From 3e6a656047bb4c92be22108e6e8a649d03d5111f Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Mon, 9 Dec 2002 21:29:44 +0000
Subject: [PATCH] Add immediate flag to job request RPC, remove the indepentent
system call REQUEST_IMMEDIATE_RESOURCE_ALLOCATION. This immediate flag also
applies to the slurm_allocate_and_run_job_step() function.
---
src/api/allocate.c | 13 ++-----
src/api/init_msg.c | 1 +
src/api/slurm.h | 7 ++--
src/api/submit.c | 1 +
src/common/slurm_protocol_defs.h | 2 --
src/common/slurm_protocol_pack.c | 6 ++--
src/slurmctld/controller.c | 61 +++++++++++++-------------------
src/srun/srun.c | 9 +++--
8 files changed, 41 insertions(+), 59 deletions(-)
diff --git a/src/api/allocate.c b/src/api/allocate.c
index f50a383db00..0c0ef4f4a63 100644
--- a/src/api/allocate.c
+++ b/src/api/allocate.c
@@ -38,15 +38,12 @@
* slurm_allocate_resources - allocate resources for a job request
* IN job_desc_msg - description of resource allocation request
* OUT slurm_alloc_msg - response to request
- * IN immediate - if set then resource allocation must be satisfied
- * immediately or fail
* RET 0 on success or slurm error code
* NOTE: free the allocated using slurm_free_resource_allocation_response_msg
*/
int
slurm_allocate_resources (job_desc_msg_t * job_desc_msg ,
- resource_allocation_response_msg_t ** slurm_alloc_msg,
- int immediate )
+ resource_allocation_response_msg_t ** slurm_alloc_msg )
{
int msg_size ;
int rc ;
@@ -63,10 +60,7 @@ slurm_allocate_resources (job_desc_msg_t * job_desc_msg ,
}
/* send request message */
- if ( immediate )
- request_msg.msg_type = REQUEST_IMMEDIATE_RESOURCE_ALLOCATION ;
- else
- request_msg . msg_type = REQUEST_RESOURCE_ALLOCATION ;
+ request_msg . msg_type = REQUEST_RESOURCE_ALLOCATION ;
request_msg . data = job_desc_msg ;
if ( ( rc = slurm_send_controller_msg ( sockfd , & request_msg ) )
== SLURM_SOCKET_ERROR ) {
@@ -104,7 +98,6 @@ slurm_allocate_resources (job_desc_msg_t * job_desc_msg ,
*slurm_alloc_msg = NULL;
break ;
case RESPONSE_RESOURCE_ALLOCATION:
- case RESPONSE_IMMEDIATE_RESOURCE_ALLOCATION:
/* Calling method is responsible to free this memory */
*slurm_alloc_msg =
( resource_allocation_response_msg_t * )
@@ -146,6 +139,7 @@ int slurm_job_will_run (job_desc_msg_t * job_desc_msg ,
}
/* send request message */
+ /* job_desc_msg.immediate = true; implicit */
request_msg . msg_type = REQUEST_JOB_WILL_RUN ;
request_msg . data = job_desc_msg ;
if ( ( rc = slurm_send_controller_msg ( sockfd , & request_msg ) )
@@ -437,4 +431,3 @@ slurm_confirm_allocation (old_job_alloc_msg_t * job_desc_msg ,
return SLURM_PROTOCOL_SUCCESS ;
}
-
diff --git a/src/api/init_msg.c b/src/api/init_msg.c
index 6fd6a5f1ac3..4008e66efd9 100644
--- a/src/api/init_msg.c
+++ b/src/api/init_msg.c
@@ -45,6 +45,7 @@ void slurm_init_job_desc_msg(job_desc_msg_t * job_desc_msg)
job_desc_msg->environment = ((char **) NULL);
job_desc_msg->env_size = 0;
job_desc_msg->features = NULL;
+ job_desc_msg->immediate = 0;
job_desc_msg->job_id = NO_VAL;
job_desc_msg->kill_on_node_fail = (uint16_t) NO_VAL;
job_desc_msg->name = NULL;
diff --git a/src/api/slurm.h b/src/api/slurm.h
index 09f73c5602a..293b5949870 100644
--- a/src/api/slurm.h
+++ b/src/api/slurm.h
