From b41b51f19bdb8de7af5d4056f4d89fa2830a983d Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 14 Jan 2011 04:32:05 +0000
Subject: [PATCH] Checkin new select and topology plugins to control node
reording functions. Based upon work by Gerrit plus bug fixes and
documentation that I've added.
---
doc/html/selectplugins.shtml | 16 +++-
doc/html/topology_plugin.shtml | 12 ++-
src/common/node_select.c | 16 ++++
src/common/node_select.h | 10 +++
src/common/slurm_topology.c | 15 ++++
src/common/slurm_topology.h | 7 ++
src/plugins/select/bgq/select_bgq.cc | 5 ++
.../select/bluegene/plugin/select_bluegene.c | 5 ++
src/plugins/select/cons_res/select_cons_res.c | 8 +-
src/plugins/select/cray/other_select.c | 1 +
src/plugins/select/cray/select_cray.c | 10 ++-
src/plugins/select/linear/select_linear.c | 7 +-
src/plugins/topology/3d_torus/hilbert_slurm.c | 65 +--------------
.../topology/3d_torus/topology_3d_torus.c | 22 +++--
.../topology/node_rank/topology_node_rank.c | 76 +++--------------
src/plugins/topology/none/topology_none.c | 10 ++-
src/plugins/topology/tree/topology_tree.c | 9 +-
src/slurmctld/read_config.c | 83 +++++++++++++++++++
18 files changed, 233 insertions(+), 144 deletions(-)
diff --git a/doc/html/selectplugins.shtml b/doc/html/selectplugins.shtml
index 8daac3ef65e..b99e0cf68e0 100644
--- a/doc/html/selectplugins.shtml
+++ b/doc/html/selectplugins.shtml
@@ -185,6 +185,18 @@ but actually registered with 2GB of memory.</p>
of the node in reference to the entire system.<br><br>
<p style="margin-left:.2in"><b>Returns</b>: SLURM_SUCCESS if successful, otherwise SLURM_ERROR</p>
+<p class="commandline">bool select_p_node_ranking(struct node_record *node_ptr, int node_cnt)</p>
+<p style="margin-left:.2in"><b>Description</b>: This function is called by the slurmctld
+daemon at start time to set node rank information for reording the nodes to
+optimize application performance. </p>
+<p style="margin-left:.2in"><b>Arguments</b>:<br>
+<span class="commandline"> node_ptr</span> (input/output) pointer
+to the node data structure. Each node's node rank field may be set.<br>
+<span class="commandline"> node_cnt</span> (input) number
+of nodes configured on the system.</p>
+<p style="margin-left:.2in"><b>Returns</b>: true if node rank information has
+been set.</p>
+
<p class="commandline">int select_p_update_node_state (int index, uint16_t state);</p>
<p style="margin-left:.2in"><b>Description</b>: push a change of state
into the plugin the index should be the index from the slurmctld of
@@ -390,7 +402,7 @@ the plugin should return SLURM_ERROR.</p>
<p class="footer"><a href="#top">top</a></p>
<h2>Versioning</h2>
-<p> This document describes version 1 of the SLURM node selection API. Future
+<p> This document describes version 100 of the SLURM node selection API. Future
releases of SLURM may revise this API. A node selection plugin conveys its ability
to implement a particular API version using the mechanism outlined for SLURM plugins.
In addition, the credential is transmitted along with the version number of the
@@ -399,6 +411,6 @@ to maintain data format compatibility across different versions of the plugin.</
<p class="footer"><a href="#top">top</a></p>
-<p style="text-align:center;">Last modified 5 October 2009</p>
+<p style="text-align:center;">Last modified 13 January 2011</p>
<!--#include virtual="footer.txt"-->
diff --git a/doc/html/topology_plugin.shtml b/doc/html/topology_plugin.shtml
index 0cf8ce1a725..3cd51fd5854 100644
--- a/doc/html/topology_plugin.shtml
+++ b/doc/html/topology_plugin.shtml
@@ -7,7 +7,7 @@
defines them.
It is intended as a resource to programmers wishing to write their own
SLURM topology plugin.
