From 0cf28d0bd46322bacec7eea912e4c6b06c0ac6b1 Mon Sep 17 00:00:00 2001
From: "Christopher J. Morrone" <morrone2@llnl.gov>
Date: Wed, 25 Jan 2006 00:30:17 +0000
Subject: [PATCH] New --enable-multiple-slurmd configure parameter to allow
running more than one copy of slurmd on a node at the same time. Only really
useful for developers.
---
NEWS | 6 +++++
configure.ac | 21 +++++++++++++++++
src/api/config_info.c | 2 ++
src/common/read_config.c | 11 +++++++--
src/common/slurm_protocol_pack.c | 4 ++++
src/slurmctld/node_mgr.c | 5 +++-
src/slurmctld/proc_req.c | 2 ++
src/slurmctld/read_config.c | 13 +++++++++++
src/slurmctld/slurmctld.h | 1 +
src/slurmd/slurmd/slurmd.c | 39 +++++++++++++++++++++++---------
src/srun/launch.c | 1 -
11 files changed, 90 insertions(+), 15 deletions(-)
diff --git a/NEWS b/NEWS
index d8040998ae1..b557e0438c4 100644
--- a/NEWS
+++ b/NEWS
@@ -1,6 +1,12 @@
This file describes changes in recent versions of SLURM. It primarily
documents those changes that are of interest to users and admins.
+* Changes in SLURM 1.1.0-pre1
+=============================
+ -- New --enable-multiple-slurmd configure parameter to allow running
+ more than one copy of slurmd on a node at the same time. Only
+ really useful for developers.
+
* Changes in SLURM 1.0.2
========================
-- Correctly report DRAINED node state as type OTHER for "sinfo --summarize".
diff --git a/configure.ac b/configure.ac
index 1b613fd196c..3c93e15da54 100644
--- a/configure.ac
+++ b/configure.ac
@@ -159,6 +159,27 @@ dnl Check for compilation of SLURM auth modules:
dnl
X_AC_MUNGE
+dnl
+dnl Check if multiple-slurmd support is requested and define MULTIPLE_SLURMD
+dnl if it is.
+dnl
+AC_MSG_CHECKING(whether to enable multiple-slurmd support)
+AC_ARG_ENABLE([multiple-slurmd],
+ AC_HELP_STRING([--enable-multiple-slurmd], [enable multiple-slurmd support]),
+ [ case "$enableval" in
+ yes) multiple_slurmd=yes ;;
+ no) multiple_slurmd=no ;;
+ *) AC_MSG_ERROR([bad value "$enableval" for --enable-multiple-slurmd]);;
+ esac ]
+)
+if test "x$multiple_slurmd" = "xyes"; then
+ AC_DEFINE([MULTIPLE_SLURMD], [1], [Enable multiple slurmd on one node])
+ AC_MSG_RESULT([yes])
+else
+ AC_MSG_RESULT([no])
+fi
+
+
AUTHD_LIBS="-lauth -le"
savedLIBS="$LIBS"
savedCFLAGS="$CFLAGS"
diff --git a/src/api/config_info.c b/src/api/config_info.c
index 4b71ff462ed..9cd4434ec09 100644
--- a/src/api/config_info.c
+++ b/src/api/config_info.c
@@ -152,8 +152,10 @@ void slurm_print_ctl_conf ( FILE* out,
slurm_ctl_conf_ptr->slurmd_logfile);
fprintf(out, "SlurmdPidFile = %s\n",
slurm_ctl_conf_ptr->slurmd_pidfile);
+#ifndef MULTIPLE_SLURMD
fprintf(out, "SlurmdPort = %u\n",
slurm_ctl_conf_ptr->slurmd_port);
+#endif
fprintf(out, "SlurmdSpoolDir = %s\n",
slurm_ctl_conf_ptr->slurmd_spooldir);
fprintf(out, "SlurmdTimeout = %u\n",
diff --git a/src/common/read_config.c b/src/common/read_config.c
index d1288feee30..6dc003b71ba 100644
--- a/src/common/read_config.c
+++ b/src/common/read_config.c
@@ -267,6 +267,7 @@ extern char *get_conf_node_name(char *node_hostname)
/* getnodename - equivalent to gethostname, but return only the first
* component of the fully qualified name
* (e.g. "linux123.foo.bar" becomes "linux123")
+ * OUT name
*/
int
getnodename (char *name, size_t len)
@@ -397,7 +398,7 @@ init_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr)
ctl_conf_ptr->slurmd_debug = (uint16_t) NO_VAL;
xfree (ctl_conf_ptr->slurmd_logfile);
xfree (ctl_conf_ptr->slurmd_pidfile);
- ctl_conf_ptr->slurmd_port = (uint32_t) NO_VAL;
+ ctl_conf_ptr->slurmd_port = (uint32_t) NO_VAL;
xfree (ctl_conf_ptr->slurmd_spooldir);
ctl_conf_ptr->slurmd_timeout = (uint16_t) NO_VAL;
xfree (ctl_conf_ptr->state_save_location);
