From 77d5598ab16d419a82585fa82b8664af2eeb9fb7 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@llnl.gov>
Date: Fri, 4 Jan 2008 20:37:49 +0000
Subject: [PATCH] mods to the bluegene plugin to include the job_ptr from the
controller
---
slurm/slurm.h.in | 3 -
src/common/node_select.c | 32 ----
src/common/node_select.h | 7 -
.../select/bluegene/plugin/bg_job_place.c | 156 +++++++++---------
.../select/bluegene/plugin/bg_job_run.c | 49 +++---
src/plugins/select/bluegene/plugin/bluegene.c | 2 +-
src/plugins/select/bluegene/plugin/bluegene.h | 9 +-
.../select/bluegene/plugin/select_bluegene.c | 43 +----
src/plugins/select/cons_res/select_cons_res.c | 5 -
src/plugins/select/linear/select_linear.c | 5 -
10 files changed, 116 insertions(+), 195 deletions(-)
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index 518c9d21320..e85be2a79c4 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -271,8 +271,6 @@ enum select_data_type {
SELECT_DATA_MLOADER_IMAGE,/* data-> char *mloaderimage */
SELECT_DATA_RAMDISK_IMAGE,/* data-> char *ramdiskimage */
SELECT_DATA_REBOOT, /* data-> uint16_t reboot */
- SELECT_DATA_EST_START /* data-> time_t etimated time job
- * will start*/
};
enum select_print_mode {
@@ -291,7 +289,6 @@ enum select_print_mode {
SELECT_PRINT_MLOADER_IMAGE,/* Print just the MLOADER IMAGE */
SELECT_PRINT_RAMDISK_IMAGE,/* Print just the RAMDISK IMAGE */
SELECT_PRINT_REBOOT, /* Print just the REBOOT */
- SELECT_PRINT_EST_START /* Print just the EST_START */
};
enum select_node_cnt {
diff --git a/src/common/node_select.c b/src/common/node_select.c
index c912edac254..69874881b4f 100644
--- a/src/common/node_select.c
+++ b/src/common/node_select.c
@@ -85,7 +85,6 @@ typedef struct slurm_select_ops {
int (*job_begin) (struct job_record *job_ptr);
int (*job_ready) (struct job_record *job_ptr);
int (*job_fini) (struct job_record *job_ptr);
- int (*job_update_end_time) (struct job_record *job_ptr);
int (*job_suspend) (struct job_record *job_ptr);
int (*job_resume) (struct job_record *job_ptr);
int (*pack_node_info) (time_t last_query_time,
@@ -145,7 +144,6 @@ struct select_jobinfo {
char *linuximage; /* LinuxImage for this block */
char *mloaderimage; /* mloaderImage for this block */
char *ramdiskimage; /* RamDiskImage for this block */
- time_t est_job_start; /* Estimated start time of job */
};
#endif
@@ -174,7 +172,6 @@ static slurm_select_ops_t * _select_get_ops(slurm_select_context_t *c)
"select_p_job_begin",
"select_p_job_ready",
"select_p_job_fini",
- "select_p_job_update_end_time",
"select_p_job_suspend",
"select_p_job_resume",
"select_p_pack_node_info",
@@ -580,19 +577,6 @@ extern int select_g_job_fini(struct job_record *job_ptr)
return (*(g_select_context->ops.job_fini))(job_ptr);
}
-/*
- * Suspend a job. Executed from slurmctld.
- * IN job_ptr - pointer to job being suspended
- * RET SLURM_SUCCESS or error code
- */
-extern int select_g_job_update_end_time(struct job_record *job_ptr)
-{
- if (slurm_select_init() < 0)
- return SLURM_ERROR;
-
- return (*(g_select_context->ops.job_update_end_time))(job_ptr);
-}
-
/*
* Suspend a job. Executed from slurmctld.