-This is version 100 of the API.</p>
+This is version 101 of the API.</p>
<p>SLURM topology plugins are SLURM plugins that implement
convey system topology information so that SLURM is able to
@@ -64,6 +64,12 @@ Functions which are not implemented should be stubbed.</p>
<p style="margin-left:.2in"><b>Returns</b>: SLURM_SUCCESS or
SLURM_ERROR on failure.</p>
+<p class="commandline">bool topo_generate_node_ranking(void)</p>
+<p style="margin-left:.2in"><b>Description</b>: Determine if this plugin will
+reorder the node records based upon each job's node rank field.</p>
+<p style="margin-left:.2in"><b>Returns</b>: true if node reording is supported,
+false otherwise.</p>
+
<p class="commandline">int topo_get_node_addr(char* node_name, char** paddr, char** ppatt);</p>
<p style="margin-left:.2in"><b>Description</b>: Get Topology address of a given node.</p>
<p style="margin-left:.2in"><b>Arguments</b>:<br>
@@ -80,10 +86,10 @@ in the hierarchy is separated by a period. The final element will always be
SLURM_ERROR on failure.</p>
<h2>Versioning</h2>
-<p> This document describes version 100 of the SLURM topology API.
+<p> This document describes version 101 of the SLURM topology API.
Future releases of SLURM may revise this API.</p>
<p class="footer"><a href="#top">top</a></p>
-<p style="text-align:center;">Last modified 27 August 2009</p>
+<p style="text-align:center;">Last modified 13 January 2011</p>
<!--#include virtual="footer.txt"-->
diff --git a/src/common/node_select.c b/src/common/node_select.c
index bc2de083a0d..67358fdd45a 100644
--- a/src/common/node_select.c
+++ b/src/common/node_select.c
@@ -79,6 +79,7 @@ static int _select_get_ops(char *select_type,
"select_p_state_save",
"select_p_state_restore",
"select_p_job_init",
+ "select_p_node_ranking",
"select_p_node_init",
"select_p_block_init",
"select_p_job_test",
@@ -456,6 +457,21 @@ extern int select_g_job_init(List job_list)
(job_list);
}
+/*
+ * Assign a 'node_rank' value to each of the node_ptr entries.
+ * IN node_ptr - current node data
+ * IN node_count - number of node entries
+ * Return true if node ranking was performed, false if not.
+ */
+extern bool select_g_node_ranking(struct node_record *node_ptr, int node_cnt)
+{
+ if (slurm_select_init(0) < 0)
+ return SLURM_ERROR;
+
+ return (*(select_context[select_context_default].ops.node_ranking))
+ (node_ptr, node_cnt);
+}
+
/*
* Note re/initialization of node record data structure
* IN node_ptr - current node data
diff --git a/src/common/node_select.h b/src/common/node_select.h
index 85dc3556266..92d846eb3d4 100644
--- a/src/common/node_select.h
+++ b/src/common/node_select.h
@@ -69,6 +69,8 @@ typedef struct slurm_select_ops {
int (*state_save) (char *dir_name);
int (*state_restore) (char *dir_name);
int (*job_init) (List job_list);
+ int (*node_ranking) (struct node_record *node_ptr,
+ int node_cnt);
int (*node_init) (struct node_record *node_ptr,
int node_cnt);
int (*block_init) (List block_list);
@@ -194,6 +196,14 @@ extern int select_g_state_restore(char *dir_name);
*/
extern int select_g_job_init(List job_list);
+/*
+ * Assign a 'node_rank' value to each of the node_ptr entries.
+ * IN node_ptr - current node data
+ * IN node_count - number of node entries
+ * Return true if node ranking was performed, false if not.
+ */
+extern bool select_g_node_ranking(struct node_record *node_ptr, int node_cnt);
+
/*
* Note re/initialization of node record data structure
* IN node_ptr - current node data
diff --git a/src/common/slurm_topology.c b/src/common/slurm_topology.c
index a3a54ec497d..932707cc5df 100644
--- a/src/common/slurm_topology.c
+++ b/src/common/slurm_topology.c
@@ -55,6 +55,7 @@ int switch_record_cnt = 0;
/* ************************************************************************ */
typedef struct slurm_topo_ops {
int (*build_config) ( void );
+ bool (*node_ranking) ( void );
int (*get_node_addr) ( char* node_name,
char** addr,
char** pattern );
@@ -87,6 +88,7 @@ slurm_topo_get_ops( slurm_topo_context_t *c )
*/
static const char *syms[] = {
"topo_build_config",
+ "topo_generate_node_ranking",
"topo_get_node_addr",
};
int n_syms = sizeof( syms ) / sizeof( char * );
@@ -268,6 +270,19 @@ slurm_topo_build_config( void )
return rc;
}
+/* *********************************************************************** */
+/* TAG( slurm_topo_generate_node_ranking ) */
+/* NOTE: This operation is only supported by those topology plugins for */
+/* which the node ordering between slurmd and slurmctld is invariant */
+/* *********************************************************************** */
+extern bool
+slurm_topo_generate_node_ranking( void )
+{
+ if ( slurm_topo_init() < 0 )
+ return SLURM_ERROR;
+
+ return (*(g_topo_context->ops.node_ranking))();
+}
/* *********************************************************************** */
/* TAG( slurm_topo_get_node_addr ) */
diff --git a/src/common/slurm_topology.h b/src/common/slurm_topology.h
index 7b144c8f593..dfda3e4b4b9 100644
--- a/src/common/slurm_topology.h
+++ b/src/common/slurm_topology.h
@@ -89,6 +89,13 @@ extern int slurm_topo_fini(void);
*/
extern int slurm_topo_build_config( void );
+/*
+ * slurm_topo_generate_node_ranking - populate node_rank fields
+ * NOTE: This operation is only supported by those topology plugins for
+ * which the node ordering between slurmd and slurmctld is invariant.