@@ -933,6 +934,7 @@ parse_config_spec (char *in_line, slurm_ctl_conf_t *ctl_conf_ptr)
ctl_conf_ptr->slurmd_logfile = slurmd_logfile;
}
+#ifndef MULTIPLE_SLURMD
if ( slurmd_port != -1) {
if ( ctl_conf_ptr->slurmd_port != (uint32_t) NO_VAL)
error (MULTIPLE_VALUE_MSG, "SlurmdPort");
@@ -941,6 +943,7 @@ parse_config_spec (char *in_line, slurm_ctl_conf_t *ctl_conf_ptr)
else
ctl_conf_ptr->slurmd_port = slurmd_port;
}
+#endif
if ( slurmd_spooldir ) {
if ( ctl_conf_ptr->slurmd_spooldir ) {
@@ -1058,12 +1061,14 @@ _parse_node_spec (char *in_line, bool slurmd_hosts)
char *state = NULL, *reason=NULL;
char *node_hostname = NULL;
int cpus_val, real_memory_val, tmp_disk_val, weight_val;
+ int port;
error_code = slurm_parser (in_line,
"Feature=", 's', &feature,
"NodeAddr=", 's', &node_addr,
"NodeName=", 's', &node_name,
"NodeHostname=", 's', &node_hostname,
+ "Port=", 'd', &port,
"Procs=", 'd', &cpus_val,
"RealMemory=", 'd', &real_memory_val,
"Reason=", 's', &reason,
@@ -1076,7 +1081,7 @@ _parse_node_spec (char *in_line, bool slurmd_hosts)
return error_code;
if (node_name
- && (node_hostname || slurmd_hosts)) {
+ && (node_hostname || slurmd_hosts)) {
all_slurmd_hosts = true;
_register_conf_node_aliases(node_name, node_hostname);
}
@@ -1433,8 +1438,10 @@ validate_config (slurm_ctl_conf_t *ctl_conf_ptr)
if (ctl_conf_ptr->slurmd_pidfile == NULL)
ctl_conf_ptr->slurmd_pidfile = xstrdup(DEFAULT_SLURMD_PIDFILE);
+#ifndef MULTIPLE_SLURMD
if (ctl_conf_ptr->slurmd_port == (uint32_t) NO_VAL)
ctl_conf_ptr->slurmd_port = SLURMD_PORT;
+#endif
if (ctl_conf_ptr->slurmd_spooldir == NULL)
ctl_conf_ptr->slurmd_spooldir = xstrdup(DEFAULT_SPOOLDIR);
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 68c14332408..fae0b90c32e 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -1920,7 +1920,9 @@ _pack_slurm_ctl_conf_msg(slurm_ctl_conf_info_msg_t * build_ptr, Buf buffer)
pack16(build_ptr->slurmd_debug, buffer);
packstr(build_ptr->slurmd_logfile, buffer);
packstr(build_ptr->slurmd_pidfile, buffer);
+#ifndef MULTIPLE_SLURMD
pack32(build_ptr->slurmd_port, buffer);
+#endif
packstr(build_ptr->slurmd_spooldir, buffer);
debug2("Packing string %s", build_ptr->slurmd_spooldir);
pack16(build_ptr->slurmd_timeout, buffer);
@@ -2006,7 +2008,9 @@ _unpack_slurm_ctl_conf_msg(slurm_ctl_conf_info_msg_t **
buffer);
safe_unpackstr_xmalloc(&build_ptr->slurmd_pidfile, &uint16_tmp,
buffer);
+#ifndef MULTIPLE_SLURMD
safe_unpack32(&build_ptr->slurmd_port, buffer);
+#endif
safe_unpackstr_xmalloc(&build_ptr->slurmd_spooldir, &uint16_tmp,
buffer);
safe_unpack16(&build_ptr->slurmd_timeout, buffer);
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index e8336ba221f..38cc10c1047 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -197,6 +197,7 @@ create_node_record (struct config_record *config_ptr, char *node_name)
strcpy (node_ptr->name, node_name);
node_ptr->node_state = default_node_record.node_state;
node_ptr->last_response = default_node_record.last_response;
+ node_ptr->port = default_node_record.port;
node_ptr->config_ptr = config_ptr;
node_ptr->part_cnt = 0;
node_ptr->part_pptr = NULL;
@@ -803,8 +804,10 @@ void set_slurmd_addr (void)
for (i = 0; i < node_record_count; i++, node_ptr++) {
if (node_ptr->name[0] == '\0')
continue;
+ if (node_ptr->port == 0)
+ node_ptr->port = slurmctld_conf.slurmd_port;
slurm_set_addr (&node_ptr->slurm_addr,
- slurmctld_conf.slurmd_port,
+ node_ptr->port,
node_ptr->comm_name);
if (node_ptr->slurm_addr.sin_port)
continue;
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index f3001888415..0d5ed7b8849 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -337,7 +337,9 @@ void _fill_ctld_conf(slurm_ctl_conf_t * conf_ptr)
slurmd_logfile);
conf_ptr->slurmd_pidfile = xstrdup(slurmctld_conf.