* IN job_ptr - pointer to job being suspended
@@ -767,7 +751,6 @@ extern int select_g_set_jobinfo (select_jobinfo_t jobinfo,
int i, rc = SLURM_SUCCESS;
uint16_t *uint16 = (uint16_t *) data;
uint32_t *uint32 = (uint32_t *) data;
- time_t *time = (time_t *) data;
char *tmp_char = (char *) data;
if (jobinfo == NULL) {
@@ -839,9 +822,6 @@ extern int select_g_set_jobinfo (select_jobinfo_t jobinfo,
xfree(jobinfo->ramdiskimage);
jobinfo->ramdiskimage = xstrdup(tmp_char);
break;
- case SELECT_DATA_EST_START:
- jobinfo->est_job_start = *time;
- break;
default:
debug("select_g_set_jobinfo data_type %d invalid",
data_type);
@@ -862,7 +842,6 @@ extern int select_g_get_jobinfo (select_jobinfo_t jobinfo,
int i, rc = SLURM_SUCCESS;
uint16_t *uint16 = (uint16_t *) data;
uint32_t *uint32 = (uint32_t *) data;
- time_t *time = (time_t *) data;
char **tmp_char = (char **) data;
if (jobinfo == NULL) {
@@ -952,9 +931,6 @@ extern int select_g_get_jobinfo (select_jobinfo_t jobinfo,
else
*tmp_char = xstrdup(jobinfo->ramdiskimage);
break;
- case SELECT_DATA_EST_START:
- *time = jobinfo->est_job_start;
- break;
default:
debug("select_g_get_jobinfo data_type %d invalid",
data_type);
@@ -999,8 +975,6 @@ extern select_jobinfo_t select_g_copy_jobinfo(select_jobinfo_t jobinfo)
rc->linuximage = xstrdup(jobinfo->linuximage);
rc->mloaderimage = xstrdup(jobinfo->mloaderimage);
rc->ramdiskimage = xstrdup(jobinfo->ramdiskimage);
- rc->est_job_start = jobinfo->est_job_start;
-
}
return rc;
@@ -1063,7 +1037,6 @@ extern int select_g_pack_jobinfo (select_jobinfo_t jobinfo, Buf buffer)
packstr(jobinfo->linuximage, buffer);
packstr(jobinfo->mloaderimage, buffer);
packstr(jobinfo->ramdiskimage, buffer);
- pack_time(jobinfo->est_job_start, buffer);
} else {
/* pack space for 3 positions for start and for geo
* then 1 for conn_type, reboot, and rotate
@@ -1081,7 +1054,6 @@ extern int select_g_pack_jobinfo (select_jobinfo_t jobinfo, Buf buffer)
packstr("", buffer); //linux
packstr("", buffer); //mloader
packstr("", buffer); //ramdisk
- pack_time(0, buffer); //est_job_start
}
return SLURM_SUCCESS;
@@ -1116,7 +1088,6 @@ extern int select_g_unpack_jobinfo(select_jobinfo_t jobinfo, Buf buffer)
safe_unpackstr_xmalloc(&(jobinfo->linuximage), &uint32_tmp, buffer);
safe_unpackstr_xmalloc(&(jobinfo->mloaderimage), &uint32_tmp, buffer);
safe_unpackstr_xmalloc(&(jobinfo->ramdiskimage), &uint32_tmp, buffer);
- safe_unpack_time(&(jobinfo->est_job_start), buffer);
return SLURM_SUCCESS;
@@ -1280,9 +1251,6 @@ extern char *select_g_sprint_jobinfo(select_jobinfo_t jobinfo,
tmp_image = jobinfo->ramdiskimage;
snprintf(buf, size, "%s", tmp_image);
break;
- case SELECT_PRINT_EST_START:
- snprintf(buf, size, "%u", (int)jobinfo->est_job_start);
- break;
default:
error("select_g_sprint_jobinfo: bad mode %d", mode);
if (size > 0)
diff --git a/src/common/node_select.h b/src/common/node_select.h
index d57e578c0c4..59e70e52b23 100644
--- a/src/common/node_select.h
+++ b/src/common/node_select.h
@@ -194,13 +194,6 @@ extern int select_g_job_ready(struct job_record *job_ptr);
*/
extern int select_g_job_fini(struct job_record *job_ptr);
-/*
- * To be called when the end time of a job is changed.
