From cc5dfa208e48a666e77c43f93bc9cacb668e0c21 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Tue, 24 Sep 2002 00:05:51 +0000
Subject: [PATCH] Added support for node states DRAINING/DRAINED.
---
src/slurmctld/node_mgr.c | 52 +++++++++++++++++++++++-----------
src/slurmctld/node_scheduler.c | 24 ++++++++++++----
src/slurmctld/read_config.c | 10 +++++--
3 files changed, 62 insertions(+), 24 deletions(-)
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index ee29608f10a..577510f586b 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -956,7 +956,7 @@ split_node_name (char *name, char *prefix, char *suffix, int *index,
int
update_node ( update_node_msg_t * update_node_msg )
{
- int error_code = 0, state_val;
+ int error_code = 0, state_val, node_inx;
char *this_node_name ;
struct node_record *node_record_point;
hostlist_t host_list;
@@ -976,6 +976,7 @@ update_node ( update_node_msg_t * update_node_msg )
last_node_update = time (NULL);
while ( (this_node_name = hostlist_shift (host_list)) ) {
node_record_point = find_node_record (this_node_name);
+ node_inx = node_record_point - node_record_table_ptr;
if (node_record_point == NULL) {
error ("update_node: node name %s does not exist, can not be updated",
this_node_name);
@@ -986,17 +987,36 @@ update_node ( update_node_msg_t * update_node_msg )
if (state_val != NO_VAL) {
if (state_val == NODE_STATE_DOWN) {
- bit_clear (up_node_bitmap,
- (int) (node_record_point - node_record_table_ptr));
- bit_clear (idle_node_bitmap,
- (int) (node_record_point - node_record_table_ptr));
+ bit_clear (up_node_bitmap, node_inx);
+ bit_clear (idle_node_bitmap, node_inx);
+ }
+ else if (state_val == NODE_STATE_UNKNOWN) {
+ bit_clear (up_node_bitmap, node_inx);
+ bit_clear (idle_node_bitmap, node_inx);
+ }
+ else if (state_val == NODE_STATE_IDLE) {
+ bit_set (up_node_bitmap, node_inx);
+ bit_set (idle_node_bitmap, node_inx);
+ }
+ else if (state_val == NODE_STATE_ALLOCATED) {
+ bit_set (up_node_bitmap, node_inx);
+ bit_clear (idle_node_bitmap, node_inx);
+ }
+ else if (state_val == NODE_STATE_DRAINED) {
+ if (~bit_test (idle_node_bitmap, node_inx))
+ state_val = NODE_STATE_DRAINING;
+ bit_clear (up_node_bitmap, node_inx);
+ }
+ else if (state_val == NODE_STATE_DRAINING) {
+ if (bit_test (idle_node_bitmap, node_inx)) {
+ state_val = NODE_STATE_DRAINED;
+ bit_clear (idle_node_bitmap, node_inx);
+ }
+ bit_clear (up_node_bitmap, node_inx);
+ }
+ else {
+ error ("Invalid node state specified %d", state_val);
}
- else if (state_val != NODE_STATE_IDLE)
- bit_clear (idle_node_bitmap,
- (int) (node_record_point - node_record_table_ptr));
- else /* (state_val == NODE_STATE_IDLE) */
- bit_set (idle_node_bitmap,
- (int) (node_record_point - node_record_table_ptr));
node_record_point->node_state = state_val;
info ("update_node: node %s state set to %s",
@@ -1081,7 +1101,7 @@ void
node_did_resp (char *name)
{
struct node_record *node_ptr;
- int i;
+ int node_inx;
node_ptr = find_node_record (name);
if (node_ptr == NULL) {
@@ -1089,16 +1109,16 @@ node_did_resp (char *name)
return;
}
- i = node_ptr - node_record_table_ptr;
+ node_inx = node_ptr - node_record_table_ptr;
last_node_update = time (NULL);
- node_record_table_ptr[i].last_response = time (NULL);
