From d9d5b71d661e32dc122bff465966b07da333afe2 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Wed, 26 Feb 2003 00:13:10 +0000
Subject: [PATCH] Added much of infrastructure for COMPLETING job state.
---
src/slurmctld/controller.c | 1 +
src/slurmctld/job_mgr.c | 3 +--
src/slurmctld/node_mgr.c | 40 +++++++++++++++++++++++-----------
src/slurmctld/node_scheduler.c | 6 ++---
src/slurmctld/read_config.c | 36 ++++++++++++++----------------
src/slurmctld/slurmctld.h | 19 +++++++++++-----
6 files changed, 60 insertions(+), 45 deletions(-)
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 6dd5b33cd9b..9afc0de9020 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -1954,6 +1954,7 @@ static void _slurm_rpc_node_registration(slurm_msg_t * msg)
node_reg_stat_msg->node_name,
(long) (clock() - start_time));
slurm_send_rc_msg(msg, SLURM_SUCCESS);
+ schedule(); /* has own locks */
}
}
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index ee13b659b29..c4c6b298af0 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -2614,8 +2614,7 @@ int update_job(job_desc_msg_t * job_specs, uid_t uid)
/*
* validate_jobs_on_node - validate that any jobs that should be on the node
* are actually running, if not clean up the job records and/or node
- * records, call this function after validate_node_specs() sets the node
- * state properly
+ * records
* IN node_name - node which should have jobs running
* IN job_count - number of jobs which should be running on specified node
* IN job_id_ptr - pointer to array of job_ids that should be on this node
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 58089475b5d..5fd1e330b1c 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -460,10 +460,10 @@ int load_all_node_state ( void )
while (remaining_buf (buffer) > 0) {
safe_unpackstr_xmalloc (&node_name, &name_len, buffer);
- safe_unpack16 (&node_state, buffer);
- safe_unpack32 (&cpus, buffer);
+ safe_unpack16 (&node_state, buffer);
+ safe_unpack32 (&cpus, buffer);
safe_unpack32 (&real_memory, buffer);
- safe_unpack32 (&tmp_disk, buffer);
+ safe_unpack32 (&tmp_disk, buffer);
/* validity test as possible */
if ((cpus == 0) ||
@@ -1059,8 +1059,7 @@ validate_node_specs (char *node_name, uint32_t cpus,
error_code = 0;
if (cpus < config_ptr->cpus) {
- error ("validate_node_specs: node %s has low cpu count %u",
- node_name, cpus);
+ error ("Node %s has low cpu count %u", node_name, cpus);
error_code = EINVAL;
}
node_ptr->cpus = cpus;
@@ -1069,22 +1068,21 @@ validate_node_specs (char *node_name, uint32_t cpus,
(cpus - config_ptr->cpus);
if (real_memory < config_ptr->real_memory) {
- error ("validate_node_specs: node %s has low real_memory size %u",
+ error ("Node %s has low real_memory size %u",
node_name, real_memory);
error_code = EINVAL;
}
node_ptr->real_memory = real_memory;
if (tmp_disk < config_ptr->tmp_disk) {
- error ("validate_node_specs: node %s has low tmp_disk size %u",
+ error ("Node %s has low tmp_disk size %u",
node_name, tmp_disk);
error_code = EINVAL;
}
node_ptr->tmp_disk = tmp_disk;
if (error_code) {
- error ("validate_node_specs: setting node %s state to DOWN",
- node_name);
+ error ("Setting node %s state to DOWN", node_name);
set_node_down(node_name);
} else if (status == ESLURMD_PROLOG_FAILED) {
error ("Prolog failure on node %s, state to DOWN",
@@ -1101,7 +1099,7 @@ validate_node_specs (char *node_name, uint32_t cpus,
* processor count at present */
if ((slurmctld_conf.fast_schedule == 0) &&
(node_ptr->config_ptr->cpus != cpus)) {
- error ("Node %s has processor count inconsistent with rest of partition",
+ error ("Node %s processor count inconsistent with rest of partition",
node_name);
return EINVAL; /* leave node down */
}
@@ -1489,29 +1487,45 @@ void msg_to_slurmd (slurm_msg_type_t msg_type)
}
+/* make_node_alloc - flag specified node as allocated to a job */
+void make_node_alloc(struct node_record *node_ptr)
+{
+ int inx = node_ptr - node_record_table_ptr;
+ uint16_t no_resp_flag;
+
+ last_node_update = time (NULL);
+ no_resp_flag = node_ptr->node_state & NODE_STATE_NO_RESPOND;
