diff --git a/src/api/config_info.c b/src/api/config_info.c
index cf1ac553feb0c3af8720272b5b1f4b486d0cfc78..e63c7286a18d6f56fe807e57b4644e9e3773d59f 100644
--- a/src/api/config_info.c
+++ b/src/api/config_info.c
@@ -43,7 +43,9 @@ void slurm_print_ctl_conf ( FILE* out, slurm_ctl_conf_info_msg_t * slurm_ctl_con
make_time_str ((time_t *)&slurm_ctl_conf_ptr->last_update, time_str);
fprintf(out, "Configuration updated at %s\n", time_str);
+ fprintf(out, "BackupAddr = %s\n", slurm_ctl_conf_ptr->backup_addr);
fprintf(out, "BackupController = %s\n", slurm_ctl_conf_ptr->backup_controller);
+ fprintf(out, "ControlAddr = %s\n", slurm_ctl_conf_ptr->control_addr);
fprintf(out, "ControlMachine = %s\n", slurm_ctl_conf_ptr->control_machine);
fprintf(out, "Epilog = %s\n", slurm_ctl_conf_ptr->epilog);
fprintf(out, "FastSchedule = %u\n", slurm_ctl_conf_ptr->fast_schedule);
diff --git a/src/common/slurm_protocol_api.c b/src/common/slurm_protocol_api.c
index 5b8ac8def1e3d9fb203d0390f8e79f5e50da4641..675fbf58a6880ed8eb1668c9790a9168a905a507 100644
--- a/src/common/slurm_protocol_api.c
+++ b/src/common/slurm_protocol_api.c
@@ -87,21 +87,21 @@ slurm_protocol_config_t *slurm_get_api_config()
int slurm_api_set_default_config()
{
- if ((slurmctld_conf.control_machine == NULL) ||
+ if ((slurmctld_conf.control_addr == NULL) ||
(slurmctld_conf.slurmctld_port == 0)) {
read_slurm_port_config();
- if ((slurmctld_conf.control_machine == NULL) ||
+ if ((slurmctld_conf.control_addr == NULL) ||
(slurmctld_conf.slurmctld_port == 0))
fatal ("Unable to establish control machine or port");
}
slurm_set_addr(&proto_conf_default.primary_controller,
slurmctld_conf.slurmctld_port,
- slurmctld_conf.control_machine);
- if (slurmctld_conf.backup_controller) {
+ slurmctld_conf.control_addr);
+ if (slurmctld_conf.backup_addr) {
slurm_set_addr(&proto_conf_default.secondary_controller,
slurmctld_conf.slurmctld_port,
- slurmctld_conf.backup_controller);
+ slurmctld_conf.backup_addr);
}
proto_conf = &proto_conf_default;
@@ -120,8 +120,8 @@ int read_slurm_port_config()
{
FILE *slurm_spec_file; /* pointer to input data file */
char in_line[BUF_SIZE]; /* input line */
- char *control_machine = NULL;
- char *backup_controller = NULL;
+ char *control_addr = NULL, *control_machine = NULL;
+ char *backup_addr = NULL, *backup_controller = NULL;
int error_code, i, j, line_num = 0;
int slurmctld_port = 0, slurmd_port = 0;
struct servent *servent;
@@ -181,8 +181,10 @@ int read_slurm_port_config()
/* parse what is left */
/* overall slurm configuration parameters */
error_code = slurm_parser(in_line,
- "ControlMachine=", 's', &control_machine,
+ "BackupAddr=", 's', &backup_addr,
"BackupController=", 's', &backup_controller,
+ "ControlAddr=", 's', &control_addr,
+ "ControlMachine=", 's', &control_machine,
"SlurmctldPort=", 'd',&slurmctld_port,
"SlurmdPort=", 'd', &slurmd_port,
"END");
@@ -193,10 +195,25 @@ int read_slurm_port_config()
if (slurmctld_conf.control_machine == NULL)
slurmctld_conf.control_machine = control_machine;
+ if (slurmctld_conf.control_addr == NULL) {
+ if (control_addr)
+ slurmctld_conf.control_addr = control_addr;
+ else
+ slurmctld_conf.control_addr = control_machine;
+ }
+
if (slurmctld_conf.backup_controller == NULL)
