From 63fdb35434a5cb2b69fa90002948b3c6799818c7 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Wed, 25 Jun 2003 20:20:30 +0000
Subject: [PATCH] Minor changes in how COMPLETING jobs and their nodes are
handled. If a node is down and not responding, don't bother to send a
KILL_JOB RPC to it. If that is the only node associated with a job, don't
have that job go through a COMPLETING state. It goes directly to a COMPLETED
state. Also preserve the NO_RESPOND flag associated with a node if its state
is changed via user request (e.g. scontrol).
---
src/slurmctld/node_mgr.c | 23 ++++++++++++-----------
src/slurmctld/node_scheduler.c | 6 +++++-
2 files changed, 17 insertions(+), 12 deletions(-)
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 730d84d8398..c93a1493535 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -912,8 +912,9 @@ int update_node ( update_node_msg_t * update_node_msg )
{
int error_code = 0, state_val, node_inx;
char *this_node_name ;
- struct node_record *node_record_point;
+ struct node_record *node_ptr;
hostlist_t host_list;
+ uint16_t no_resp_flag = 0;
if (update_node_msg -> node_names == NULL ) {
error ("update_node: invalid node name %s",
@@ -932,9 +933,9 @@ int 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) {
+ node_ptr = find_node_record (this_node_name);
+ node_inx = node_ptr - node_record_table_ptr;
+ if (node_ptr == NULL) {
error ("update_node: node %s does not exist",
this_node_name);
error_code = ESLURM_INVALID_NODE_NAME;
@@ -944,8 +945,8 @@ int 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, node_inx);
- bit_clear (idle_node_bitmap, node_inx);
+ /* We must set node down before killing its jobs */
+ _make_node_down(node_ptr);
kill_running_job_by_node_name (this_node_name,
false);
}
@@ -977,8 +978,7 @@ int update_node ( update_node_msg_t * update_node_msg )
else if (state_val == NODE_STATE_NO_RESPOND) {
bit_clear (up_node_bitmap, node_inx);
bit_clear (idle_node_bitmap, node_inx);
- node_record_point->node_state |=
- NODE_STATE_NO_RESPOND;
+ node_ptr->node_state |= NODE_STATE_NO_RESPOND;
info ("update_node: node %s state set to %s",
this_node_name, "NoResp");
continue;
@@ -989,7 +989,8 @@ int update_node ( update_node_msg_t * update_node_msg )
continue;
}
- node_record_point->node_state = state_val;
+ no_resp_flag = node_ptr->node_state & NODE_STATE_NO_RESPOND;
+ node_ptr->node_state = state_val | no_resp_flag;
info ("update_node: node %s state set to %s",
this_node_name, node_state_string(state_val));
}
@@ -1209,8 +1210,8 @@ void set_node_down (char *name)
return;
}
- (void) kill_running_job_by_node_name(name, false);
_make_node_down(node_ptr);
+ (void) kill_running_job_by_node_name(name, false);
return;
}
@@ -1278,9 +1279,9 @@ void ping_nodes (void)
(base_state != NODE_STATE_DOWN)) {
error ("Node %s not responding, setting DOWN",
node_record_table_ptr[i].name);
+ _make_node_down(&node_record_table_ptr[i]);
kill_running_job_by_node_name (
node_record_table_ptr[i].name, false);
- _make_node_down(&node_record_table_ptr[i]);
continue;
}
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 471bfec42c9..eec6ed24ed5 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -126,7 +126,7 @@ int count_cpus(unsigned *bitmap)
/*
* deallocate_nodes - for a given job, deallocate its nodes and make
* their state NODE_STATE_COMPLETING
- * IN job_ptr - pointer to terminating job
+ * IN job_ptr - pointer to terminating job (already in some COMPLETING state)
* IN timeout - true of job exhausted time limit, send REQUEST_KILL_TIMELIMIT
* RPC instead of REQUEST_KILL_JOB
* globals: node_record_count - number of nodes in the system
@@ -159,6 +159,9 @@ void deallocate_nodes(struct job_record *job_ptr, bool timeout)
for (i = 0; i < node_record_count; i++) {
if (bit_test(job_ptr->node_bitmap, i) == 0)
continue;
+ if (node_record_table_ptr[i].node_state ==
+ (NODE_STATE_DOWN | NODE_STATE_NO_RESPOND))
+ continue; /* don't bother with dead nodes */
if ((agent_args->node_count + 1) > buf_rec_size) {
buf_rec_size += 32;
xrealloc((agent_args->slurm_addr),
@@ -180,6 +183,7 @@ void deallocate_nodes(struct job_record *job_ptr, bool timeout)
error("Job %u allocated no nodes to be killed on",
job_ptr->job_id);
xfree(agent_args);
+ job_ptr->job_state &= (~JOB_COMPLETING);
return;
}
--
GitLab