_xmalloc(&tmp_ptr->node_list, &uint32_tmp,
+ buffer);
- safe_unpack32(&tmp_ptr->node_cnt, buffer);
+ safe_unpack32(&tmp_ptr->num_cpu_groups, buffer);
+ if (tmp_ptr->num_cpu_groups > 0) {
+ safe_unpack16_array(&tmp_ptr->cpus_per_node,
+ &uint32_tmp, buffer);
+ if (tmp_ptr->num_cpu_groups != uint32_tmp)
+ goto unpack_error;
+ safe_unpack32_array(&tmp_ptr->cpu_count_reps,
+ &uint32_tmp, buffer);
+ if (tmp_ptr->num_cpu_groups != uint32_tmp)
+ goto unpack_error;
+ } else {
+ tmp_ptr->cpus_per_node = NULL;
+ tmp_ptr->cpu_count_reps = NULL;
+ }
- if (select_g_select_jobinfo_unpack(&tmp_ptr->select_jobinfo, buffer,
- protocol_version))
- goto unpack_error;
+ safe_unpack32(&tmp_ptr->node_cnt, buffer);
+ if (select_g_select_jobinfo_unpack(&tmp_ptr->select_jobinfo,
+ buffer, protocol_version))
+ goto unpack_error;
+ }
return SLURM_SUCCESS;
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 1e7d408b62c2548845c49e1231d2eb491c84c1dc..4f080efa12c75bb5dfb324fb206e27632daa1a13 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -5378,6 +5378,7 @@ static void _list_delete_job(void *job_entry)
delete_job_details(job_ptr);
xfree(job_ptr->account);
+ xfree(job_ptr->alias_list);
xfree(job_ptr->alloc_node);
xfree(job_ptr->batch_host);
xfree(job_ptr->comment);
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 1d278e4d98376e76c93139fc32ade6e77e3bd00e..5ec80a4b01617b53cf4d481a682105250c7b7682 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -129,6 +129,7 @@ static bitstr_t *_valid_features(struct job_details *detail_ptr,
extern void allocate_nodes(struct job_record *job_ptr)
{
int i;
+ struct node_record *node_ptr;
#ifdef HAVE_FRONT_END
job_ptr->front_end_ptr = assign_front_end();
@@ -137,13 +138,22 @@ extern void allocate_nodes(struct job_record *job_ptr)
job_ptr->batch_host = xstrdup(job_ptr->front_end_ptr->name);
#endif
- for (i = 0; i < node_record_count; i++) {
+ xfree(job_ptr->alias_list);
+ for (i = 0, node_ptr = node_record_table_ptr; i < node_record_count;
+ i++, node_ptr++) {
if (!bit_test(job_ptr->node_bitmap, i))
continue;
+ if (IS_NODE_CLOUD(node_ptr)) {
+ if (job_ptr->alias_list)
+ xstrcat(job_ptr->alias_list, ",");
+ xstrcat(job_ptr->alias_list, node_ptr->name);
+ xstrcat(job_ptr->alias_list, ":");
+ xstrcat(job_ptr->alias_list, node_ptr->comm_name);
+ }
make_node_alloc(&node_record_table_ptr[i], job_ptr);
if (job_ptr->batch_host)
continue;
- job_ptr->batch_host = xstrdup(node_record_table_ptr[i].name);
+ job_ptr->batch_host = xstrdup(node_ptr->name);
}
last_node_update = time(NULL);
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index ba6958cfbd3008955f33d7f584f65a4b0531559c..463ca91fe8f23271051154fe3c91f6a618a0e20e 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -858,6 +858,7 @@ static void _slurm_rpc_allocate_resources(slurm_msg_t * msg)
alloc_msg.job_id = job_ptr->job_id;
alloc_msg.node_cnt = job_ptr->node_cnt;
alloc_msg.node_list = xstrdup(job_ptr->nodes);
+ alloc_msg.alias_list = xstrdup(job_ptr->alias_list);
alloc_msg.select_jobinfo =
select_g_select_jobinfo_copy(job_ptr->select_jobinfo);
if (job_ptr->details) {
@@ -2074,6 +2075,7 @@ static void _slurm_rpc_job_alloc_info_lite(slurm_msg_t * msg)
job_info_resp_msg.job_id = job_info_msg->job_id;
job_info_resp_msg.node_cnt = job_ptr->node_cnt;
job_info_resp_msg.node_list = xstrdup(job_ptr->nodes);
+ job_info_resp_msg.alias_list = xstrdup(job_ptr->alias_list);
job_info_resp_msg.select_jobinfo =
select_g_select_jobinfo_copy(job_ptr->select_jobinfo);
unlock_slurmctld(job_read_lock);
@@ -2125,7 +2127,8 @@ static void _slurm_rpc_job_sbcast_cred(slurm_msg_t * msg)
slurm_strerror(error_code));
slurm_send_rc_msg(msg, error_code);
} else if ((sbcast_cred = create_sbcast_cred(slurmctld_config.cred_ctx,
- job_ptr->job_id, job_ptr->nodes)) == NULL) {
+ job_ptr->job_id,
+ job_ptr->nodes)) == NULL){
unlock_slurmctld(job_read_lock);
error("_slurm_rpc_job_sbcast_cred JobId=%u cred create error",
job_info_msg->job_id);