slurmctld_conf.backup_controller = backup_controller;
+ if (slurmctld_conf.backup_addr == NULL) {
+ if (backup_addr)
+ slurmctld_conf.backup_addr = backup_addr;
+ else
+ slurmctld_conf.backup_addr = backup_controller;
+ }
+
if (slurmctld_port)
slurmctld_conf.slurmctld_port = slurmctld_port;
+
if (slurmd_port)
slurmctld_conf.slurmd_port = slurmd_port;
}
diff --git a/src/common/slurm_protocol_defs.c b/src/common/slurm_protocol_defs.c
index 662d1d29558360bb9e45c9829c55c11ca15f5b3f..27ad29e3017ba867a74ed69766abbfd55ee2915b 100644
--- a/src/common/slurm_protocol_defs.c
+++ b/src/common/slurm_protocol_defs.c
@@ -68,8 +68,12 @@ void slurm_free_return_code_msg(return_code_msg_t * msg)
void slurm_free_ctl_conf(slurm_ctl_conf_info_msg_t * build_ptr)
{
if (build_ptr) {
+ if (build_ptr->backup_addr)
+ xfree(build_ptr->backup_addr);
if (build_ptr->backup_controller)
xfree(build_ptr->backup_controller);
+ if (build_ptr->control_addr)
+ xfree(build_ptr->control_addr);
if (build_ptr->control_machine)
xfree(build_ptr->control_machine);
if (build_ptr->epilog)
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index 0d35ed24cccf9eab9feddd22669e7ec3a8dcda22..9bd788339b2b759797781aa96dc5b81db5f73ee9 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -343,7 +343,9 @@ typedef struct partition_info {
struct slurm_ctl_conf {
uint32_t last_update; /* last update time of the build parameters */
+ char *backup_addr; /* comm path of slurmctld secondary server */
char *backup_controller; /* name of slurmctld secondary server */
+ char *control_addr; /* comm path of slurmctld primary server */
char *control_machine; /* name of slurmctld primary server */
char *epilog; /* pathname of job epilog */
uint32_t first_job_id; /* first slurm generated job_id to assign */
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 9693b7a27c3a15529f3d17f25a49a0caa10fa2b7..f07b309cf53fbc90b86b913a632791aad38eaa23 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -1069,7 +1069,9 @@ int unpack_job_info_members ( job_info_t * job , Buf buffer )
void pack_slurm_ctl_conf ( slurm_ctl_conf_info_msg_t * build_ptr, Buf buffer )
{
pack_time (build_ptr->last_update, buffer);
+ packstr (build_ptr->backup_addr, buffer);
packstr (build_ptr->backup_controller, buffer);
+ packstr (build_ptr->control_addr, buffer);
packstr (build_ptr->control_machine, buffer);
packstr (build_ptr->epilog, buffer);
pack16 (build_ptr->fast_schedule, buffer);
@@ -1101,7 +1103,9 @@ int unpack_slurm_ctl_conf ( slurm_ctl_conf_info_msg_t **build_buffer_ptr, Buf bu
/* load the data values */
/* unpack timestamp of snapshot */
unpack_time (&build_ptr->last_update, buffer);
+ unpackstr_xmalloc (&build_ptr->backup_addr, &uint16_tmp, buffer);
unpackstr_xmalloc (&build_ptr->backup_controller, &uint16_tmp, buffer);
+ unpackstr_xmalloc (&build_ptr->control_addr, &uint16_tmp, buffer);
unpackstr_xmalloc (&build_ptr->control_machine, &uint16_tmp, buffer);
unpackstr_xmalloc (&build_ptr->epilog, &uint16_tmp, buffer);
unpack16 (&build_ptr->fast_schedule, buffer);
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 2ebfaa0b35c1d97bb079a718eaf8607613d6aa3c..40a03065a7adb05d57186638de825ef35bd30233 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -1810,7 +1810,9 @@ init_ctld_conf ( slurm_ctl_conf_t * conf_ptr )
