diff --git a/doc/html/heterogeneous_jobs.shtml b/doc/html/heterogeneous_jobs.shtml
index ff57932a3f56368543dc206393339685eeff8ae1..5b37a64a3a8c3affae68a944f2276a9842bce643 100644
--- a/doc/html/heterogeneous_jobs.shtml
+++ b/doc/html/heterogeneous_jobs.shtml
@@ -46,7 +46,7 @@ A list of propogated options follows.</p>
<li>--clusters</li>
<li>--comment</li>
<li>--deadline</li>
-<li>delay-boot</li>
+<li>--delay-boot</li>
<li>--dependency</li>
<li>--error</li>
<li>--export</li>
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index 382cf7c6ef59811ec5fb8aab481400b1cbcbad4b..31338e2fb3d1e51e8ec39a0400d458c4c935fc39 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -3172,7 +3172,7 @@ extern void slurm_free_resource_allocation_response_msg(resource_allocation_resp
* allocate resources for a list of job requests. This call will block
* until the entire allocation is granted, or the specified timeout limit
* is reached.
- * IN req - List of resource allocation requests
+ * IN job_req_list - List of resource allocation requests, type job_desc_msg_t
* IN timeout - amount of time, in seconds, to wait for a response before
* giving up.
* A timeout of zero will wait indefinitely.
@@ -3198,17 +3198,18 @@ List slurm_allocate_pack_job_blocking(List job_req_list, time_t timeout,
* NOTE: free the response using slurm_free_resource_allocation_response_msg()
*/
extern int slurm_allocation_lookup(uint32_t job_id,
- resource_allocation_response_msg_t **info);
+ resource_allocation_response_msg_t **resp);
/*
* slurm_pack_job_lookup - retrieve info for an existing heterogeneous job
* allocation without the addrs and such
* IN jobid - job allocation identifier
- * OUT info - job allocation information
+ * OUT resp - list of job allocation information, type
+ * resource_allocation_response_msg_t
* RET 0 on success, otherwise return -1 and set errno to indicate the error
* NOTE: free the response using list_destroy()
*/
-extern int slurm_pack_job_lookup(uint32_t jobid, List* info);
+extern int slurm_pack_job_lookup(uint32_t jobid, List *resp);
/*
* slurm_read_hostfile - Read a SLURM hostfile specified by "filename".
@@ -3257,6 +3258,17 @@ extern void slurm_allocation_msg_thr_destroy(allocation_msg_thread_t *msg_thr);
extern int slurm_submit_batch_job(job_desc_msg_t *job_desc_msg,
submit_response_msg_t **slurm_alloc_msg);
+/*
+ * slurm_submit_batch_pack_job - issue RPC to submit a heterogeneous job for
+ * later execution
+ * NOTE: free the response using slurm_free_submit_response_response_msg
+ * IN job_req_list - List of resource allocation requests, type job_desc_msg_t
+ * OUT slurm_alloc_msg - response to request
+ * RET 0 on success, otherwise return -1 and set errno to indicate the error
+ */
+extern int slurm_submit_batch_pack_job(List job_req_list,
+ submit_response_msg_t **slurm_alloc_msg);
+
/*
* slurm_free_submit_response_response_msg - free slurm
* job submit response message
diff --git a/src/api/submit.c b/src/api/submit.c
index 0d53ff5f342ee648a8e51281944626d136b44fc5..29cc815d5f49773d10a9392a790cdb7ec67dabab 100644
--- a/src/api/submit.c
