From 1fd5eaf54a3959c6c6941f4f54d35d39d5051b67 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Thu, 16 Sep 2004 16:06:34 +0000
Subject: [PATCH] Add squeue options to print BGL specific options (geometry,
node_use, conn_type, and rotate). slurmctld to set min_nodes for job based
upon geometry (if supplied). This avoid trying to schedule jobs when
insufficient resources are available.
---
doc/man/man1/squeue.1 | 14 +++++-
doc/man/man1/srun.1 | 12 +++++
src/api/job_info.c | 6 +--
src/slurmctld/job_mgr.c | 54 +++++++++++++++++++--
src/slurmctld/node_mgr.c | 2 +-
src/slurmctld/node_scheduler.c | 2 +-
src/squeue/opts.c | 20 ++++++++
src/squeue/print.c | 88 ++++++++++++++++++++++++++++++++++
src/squeue/print.h | 16 +++++++
src/srun/opt.c | 50 +++++++++++++++----
10 files changed, 245 insertions(+), 19 deletions(-)
diff --git a/doc/man/man1/squeue.1 b/doc/man/man1/squeue.1
index 7cf82414095..81c41a8cc18 100644
--- a/doc/man/man1/squeue.1
+++ b/doc/man/man1/squeue.1
@@ -1,4 +1,4 @@
-.TH SQUEUE "1" "August 2004" "squeue 0.4" "Slurm components"
+.TH SQUEUE "1" "September 2004" "squeue 0.4" "Slurm components"
.SH "NAME"
squeue \- view information about jobs located in the SLURM scheduling queue.
@@ -185,6 +185,18 @@ User ID
\fB%x\fR
List of node names explicitly excluded by the job
.TP
+\fB%y\fR
+Geometry requirement of resource allocation (X,Y,Z dimensions).
+.TP
+\fB%Y\fR
+Connection type: TORUS, MESH, or NAV (torus else mesh).
+.TP
+\fB%z\fR
+Permit rotation of geometry (yes or no).
+.TP
+\fB%Z\fR
+Node use: VIRTUAL or COPROCESSOR.
+.TP
\fB%.<*>\fR
right justification of the field
.TP
diff --git a/doc/man/man1/srun.1 b/doc/man/man1/srun.1
index 93ea9514682..484df3a85a3 100644
--- a/doc/man/man1/srun.1
+++ b/doc/man/man1/srun.1
@@ -603,6 +603,9 @@ SLURM_ACCOUNT
SLURM_CPUS_PER_TASK
\fB\-c, \-\-ncpus\-per\-task\fR=\fIn\fR
.TP
+SLURM_CONN_TYPE
+\fB\-\-conn\-type\fR=(\fImesh|nav|torus\fR)
+.TP
SLURM_CORE_FORMAT
\fB\-\-core\fR=\fIformat\fR
.TP
@@ -615,12 +618,21 @@ SLURMD_DEBUG
SLURM_DISTRIBUTION
\fB\-m, \-\-distribution\fR=(\fIblock|cyclic\fR)
.TP
+SLURM_GEOMETRY
+\fB\-g, \-\-geometry\fR=\fIX,Y,Z\fR
+.TP
SLURM_LABELIO
\fB-l, --label\fR
.TP
SLURM_NNODES
\fB\-N, \-\-nodes\fR=(\fIn|min-max\fR)
.TP
+SLURM_NO_ROTATE
+\fB\-\-no\-rotate\fR
+.TP
+SLURM_NODE_USE
+\fB\-\-node\-use\fR=(\fIcoprocessor|virtual\fR)
+.TP
SLURM_NPROCS
\fB\-n, \-\-ntasks\fR=\fIn\fR
.TP
diff --git a/src/api/job_info.c b/src/api/job_info.c
index 0d6b3a1111c..ea9e3fdc279 100644
--- a/src/api/job_info.c
+++ b/src/api/job_info.c
@@ -237,10 +237,10 @@ slurm_print_job_info ( FILE* out, job_info_t * job_ptr, int one_liner )
fprintf ( out, "Rotate=YES ");
else
fprintf ( out, "Rotate=NO ");
- if (job_ptr->node_use == RM_VIRTUAL)
- fprintf ( out, "NodeUse=VIRTUAL ");
- else
+ if (job_ptr->node_use == RM_COPROCESSOR)
fprintf ( out, "NodeUse=COPROCESSOR ");
+ else
+ fprintf ( out, "NodeUse=VIRTUAL ");
#endif
fprintf( out, "\n\n");
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 17e1a680191..785db65772c 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -1147,10 +1147,41 @@ void dump_job_desc(job_desc_msg_t * job_specs)
dependency, job_specs->account);
#ifdef HAVE_BGL
- debug3(" conn_type=%u rotate=%u node_use=%u geometry=%u,%u,%u",
- job_specs->conn_type, job_specs->rotate, job_specs->node_use,
- job_specs->geometry[0], job_specs->geometry[1],
- job_specs->geometry[2]);
+{
+ char *conn_type, *rotate, *node_use;
+ int geometry[SYSTEM_DIMENSIONS];