- * IN job_ptr - pointer to job being suspended
- * RET SLURM_SUCCESS or error code
- */
-extern int select_p_job_update_end_time(struct job_record *job_ptr);
-
/*
* Suspend a job. Executed from slurmctld.
* IN job_ptr - pointer to job being suspended
diff --git a/src/plugins/select/bluegene/plugin/bg_job_place.c b/src/plugins/select/bluegene/plugin/bg_job_place.c
index 00277f0bb98..dd815433c7d 100644
--- a/src/plugins/select/bluegene/plugin/bg_job_place.c
+++ b/src/plugins/select/bluegene/plugin/bg_job_place.c
@@ -114,7 +114,7 @@ static void _rotate_geo(uint16_t *req_geometry, int rot_cnt)
/*
* Comparator used for sorting blocks smallest to largest
*
- * returns: -1: rec_a >rec_b 0: rec_a == rec_b 1: rec_a < rec_b
+ * returns: -1: rec_a < rec_b 0: rec_a == rec_b 1: rec_a > rec_b
*
*/
static int _bg_record_sort_aval_inc(bg_record_t* rec_a, bg_record_t* rec_b)
@@ -122,10 +122,16 @@ static int _bg_record_sort_aval_inc(bg_record_t* rec_a, bg_record_t* rec_b)
int size_a = rec_a->node_cnt;
int size_b = rec_b->node_cnt;
- if(rec_a->est_job_end < rec_b->est_job_end)
+ if(rec_a->job_ptr && !rec_b->job_ptr)
return -1;
- else if(rec_a->est_job_end > rec_b->est_job_end)
+ else if(!rec_a->job_ptr && rec_b->job_ptr)
return 1;
+ else if(rec_a->job_ptr && rec_b->job_ptr) {
+ if(rec_a->job_ptr->start_time > rec_b->job_ptr->start_time)
+ return 1;
+ else if(rec_a->job_ptr->start_time < rec_b->job_ptr->start_time)
+ return -1;
+ }
if (size_a < size_b)
return -1;
@@ -162,10 +168,16 @@ static int _bg_record_sort_aval_dec(bg_record_t* rec_a, bg_record_t* rec_b)
int size_a = rec_a->node_cnt;
int size_b = rec_b->node_cnt;
- if(rec_a->est_job_end > rec_b->est_job_end)
- return -1;
- else if(rec_a->est_job_end < rec_b->est_job_end)
+ if(rec_a->job_ptr && !rec_b->job_ptr)
return 1;
+ else if(!rec_a->job_ptr && rec_b->job_ptr)
+ return -1;
+ else if(rec_a->job_ptr && rec_b->job_ptr) {
+ if(rec_a->job_ptr->start_time > rec_b->job_ptr->start_time)
+ return -1;
+ else if(rec_a->job_ptr->start_time < rec_b->job_ptr->start_time)
+ return 1;
+ }
if (size_a < size_b)
return -1;
@@ -1041,7 +1053,7 @@ static int _find_best_block_match(List block_list,
debug2("taking off %d(%s) ends at %d",
bg_record->job_running,
bg_record->bg_block_id,
- bg_record->est_job_end);
+ bg_record->job_ptr->start_time);
if(!(new_blocks = create_dynamic_block(
block_list, &request, job_list))) {
destroy_bg_record(bg_record);
@@ -1056,8 +1068,8 @@ static int _find_best_block_match(List block_list,
rc = SLURM_SUCCESS;
(*found_bg_record) = list_pop(new_blocks);
if(bg_record) {
- (*found_bg_record)->est_job_end
- = bg_record->est_job_end;
+ (*found_bg_record)->job_ptr
+ = bg_record->job_ptr;
destroy_bg_record(bg_record);
}
list_destroy(new_blocks);
@@ -1138,12 +1150,12 @@ extern int submit_job(struct job_record *job_ptr, bitstr_t *slurm_block_bitmap,
bg_record_t* bg_record = NULL;
char buf[100];
int i, rc = SLURM_SUCCESS;
- uint16_t geo[BA_SYSTEM_DIMENSIONS];
uint16_t tmp16 = (uint16_t)NO_VAL;