+++ b/src/api/submit.c
@@ -49,56 +49,106 @@ extern pid_t getsid(pid_t pid); /* missing from <unistd.h> */
#include "src/common/read_config.h"
#include "src/common/slurm_protocol_api.h"
+#include "src/common/xmalloc.h"
+#include "src/common/xstring.h"
/*
* slurm_submit_batch_job - issue RPC to submit a job for later execution
* NOTE: free the response using slurm_free_submit_response_response_msg
* IN job_desc_msg - description of batch job request
- * OUT slurm_alloc_msg - response to request
+ * OUT resp - response to request
* RET 0 on success, otherwise return -1 and set errno to indicate the error
*/
-int
-slurm_submit_batch_job (job_desc_msg_t *req,
- submit_response_msg_t **resp)
+extern int slurm_submit_batch_job(job_desc_msg_t *req,
+ submit_response_msg_t **resp)
{
- int rc;
- slurm_msg_t req_msg;
- slurm_msg_t resp_msg;
- bool host_set = false;
- char host[64];
+ int rc;
+ slurm_msg_t req_msg;
+ slurm_msg_t resp_msg;
+ char *local_hostname = NULL;
slurm_msg_t_init(&req_msg);
slurm_msg_t_init(&resp_msg);
-
/*
* set Node and session id for this request
*/
if (req->alloc_sid == NO_VAL)
req->alloc_sid = getsid(0);
- if ( (req->alloc_node == NULL)
- && (gethostname_short(host, sizeof(host)) == 0) ) {
- req->alloc_node = host;
- host_set = true;
+ if (req->alloc_node == NULL) {
+ local_hostname = xshort_hostname();
+ req->alloc_node = local_hostname;
}
- req_msg.msg_type = REQUEST_SUBMIT_BATCH_JOB ;
+ req_msg.msg_type = REQUEST_SUBMIT_BATCH_JOB;
req_msg.data = req;
rc = slurm_send_recv_controller_msg(&req_msg, &resp_msg,
working_cluster_rec);
+ xfree(local_hostname);
+ if (rc == SLURM_SOCKET_ERROR)
+ return SLURM_ERROR;
+
+ switch (resp_msg.msg_type) {
+ case RESPONSE_SLURM_RC:
+ rc = ((return_code_msg_t *) resp_msg.data)->return_code;
+ if (rc)
+ slurm_seterrno_ret(rc);
+ *resp = NULL;
+ break;
+ case RESPONSE_SUBMIT_BATCH_JOB:
+ *resp = (submit_response_msg_t *) resp_msg.data;
+ break;
+ default:
+ slurm_seterrno_ret(SLURM_UNEXPECTED_MSG_ERROR);
+ }
+
+ return SLURM_PROTOCOL_SUCCESS;
+}
+
+/*
+ * slurm_submit_batch_pack_job - issue RPC to submit a heterogeneous job for
+ * later execution
+ * NOTE: free the response using slurm_free_submit_response_response_msg
+ * IN job_req_list - List of resource allocation requests, type job_desc_msg_t
+ * OUT resp - response to request
+ * RET 0 on success, otherwise return -1 and set errno to indicate the error
+ */
+extern int slurm_submit_batch_pack_job(List job_req_list,
+ submit_response_msg_t **resp)
+{
+ int rc;
+ job_desc_msg_t *req;
+ slurm_msg_t req_msg;
+ slurm_msg_t resp_msg;
+ char *local_hostname = NULL;
+ ListIterator iter;
+
+ slurm_msg_t_init(&req_msg);
+ slurm_msg_t_init(&resp_msg);
/*
- * Clear this hostname if set internally to this function
- * (memory is on the stack)
+ * set Node and session id for this request
*/
- if (host_set)
- req->alloc_node = NULL;
+ local_hostname = xshort_hostname();
+ iter = list_iterator_create(job_req_list);
+ while ((req = (job_desc_msg_t *) list_next(iter))) {
+ if (req->alloc_sid == NO_VAL)
+ req->alloc_sid = getsid(0);
+ if (!req->alloc_node)