+
+ if (job_specs->conn_type == RM_MESH)
+ conn_type = "MESH";
+ else if (job_specs->conn_type == RM_TORUS)
+ conn_type = "TORUS";
+ else
+ conn_type = "NAV";
+
+ if (job_specs->rotate == 0)
+ rotate = "NO";
+ else
+ rotate = "YES";
+
+ if (job_specs->node_use == RM_VIRTUAL)
+ node_use = "VIRTUAL";
+ else
+ node_use = "COPROCESSOR";
+
+ if (job_specs->geometry[0] == (uint16_t) NO_VAL) {
+ geometry[0] = -1;
+ geometry[1] = -1;
+ geometry[2] = -1;
+ } else {
+ geometry[0] = job_specs->geometry[0];
+ geometry[1] = job_specs->geometry[1];
+ geometry[2] = job_specs->geometry[2];
+ }
+
+ debug3(" conn_type=%s rotate=%s node_use=%s geometry=%d,%d,%d",
+ conn_type, rotate, node_use,
+ geometry[0], geometry[1], geometry[2]);
+}
#endif
}
@@ -1653,6 +1684,21 @@ static int _job_create(job_desc_msg_t * job_desc, uint32_t * new_job_id,
if (job_desc->min_nodes == NO_VAL)
job_desc->min_nodes = 1;
+#ifdef SYSTEM_DIMENSIONS
+ if ((job_desc->geometry[0] != (uint16_t) NO_VAL)
+ && (job_desc->geometry[0] != 0)) {
+ int i, tot = 1;
+ for (i=0; i<SYSTEM_DIMENSIONS; i++)
+ tot *= job_desc->geometry[i];
+ if (job_desc->min_nodes > tot) {
+ info("MinNodes(%d) > GeometryNodes(%d)",
+ job_desc->min_nodes, tot);
+ error_code = ESLURM_TOO_MANY_REQUESTED_CPUS;
+ goto cleanup;
+ }
+ job_desc->min_nodes = tot;
+ }
+#endif
if (job_desc->max_nodes == NO_VAL)
job_desc->max_nodes = 0;
if (job_desc->num_procs > part_ptr->total_cpus) {
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 6fb7e8fe4aa..a6a26950583 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -111,7 +111,7 @@ char * bitmap2node_name (bitstr_t *bitmap)
hostlist_push(hl, node_record_table_ptr[i].name);
}
hostlist_uniq(hl);
- hostlist_ranged_string(hl, 8192, buf);
+ hostlist_ranged_string(hl, sizeof(buf), buf);
hostlist_destroy(hl);
return xstrdup(buf);
}
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index dff0575cc22..615d1cd1874 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -386,7 +386,7 @@ _pick_best_nodes(struct node_set *node_set_ptr, int node_set_size,
#ifdef HAVE_BGL
if (shared) {
- error("attempt to share Blue Gene nodes");
+ error("attempt to share Blue Gene nodes ignored");
shared = 0;
}
#endif
diff --git a/src/squeue/opts.c b/src/squeue/opts.c
index e8c0b3a7771..4bd141fff1e 100644
--- a/src/squeue/opts.c
+++ b/src/squeue/opts.c
@@ -523,6 +523,26 @@ extern int parse_format( char* format )
field_size,
right_justify,
suffix );
+ else if (field[0] == 'y')
+ job_format_add_geometry( params.format_list,
+ field_size,
+ right_justify,
+ suffix );
+ else if (field[0] == 'Y')
+ job_format_add_conn_type( params.format_list,
+ field_size,
+ right_justify,
+ suffix );
+ else if (field[0] == 'z')
+ job_format_add_rotate( params.format_list,
+ field_size,
+ right_justify,
+ suffix );
+ else if (field[0] == 'Z')
+ job_format_add_node_use( params.format_list,
+ field_size,
+ right_justify,
+ suffix );
else
error( "Invalid job format specification: %c",
field[0] );
diff --git a/src/squeue/print.c b/src/squeue/print.c
index 382277eeccd..59cf99d3575 100644
--- a/src/squeue/print.c
+++ b/src/squeue/print.c
@@ -792,6 +792,94 @@ int _print_job_dependency(job_info_t * job, int width, bool right_justify,
return SLURM_SUCCESS;
}
+int _print_job_geometry(job_info_t * job, int width, bool right_justify,
+ char* suffix)
+{
+ if (job == NULL) /* Print the Header instead */
+ _print_str("GEOMETRY", width, right_justify, true);
+ else {
+#ifdef HAVE_BGL