List block_list = NULL;
int block_list_count = 0;
int blocks_added = 0;
-
+ int starttime = time(NULL);
+
if(bluegene_layout_mode == LAYOUT_DYNAMIC)
slurm_mutex_lock(&create_dynamic_mutex);
@@ -1177,85 +1189,73 @@ extern int submit_job(struct job_record *job_ptr, bitstr_t *slurm_block_bitmap,
&bg_record, test_only);
if(rc == SLURM_SUCCESS) {
- if(bg_record && !bg_record->bg_block_id) {
- //ListIterator itr = NULL;
-
- debug2("%d can start job at %u on %s on unmade block",
- test_only, bg_record->est_job_end,
- bg_record->nodes);
- select_g_set_jobinfo(job_ptr->select_jobinfo,
+ if(bg_record) {
+ if(bg_record->job_ptr && bg_record->job_ptr->end_time)
+ starttime = bg_record->job_ptr->end_time;
+
+ job_ptr->start_time = starttime;
+
+ if(!bg_record->bg_block_id) {
+ uint16_t geo[BA_SYSTEM_DIMENSIONS];
+
+ debug2("%d can start job at "
+ "%u on %s on unmade block",
+ test_only, starttime,
+ bg_record->nodes);
+ select_g_set_jobinfo(job_ptr->select_jobinfo,
SELECT_DATA_BLOCK_ID,
"unassigned");
- if(job_ptr->num_procs < bluegene_bp_node_cnt
- && job_ptr->num_procs > 0) {
- i = procs_per_node/job_ptr->num_procs;
- debug2("divide by %d", i);
- } else
- i = 1;
- min_nodes *= bluegene_bp_node_cnt/i;
-
- /* itr = list_iterator_create(block_list); */
-/* while(bg_record = list_next(itr)) { */
-
-/* } */
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_EST_START,
- &bg_record->est_job_end);
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_NODES,
- bg_record->nodes);
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_IONODES,
- bg_record->ionodes);
-
- select_g_set_jobinfo(job_ptr->select_jobinfo,
+ if(job_ptr->num_procs < bluegene_bp_node_cnt
+ && job_ptr->num_procs > 0) {
+ i = procs_per_node/job_ptr->num_procs;
+ debug2("divide by %d", i);
+ } else
+ i = 1;
+ min_nodes *= bluegene_bp_node_cnt/i;
+ select_g_set_jobinfo(job_ptr->select_jobinfo,
SELECT_DATA_NODE_CNT,
&min_nodes);
+ memset(geo, 0,
+ sizeof(uint16_t) * BA_SYSTEM_DIMENSIONS);
+ select_g_set_jobinfo(job_ptr->select_jobinfo,
+ SELECT_DATA_GEOMETRY,
+ &geo);
+ } else {
+ if((bg_record->ionodes)
+ && (job_ptr->part_ptr->max_share <= 1))
+ error("Small block used in "
+ "non-shared partition");
+
+ debug2("%d can start job at %u on %s",
+ test_only, starttime,
+ bg_record->nodes);
+
+ select_g_set_jobinfo(job_ptr->select_jobinfo,
+ SELECT_DATA_BLOCK_ID,
+ bg_record->bg_block_id);
+ select_g_set_jobinfo(job_ptr->select_jobinfo,
+ SELECT_DATA_NODE_CNT,
+ &bg_record->node_cnt);
+ select_g_set_jobinfo(job_ptr->select_jobinfo,
+ SELECT_DATA_GEOMETRY,
+ &bg_record->geo);
+
+ tmp16 = bg_record->conn_type;
+ select_g_set_jobinfo(job_ptr->select_jobinfo,
+ SELECT_DATA_CONN_TYPE,
+ &tmp16);
+ }
- memset(geo, 0, sizeof(uint16_t) * BA_SYSTEM_DIMENSIONS);
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_GEOMETRY,
- &geo);
- } else if(bg_record) {
- if((bg_record->ionodes)
- && (job_ptr->part_ptr->max_share <= 1))
- error("Small block used in "
- "non-shared partition");