+ */
+extern bool slurm_topo_generate_node_ranking( void );
+
/*
* slurm_topo_get_node_addr - build node address and the associated pattern
* based on the topology information
diff --git a/src/plugins/select/bgq/select_bgq.cc b/src/plugins/select/bgq/select_bgq.cc
index d017bc01195..50d4ac6d690 100644
--- a/src/plugins/select/bgq/select_bgq.cc
+++ b/src/plugins/select/bgq/select_bgq.cc
@@ -154,6 +154,11 @@ extern int select_p_job_init(List job_list)
#endif
}
+extern bool select_p_node_ranking(struct node_record *node_ptr, int node_cnt)
+{
+ return false;
+}
+
/* All initialization is performed by init() */
extern int select_p_node_init(struct node_record *node_ptr, int node_cnt)
{
diff --git a/src/plugins/select/bluegene/plugin/select_bluegene.c b/src/plugins/select/bluegene/plugin/select_bluegene.c
index fb4aadc6ac2..f72967454cc 100644
--- a/src/plugins/select/bluegene/plugin/select_bluegene.c
+++ b/src/plugins/select/bluegene/plugin/select_bluegene.c
@@ -462,6 +462,11 @@ extern int select_p_job_init(List job_list)
#endif
}
+extern bool select_p_node_ranking(struct node_record *node_ptr, int node_cnt)
+{
+ return false;
+}
+
/* All initialization is performed by init() */
extern int select_p_node_init(struct node_record *node_ptr, int node_cnt)
{
diff --git a/src/plugins/select/cons_res/select_cons_res.c b/src/plugins/select/cons_res/select_cons_res.c
index c3f23c5d31a..b6a0527fe71 100644
--- a/src/plugins/select/cons_res/select_cons_res.c
+++ b/src/plugins/select/cons_res/select_cons_res.c
@@ -168,7 +168,7 @@ bitstr_t *idle_node_bitmap;
const char plugin_name[] = "Consumable Resources (CR) Node Selection plugin";
const char plugin_type[] = "select/cons_res";
const uint32_t plugin_id = 101;
-const uint32_t plugin_version = 91;
+const uint32_t plugin_version = 100;
const uint32_t pstate_version = 7; /* version control on saved state */
uint16_t cr_type = CR_CPU; /* cr_type is overwritten in init() */
@@ -1676,6 +1676,12 @@ extern int select_p_job_init(List job_list)
return SLURM_SUCCESS;
}
+/* This plugin does not generate a node ranking. */
+extern bool select_p_node_ranking(struct node_record *node_ptr, int node_cnt)
+{
+ return false;
+}
+
/* This is Part 1 of a 4-part procedure which can be found in
* src/slurmctld/read_config.c. The whole story goes like this:
*
diff --git a/src/plugins/select/cray/other_select.c b/src/plugins/select/cray/other_select.c
index 33f79feb90d..29d1c10071a 100644
--- a/src/plugins/select/cray/other_select.c
+++ b/src/plugins/select/cray/other_select.c
@@ -78,6 +78,7 @@ static slurm_select_ops_t *_other_select_get_ops(slurm_select_context_t *c)
"select_p_state_save",
"select_p_state_restore",
"select_p_job_init",
+ "select_p_node_ranking",
"select_p_node_init",
"select_p_block_init",
"select_p_job_test",
diff --git a/src/plugins/select/cray/select_cray.c b/src/plugins/select/cray/select_cray.c
index 5eff8fbacbc..b8ceabc0669 100644
--- a/src/plugins/select/cray/select_cray.c
+++ b/src/plugins/select/cray/select_cray.c
@@ -126,7 +126,7 @@ struct select_nodeinfo {
const char plugin_name[] = "Cray node selection plugin";
const char plugin_type[] = "select/cray";
uint32_t plugin_id = 104;
-const uint32_t plugin_version = 1;
+const uint32_t plugin_version = 100;
/*
@@ -179,6 +179,14 @@ extern int select_p_job_init(List job_list)
return other_job_init(job_list);
}
+/*
+ * select_p_node_ranking - generate node ranking for Cray nodes
+ */
+extern bool select_p_node_ranking(struct node_record *node_ptr, int node_cnt)