struct servent *servent;
conf_ptr->last_update = time (NULL) ;
+ conf_ptr->backup_addr = NULL ;
conf_ptr->backup_controller = NULL ;
+ conf_ptr->control_addr = NULL ;
conf_ptr->control_machine = NULL ;
conf_ptr->epilog = NULL ;
conf_ptr->fast_schedule = 1 ;
@@ -1848,7 +1850,9 @@ void
fill_ctld_conf ( slurm_ctl_conf_t * conf_ptr )
{
conf_ptr->last_update = slurmctld_conf.last_update ;
+ conf_ptr->backup_addr = slurmctld_conf.backup_addr ;
conf_ptr->backup_controller = slurmctld_conf.backup_controller ;
+ conf_ptr->control_addr = slurmctld_conf.control_addr ;
conf_ptr->control_machine = slurmctld_conf.control_machine ;
conf_ptr->epilog = slurmctld_conf.epilog ;
conf_ptr->fast_schedule = slurmctld_conf.fast_schedule ;
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index ba112aec9e0af7df76a06486e9844e983f172b08..8426ea0ac94c5c678feb3fe9217c55a70072386f 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -886,7 +886,8 @@ set_slurmd_addr (void)
if (strlen (node_record_table_ptr[i].name) == 0)
continue;
slurm_set_addr (& node_record_table_ptr[i].slurm_addr,
- slurmctld_conf.slurmd_port, node_record_table_ptr[i].name);
+ slurmctld_conf.slurmd_port,
+ node_record_table_ptr[i].comm_name);
}
return;
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index f8a318fe60d8c917ef9884956062eae9521242f8..397314dd625ef1df4eb88d88be0e662ac18e9973 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -43,6 +43,7 @@
#include <src/common/list.h>
#include <src/common/macros.h>
#include <src/common/parse_spec.h>
+#include <src/common/xstring.h>
#include <src/slurmctld/locks.h>
#include <src/slurmctld/slurmctld.h>
@@ -248,7 +249,8 @@ parse_config_spec (char *in_line)
int fast_schedule = -1, hash_base = -1, heartbeat_interval = -1;
int inactive_limit = -1, kill_wait = -1;
int ret2service = -1, slurmctld_timeout = -1, slurmd_timeout = -1;
- char *backup_controller = NULL, *control_machine = NULL, *epilog = NULL;
+ char *backup_addr = NULL, *backup_controller = NULL;
+ char *control_addr = 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;
char *job_credential_private_key = NULL , *job_credential_public_certificate = NULL;
@@ -257,7 +259,9 @@ parse_config_spec (char *in_line)
struct stat sbuf;
error_code = slurm_parser(in_line,
+ "BackupAddr=", 's', &backup_addr,
"BackupController=", 's', &backup_controller,
+ "ControlAddr=", 's', &control_addr,
"ControlMachine=", 's', &control_machine,
"Epilog=", 's', &epilog,
"FastSchedule=", 'd', &fast_schedule,
@@ -281,12 +285,32 @@ parse_config_spec (char *in_line)
if (error_code)
return error_code;
+ if ( backup_addr ) {
+ if ( slurmctld_conf.backup_addr )
+ xfree (slurmctld_conf.backup_addr);
+ slurmctld_conf.backup_addr = backup_addr;
+ } else if ( backup_controller ) {
+ if ( slurmctld_conf.backup_addr )
+ xfree (slurmctld_conf.backup_addr);
+ slurmctld_conf.backup_addr = xstrdup (backup_controller);
+ }
+
if ( backup_controller ) {
if ( slurmctld_conf.backup_controller )
xfree (slurmctld_conf.backup_controller);
slurmctld_conf.backup_controller = backup_controller;
}
+ if ( control_addr ) {
+ if ( slurmctld_conf.control_addr )
+ xfree (slurmctld_conf.control_addr);
+ slurmctld_conf.control_addr = control_addr;