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 98d7651cd1d629587e8b06e7fd9495738dc2b58a..e1d707db75b6426cbe819db7490a5e4fc03bfba6 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -422,6 +422,7 @@ struct job_details {
struct job_record {
char *account; /* account number to charge */
+ char *alias_list; /* node name to address aliases */
char *alloc_node; /* local node making resource alloc */
uint16_t alloc_resp_port; /* RESPONSE_RESOURCE_ALLOCATION port */
uint32_t alloc_sid; /* local sid making resource alloc */
diff --git a/src/slurmctld/srun_comm.c b/src/slurmctld/srun_comm.c
index 3108bca7f52d4172fed6539efffe5d05399c311f..2032d116ed9e2048f12a1fc7c7fd3a3278a2e20f 100644
--- a/src/slurmctld/srun_comm.c
+++ b/src/slurmctld/srun_comm.c
@@ -93,6 +93,7 @@ extern void srun_allocate (uint32_t job_id)
msg_arg = xmalloc(sizeof(resource_allocation_response_msg_t));
msg_arg->job_id = job_ptr->job_id;
msg_arg->node_list = xstrdup(job_ptr->nodes);
+ msg_arg->alias_list = xstrdup(job_ptr->alias_list);
msg_arg->num_cpu_groups = job_resrcs_ptr->cpu_array_cnt;
msg_arg->cpus_per_node = xmalloc(sizeof(uint16_t) *
job_resrcs_ptr->cpu_array_cnt);
diff --git a/src/srun/srun.c b/src/srun/srun.c
index ae31f92e372c4546edc2d4e19aaaa38bca88a319..a39dba1d5fd2a20f8d8e6176358e998105ad0db4 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -149,8 +149,9 @@ static void _pty_restore(void);
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_env_vars(resource_allocation_response_msg_t *resp);
static void _set_exit_code(void);
+static void _set_node_alias(void);
static void _step_opt_exclusive(void);
static void _set_stdio_fds(srun_job_t *job, slurm_step_io_fds_t *cio_fds);
static void _set_submit_dir_env(void);
@@ -309,7 +310,7 @@ int srun(int ac, char **av)
opt.alloc_nodelist = xstrdup(resp->node_list);
if (opt.exclusive)
_step_opt_exclusive();
- _set_cpu_env_var(resp);
+ _set_env_vars(resp);
if (_validate_relative(resp))
exit(error_exit);
job = job_step_create_allocation(resp);
@@ -345,7 +346,7 @@ int srun(int ac, char **av)
exit(error_exit);
got_alloc = 1;
_print_job_information(resp);
- _set_cpu_env_var(resp);
+ _set_env_vars(resp);
if (_validate_relative(resp)) {
slurm_complete_job(resp->job_id, 1);
exit(error_exit);
@@ -438,6 +439,7 @@ int srun(int ac, char **av)
setup_env(env, opt.preserve_env);
xfree(env->task_count);
xfree(env);
+ _set_node_alias();
re_launch:
#if defined HAVE_BGQ
@@ -779,19 +781,29 @@ static void _set_exit_code(void)
}
}
-static void _set_cpu_env_var(resource_allocation_response_msg_t *resp)
+static void _set_env_vars(resource_allocation_response_msg_t *resp)
{
char *tmp;
- if (getenv("SLURM_JOB_CPUS_PER_NODE"))
- return;
+ if (!getenv("SLURM_JOB_CPUS_PER_NODE")) {
+ tmp = uint32_compressed_to_str(resp->num_cpu_groups,
+ resp->cpus_per_node,
+ resp->cpu_count_reps);
+ if (setenvf(NULL, "SLURM_JOB_CPUS_PER_NODE", "%s", tmp) < 0) {
+ error("unable to set SLURM_JOB_CPUS_PER_NODE in "
+ "environment");
+ }
+ xfree(tmp);
+ }
+
+ if (!getenv("SLURM_NODE_ALIASES") && resp->alias_list) {
+ if (setenvf(NULL, "SLURM_NODE_ALIASES", "%s",
+ resp->alias_list) < 0) {
+ error("unable to set SLURM_NODE_ALIASES in "
+ "environment");
+ }
+ }
- tmp = uint32_compressed_to_str(resp->num_cpu_groups,
- resp->cpus_per_node,
- resp->cpu_count_reps);
- if (setenvf(NULL, "SLURM_JOB_CPUS_PER_NODE", "%s", tmp) < 0)
- error("unable to set SLURM_JOB_CPUS_PER_NODE in environment");
- xfree(tmp);
return;
}
@@ -875,6 +887,27 @@ static int _set_rlimit_env(void)
return rc;
}
+static void _set_node_alias(void)
+{
+ char *aliases, *save_ptr = NULL, *tmp;
+ char *addr, *name;
+
+ tmp = getenv("SLURM_NODE_ALIASES");
+ if (!tmp)
+ return;
+ aliases = xstrdup(tmp);
+ name = strtok_r(aliases, ":", &save_ptr);
+ while (name) {
+ addr = strtok_r(NULL, ",", &save_ptr);
+ if (addr) {
+ slurm_reset_alias(name, addr, addr);
+ name = strtok_r(NULL, ":", &save_ptr);
+ } else
+ name = NULL;
+ }
+ xfree(aliases);
+}
+
static int _become_user (void)
{
char *user = uid_to_string(opt.uid);