@@ -132,6 +132,8 @@ typedef struct job_descriptor { /* For submit, allocate, and update requests */
uint16_t env_size; /* element count in environment */
char *features; /* comma separated list of required features,
* default NONE */
+ uint16_t immediate; /* 1 if allocate to run or fail immediately,
+ * 0 if to be queued awaiting resources */
uint32_t job_id; /* job ID, default set by SLURM */
char *name; /* name of the job, default "" */
uint32_t min_procs; /* minimum processors per node, default=0 */
@@ -377,14 +379,11 @@ extern void slurm_init_job_desc_msg (job_desc_msg_t * job_desc_msg);
* slurm_allocate_resources - allocate resources for a job request
* IN job_desc_msg - description of resource allocation request
* OUT slurm_alloc_msg - response to request
- * IN immediate - if set then resource allocation must be satisfied
- * immediately or fail
* RET 0 on success or slurm error code
* NOTE: free the allocated using slurm_free_resource_allocation_response_msg
*/
extern int slurm_allocate_resources (job_desc_msg_t * job_desc_msg ,
- resource_allocation_response_msg_t ** job_alloc_resp_msg,
- int immediate ) ;
+ resource_allocation_response_msg_t ** job_alloc_resp_msg ) ;
/*
* slurm_free_resource_allocation_response_msg - free slurm resource
diff --git a/src/api/submit.c b/src/api/submit.c
index 9c8b98ccb50..875e0b757d3 100644
--- a/src/api/submit.c
+++ b/src/api/submit.c
@@ -61,6 +61,7 @@ slurm_submit_batch_job (job_desc_msg_t * job_desc_msg,
}
/* send request message */
+ /* job_desc_msg.immediate = false; implicit */
request_msg . msg_type = REQUEST_SUBMIT_BATCH_JOB ;
request_msg . data = job_desc_msg ;
if ( ( rc = slurm_send_controller_msg ( sockfd , & request_msg ) )
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index 1147d93a2fb..50d7a87c499 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -112,8 +112,6 @@ typedef enum {
RESPONSE_JOB_RESOURCE,
REQUEST_JOB_ATTACH,
RESPONSE_JOB_ATTACH,
- REQUEST_IMMEDIATE_RESOURCE_ALLOCATION,
- RESPONSE_IMMEDIATE_RESOURCE_ALLOCATION,
REQUEST_JOB_WILL_RUN,
RESPONSE_JOB_WILL_RUN,
REQUEST_REVOKE_JOB_CREDENTIAL,
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 091844a398f..118d7049d6b 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -320,7 +320,6 @@ pack_msg(slurm_msg_t const *msg, Buf buffer)
break;
case REQUEST_RESOURCE_ALLOCATION:
case REQUEST_SUBMIT_BATCH_JOB:
- case REQUEST_IMMEDIATE_RESOURCE_ALLOCATION:
case REQUEST_JOB_WILL_RUN:
case REQUEST_ALLOCATION_AND_RUN_JOB_STEP:
case REQUEST_UPDATE_JOB:
@@ -346,7 +345,6 @@ pack_msg(slurm_msg_t const *msg, Buf buffer)
msg->data, buffer);
break;
case RESPONSE_RESOURCE_ALLOCATION:
- case RESPONSE_IMMEDIATE_RESOURCE_ALLOCATION:
case RESPONSE_JOB_WILL_RUN:
_pack_resource_allocation_response_msg
((resource_allocation_response_msg_t *) msg->data,
@@ -507,7 +505,6 @@ unpack_msg(slurm_msg_t * msg, Buf buffer)
break;
case REQUEST_RESOURCE_ALLOCATION:
case REQUEST_SUBMIT_BATCH_JOB:
- case REQUEST_IMMEDIATE_RESOURCE_ALLOCATION:
case REQUEST_JOB_WILL_RUN:
case REQUEST_ALLOCATION_AND_RUN_JOB_STEP:
case REQUEST_UPDATE_JOB:
@@ -534,7 +531,6 @@ unpack_msg(slurm_msg_t * msg, Buf buffer)
& (msg->data), buffer);
break;
case RESPONSE_RESOURCE_ALLOCATION:
- case RESPONSE_IMMEDIATE_RESOURCE_ALLOCATION:
case RESPONSE_JOB_WILL_RUN:
rc = _unpack_resource_allocation_response_msg(
(resource_allocation_response_msg_t **)
@@ -1683,6 +1679,7 @@ _pack_job_desc_msg(job_desc_msg_t * job_desc_ptr, Buf buffer)
packstr(job_desc_ptr->out, buffer);
packstr(job_desc_ptr->work_dir, buffer);
+ pack16(job_desc_ptr->immediate, buffer);
pack16(job_desc_ptr->shared, buffer);
pack32(job_desc_ptr->time_limit, buffer);
@@ -1733,6 +1730,7 @@ _unpack_job_desc_msg(job_desc_msg_t ** job_desc_buffer_ptr, Buf buffer)
safe_unpackstr_xmalloc(&job_desc_ptr->out, &uint16_tmp, buffer);
safe_unpackstr_xmalloc(&job_desc_ptr->work_dir, &uint16_tmp, buffer);
+ safe_unpack16(&job_desc_ptr->immediate, buffer);
safe_unpack16(&job_desc_ptr->shared, buffer);
safe_unpack32(&job_desc_ptr->time_limit, buffer);
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index b3b1f1f520c..3246e657c34 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -114,8 +114,7 @@ static void _run_backup(void);
inline static void _save_all_state(void);
static void * _service_connection(void *arg);
static int _shutdown_backup_controller(void);
-inline static void _slurm_rpc_allocate_resources(slurm_msg_t * msg,
- uint8_t immediate);
+inline static void _slurm_rpc_allocate_resources(slurm_msg_t * msg);
inline static void _slurm_rpc_allocate_and_run(slurm_msg_t * msg);
inline static void _slurm_rpc_dump_build(slurm_msg_t * msg);
inline static void _slurm_rpc_dump_nodes(slurm_msg_t * msg);
@@ -692,11 +691,7 @@ static void _slurmctld_req (slurm_msg_t * msg)
slurm_free_last_update_msg(msg->data);
break;
case REQUEST_RESOURCE_ALLOCATION:
- _slurm_rpc_allocate_resources(msg, false);
- slurm_free_job_desc_msg(msg->data);
- break;
- case REQUEST_IMMEDIATE_RESOURCE_ALLOCATION:
- _slurm_rpc_allocate_resources(msg, true);
+ _slurm_rpc_allocate_resources(msg);
slurm_free_job_desc_msg(msg->data);
break;
case REQUEST_ALLOCATION_AND_RUN_JOB_STEP:
@@ -1366,7 +1361,7 @@ static void _slurm_rpc_submit_batch_job(slurm_msg_t * msg)
/* _slurm_rpc_allocate_resources: process RPC to allocate resources for
* a job */
-static void _slurm_rpc_allocate_resources(slurm_msg_t * msg, uint8_t immediate)
+static void _slurm_rpc_allocate_resources(slurm_msg_t * msg)
{
/* init */
int error_code = 0;
@@ -1387,11 +1382,7 @@ static void _slurm_rpc_allocate_resources(slurm_msg_t * msg, uint8_t immediate)
slurm_addr *node_addr;
start_time = clock();
- if (immediate)
- debug
- ("Processing RPC: REQUEST_IMMEDIATE_RESOURCE_ALLOCATION");
- else
- debug("Processing RPC: REQUEST_RESOURCE_ALLOCATION");
+ debug("Processing RPC: REQUEST_RESOURCE_ALLOCATION");
/* do RPC call */
dump_job_desc(job_desc_msg);
@@ -1403,6 +1394,7 @@ static void _slurm_rpc_allocate_resources(slurm_msg_t * msg, uint8_t immediate)
(unsigned int) uid);
}
if (error_code == 0) {
+ int immediate = job_desc_msg->immediate;
lock_slurmctld(job_write_lock);
error_code = job_allocate(job_desc_msg, &job_id,
&node_list_ptr, &num_cpu_groups,
@@ -1432,9 +1424,7 @@ static void _slurm_rpc_allocate_resources(slurm_msg_t * msg, uint8_t immediate)
alloc_msg.cpu_count_reps = cpu_count_reps;
alloc_msg.node_cnt = node_cnt;
alloc_msg.node_addr = node_addr;
- response_msg.msg_type = (immediate) ?