- debug2("%d can start job at %u on %s",
- test_only, bg_record->est_job_end,
- bg_record->nodes);
- /* set the block id and info about block */
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_NODE_CNT,
- &bg_record->node_cnt);
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_GEOMETRY,
- &bg_record->geo);
-
select_g_set_jobinfo(job_ptr->select_jobinfo,
SELECT_DATA_NODES,
bg_record->nodes);
select_g_set_jobinfo(job_ptr->select_jobinfo,
SELECT_DATA_IONODES,
bg_record->ionodes);
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_EST_START,
- &bg_record->est_job_end);
- tmp16 = bg_record->conn_type;
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_CONN_TYPE,
- &tmp16);
- if(test_only) {
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_BLOCK_ID,
- "unassigned");
- } else {
- select_g_set_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_BLOCK_ID,
- bg_record->bg_block_id);
- bg_record->job_running = job_ptr->job_id;
- }
+
+ } else {
+ error("we got a success, but no block back");
}
}
diff --git a/src/plugins/select/bluegene/plugin/bg_job_run.c b/src/plugins/select/bluegene/plugin/bg_job_run.c
index bac6f84dd76..0393c765988 100644
--- a/src/plugins/select/bluegene/plugin/bg_job_run.c
+++ b/src/plugins/select/bluegene/plugin/bg_job_run.c
@@ -70,9 +70,8 @@ enum update_op {START_OP, TERM_OP, SYNC_OP};
typedef struct bg_update {
enum update_op op; /* start | terminate | sync */
- uid_t uid; /* new user */
- uint32_t job_id; /* SLURM job id */
- time_t end_time; /* estimated time job will end */
+ struct job_record *job_ptr; /* pointer to job running on
+ * block or NULL if no job */
uint16_t reboot; /* reboot block before starting job */
pm_partition_id_t bg_block_id;
char *blrtsimage; /* BlrtsImage for this block */
@@ -212,8 +211,10 @@ static void _sync_agent(bg_update_t *bg_update_ptr)
return;
}
slurm_mutex_lock(&block_state_mutex);
- bg_record->job_running = bg_update_ptr->job_id;
- bg_record->est_job_end = bg_update_ptr->end_time;
+
+ bg_record->job_running = bg_update_ptr->job_ptr->job_id;
+ bg_record->job_ptr = bg_update_ptr->job_ptr;
+
if(!block_exist_in_list(bg_job_block_list, bg_record)) {
list_push(bg_job_block_list, bg_record);
num_unused_cpus -= bg_record->bp_count*bg_record->cpus_per_bp;
@@ -223,17 +224,17 @@ static void _sync_agent(bg_update_t *bg_update_ptr)
slurm_mutex_unlock(&block_state_mutex);
if(bg_record->state == RM_PARTITION_READY) {
- if(bg_record->user_uid != bg_update_ptr->uid) {
+ if(bg_record->user_uid != bg_update_ptr->job_ptr->user_id) {
int set_user_rc = SLURM_SUCCESS;
slurm_mutex_lock(&block_state_mutex);
debug("User isn't correct for job %d on %s, "
"fixing...",
- bg_update_ptr->job_id,
+ bg_update_ptr->job_ptr->job_id,
bg_update_ptr->bg_block_id);
xfree(bg_record->target_name);
- bg_record->target_name =
- xstrdup(uid_to_string(bg_update_ptr->uid));
+ bg_record->target_name = xstrdup(
+ uid_to_string(bg_update_ptr->job_ptr->user_id));
set_user_rc = set_block_user(bg_record);