+ req->alloc_node = local_hostname;
+ }
+ list_iterator_destroy(iter);
+ req_msg.msg_type = REQUEST_SUBMIT_BATCH_JOB_PACK;
+ req_msg.data = job_req_list;
+
+ rc = slurm_send_recv_controller_msg(&req_msg, &resp_msg,
+ working_cluster_rec);
+ xfree(local_hostname);
if (rc == SLURM_SOCKET_ERROR)
return SLURM_ERROR;
-
switch (resp_msg.msg_type) {
case RESPONSE_SLURM_RC:
rc = ((return_code_msg_t *) resp_msg.data)->return_code;
diff --git a/src/common/slurm_protocol_defs.c b/src/common/slurm_protocol_defs.c
index 0fe67fe986c49968d9dee17e7920ab37eb26a20f..e8a98d672bf0e75ec774c4ce2457ecce1a8b7e10 100644
--- a/src/common/slurm_protocol_defs.c
+++ b/src/common/slurm_protocol_defs.c
@@ -4300,6 +4300,7 @@ extern int slurm_free_msg_data(slurm_msg_type_t type, void *data)
slurm_free_job_info(data);
break;
case REQUEST_JOB_PACK_ALLOCATION:
+ case REQUEST_SUBMIT_BATCH_JOB_PACK:
case RESPONSE_JOB_PACK_ALLOCATION:
FREE_NULL_LIST(data);
break;
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index cb814f987a4da28cc634b10938a90a38f13eea99..99c91d5bbbbee45ee455569d157d80978dfee31d 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -308,6 +308,7 @@ typedef enum {
REQUEST_CTLD_MULT_MSG,
RESPONSE_CTLD_MULT_MSG,
REQUEST_JOB_PACK_ALLOC_INFO,
+ REQUEST_SUBMIT_BATCH_JOB_PACK,
REQUEST_JOB_STEP_CREATE = 5001,
RESPONSE_JOB_STEP_CREATE,
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 49df8c0627b3334167b959c99cc49a19b59a00e3..7ddf1fabf22676d87a1cb7585518e57241883826 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -943,6 +943,7 @@ pack_msg(slurm_msg_t const *msg, Buf buffer)
msg->protocol_version);
break;
case REQUEST_JOB_PACK_ALLOCATION:
+ case REQUEST_SUBMIT_BATCH_JOB_PACK:
_pack_job_desc_list_msg((List) msg->data, buffer,
msg->protocol_version);
break;
@@ -1633,6 +1634,7 @@ unpack_msg(slurm_msg_t * msg, Buf buffer)
buffer, msg->protocol_version);
break;
case REQUEST_JOB_PACK_ALLOCATION:
+ case REQUEST_SUBMIT_BATCH_JOB_PACK:
rc = _unpack_job_desc_list_msg((List *) &(msg->data),
buffer, msg->protocol_version);
break;
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 78be24585176dd12b3b8fdba719cbc81488d5061..d7050aba909f37dd98980e043f272a2fa052970e 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -213,6 +213,7 @@ inline static void _slurm_rpc_step_layout(slurm_msg_t * msg);
inline static void _slurm_rpc_step_update(slurm_msg_t * msg);
inline static void _slurm_rpc_submit_batch_job(slurm_msg_t * msg,
bool is_sib_job);
+inline static void _slurm_rpc_submit_batch_pack_job(slurm_msg_t * msg);
inline static void _slurm_rpc_suspend(slurm_msg_t * msg);
inline static void _slurm_rpc_top_job(slurm_msg_t * msg);
inline static void _slurm_rpc_trigger_clear(slurm_msg_t * msg);
@@ -445,6 +446,9 @@ void slurmctld_req(slurm_msg_t *msg, connection_arg_t *arg)
case REQUEST_SUBMIT_BATCH_JOB:
_slurm_rpc_submit_batch_job(msg, false);
break;
+ case REQUEST_SUBMIT_BATCH_JOB_PACK:
+ _slurm_rpc_submit_batch_pack_job(msg);
+ break;
case REQUEST_UPDATE_FRONT_END:
_slurm_rpc_update_front_end(msg);
break;
@@ -3687,7 +3691,7 @@ static void _slurm_rpc_step_update(slurm_msg_t *msg)