- RESPONSE_IMMEDIATE_RESOURCE_ALLOCATION :
- RESPONSE_RESOURCE_ALLOCATION;
+ response_msg.msg_type = RESPONSE_RESOURCE_ALLOCATION;
response_msg.data = &alloc_msg;
slurm_send_node_msg(msg->conn_fd, &response_msg);
@@ -1481,11 +1471,12 @@ static void _slurm_rpc_allocate_and_run(slurm_msg_t * msg)
}
if (error_code == 0) {
+ int immediate = job_desc_msg->immediate;
lock_slurmctld(job_write_lock);
error_code = job_allocate(job_desc_msg, &job_id,
&node_list_ptr, &num_cpu_groups,
&cpus_per_node, &cpu_count_reps,
- true, false, true, uid,
+ immediate, false, true, uid,
&node_cnt, &node_addr);
}
@@ -1520,22 +1511,22 @@ static void _slurm_rpc_allocate_and_run(slurm_msg_t * msg)
node_list_ptr, job_id, (long) (clock() - start_time));
/* send job_ID and node_name_ptr */
- alloc_msg.job_id = job_id;
- alloc_msg.node_list = node_list_ptr;
+ alloc_msg.job_id = job_id;
+ alloc_msg.node_list = node_list_ptr;
alloc_msg.num_cpu_groups = num_cpu_groups;
- alloc_msg.cpus_per_node = cpus_per_node;
+ alloc_msg.cpus_per_node = cpus_per_node;
alloc_msg.cpu_count_reps = cpu_count_reps;
- alloc_msg.job_step_id = step_rec->step_id;
- alloc_msg.node_cnt = node_cnt;
- alloc_msg.node_addr = node_addr;
- alloc_msg.credentials =
- &step_rec->job_ptr->details->credential;
+ alloc_msg.job_step_id = step_rec->step_id;
+ alloc_msg.node_cnt = node_cnt;
+ alloc_msg.node_addr = node_addr;
+ alloc_msg.credentials =
+ &step_rec->job_ptr->details->credential;
#ifdef HAVE_LIBELAN3
alloc_msg.qsw_job = step_rec->qsw_job;
#endif
unlock_slurmctld(job_write_lock);
response_msg.msg_type =
- RESPONSE_ALLOCATION_AND_RUN_JOB_STEP;
+ RESPONSE_ALLOCATION_AND_RUN_JOB_STEP;
response_msg.data = &alloc_msg;
slurm_send_node_msg(msg->conn_fd, &response_msg);
@@ -1557,8 +1548,7 @@ static void _slurm_rpc_old_job_alloc(slurm_msg_t * msg)
resource_allocation_response_msg_t alloc_msg;
/* Locks: Read job, read node */
slurmctld_lock_t job_read_lock = { NO_LOCK, READ_LOCK,
- READ_LOCK, NO_LOCK
- };
+ READ_LOCK, NO_LOCK };
uint16_t node_cnt;
slurm_addr *node_addr;
uid_t uid;
@@ -1598,18 +1588,17 @@ static void _slurm_rpc_old_job_alloc(slurm_msg_t * msg)
/* send job_ID and node_name_ptr */
- alloc_msg.job_id = job_desc_msg->job_id;
- alloc_msg.node_list = node_list_ptr;
+ alloc_msg.job_id = job_desc_msg->job_id;
+ alloc_msg.node_list = node_list_ptr;
alloc_msg.num_cpu_groups = num_cpu_groups;
- alloc_msg.cpus_per_node = cpus_per_node;
+ alloc_msg.cpus_per_node = cpus_per_node;
alloc_msg.cpu_count_reps = cpu_count_reps;
- alloc_msg.node_cnt = node_cnt;
- alloc_msg.node_addr = node_addr;
- response_msg.msg_type = RESPONSE_RESOURCE_ALLOCATION;
- response_msg.data = &alloc_msg;
+ alloc_msg.node_cnt = node_cnt;
+ alloc_msg.node_addr = node_addr;
+ response_msg.msg_type = RESPONSE_RESOURCE_ALLOCATION;
+ response_msg.data = &alloc_msg;
slurm_send_node_msg(msg->conn_fd, &response_msg);
- (void) dump_all_job_state();
}
}
diff --git a/src/srun/srun.c b/src/srun/srun.c
index fdd28dfe9fd..95659ef7228 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -307,6 +307,7 @@ _allocate_nodes(void)
job.contiguous = opt.contiguous;
job.features = opt.constraints;
+ job.immediate = opt.immediate;
job.name = opt.job_name;
job.req_nodes = opt.nodelist;
job.partition = opt.partition;
@@ -333,7 +334,7 @@ _allocate_nodes(void)
job.shared = 1;
retries = 0;
- while ((rc = slurm_allocate_resources(&job, &resp, opt.immediate))
+ while ((rc = slurm_allocate_resources(&job, &resp))
== SLURM_FAILURE) {
if ((slurm_get_errno() == ESLURM_ERROR_ON_DESC_TO_RECORD_COPY)
&& (retries < MAX_RETRIES)) {
@@ -366,9 +367,10 @@ _allocate_nodes(void)
/* Keep polling until the job is allocated resources */
while (slurm_confirm_allocation(&old_job, &resp) ==
SLURM_FAILURE) {
- if (slurm_get_errno() == ESLURM_JOB_PENDING)
+ if (slurm_get_errno() == ESLURM_JOB_PENDING) {
+ debug3("Still waiting for allocation");
sleep (10);
- else {
+ } else {
error("Unable to confirm resource allocation for job %u: %s",
old_job.job_id, slurm_strerror(errno));
exit (1);
@@ -395,6 +397,7 @@ _alloc_nodes_step(void)
job.contiguous = opt.contiguous;
job.features = opt.constraints;
+ job.immediate = opt.immediate;
job.name = opt.job_name;
job.req_nodes = opt.nodelist;
job.partition = opt.partition;
--
GitLab