From c2444b3d146768de5db43498efe17582f2960b3f Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Wed, 7 Apr 2010 22:31:59 +0000
Subject: [PATCH] In the job_resources structure, If a job changes size
(reliquishes nodes), the node_bitmap will remain unchanged, but cpus,
cpus_used, cpus_array_*, and memory_used will be updated (e.g. cpus and
mem_used on that node cleared).
---
src/common/job_resources.c | 33 +++++-----
src/common/job_resources.h | 10 ++-
src/plugins/select/cons_res/select_cons_res.c | 11 ++--
src/scontrol/update_job.c | 21 +++----
src/slurmctld/job_mgr.c | 7 +--
src/slurmctld/proc_req.c | 63 ++++++++++++++-----
6 files changed, 86 insertions(+), 59 deletions(-)
diff --git a/src/common/job_resources.c b/src/common/job_resources.c
index e5a8f90e2e1..8ce8c82fbeb 100644
--- a/src/common/job_resources.c
+++ b/src/common/job_resources.c
@@ -968,47 +968,46 @@ extern void add_job_to_cores(job_resources_t *job_resrcs_ptr,
}
/* Given a job pointer and a global node index, return the index of that
- * node in the job_resrcs_ptr->cpu_array_value. Return -1 if invalid */
-extern int job_resources_node_inx_to_cpu_array_inx(
- job_resources_t *job_resrcs_ptr, int node_inx)
+ * node in the job_resrcs_ptr->cpus. Return -1 if invalid */
+extern int job_resources_node_inx_to_cpu_inx(job_resources_t *job_resrcs_ptr,
+ int node_inx)
{
- int first_inx, i, node_cnt, node_sum;
+ int first_inx, i, node_offset;
/* Test for error cases */
if (!job_resrcs_ptr || !job_resrcs_ptr->node_bitmap) {
- error("job_resources_node_inx_to_cpu_array_inx: "
+ error("job_resources_node_inx_to_cpu_inx: "
"no job_resrcs or node_bitmap");
return -1;
}
if (!bit_test(job_resrcs_ptr->node_bitmap, node_inx)) {
- error("job_resources_node_inx_to_cpu_array_inx: "
+ error("job_resources_node_inx_to_cpu_inx: "
"Invalid node_inx");
return -1;
}
if (job_resrcs_ptr->cpu_array_cnt == 0) {
- error("job_resources_node_inx_to_cpu_array_inx: "
+ error("job_resources_node_inx_to_cpu_inx: "
"Invalid cpu_array_cnt");
return -1;
}
/* Only one record, no need to search */
- if (job_resrcs_ptr->cpu_array_cnt == 1)
+ if (job_resrcs_ptr->nhosts == 1)
return 0;
/* Scan bitmap, convert node_inx to node_cnt within job's allocation */
first_inx = bit_ffs(job_resrcs_ptr->node_bitmap);
- for (i=first_inx, node_cnt=0; i<node_inx; i++) {
+ for (i=first_inx, node_offset=-1; i<=node_inx; i++) {
if (bit_test(job_resrcs_ptr->node_bitmap, i))
- node_cnt++;
+ node_offset++;
}
- /* if (bit_test(job_resrcs_ptr->node_bitmap, node_inx)) */
- node_cnt++;
- for (i=0, node_sum=0; i<job_resrcs_ptr->cpu_array_cnt; i++) {
- node_sum += job_resrcs_ptr->cpu_array_reps[i];
- if (node_sum >= node_cnt)
- return i;
+ if (node_offset >= job_resrcs_ptr->nhosts) {
+ error("job_resources_node_inx_to_cpu_inx: "
+ "Found %d of %d nodes",
+ job_resrcs_ptr->nhosts, node_offset);
+ return -1;
}
- return -1;
+ return node_offset;
}
diff --git a/src/common/job_resources.h b/src/common/job_resources.h
index 6f812b3fd97..80f70781871 100644
--- a/src/common/job_resources.h
+++ b/src/common/job_resources.h
@@ -100,6 +100,10 @@
* | Sock_0 | Sock_1 | Sock_0 | Sock_1 |
* | Core_0 | Core_1 | Core_0 | Core_1 | Core_0 | Core_1 | Core_0 | Core_1 |
* | Bit_0 | Bit_1 | Bit_2 | Bit_3 | Bit_4 | Bit_5 | Bit_6 | Bit_7 |
+ *
+ * If a job changes size (reliquishes nodes), the node_bitmap will remain
+ * unchanged, but cpus, cpus_used, cpus_array_*, and memory_used will be
+ * updated (e.g. cpus and mem_used on that node cleared).