+ node_ptr->node_state = NODE_STATE_ALLOCATED | no_resp_flag;
+ (node_ptr->job_cnt)++;
+ bit_clear(idle_node_bitmap, inx);
+}
+
/* make_node_comp - flag specified node as completing a job */
void make_node_comp(struct node_record *node_ptr)
{
uint16_t no_resp_flag, base_state;
+ last_node_update = time (NULL);
base_state = node_ptr->node_state & (~NODE_STATE_NO_RESPOND);
no_resp_flag = node_ptr->node_state & NODE_STATE_NO_RESPOND;
if ((base_state == NODE_STATE_DOWN) ||
(base_state == NODE_STATE_DRAINED) ||
(base_state == NODE_STATE_DRAINING)) {
- debug3("Node %s being left in state %s",
- node_state_string((enum node_states)node_ptr->name));
+ debug3("Node %s being left in state %s", node_ptr->name,
+ node_state_string((enum node_states)
+ node_ptr->node_state));
} else {
node_ptr->node_state = NODE_STATE_COMPLETING | no_resp_flag;
}
}
-/* make_node_idle - flag specified node as no longer being in use */
+/* make_node_idle - flag specified node as having completed a job */
void make_node_idle(struct node_record *node_ptr)
{
int inx = node_ptr - node_record_table_ptr;
uint16_t no_resp_flag, base_state;
+ last_node_update = time (NULL);
base_state = node_ptr->node_state & (~NODE_STATE_NO_RESPOND);
no_resp_flag = node_ptr->node_state & NODE_STATE_NO_RESPOND;
no_resp_flag = node_ptr->node_state & NODE_STATE_NO_RESPOND;
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 5153ea25bca..41d9b4ed56a 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -94,10 +94,8 @@ void allocate_nodes(unsigned *bitmap)
last_node_update = time(NULL);
for (i = 0; i < node_record_count; i++) {
- if (bit_test(bitmap, i) == 0)
- continue;
- node_record_table_ptr[i].node_state = NODE_STATE_ALLOCATED;
- bit_clear(idle_node_bitmap, i);
+ if (bit_test(bitmap, i))
+ make_node_alloc(&node_record_table_ptr[i]);
}
return;
}
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 538709067b7..f72ec18c7f4 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -125,11 +125,11 @@ static int _build_bitmaps(void)
if (node_record_table_ptr[i].name[0] == '\0')
continue; /* defunct */
base_state =
- node_record_table_ptr[i].
- node_state & (~NODE_STATE_NO_RESPOND);
+ node_record_table_ptr[i].node_state &
+ (~NODE_STATE_NO_RESPOND);
no_resp_flag =
- node_record_table_ptr[i].
- node_state & NODE_STATE_NO_RESPOND;
+ node_record_table_ptr[i].node_state &
+ NODE_STATE_NO_RESPOND;
if (base_state == NODE_STATE_IDLE)
bit_set(idle_node_bitmap, i);
if ((base_state != NODE_STATE_DOWN) &&
@@ -284,9 +284,10 @@ static int _parse_node_spec(char *in_line)
break;
}
}
- if (state_val == NO_VAL) {
+ if ((state_val == NO_VAL) ||
+ (state_val == NODE_STATE_COMPLETING)) {
error
- ("_parse_node_spec: invalid state %s for node_name %s",
+ ("_parse_node_spec: invalid initial state %s for node %s",
state, node_name);
error_code = EINVAL;
goto cleanup;
@@ -374,8 +375,8 @@ static int _parse_node_spec(char *in_line)
node_record_point =
create_node_record(config_point,
this_node_name);
- if ((state_val != NO_VAL)
- && (state_val != NODE_STATE_UNKNOWN))
+ if ((state_val != NO_VAL) &&
+ (state_val != NODE_STATE_UNKNOWN))
node_record_point->node_state = state_val;
node_record_point->last_response = (time_t) 0;
if (node_addr)
@@ -827,32 +828,27 @@ static int _sync_nodes_to_jobs(void)
struct job_record *job_ptr;
ListIterator job_record_iterator;
int i, update_cnt = 0;
+ uint16_t no_resp_flag;
job_record_iterator = list_iterator_create(job_list);
while ((job_ptr =
(struct job_record *) list_next(job_record_iterator))) {
- if ((job_ptr->job_state == JOB_PENDING)
- || (job_ptr->job_state == JOB_COMPLETE)
- || (job_ptr->job_state == JOB_FAILED)
- || (job_ptr->job_state == JOB_TIMEOUT))
+ if (job_ptr->job_state > JOB_COMPLETING)
continue;
if (job_ptr->node_bitmap == NULL)
continue;
for (i = 0; i < node_record_count; i++) {
if (bit_test(job_ptr->node_bitmap, i) == 0)
continue;
+ node_record_table_ptr[i].job_cnt++;
if (node_record_table_ptr[i].node_state ==
NODE_STATE_ALLOCATED)
continue; /* already in proper state */
update_cnt++;
- if (node_record_table_ptr[i].