+ char id[FORMAT_STRING_SIZE];
+ snprintf(id, FORMAT_STRING_SIZE, "%u,%u,%u",
+ job->geometry[0], job->geometry[1], job->geometry[2]);
+ _print_str(id, width, right_justify, true);
+#else
+ _print_str("n/a", width, right_justify, true);
+#endif
+ }
+ if (suffix)
+ printf("%s", suffix);
+ return SLURM_SUCCESS;
+}
+
+int _print_job_conn_type(job_info_t * job, int width, bool right_justify,
+ char* suffix)
+{
+ if (job == NULL) /* Print the Header instead */
+ _print_str("CONN_TYPE", width, right_justify, true);
+ else {
+ char *id;
+#ifdef HAVE_BGL
+ if (job->conn_type == RM_TORUS)
+ id = "torus";
+ else if (job->conn_type == RM_MESH)
+ id = "mesh";
+ else
+ id = "nav";
+#else
+ id = "n/a";
+#endif
+ _print_str(id, width, right_justify, true);
+ }
+ if (suffix)
+ printf("%s", suffix);
+ return SLURM_SUCCESS;
+}
+
+int _print_job_node_use(job_info_t * job, int width, bool right_justify,
+ char* suffix)
+{
+ if (job == NULL) /* Print the Header instead */
+ _print_str("NODE_USE", width, right_justify, true);
+ else {
+ char *id;
+#ifdef HAVE_BGL
+ if (job->node_use == RM_COPROCESSOR)
+ id = "coprocessor";
+ else
+ id = "virtual";
+#else
+ id = "n/a";
+#endif
+ _print_str(id, width, right_justify, true);
+ }
+ if (suffix)
+ printf("%s", suffix);
+ return SLURM_SUCCESS;
+}
+
+int _print_job_rotate(job_info_t * job, int width, bool right_justify,
+ char* suffix)
+{
+ if (job == NULL) /* Print the Header instead */
+ _print_str("ROTATE", width, right_justify, true);
+ else {
+ char *id;
+#ifdef HAVE_BGL
+ if (job->rotate == 0)
+ id = "no";
+ else
+ id = "yes";
+#else
+ id = "n/a";
+#endif
+ _print_str(id, width, right_justify, true);
+ }
+ if (suffix)
+ printf("%s", suffix);
+ return SLURM_SUCCESS;
+}
+
/*****************************************************************************
* Job Step Print Functions
*****************************************************************************/
diff --git a/src/squeue/print.h b/src/squeue/print.h
index 68aea253eea..6728541f22b 100644
--- a/src/squeue/print.h
+++ b/src/squeue/print.h
@@ -136,6 +136,14 @@ int job_format_add_function(List list, int width, bool right_justify,
job_format_add_function(list,wid,right,suffix,_print_job_account)
#define job_format_add_dependency(list,wid,right,suffix) \
job_format_add_function(list,wid,right,suffix,_print_job_dependency)
+#define job_format_add_geometry(list,wid,right,suffix) \
+ job_format_add_function(list,wid,right,suffix,_print_job_geometry)
+#define job_format_add_rotate(list,wid,right,suffix) \
+ job_format_add_function(list,wid,right,suffix,_print_job_rotate)
+#define job_format_add_conn_type(list,wid,right,suffix) \
+ job_format_add_function(list,wid,right,suffix,_print_job_conn_type)
+#define job_format_add_node_use(list,wid,right,suffix) \
+ job_format_add_function(list,wid,right,suffix,_print_job_node_use)
/*****************************************************************************
* Job Line Print Functions
@@ -208,6 +216,14 @@ int _print_job_account(job_info_t * job, int width, bool right_justify,
char* suffix);
int _print_job_dependency(job_info_t * job, int width, bool right_justify,
char* suffix);
+int _print_job_geometry(job_info_t * job, int width, bool right_justify,
+ char* suffix);
+int _print_job_rotate(job_info_t * job, int width, bool right_justify,
+ char* suffix);
+int _print_job_conn_type(job_info_t * job, int width, bool right_justify,
+ char* suffix);
+int _print_job_node_use(job_info_t * job, int width, bool right_justify,
