From d40ca77e97d5cc45be5581609ee2184ed5c54e91 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Wed, 16 Oct 2002 16:22:37 +0000
Subject: [PATCH] Require that all nodes in a partition on a computer with Elan
have the same processor count (as used for scheduling, configured or actual).
This will be required until the Elan interface is more sophisticated.
---
src/slurmctld/node_mgr.c | 13 ++++++++
src/slurmctld/read_config.c | 59 ++++++++++++++++++++++++++++++-------
2 files changed, 62 insertions(+), 10 deletions(-)
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 61a4f27e76b..2d0420ecb21 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -1051,12 +1051,25 @@ validate_node_specs (char *node_name, uint32_t cpus,
}
else {
info ("validate_node_specs: node %s has registered", node_name);
+ node_ptr->cpus = cpus;
+ node_ptr->real_memory = real_memory;
+ node_ptr->tmp_disk = tmp_disk;
+#ifdef HAVE_LIBELAN3
+ /* Every node in a given partition must have the same 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",
+ node_name);
+ return EINVAL; /* leave node down */
+ }
+#endif
resp_state = node_ptr->node_state & NODE_STATE_NO_RESPOND;
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;
else if ((node_ptr->node_state == NODE_STATE_DOWN) &&
(slurmctld_conf.ret2service == 1)) {
+
if (job_count)
node_ptr->node_state = NODE_STATE_ALLOCATED;
else
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 4e1249e3508..c0688f0d628 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -48,10 +48,13 @@
#define BUF_SIZE 1024
-static int init_slurm_conf ();
+static int init_slurm_conf (void);
static int parse_config_spec (char *in_line);
static int parse_node_spec (char *in_line);
static int parse_part_spec (char *in_line);
+#ifdef HAVE_LIBELAN3
+static void validate_node_proc_count (void);
+#endif
static char highest_node_name[MAX_NAME_LEN] = "";
int node_record_count = 0;
@@ -242,8 +245,8 @@ static int
parse_config_spec (char *in_line)
{
int error_code;
- int fast_schedule = 0, hash_base = 0, heartbeat_interval = 0, kill_wait = 0;
- int ret2service = 0, slurmctld_timeout = 0, slurmd_timeout = 0;
+ int fast_schedule = -1, hash_base = -1, heartbeat_interval = -1, kill_wait = -1;
+ int ret2service = -1, slurmctld_timeout = -1, slurmd_timeout = -1;
char *backup_controller = NULL, *control_machine = NULL, *epilog = NULL;
char *prioritize = NULL, *prolog = NULL, *state_save_location = NULL, *tmp_fs = NULL;
char *slurmctld_port = NULL, *slurmd_port = NULL;
@@ -294,19 +297,19 @@ parse_config_spec (char *in_line)
slurmctld_conf.epilog = epilog;
}
- if ( fast_schedule )
+ if ( fast_schedule != -1)
slurmctld_conf.fast_schedule = fast_schedule;
if ( first_job_id )
slurmctld_conf.first_job_id = first_job_id;
- if ( hash_base )
+ if ( hash_base != -1)
slurmctld_conf.hash_base = hash_base;
- if ( heartbeat_interval )
+ if ( heartbeat_interval != -1)
slurmctld_conf.heartbeat_interval = heartbeat_interval;
- if ( kill_wait )
+ if ( kill_wait != -1)
slurmctld_conf.kill_wait = kill_wait;
if ( prioritize ) {
@@ -321,7 +324,7 @@ parse_config_spec (char *in_line)
slurmctld_conf.prolog = prolog;
}
- if ( ret2service )
+ if ( ret2service != -1)
slurmctld_conf.ret2service = ret2service;
if ( slurmctld_port ) {
@@ -333,7 +336,7 @@ parse_config_spec (char *in_line)
endservent ();
}
- if ( slurmctld_timeout )
+ if ( slurmctld_timeout != -1)
slurmctld_conf.slurmctld_timeout = slurmctld_timeout;
if ( slurmd_port ) {
@@ -345,7 +348,7 @@ parse_config_spec (char *in_line)
endservent ();
}
- if ( slurmd_timeout )
+ if ( slurmd_timeout != -1)
slurmctld_conf.slurmd_timeout = slurmd_timeout;
if ( state_save_location ) {
@@ -866,6 +869,9 @@ read_slurm_conf (int recover) {
if ((error_code = build_bitmaps ()))
return error_code;
+#ifdef HAVE_LIBELAN3
+ validate_node_proc_count ();
+#endif
if (recover) {
(void) sync_nodes_to_jobs ();
}
@@ -923,3 +929,36 @@ sync_nodes_to_jobs (void)
info ("sync_nodes_to_jobs updated state of %d nodes", update_cnt);
return update_cnt;
}
+
+#ifdef HAVE_LIBELAN3
+/* Every node in a given partition must have the same processor count at present */
+void validate_node_proc_count (void)
+{
+ ListIterator part_record_iterator;
+ struct part_record *part_record_point;
+ int first_bit, last_bit, i, node_size, part_size;
+
+ part_record_iterator = list_iterator_create (part_list);
+ while ((part_record_point = (struct part_record *) list_next (part_record_iterator))) {
+ first_bit = bit_ffs (part_record_point->node_bitmap);
+ last_bit = bit_fls (part_record_point->node_bitmap);
+ part_size = -1;
+ for (i=first_bit; i<=last_bit; i++) {
+ if (bit_test (part_record_point->node_bitmap, i) == 0)
+ continue;
+
+ if (slurmctld_conf.fast_schedule)
+ node_size = node_record_table_ptr[i].config_ptr->cpus;
+ else
+ node_size = node_record_table_ptr[i].cpus;
+
+ if (part_size == -1)
+ part_size = node_size;
+ else if (part_size != node_size)
+ fatal ("Partition %s has inconsisent processor count",
+ part_record_point->name);
+ }
+ }
+ list_iterator_destroy (part_record_iterator);
+}
+#endif
--
GitLab