}
/* _slurm_rpc_submit_batch_job - process RPC to submit a batch job */
-static void _slurm_rpc_submit_batch_job(slurm_msg_t * msg, bool is_sib_job)
+static void _slurm_rpc_submit_batch_job(slurm_msg_t *msg, bool is_sib_job)
{
static int active_rpc_cnt = 0;
int error_code = SLURM_SUCCESS;
@@ -3710,7 +3714,7 @@ static void _slurm_rpc_submit_batch_job(slurm_msg_t * msg, bool is_sib_job)
bool reject_job = false;
START_TIMER;
-
+ debug2("Processing RPC: REQUEST_SUBMIT_BATCH_JOB from uid=%d", uid);
if (slurmctld_config.submissions_disabled) {
info("Submissions disabled on system");
error_code = ESLURM_SUBMISSIONS_DISABLED;
@@ -3718,8 +3722,6 @@ static void _slurm_rpc_submit_batch_job(slurm_msg_t * msg, bool is_sib_job)
goto send_msg;
}
- debug2("Processing RPC: REQUEST_SUBMIT_BATCH_JOB from uid=%d", uid);
-
slurm_msg_t_init(&response_msg);
response_msg.flags = msg->flags;
response_msg.protocol_version = msg->protocol_version;
@@ -3734,7 +3736,8 @@ static void _slurm_rpc_submit_batch_job(slurm_msg_t * msg, bool is_sib_job)
if ((job_desc_msg->alloc_node == NULL) ||
(job_desc_msg->alloc_node[0] == '\0')) {
error_code = ESLURM_INVALID_NODE_NAME;
- error("REQUEST_SUBMIT_BATCH_JOB lacks alloc_node from uid=%d", uid);
+ error("REQUEST_SUBMIT_BATCH_JOB lacks alloc_node from uid=%d",
+ uid);
}
dump_job_desc(job_desc_msg);
@@ -3786,15 +3789,13 @@ send_msg:
END_TIMER2("_slurm_rpc_submit_batch_job");
if (reject_job) {
- info("_slurm_rpc_submit_batch_job: %s",
- slurm_strerror(error_code));
+ info("%s: %s", __func__, slurm_strerror(error_code));
if (err_msg)
slurm_send_rc_err_msg(msg, error_code, err_msg);
else
slurm_send_rc_msg(msg, error_code);
} else {
- info("_slurm_rpc_submit_batch_job JobId=%u %s",
- job_id, TIME_STR);
+ info("%s: JobId=%u %s", __func__, job_id, TIME_STR);
/* send job_ID */
submit_msg.job_id = job_id;
submit_msg.step_id = step_id;
@@ -3813,6 +3814,41 @@ send_msg:
xfree(err_msg);
}
+/* _slurm_rpc_submit_batch_pack_job - process RPC to submit a batch pack job */
+static void _slurm_rpc_submit_batch_pack_job(slurm_msg_t *msg)
+{
+ ListIterator iter;
+ int error_code = SLURM_SUCCESS;
+ job_desc_msg_t *job_desc_msg;
+ List job_req_list = (List) msg->data;
+ uid_t uid = g_slurm_auth_get_uid(msg->auth_cred,
+ slurmctld_config.auth_info);
+
+ info("Processing RPC: REQUEST_SUBMIT_BATCH_PACK_JOB from uid=%d", uid);
+ if (!job_req_list || (list_count(job_req_list) == 0)) {
+ info("REQUEST_SUBMIT_BATCH_PACK_JOB from uid=%d with empty job list",
+ uid);
+ error_code = SLURM_ERROR;
+ goto send_msg;
+ }
+ if (slurmctld_config.submissions_disabled) {
+ info("Submissions disabled on system");
+ error_code = ESLURM_SUBMISSIONS_DISABLED;
+ goto send_msg;
+ }
+
+ iter = list_iterator_create(job_req_list);
+ while ((job_desc_msg = (job_desc_msg_t *) list_next(iter))) {
+//FIXME: FLesh out the logic here
+ dump_job_desc(job_desc_msg);
+ }
+ list_iterator_destroy(iter);
+
+error_code = SLURM_ERROR;
+send_msg:
+ slurm_send_rc_msg(msg, error_code);
+}
+
/* _slurm_rpc_update_job - process RPC to update the configuration of a
* job (e.g. priority)
*/