+ char* suffix);
/*****************************************************************************
* Step Print Format Functions
diff --git a/src/srun/opt.c b/src/srun/opt.c
index 9863c5cbba8..2fc9faee868 100644
--- a/src/srun/opt.c
+++ b/src/srun/opt.c
@@ -74,6 +74,10 @@
#define OPT_NODES 0x05
#define OPT_OVERCOMMIT 0x06
#define OPT_CORE 0x07
+#define OPT_CONN_TYPE 0x08
+#define OPT_NODE_USE 0x09
+#define OPT_NO_ROTATE 0x0a
+#define OPT_GEOMETRY 0x0b
/* generic getopt_long flags, integers and *not* valid characters */
#define LONG_OPT_HELP 0x100
@@ -214,11 +218,12 @@ static bool _valid_node_list(char **node_list_pptr)
*/
static enum distribution_t _verify_dist_type(const char *arg)
{
+ int len = strlen(arg);
enum distribution_t result = SRUN_DIST_UNKNOWN;
- if (strncasecmp(arg, "cyclic", strlen(arg)) == 0)
+ if (strncasecmp(arg, "cyclic", len) == 0)
result = SRUN_DIST_CYCLIC;
- else if (strncasecmp(arg, "block", strlen(arg)) == 0)
+ else if (strncasecmp(arg, "block", len) == 0)
result = SRUN_DIST_BLOCK;
return result;
@@ -230,11 +235,13 @@ static enum distribution_t _verify_dist_type(const char *arg)
*/
static int _verify_conn_type(const char *arg)
{
- if (!strcasecmp(arg, "MESH"))
+ int len = strlen(arg);
+
+ if (!strncasecmp(arg, "MESH", len))
return RM_MESH;
- else if (!strcasecmp(arg, "TORUS"))
+ else if (!strncasecmp(arg, "TORUS", len))
return RM_TORUS;
- else if (!strcasecmp(arg, "NAV"))
+ else if (!strncasecmp(arg, "NAV", len))
return RM_NAV;
error("invalid --conn-type argument %s ignored.", arg);
@@ -247,9 +254,11 @@ static int _verify_conn_type(const char *arg)
*/
static int _verify_node_use(const char *arg)
{
- if (!strcasecmp(arg, "VIRTUAL"))
+ int len = strlen(arg);
+
+ if (!strncasecmp(arg, "VIRTUAL", len))
return RM_VIRTUAL;
- else if (!strcasecmp(arg, "COPROCESSOR"))
+ else if (!strncasecmp(arg, "COPROCESSOR", len))
return RM_COPROCESSOR;
error("invalid --node-use argument %s ignored.", arg);
@@ -543,13 +552,17 @@ env_vars_t env_vars[] = {
{"SLURM_ACCOUNT", OPT_STRING, &opt.account, NULL },
{"SLURMD_DEBUG", OPT_INT, &opt.slurmd_debug, NULL },
{"SLURM_CPUS_PER_TASK", OPT_INT, &opt.cpus_per_task, &opt.cpus_set },
- {"SLURM_CORE_FORMAT", OPT_CORE, NULL, NULL, },
+ {"SLURM_CONN_TYPE", OPT_CONN_TYPE, NULL, NULL },
+ {"SLURM_CORE_FORMAT", OPT_CORE, NULL, NULL },
{"SLURM_DEBUG", OPT_DEBUG, NULL, NULL },
{"SLURM_DISTRIBUTION", OPT_DISTRIB, NULL, NULL },
+ {"SLURM_GEOMETRY", OPT_GEOMETRY, NULL, NULL },
{"SLURM_IMMEDIATE", OPT_INT, &opt.immediate, NULL },
{"SLURM_JOBID", OPT_INT, &opt.jobid, NULL },
{"SLURM_LABELIO", OPT_INT, &opt.labelio, NULL },
{"SLURM_NNODES", OPT_NODES, NULL, NULL },
+ {"SLURM_NO_ROTATE", OPT_NO_ROTATE, NULL, NULL },
+ {"SLURM_NODE_USE", OPT_NODE_USE, NULL, NULL },
{"SLURM_NPROCS", OPT_INT, &opt.nprocs, &opt.nprocs_set},
{"SLURM_OVERCOMMIT", OPT_OVERCOMMIT, NULL, NULL },
{"SLURM_PARTITION", OPT_STRING, &opt.partition, NULL },
@@ -639,7 +652,26 @@ _process_env_var(env_vars_t *e, const char *val)
case OPT_CORE:
opt.core_type = core_format_type (val);
break;
-
+
+ case OPT_CONN_TYPE:
+ opt.conn_type = _verify_conn_type(val);
+ break;
+
+ case OPT_NODE_USE:
+ opt.node_use = _verify_node_use(val);
+ break;
+
+ case OPT_NO_ROTATE:
+ opt.no_rotate = true;
+ break;
+
+ case OPT_GEOMETRY:
+ if (_verify_geometry(val, opt.geometry)) {
+ error("\"%s=%s\" -- invalid geometry, ignoring...",
+ e->var, val);
+ }
+ break;
+
default:
/* do nothing */
break;
--
GitLab