slurm_mutex_unlock(&block_state_mutex);
@@ -270,7 +271,7 @@ static void _start_agent(bg_update_t *bg_update_ptr)
if(!bg_record) {
error("block %s not found in bg_list",
bg_update_ptr->bg_block_id);
- (void) slurm_fail_job(bg_update_ptr->job_id);
+ (void) slurm_fail_job(bg_update_ptr->job_ptr->job_id);
slurm_mutex_unlock(&job_start_mutex);
return;
}
@@ -280,7 +281,7 @@ static void _start_agent(bg_update_t *bg_update_ptr)
slurm_mutex_unlock(&job_start_mutex);
debug("job %d finished during the queueing job "
"(everything is ok)",
- bg_update_ptr->job_id);
+ bg_update_ptr->job_ptr->job_id);
return;
}
if(bg_record->state == RM_PARTITION_DEALLOCATING) {
@@ -337,7 +338,7 @@ static void _start_agent(bg_update_t *bg_update_ptr)
slurm_mutex_unlock(&block_state_mutex);
slurm_mutex_unlock(&job_start_mutex);
debug("job %d already finished before boot",
- bg_update_ptr->job_id);
+ bg_update_ptr->job_ptr->job_id);
return;
}
@@ -427,7 +428,7 @@ static void _start_agent(bg_update_t *bg_update_ptr)
is a no-op if issued prior
to the script initiation do clean up just
incase the fail job isn't ran */
- (void) slurm_fail_job(bg_update_ptr->job_id);
+ (void) slurm_fail_job(bg_update_ptr->job_ptr->job_id);
slurm_mutex_lock(&block_state_mutex);
if (remove_from_bg_list(bg_job_block_list, bg_record)
== SLURM_SUCCESS) {
@@ -446,14 +447,15 @@ static void _start_agent(bg_update_t *bg_update_ptr)
slurm_mutex_unlock(&job_start_mutex);
debug("job %d finished during the start of the boot "
"(everything is ok)",
- bg_update_ptr->job_id);
+ bg_update_ptr->job_ptr->job_id);
return;
}
slurm_mutex_lock(&block_state_mutex);
bg_record->boot_count = 0;
xfree(bg_record->target_name);
- bg_record->target_name = xstrdup(uid_to_string(bg_update_ptr->uid));
+ bg_record->target_name = xstrdup(
+ uid_to_string(bg_update_ptr->job_ptr->user_id));
debug("setting the target_name for Block %s to %s",
bg_record->bg_block_id,
bg_record->target_name);
@@ -612,7 +614,7 @@ static void _term_agent(bg_update_t *bg_update_ptr)
slurm_mutex_lock(&block_state_mutex);
if(bg_record->job_running > NO_JOB_RUNNING) {
bg_record->job_running = NO_JOB_RUNNING;
- bg_record->est_job_end = 0;
+ bg_record->job_ptr = NULL;
}
/* remove user from list */
@@ -847,9 +849,7 @@ extern int start_job(struct job_record *job_ptr)
bg_update_ptr = xmalloc(sizeof(bg_update_t));
bg_update_ptr->op = START_OP;
- bg_update_ptr->uid = job_ptr->user_id;
- bg_update_ptr->job_id = job_ptr->job_id;
- bg_update_ptr->end_time = job_ptr->end_time;
+ bg_update_ptr->job_ptr = job_ptr;
select_g_get_jobinfo(job_ptr->select_jobinfo,
SELECT_DATA_BLOCK_ID,
@@ -899,8 +899,8 @@ extern int start_job(struct job_record *job_ptr)
slurm_mutex_lock(&block_state_mutex);
job_ptr->num_procs = (bg_record->cpus_per_bp *
bg_record->bp_count);
- bg_record->job_running = bg_update_ptr->job_id;
- bg_record->est_job_end = bg_update_ptr->end_time;
+ bg_record->job_running = bg_update_ptr->job_ptr->job_id;
+ bg_record->job_ptr = bg_update_ptr->job_ptr;