+ } else if ( control_machine ) {
+ if ( slurmctld_conf.control_addr )
+ xfree (slurmctld_conf.control_addr);
+ slurmctld_conf.control_addr = xstrdup (control_machine);
+ }
+
if ( control_machine ) {
if ( slurmctld_conf.control_machine )
xfree (slurmctld_conf.control_machine);
@@ -397,14 +421,15 @@ parse_config_spec (char *in_line)
*/
static int
parse_node_spec (char *in_line) {
- char *node_name, *state, *feature, *this_node_name;
+ char *node_addr, *node_name, *state, *feature;
+ char *this_node_addr , *this_node_name;
int error_code, first, i;
int state_val, cpus_val, real_memory_val, tmp_disk_val, weight_val;
struct node_record *node_record_point;
struct config_record *config_point = NULL;
- hostlist_t host_list = NULL;
+ hostlist_t addr_list = NULL, host_list = NULL;
- node_name = state = feature = (char *) NULL;
+ node_addr = node_name = state = feature = (char *) NULL;
cpus_val = real_memory_val = state_val = NO_VAL;
tmp_disk_val = weight_val = NO_VAL;
if ((error_code = load_string (&node_name, "NodeName=", in_line)))
@@ -413,8 +438,9 @@ parse_node_spec (char *in_line) {
return 0; /* no node info */
error_code = slurm_parser(in_line,
- "Procs=", 'd', &cpus_val,
"Feature=", 's', &feature,
+ "NodeAddr=", 's', &node_addr,
+ "Procs=", 'd', &cpus_val,
"RealMemory=", 'd', &real_memory_val,
"State=", 's', &state,
"TmpDisk=", 'd', &tmp_disk_val,
@@ -442,8 +468,15 @@ parse_node_spec (char *in_line) {
}
}
+ if ( node_addr &&
+ ((addr_list = hostlist_create (node_addr)) == NULL)) {
+ error ("hostlist_create error for %s: %m", node_addr);
+ error_code = errno;
+ goto cleanup;
+ }
+
if ( (host_list = hostlist_create (node_name)) == NULL) {
- error ("hostlist_create error for %s, %m", node_name);
+ error ("hostlist_create error for %s: %m", node_name);
error_code = errno;
goto cleanup;
}
@@ -513,6 +546,18 @@ parse_node_spec (char *in_line) {
(state_val != NODE_STATE_UNKNOWN))
node_record_point->node_state = state_val;
node_record_point->last_response = time (NULL);
+ if (node_addr)
+ this_node_addr = hostlist_shift (addr_list);
+ else
+ this_node_addr = NULL;
+ if (this_node_addr) {
+ strncpy (node_record_point->comm_name,
+ this_node_addr, MAX_NAME_LEN);
+ free (this_node_addr);
+ } else {
+ strncpy (node_record_point->comm_name,
+ node_record_point->name, MAX_NAME_LEN);
+ }
}
else {
error ("parse_node_spec: reconfiguration for node %s ignored.",
@@ -524,6 +569,8 @@ parse_node_spec (char *in_line) {
/* xfree allocated storage */
if (state)
xfree(state);
+ if (addr_list)
+ hostlist_destroy (addr_list);
hostlist_destroy (host_list);
return error_code;
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 75aa952d163900b81e846645d6b9e57254ca30e1..a7dd85ead9eabb6142791d1509d8c55324182d29 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -123,6 +123,7 @@ struct node_record {
uint32_t tmp_disk; /* actual megabytes of total disk in TMP_FS */
struct config_record *config_ptr; /* configuration specification for this node */
struct part_record *partition_ptr; /* partition for this node */
+ char comm_name[MAX_NAME_LEN]; /* communications path name of the node */
struct sockaddr_in slurm_addr; /* network address */
};
extern struct node_record *node_record_table_ptr; /* location of the node records */