- node_state & NODE_STATE_NO_RESPOND)
- node_record_table_ptr[i].node_state =
- NODE_STATE_ALLOCATED |
- NODE_STATE_NO_RESPOND;
- else
- node_record_table_ptr[i].node_state =
- NODE_STATE_ALLOCATED;
+ no_resp_flag = node_record_table_ptr[i].node_state &
+ NODE_STATE_NO_RESPOND;
+ node_record_table_ptr[i].node_state =
+ NODE_STATE_ALLOCATED | no_resp_flag;
}
}
if (update_cnt)
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index d724305a7dc..482f85fbe87 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -122,7 +122,7 @@ struct config_record {
uint32_t real_memory; /* MB real memory on the node */
uint32_t tmp_disk; /* MB total storage in TMP_FS file system */
uint32_t weight; /* arbitrary priority of node for
- scheduling work on */
+ * scheduling work on */
char *feature; /* arbitrary list of features associated */
char *nodes; /* name of nodes with this configuration */
bitstr_t *node_bitmap; /* bitmap of nodes with this configuration */
@@ -134,8 +134,8 @@ struct node_record {
uint32_t magic; /* magic cookie for data integrity */
char name[MAX_NAME_LEN]; /* name of the node. NULL==defunct */
uint16_t node_state; /* enum node_states, ORed with
- NODE_STATE_NO_RESPOND if not
- responding */
+ * NODE_STATE_NO_RESPOND if not
+ * responding */
time_t last_response; /* last response from the node */
uint32_t cpus; /* count of cpus on the node */
uint32_t real_memory; /* MB real memory on the node */
@@ -144,15 +144,16 @@ struct node_record {
struct part_record *partition_ptr; /* partition for this node */
char comm_name[MAX_NAME_LEN]; /* communications path name to node */
slurm_addr slurm_addr; /* network address */
+ uint16_t job_cnt; /* count of jobs allocated to node */
};
extern struct node_record *node_record_table_ptr; /* ptr to node records */
extern time_t last_bitmap_update; /* time of last node creation or
- deletion */
+ * deletion */
extern time_t last_node_update; /* time of last node record update */
extern int node_record_count; /* count in node_record_table_ptr */
extern int *hash_table; /* table of hashed indicies into
- node_record_table_ptr */
+ * node_record_table_ptr */
extern bitstr_t *up_node_bitmap; /* bitmap of nodes are up */
extern bitstr_t *idle_node_bitmap; /* bitmap of nodes are idle */
extern struct config_record default_config_record;
@@ -236,7 +237,10 @@ struct job_record {
struct part_record *part_ptr; /* pointer to the partition record */
uint16_t batch_flag; /* 1 if batch job (with script) */
uint32_t user_id; /* user the job runs as */
- enum job_states job_state; /* state of the job */
+ enum job_states job_state; /* state of the job, NOTE: state
+ * JOB_COMPLETING is set in pack_job
+ * when (job state > JOB_RUNNING) &&
+ * (node_count > 0), its artificial */
uint16_t kill_on_node_fail; /* 1 if job should be killed on
node failure */
uint16_t kill_on_step_done; /* 1 if job should be killed when
@@ -701,6 +705,9 @@ extern void load_part_uid_allow_list ( int force );
*/
extern int load_all_part_state ( void );
+/* make_node_alloc - flag specified node as allocated to a job */
+extern void make_node_alloc(struct node_record *node_ptr);
+
/* make_node_comp - flag specified node as completing a job */
extern void make_node_comp(struct node_record *node_ptr);
--
GitLab