if(!block_exist_in_list(bg_job_block_list, bg_record)) {
list_push(bg_job_block_list, bg_record);
num_unused_cpus -=
@@ -939,8 +939,7 @@ int term_job(struct job_record *job_ptr)
bg_update_ptr = xmalloc(sizeof(bg_update_t));
bg_update_ptr->op = TERM_OP;
- bg_update_ptr->uid = job_ptr->user_id;
- bg_update_ptr->job_id = job_ptr->job_id;
+ bg_update_ptr->job_ptr = job_ptr;
select_g_get_jobinfo(job_ptr->select_jobinfo,
SELECT_DATA_BLOCK_ID,
&(bg_update_ptr->bg_block_id));
@@ -1030,9 +1029,7 @@ extern int sync_jobs(List job_list)
bg_update_ptr->bg_block_id,
job_ptr->end_time);
bg_update_ptr->op = SYNC_OP;
- bg_update_ptr->uid = job_ptr->user_id;
- bg_update_ptr->job_id = job_ptr->job_id;
- bg_update_ptr->end_time = job_ptr->end_time;
+ bg_update_ptr->job_ptr = job_ptr;
_block_op(bg_update_ptr);
}
list_iterator_destroy(job_iterator);
diff --git a/src/plugins/select/bluegene/plugin/bluegene.c b/src/plugins/select/bluegene/plugin/bluegene.c
index c44a405c103..1a61b7bfb49 100644
--- a/src/plugins/select/bluegene/plugin/bluegene.c
+++ b/src/plugins/select/bluegene/plugin/bluegene.c
@@ -567,7 +567,7 @@ extern void copy_bg_record(bg_record_t *fir_record, bg_record_t *sec_record)
list_iterator_destroy(itr);
}
sec_record->job_running = fir_record->job_running;
- sec_record->est_job_end = fir_record->est_job_end;
+ sec_record->job_ptr = fir_record->job_ptr;
sec_record->cpus_per_bp = fir_record->cpus_per_bp;
sec_record->node_cnt = fir_record->node_cnt;
sec_record->quarter = fir_record->quarter;
diff --git a/src/plugins/select/bluegene/plugin/bluegene.h b/src/plugins/select/bluegene/plugin/bluegene.h
index 45f123fc051..1a1d11493ba 100644
--- a/src/plugins/select/bluegene/plugin/bluegene.h
+++ b/src/plugins/select/bluegene/plugin/bluegene.h
@@ -98,10 +98,11 @@ typedef struct bg_record {
bitstr_t *ionode_bitmap; /* for small blocks bitmap to
keep track which ionodes we
are on. NULL if not a small block*/
- int job_running; /* job id if there is a job running
- on the block */
- time_t est_job_end; /* if job_running time job is
- estimated to end */
+ struct job_record *job_ptr; /* pointer to job running on
+ * block or NULL if no job */
+ int job_running; /* job id of job running of if
+ * block is in an error state
+ * BLOCK_ERROR_STATE */
int cpus_per_bp; /* count of cpus per base part */
uint32_t node_cnt; /* count of nodes per block */
uint16_t quarter; /* used for small blocks
diff --git a/src/plugins/select/bluegene/plugin/select_bluegene.c b/src/plugins/select/bluegene/plugin/select_bluegene.c
index 0aa2fdbdb06..a18bcd8d026 100644
--- a/src/plugins/select/bluegene/plugin/select_bluegene.c
+++ b/src/plugins/select/bluegene/plugin/select_bluegene.c
@@ -618,7 +618,8 @@ extern int select_p_node_init(struct node_record *node_ptr, int node_cnt)
* identify the nodes which "best" satify the request. The specified
* nodes may be DOWN or BUSY at the time of this test as may be used
* to deterime if a job could ever run.
