diff --git a/src/plugins/select/bluegene/bgl_job_run.c b/src/plugins/select/bluegene/bgl_job_run.c
index e224ea55417c33a08371840b4b2326893b666a03..21786550cbbd411ef243d9f22dc6ad7b5f362e55 100644
--- a/src/plugins/select/bluegene/bgl_job_run.c
+++ b/src/plugins/select/bluegene/bgl_job_run.c
@@ -135,7 +135,7 @@ static int _remove_job(db_job_id_t job_id)
if ((rc = rm_free_job(job_rec)) != STATUS_OK)
error("rm_free_job: %s", bgl_err_str(rc));
- debug("job %d is in state %d", job_id, job_state);
+ debug2("job %d is in state %d", job_id, job_state);
/* check the state and process accordingly */
if(job_state == RM_JOB_TERMINATED)
@@ -339,7 +339,7 @@ static void _term_agent(bgl_update_t *bgl_update_ptr)
db_job_id_t job_id;
bgl_record_t *bgl_record = NULL;
- //debug("getting the job info");
+ debug2("getting the job info");
live_states = JOB_ALL_FLAG
& (~JOB_TERMINATED_FLAG)
& (~JOB_KILLED_FLAG);
@@ -353,7 +353,7 @@ static void _term_agent(bgl_update_t *bgl_update_ptr)
jobs = 0;
} else if (jobs > 300)
fatal("Active job count (%d) invalid, restart MMCS", jobs);
- //debug("job count %d",jobs);
+ debug2("job count %d",jobs);
for (i=0; i<jobs; i++) {
if (i) {
@@ -383,7 +383,7 @@ static void _term_agent(bgl_update_t *bgl_update_ptr)
part_id, bgl_err_str(rc));
continue;
}
- debug("looking at partition %s looking for %s\n",
+ debug2("looking at partition %s looking for %s\n",
part_id, bgl_update_ptr->bgl_part_id);
if (strcmp(part_id, bgl_update_ptr->bgl_part_id) != 0)
continue;
@@ -393,7 +393,7 @@ static void _term_agent(bgl_update_t *bgl_update_ptr)
bgl_err_str(rc));
continue;
}
- debug("got job_id %d",job_id);
+ debug2("got job_id %d",job_id);
if((rc = _remove_job(job_id))
== INTERNAL_ERROR)
goto not_removed;
@@ -404,7 +404,7 @@ static void _term_agent(bgl_update_t *bgl_update_ptr)
bgl_record = find_bgl_record(bgl_update_ptr->bgl_part_id);
if(bgl_record) {
- debug("got the record %s user is %s",
+ debug2("got the record %s user is %s",
bgl_record->bgl_part_id,
bgl_record->user_name);
@@ -600,7 +600,7 @@ int term_jobs_on_part(pm_partition_id_t bgl_part_id)
bgl_update_ptr = xmalloc(sizeof(bgl_update_t));
bgl_update_ptr->op = TERM_OP;
bgl_update_ptr->bgl_part_id = xstrdup(bgl_part_id);
- _term_agent(bgl_update_ptr);
+ _part_op(bgl_update_ptr);
return rc;
}
diff --git a/src/plugins/select/bluegene/bgl_part_info.c b/src/plugins/select/bluegene/bgl_part_info.c
index f002702a2ef2876ded77af5b72129111035a0750..6ce2194f5ff8440848a14f6684b2cf815cf2b7c8 100644
--- a/src/plugins/select/bluegene/bgl_part_info.c
+++ b/src/plugins/select/bluegene/bgl_part_info.c
@@ -56,6 +56,61 @@
#define _DEBUG 0
#define RETRY_BOOT_COUNT 3
+static int _partition_is_deallocating(bgl_record_t *bgl_record);
+
+static int _partition_is_deallocating(bgl_record_t *bgl_record)
+{
+ if(remove_all_users(bgl_record->bgl_part_id, NULL)
+ == REMOVE_USER_ERR) {
+ error("Something happened removing "
+ "users from partition %s",
+ bgl_record->bgl_part_id);
+ }
+
+ if(bgl_record->target_name
+ && bgl_record->user_name) {
+ if(!strcmp(bgl_record->target_name,
+ slurmctld_conf.slurm_user_name)) {
+ if(strcmp(bgl_record->target_name,
+ bgl_record->user_name)) {
+ error("Partition %s was in a ready state "
+ "for user %s but is being freed. "
+ "Job was lost.",
+ bgl_record->bgl_part_id,
+ bgl_record->user_name);
+ xfree(bgl_record->target_name);
+ bgl_record->target_name =
+ xstrdup(bgl_record->user_name);
+ //term_jobs_on_part(bgl_record->bgl_part_id);
+ } else {
+ debug("Partition %s was in a ready state "
+ "but is being freed. No job running.",
+ bgl_record->bgl_part_id);
+ }
+ } else {
+ error("State went to free on a boot "
+ "for partition %s.",
+ bgl_record->bgl_part_id);
+ }
+
+ } else if(bgl_record->user_name) {
+ error("Target Name was not set "
+ "not set for partition %s.",
+ bgl_record->bgl_part_id);
+ bgl_record->target_name =
+ xstrdup(bgl_record->user_name);
+ } else {