+ node_record_table_ptr[node_inx].last_response = time (NULL);
node_ptr->node_state &= (uint16_t) (~NODE_STATE_NO_RESPOND);
if (node_ptr->node_state == NODE_STATE_UNKNOWN)
node_ptr->node_state = NODE_STATE_IDLE;
if (node_ptr->node_state == NODE_STATE_IDLE)
- bit_set (idle_node_bitmap, (node_ptr - node_record_table_ptr));
+ bit_set (idle_node_bitmap, node_inx);
if (node_ptr->node_state != NODE_STATE_DOWN)
- bit_set (up_node_bitmap, (node_ptr - node_record_table_ptr));
+ bit_set (up_node_bitmap, node_inx);
return;
}
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 294bd41cbfa..c7336589aee 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -118,6 +118,7 @@ deallocate_nodes (struct job_record * job_ptr)
pthread_attr_t attr_agent;
pthread_t thread_agent;
int buf_rec_size = 0;
+ uint16_t no_resp_flag, base_state;
agent_args = xmalloc (sizeof (agent_arg_t));
agent_args->msg_type = REQUEST_REVOKE_JOB_CREDENTIAL;
@@ -132,15 +133,28 @@ deallocate_nodes (struct job_record * job_ptr)
continue;
if ((agent_args->addr_count+1) > buf_rec_size) {
buf_rec_size += 32;
- xrealloc ((agent_args->slurm_addr), (sizeof (struct sockaddr_in) * buf_rec_size));
- xrealloc ((agent_args->node_names), (MAX_NAME_LEN * buf_rec_size));
+ xrealloc ((agent_args->slurm_addr),
+ (sizeof (struct sockaddr_in) * buf_rec_size));
+ xrealloc ((agent_args->node_names),
+ (MAX_NAME_LEN * buf_rec_size));
}
- agent_args->slurm_addr[agent_args->addr_count] = node_record_table_ptr[i].slurm_addr;
+ agent_args->slurm_addr[agent_args->addr_count] =
+ node_record_table_ptr[i].slurm_addr;
strncpy (&agent_args->node_names[MAX_NAME_LEN*agent_args->addr_count],
node_record_table_ptr[i].name, MAX_NAME_LEN);
agent_args->addr_count++;
- node_record_table_ptr[i].node_state = NODE_STATE_IDLE;
- bit_set (idle_node_bitmap, i);
+ base_state = 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;
+ if (base_state == NODE_STATE_DRAINING) {
+ node_record_table_ptr[i].node_state = NODE_STATE_DRAINED;
+ bit_clear (idle_node_bitmap, i);
+ bit_clear (up_node_bitmap, i);
+ }
+ else {
+ node_record_table_ptr[i].node_state = NODE_STATE_IDLE | no_resp_flag;
+ if (no_resp_flag == 0)
+ bit_set (idle_node_bitmap, i);
+ }
}
agent_args->msg_args = revoke_job_cred;
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 6cafc7a8ff5..d81f0e4d622 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -139,12 +139,16 @@ build_bitmaps ()
/* scan all nodes and identify which are up and idle and their configuration */
for (i = 0; i < node_record_count; i++) {
+ uint16_t base_state;
+
if (node_record_table_ptr[i].name[0] == '\0')
continue; /* defunct */
- if (node_record_table_ptr[i].node_state == NODE_STATE_IDLE)
+ base_state = node_record_table_ptr[i].node_state & (~NODE_STATE_NO_RESPOND);
+ if (base_state == NODE_STATE_IDLE)
bit_set (idle_node_bitmap, i);
- if ((node_record_table_ptr[i].node_state != NODE_STATE_DOWN) &&
- (node_record_table_ptr[i].node_state != NODE_STATE_UNKNOWN) &&
+ if ((base_state != NODE_STATE_DOWN) &&
+ (base_state != NODE_STATE_UNKNOWN) &&
+ (base_state != NODE_STATE_DRAINED) &&
((node_record_table_ptr[i].node_state & NODE_STATE_NO_RESPOND) == 0))
bit_set (up_node_bitmap, i);
if (node_record_table_ptr[i].config_ptr)
--
GitLab