- * IN job_ptr - pointer to job being scheduled
+ * IN/OUT job_ptr - pointer to job being scheduled start_time is set
+ * when we can possibly start job.
* IN/OUT bitmap - usable nodes are set on input, nodes not required to
* satisfy the request are cleared, other left set
* IN min_nodes - minimum count of nodes
@@ -632,19 +633,19 @@ extern int select_p_node_init(struct node_record *node_ptr, int node_cnt)
* select_p_job_test is called
*/
extern int select_p_job_test(struct job_record *job_ptr, bitstr_t *bitmap,
- uint32_t min_nodes, uint32_t max_nodes,
- uint32_t req_nodes, int mode)
+ uint32_t min_nodes, uint32_t max_nodes,
+ uint32_t req_nodes, int mode)
{
bool test_only;
- if (mode == SELECT_MODE_TEST_ONLY)
+ if (mode == SELECT_MODE_TEST_ONLY || mode == SELECT_MODE_WILL_RUN)
test_only = true;
- else if (mode == SELECT_MODE_TEST_ONLY)
+ else if (mode == SELECT_MODE_RUN_NOW)
test_only = false;
- else /* SELECT_MODE_WILL_RUN */
- return EINVAL; /* not yet supported */
+ else
+ return EINVAL; /* something not yet supported */
- /* bg block test - is there a block where we have:
+ /* submit_job - is there a block where we have:
* 1) geometry requested
* 2) min/max nodes (BPs) requested
* 3) type: TORUS or MESH or NAV (torus else mesh)
@@ -667,32 +668,6 @@ extern int select_p_job_fini(struct job_record *job_ptr)
return term_job(job_ptr);
}
-extern int select_p_job_update_end_time(struct job_record *job_ptr)
-{
- int rc = SLURM_ERROR;
- char *block_id = NULL;
- bg_record_t *bg_record = NULL;
-
- rc = select_g_get_jobinfo(job_ptr->select_jobinfo,
- SELECT_DATA_BLOCK_ID, &block_id);
- if (rc == SLURM_SUCCESS) {
- bg_record = find_bg_record_in_list(bg_list, block_id);
- slurm_mutex_lock(&block_state_mutex);
-
- if(bg_record) {
- if(bg_record->job_running != job_ptr->job_id)
- rc = SLURM_ERROR;
- else
- bg_record->est_job_end = job_ptr->end_time;
- } else
- rc = SLURM_ERROR;
- slurm_mutex_unlock(&block_state_mutex);
- xfree(block_id);
- }
-
- return rc;
-}
-
extern int select_p_job_suspend(struct job_record *job_ptr)
{
return ESLURM_NOT_SUPPORTED;
diff --git a/src/plugins/select/cons_res/select_cons_res.c b/src/plugins/select/cons_res/select_cons_res.c
index d65a4442090..e5b01afacf1 100644
--- a/src/plugins/select/cons_res/select_cons_res.c
+++ b/src/plugins/select/cons_res/select_cons_res.c
@@ -2680,11 +2680,6 @@ extern int select_p_job_fini(struct job_record *job_ptr)
return SLURM_SUCCESS;
}
-extern int select_p_job_update_end_time(struct job_record *job_ptr)
-{
- return SLURM_SUCCESS;
-}
-
extern int select_p_job_suspend(struct job_record *job_ptr)
{
ListIterator job_iterator;
diff --git a/src/plugins/select/linear/select_linear.c b/src/plugins/select/linear/select_linear.c
index be95f078446..038b0064066 100644
--- a/src/plugins/select/linear/select_linear.c
+++ b/src/plugins/select/linear/select_linear.c
@@ -867,11 +867,6 @@ extern int select_p_job_fini(struct job_record *job_ptr)
return rc;
}
-extern int select_p_job_update_end_time(struct job_record *job_ptr)
-{
- return SLURM_SUCCESS;
-}
-
extern int select_p_job_suspend(struct job_record *job_ptr)
{
slurm_mutex_lock(&cr_mutex);
--
GitLab