slurmd_pidfile);
+#ifndef MULTIPLE_SLURMD
conf_ptr->slurmd_port = slurmctld_conf.slurmd_port;
+#endif
conf_ptr->slurmd_spooldir = xstrdup(slurmctld_conf.
slurmd_spooldir);
conf_ptr->slurmd_timeout = slurmctld_conf.slurmd_timeout;
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 14eddbc0177..bb6a7a6fad5 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -289,10 +289,12 @@ static int _parse_node_spec(char *in_line)
#ifndef HAVE_FRONT_END /* Fake node addresses for front-end */
char *this_node_addr;
#endif
+ int port;
node_addr = node_name = state = feature = (char *) NULL;
cpus_val = real_memory_val = state_val = NO_VAL;
tmp_disk_val = weight_val = NO_VAL;
+ port = NO_VAL;
if ((error_code = load_string(&node_name, "NodeName=", in_line)))
return error_code;
if (node_name == NULL)
@@ -307,6 +309,9 @@ static int _parse_node_spec(char *in_line)
"Feature=", 's', &feature,
"NodeAddr=", 's', &node_addr,
"NodeHostname=", 's', &node_hostname,
+#ifdef MULTIPLE_SLURMD
+ "Port=", 'd', &port,
+#endif
"Procs=", 'd', &cpus_val,
"RealMemory=", 'd', &real_memory_val,
"Reason=", 's', &reason,
@@ -396,6 +401,10 @@ static int _parse_node_spec(char *in_line)
default_config_record.feature = feature;
feature = NULL;
}
+#ifdef MULTIPLE_SLURMD
+ if (port != NO_VAL)
+ default_node_record.port = port;
+#endif
free(this_node_name);
break;
}
@@ -457,6 +466,10 @@ static int _parse_node_spec(char *in_line)
} else
strncpy(node_ptr->comm_name,
node_ptr->name, MAX_NAME_LEN);
+#endif
+#ifdef MULTIPLE_SLURMD
+ if (port != NO_VAL)
+ node_ptr->port = port;
#endif
node_ptr->reason = xstrdup(reason);
} else {
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 98a009ffc8f..a860c5dbf3d 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -171,6 +171,7 @@ struct node_record {
struct part_record **part_pptr; /* array of pointers to partitions
* associated with this node*/
char comm_name[MAX_NAME_LEN]; /* communications path name to node */
+ uint16_t port; /* TCP port number of the slurmd */
slurm_addr slurm_addr; /* network address */
uint16_t comp_job_cnt; /* count of jobs completing on node */
uint16_t run_job_cnt; /* count of jobs running on node */
diff --git a/src/slurmd/slurmd/slurmd.c b/src/slurmd/slurmd/slurmd.c
index b977908d784..587d1046cd4 100644
--- a/src/slurmd/slurmd/slurmd.c
+++ b/src/slurmd/slurmd/slurmd.c
@@ -67,7 +67,11 @@
#include "src/slurmd/common/proctrack.h"
#include "src/slurmd/common/task_plugin.h"
+#ifdef MULTIPLE_SLURMD
+#define GETOPT_ARGS "L:Dvhcf:MN:P:"
+#else
#define GETOPT_ARGS "L:Dvhcf:M"
+#endif
#ifndef MAXHOSTNAMELEN
# define MAXHOSTNAMELEN 64
@@ -105,7 +109,6 @@ static void _create_msg_socket();
static void _msg_engine();
static int _slurmd_init();
static int _slurmd_fini();
-static void _create_conf();
static void _init_conf();
static void _print_conf();
static void _read_config();
@@ -141,7 +144,7 @@ main (int argc, char *argv[])
* Create and set default values for the slurmd global
* config variable "conf"
*/
- _create_conf();
+ conf = xmalloc(sizeof(slurmd_conf_t));
_init_conf();
conf->argv = &argv;
conf->argc = &argc;
@@ -473,7 +476,6 @@ _free_and_set(char **confvar, char *newval)
/*
* Read the slurm configuration file (slurm.conf) and substitute some
* values into the slurmd configuration in preference of the defaults.