*/
struct job_resources {
bitstr_t * core_bitmap;
@@ -237,8 +241,8 @@ extern void add_job_to_cores(job_resources_t *job_resrcs_ptr,
const uint32_t *core_rep_count);
/* Given a job pointer and a global node index, return the index of that
- * node in the job_resrcs_ptr->cpu_array_value. Return -1 if invalid */
-extern int job_resources_node_inx_to_cpu_array_inx(
- job_resources_t *job_resrcs_ptr, int node_inx);
+ * node in the job_resrcs_ptr->cpus. Return -1 if invalid */
+extern int job_resources_node_inx_to_cpu_inx(job_resources_t *job_resrcs_ptr,
+ int node_inx);
#endif /* !_JOB_RESOURCES_H */
diff --git a/src/plugins/select/cons_res/select_cons_res.c b/src/plugins/select/cons_res/select_cons_res.c
index a2d5768ef49..014b3d0fbaf 100644
--- a/src/plugins/select/cons_res/select_cons_res.c
+++ b/src/plugins/select/cons_res/select_cons_res.c
@@ -1013,9 +1013,8 @@ static int _rm_job_from_res(struct part_res_record *part_record_ptr,
return SLURM_ERROR;
}
- if (!p_ptr->row) {
+ if (!p_ptr->row)
return SLURM_SUCCESS;
- }
/* remove the job from the job_list */
n = 0;
@@ -1104,6 +1103,11 @@ static int _rm_job_from_one_node(struct job_record *job_ptr,
if (i != node_inx)
continue;
+ if (job->cpus[n] == 0) {
+ info("attempt to remove node %s from job %u again",
+ node_ptr->name, job_ptr->job_id);
+ return SLURM_SUCCESS;
+ }
job->cpus[n] = 0;
job->nprocs = build_job_resources_cpu_array(job);
clear_job_resources_node(job, n);
@@ -1139,9 +1143,8 @@ static int _rm_job_from_one_node(struct job_record *job_ptr,
return SLURM_ERROR;
}
- if (!p_ptr->row) {
+ if (!p_ptr->row)
return SLURM_SUCCESS;
- }
/* look for the job in the partition's job_list */
n = 0;
diff --git a/src/scontrol/update_job.c b/src/scontrol/update_job.c
index 2ea03f68f19..37116d83be7 100644
--- a/src/scontrol/update_job.c
+++ b/src/scontrol/update_job.c
@@ -629,6 +629,9 @@ static void _update_job_size(uint32_t job_id)
char *fname_csh = NULL, *fname_sh = NULL;
FILE *resize_csh = NULL, *resize_sh = NULL;
+ if (!getenv("SLURM_JOBID"))
+ return; /*No job environment here to update */
+
if (slurm_allocation_lookup_lite(job_id, &alloc_info) !=
SLURM_SUCCESS) {
slurm_perror("slurm_allocation_lookup_lite");
@@ -649,8 +652,8 @@ static void _update_job_size(uint32_t job_id)
strerror(errno));
goto fini;
}
- chmod(fname_csh, 0500); /* Make file executable */
- chmod(fname_sh, 0500);
+ chmod(fname_csh, 0700); /* Make file executable */
+ chmod(fname_sh, 0700);
if (getenv("SLURM_NODELIST")) {
fprintf(resize_sh, "export SLURM_NODELIST=\"%s\"\n",
@@ -677,15 +680,6 @@ static void _update_job_size(uint32_t job_id)
alloc_info->node_cnt);
}
if (getenv("SLURM_JOB_CPUS_PER_NODE")) {
-#if 1
- fprintf(resize_sh, "unset SLURM_JOB_CPUS_PER_NODE\n");
- fprintf(resize_csh, "unsetenv SLURM_JOB_CPUS_PER_NODE\n");
-#else
- /* The job resource structure is currently based upon the
- * original job allocation, so we don't have sufficient
- * information to recreate this environment variable today.