+ error("Target Name and User Name are "
+ "not set for partition %s.",
+ bgl_record->bgl_part_id);
+ bgl_record->user_name =
+ xstrdup(slurmctld_conf.slurm_user_name);
+ bgl_record->target_name =
+ xstrdup(bgl_record->user_name);
+ }
+ return SLURM_SUCCESS;
+}
+
/*
* check to see if partition is ready to execute. Meaning
* User is added to the list of users able to run, and no one
@@ -124,7 +179,8 @@ extern int update_partition_list()
time_t now;
struct tm *time_ptr;
char reason[128];
-
+ int skipped_dealloc = 0;
+
if(bgl_list == NULL && !last_bgl_update)
return 0;
@@ -212,62 +268,19 @@ extern int update_partition_list()
} else if(bgl_record->state != state) {
debug("state of Partition %s was %d and now is %d",
name, bgl_record->state, state);
+ /*
+ check to make sure partition went
+ through freeing correctly
+ */
+ if(bgl_record->state != RM_PARTITION_DEALLOCATING
+ && state == RM_PARTITION_FREE)
+ skipped_dealloc = 1;
bgl_record->state = state;
- error("Yeah, the state changed");
- if(bgl_record->state == RM_PARTITION_FREE) {
- if((rc = remove_all_users(
- bgl_record->bgl_part_id,
- NULL))
- == REMOVE_USER_ERR) {
- error("Something happened removing "
- "users from partition %s",
- bgl_record->bgl_part_id);
- }
-
- if(bgl_record->target_name
- && bgl_record->user_name) {
- if(!strcmp(bgl_record->target_name,
- slurmctld_conf.
- slurm_user_name)) {
- if(strcmp(bgl_record->target_name,
- bgl_record->user_name)) {
- error("Partition %s was in a "
- "ready state for user %s but got "
- "freed. Job was probably lost.",
- bgl_record->user_name,
- bgl_record->bgl_part_id);
- xfree(bgl_record->
- user_name);
- bgl_record->user_name =
- xstrdup(bgl_record->
- target_name);
- } else {
- error("Partition %s was in a "
- "ready state for user %s but got "
- "freed. No job running.",
- bgl_record->bgl_part_id);
- }
- } else {
- error("State went to free on a boot "
- "for partition %s.",
- bgl_record->bgl_part_id);
- }
-
- } else if(bgl_record->user_name) {
- error("Target Name was not set "
- "not set for partition %s.",
- bgl_record->bgl_part_id);
- bgl_record->target_name =
- xstrdup(bgl_record->user_name);
- } else {
- error("Target Name and User Name are "
- "not set for partition %s.",
- bgl_record->bgl_part_id);
- bgl_record->user_name =
- xstrdup(slurmctld_conf.slurm_user_name);
- bgl_record->target_name =
- xstrdup(bgl_record->user_name);
- }
+ if(bgl_record->state == RM_PARTITION_DEALLOCATING) {
+ _partition_is_deallocating(bgl_record);
+ } else if(skipped_dealloc) {
+ _partition_is_deallocating(bgl_record);
+ skipped_dealloc = 0;
} else if(bgl_record->state
== RM_PARTITION_CONFIGURING)
bgl_record->boot_state = 1;
diff --git a/src/plugins/select/bluegene/bluegene.c b/src/plugins/select/bluegene/bluegene.c
index 8d4d93d30ae4645cd2b1ff97a6cb6947a21bba92..e2b32e645489df16ea5aaa7f661981e83c2debea 100644
--- a/src/plugins/select/bluegene/bluegene.c
+++ b/src/plugins/select/bluegene/bluegene.c
@@ -383,7 +383,7 @@ extern int remove_all_users(char *bgl_part_id, char *user_name)
returnc = REMOVE_USER_ERR;
user_count = 0;
} else
- debug("got %d users for %s",user_count, bgl_part_id);
+ debug2("got %d users for %s",user_count, bgl_part_id);
for(i=0; i<user_count; i++) {
if(i) {
if ((rc = rm_get_data(part_ptr,
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 3c9a29d6352f1fe1b5889337a392a1428a2b7a16..2029d942980e44e359dd752dee827270a71fdeed 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -1792,11 +1792,11 @@ static void _slurm_rpc_job_ready(slurm_msg_t * msg)
END_TIMER;
if (error_code) {
- debug("_slurm_rpc_job_ready: %s",
+ debug2("_slurm_rpc_job_ready: %s",
slurm_strerror(error_code));
slurm_send_rc_msg(msg, error_code);
} else {
- debug("_slurm_rpc_job_ready(%u)=%d %s", id_msg->job_id,
+ debug2("_slurm_rpc_job_ready(%u)=%d %s", id_msg->job_id,
result, TIME_STR);
rc_msg.return_code = result;
response_msg.address = msg->address;