- *
*/
static void
_read_config()
@@ -489,7 +491,9 @@ _read_config()
if (conf->conffile == NULL)
conf->conffile = xstrdup(conf->cf.slurm_conf);
+#ifndef MULTIPLE_SLURMD
conf->port = conf->cf.slurmd_port;
+#endif
conf->slurm_user_id = conf->cf.slurm_user_id;
path_pubkey = xstrdup(conf->cf.job_credential_public_certificate);
@@ -497,12 +501,23 @@ _read_config()
if (!conf->logfile)
conf->logfile = xstrdup(conf->cf.slurmd_logfile);
- _free_and_set(&conf->node_name, get_conf_node_name(conf->hostname));
+ /* node_name may already be set from a command line parameter */
+ if (conf->node_name == NULL)
+ _free_and_set(&conf->node_name,
+ get_conf_node_name(conf->hostname));
_free_and_set(&conf->epilog, xstrdup(conf->cf.epilog));
_free_and_set(&conf->prolog, xstrdup(conf->cf.prolog));
_free_and_set(&conf->tmpfs, xstrdup(conf->cf.tmp_fs));
_free_and_set(&conf->spooldir, xstrdup(conf->cf.slurmd_spooldir));
+#ifdef MULTIPLE_SLURMD
+ /* append the NodeName to the spooldir to make it unique */
+ xstrfmtcat(conf->spooldir, ".%s", conf->node_name);
+#endif
_free_and_set(&conf->pidfile, xstrdup(conf->cf.slurmd_pidfile));
+#ifdef MULTIPLE_SLURMD
+ /* append the NodeName to the pidfile name to make it unique */
+ xstrfmtcat(conf->pidfile, ".%s", conf->node_name);
+#endif
_free_and_set(&conf->task_prolog, xstrdup(conf->cf.task_prolog));
_free_and_set(&conf->task_epilog, xstrdup(conf->cf.task_epilog));
_free_and_set(&conf->pubkey, path_pubkey);
@@ -562,12 +577,6 @@ _print_conf()
debug3("TaskEpilog = `%s'", conf->task_epilog);
}
-static void
-_create_conf()
-{
- conf = xmalloc(sizeof(slurmd_conf_t));
-}
-
static void
_init_conf()
{
@@ -632,6 +641,14 @@ _process_cmdline(int ac, char **av)
case 'M':
conf->mlock_pages = 1;
break;
+#ifdef MULTIPLE_SLURMD
+ case 'N':
+ conf->node_name = xstrdup(optarg);
+ break;
+ case 'P':
+ conf->port = (uint16_t)atoi(optarg);
+ break;
+#endif
default:
_usage(c);
exit(1);
@@ -721,7 +738,7 @@ _slurmd_init()
/*
* Need to kill any running slurmd's here
*/
- _kill_old_slurmd();
+ _kill_old_slurmd();
stepd_cleanup_sockets(conf->spooldir, conf->node_name);
}
diff --git a/src/srun/launch.c b/src/srun/launch.c
index f045455e4fd..d516cf176ea 100644
--- a/src/srun/launch.c
+++ b/src/srun/launch.c
@@ -115,7 +115,6 @@ launch(void *arg)
debug("going to launch %d tasks on %d hosts",
opt.nprocs, job->step_layout->num_hosts);
- debug("sending to slurmd port %d", slurm_get_slurmd_port());
msg_array_ptr = xmalloc(sizeof(launch_tasks_request_msg_t)
* job->step_layout->num_hosts);
--
GitLab