- * This is a possible future enhancement if/when job resource
- * information is corrected after job resize */
char *tmp;
tmp = uint32_compressed_to_str(alloc_info->num_cpu_groups,
alloc_info->cpus_per_node,
@@ -695,7 +689,6 @@ static void _update_job_size(uint32_t job_id)
fprintf(resize_csh, "setenv SLURM_JOB_CPUS_PER_NODE \"%s\"\n",
tmp);
xfree(tmp);
-#endif
}
if (getenv("SLURM_TASKS_PER_NODE")) {
/* We don't have sufficient information to recreate this */
@@ -704,8 +697,8 @@ static void _update_job_size(uint32_t job_id)
}
printf("To reset SLURM environment variables, execute\n");
- printf(" For bash or sh shells: . ./%s\n", fname_sh);
- printf(" For csh shells: source ./%s\n", fname_csh);
+ printf(" For bash or sh shells: . ./%s\n", fname_sh);
+ printf(" For csh shells: source ./%s\n", fname_csh);
fini: slurm_free_resource_allocation_response_msg(alloc_info);
xfree(fname_csh);
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index f80a2cd41ea..a0038894d02 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -4147,18 +4147,15 @@ extern int job_update_cpu_cnt(struct job_record *job_ptr, int node_inx)
* cpu count isn't set up on that system. */
return SLURM_SUCCESS;
#endif
- if((offset = job_resources_node_inx_to_cpu_array_inx(
+ if ((offset = job_resources_node_inx_to_cpu_inx(
job_ptr->job_resrcs, node_inx)) < 0) {
error("job_update_cpu_cnt: problem getting offset of job %u",
job_ptr->job_id);
job_ptr->cpu_cnt = 0;
return SLURM_ERROR;
}
- /* info("cpu for %d is %d out of %d", */
- /* node_inx, job_ptr->job_resrcs->cpu_array_value[offset], */
- /* job_ptr->cpu_cnt); */
- cnt = job_ptr->job_resrcs->cpu_array_value[offset];
+ cnt = job_ptr->job_resrcs->cpus[offset];
if (cnt > job_ptr->cpu_cnt) {
error("job_update_cpu_cnt: cpu_cnt underflow on job_id %u",
job_ptr->job_id);
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 824c4240165..624ffc7f7f7 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -1743,7 +1743,7 @@ static void _slurm_rpc_job_alloc_info(slurm_msg_t * msg)
on existing job */
static void _slurm_rpc_job_alloc_info_lite(slurm_msg_t * msg)
{
- int error_code = SLURM_SUCCESS;
+ int error_code = SLURM_SUCCESS, i, j;
slurm_msg_t response_msg;
struct job_record *job_ptr;
DEF_TIMERS;
@@ -1777,21 +1777,52 @@ static void _slurm_rpc_job_alloc_info_lite(slurm_msg_t * msg)
debug("_slurm_rpc_job_alloc_info_lite JobId=%u NodeList=%s %s",
job_info_msg->job_id, job_ptr->nodes, TIME_STR);
- /* send job_ID and node_name_ptr */
- job_info_resp_msg.num_cpu_groups = job_ptr->job_resrcs->
- cpu_array_cnt;
- job_info_resp_msg.cpu_count_reps =
- xmalloc(sizeof(uint32_t) *
- job_ptr->job_resrcs->cpu_array_cnt);
- memcpy(job_info_resp_msg.cpu_count_reps,
- job_ptr->job_resrcs->cpu_array_reps,
- (sizeof(uint32_t) * job_ptr->job_resrcs->cpu_array_cnt));
- job_info_resp_msg.cpus_per_node =
- xmalloc(sizeof(uint16_t) *
- job_ptr->job_resrcs->cpu_array_cnt);
- memcpy(job_info_resp_msg.cpus_per_node,
- job_ptr->job_resrcs->cpu_array_value,
- (sizeof(uint16_t) * job_ptr->job_resrcs->cpu_array_cnt));
+ /* send job_ID and node_name_ptr */
+ if (bit_equal(job_ptr->node_bitmap,
+ job_ptr->job_resrcs->node_bitmap)) {
+ job_info_resp_msg.num_cpu_groups = job_ptr->job_resrcs->
+ cpu_array_cnt;
+ job_info_resp_msg.cpu_count_reps =
+ xmalloc(sizeof(uint32_t) *
+ job_ptr->job_resrcs->
+ cpu_array_cnt);
+ memcpy(job_info_resp_msg.cpu_count_reps,
+ job_ptr->job_resrcs->cpu_array_reps,
+ (sizeof(uint32_t) *
+ job_ptr->job_resrcs->cpu_array_cnt));
+ job_info_resp_msg.cpus_per_node =
+ xmalloc(sizeof(uint16_t) *
+ job_ptr->job_resrcs->
+ cpu_array_cnt);
+ memcpy(job_info_resp_msg.cpus_per_node,
+ job_ptr->job_resrcs->cpu_array_value,
+ (sizeof(uint16_t) *
+ job_ptr->job_resrcs->cpu_array_cnt));
+ } else {
+ /* Job has changed size, rebuild CPU count info */
+ job_info_resp_msg.num_cpu_groups = job_ptr->node_cnt;
+ job_info_resp_msg.cpu_count_reps =
+ xmalloc(sizeof(uint32_t) *
+ job_ptr->node_cnt);
+ job_info_resp_msg.cpus_per_node =
+ xmalloc(sizeof(uint32_t) *
+ job_ptr->node_cnt);
+ for (i=0, j=-1; i<job_ptr->job_resrcs->nhosts; i++) {
+ if (job_ptr->job_resrcs->cpus[i] == 0)
+ continue;
+ if ((j == -1) ||
+ (job_info_resp_msg.cpus_per_node[j] !=
+ job_ptr->job_resrcs->cpus[i])) {
+ j++;
+ job_info_resp_msg.cpus_per_node[j] =
+ job_ptr->job_resrcs->cpus[i];
+ job_info_resp_msg.cpu_count_reps[j] = 1;
+ } else {
+ job_info_resp_msg.cpu_count_reps[j]++;
+ }
+ }
+ job_info_resp_msg.num_cpu_groups = j + 1;
+ }
job_info_resp_msg.error_code = error_code;
job_info_resp_msg.job_id = job_info_msg->job_id;
job_info_resp_msg.node_cnt = job_ptr->node_cnt;
--
GitLab