+{
+ return false; /* FIXME - to be filled in */
+}
+
extern int select_p_node_init(struct node_record *node_ptr, int node_cnt)
{
return other_node_init(node_ptr, node_cnt);
diff --git a/src/plugins/select/linear/select_linear.c b/src/plugins/select/linear/select_linear.c
index d4d31875d21..412b8ba3f7a 100644
--- a/src/plugins/select/linear/select_linear.c
+++ b/src/plugins/select/linear/select_linear.c
@@ -191,7 +191,7 @@ extern int select_p_select_nodeinfo_free(select_nodeinfo_t *nodeinfo);
const char plugin_name[] = "Linear node selection plugin";
const char plugin_type[] = "select/linear";
const uint32_t plugin_id = 102;
-const uint32_t plugin_version = 90;
+const uint32_t plugin_version = 100;
static struct node_record *select_node_ptr = NULL;
static int select_node_cnt = 0;
@@ -2352,6 +2352,11 @@ extern int select_p_job_init(List job_list)
return SLURM_SUCCESS;
}
+extern bool select_p_node_ranking(struct node_record *node_ptr, int node_cnt)
+{
+ return false;
+}
+
extern int select_p_node_init(struct node_record *node_ptr, int node_cnt)
{
if (node_ptr == NULL) {
diff --git a/src/plugins/topology/3d_torus/hilbert_slurm.c b/src/plugins/topology/3d_torus/hilbert_slurm.c
index 1ecf8ba1021..7e68c4c0160 100644
--- a/src/plugins/topology/3d_torus/hilbert_slurm.c
+++ b/src/plugins/topology/3d_torus/hilbert_slurm.c
@@ -61,10 +61,9 @@ static int _coord(char coord)
* be called once, immediately after reading the slurm.conf file. */
extern void nodes_to_hilbert_curve(void)
{
- int coord_inx, i, j, k, max_coord = 0, min_inx;
- uint32_t min_val;
+ int coord_inx, i, j, k, max_coord = 0;
int *coords;
- struct node_record *node_ptr, *node_ptr2;
+ struct node_record *node_ptr;
coord_t hilbert[3];
int dims = 3;
#ifdef HAVE_SUN_CONST
@@ -127,64 +126,4 @@ extern void nodes_to_hilbert_curve(void)
((hilbert[1]>>0 & 1) << 1) +
((hilbert[2]>>0 & 1) << 0);
}
-
- /* Now we need to sort the node records. We only need to move a few
- * fields since the others were all initialized to identical values.
- * The fields needing to be copied are those set by the function
- * _build_single_nodeline_info() in src/common/read_conf.c */
- for (i=0; i<node_record_count; i++) {
- min_val = node_record_table_ptr[i].node_rank;
- min_inx = i;
- for (j=(i+1); j<node_record_count; j++) {
- if (node_record_table_ptr[j].node_rank < min_val) {
- min_val = node_record_table_ptr[j].node_rank;
- min_inx = j;
- }
- }
- if (min_inx != i) { /* swap records */
- char *tmp_str;
- uint16_t tmp_uint16;
- uint32_t tmp_uint32;
-
- node_ptr = node_record_table_ptr + i;
- node_ptr2 = node_record_table_ptr + min_inx;
-
- tmp_str = node_ptr->name;
- node_ptr->name = node_ptr2->name;
- node_ptr2->name = tmp_str;
-
- tmp_str = node_ptr->comm_name;
- node_ptr->comm_name = node_ptr2->comm_name;
- node_ptr2->comm_name = tmp_str;
-
- tmp_uint32 = node_ptr->node_rank;
- node_ptr->node_rank = node_ptr2->node_rank;
- node_ptr2->node_rank = tmp_uint32;
-
- tmp_str = node_ptr->features;
- node_ptr->features = node_ptr2->features;
- node_ptr2->features = tmp_str;
-
- tmp_uint16 = node_ptr->port;
- node_ptr->port = node_ptr2->port;
- node_ptr2->port = tmp_uint16;
-
- tmp_str = node_ptr->reason;
- node_ptr->reason = node_ptr2->reason;
- node_ptr2->reason = tmp_str;
-
- tmp_uint32 = node_ptr->weight;
- node_ptr->weight = node_ptr2->weight;
- node_ptr2->weight = tmp_uint32;
- }
- }
-
-#if _DEBUG
- /* Log the results */
- for (i=0, node_ptr=node_record_table_ptr; i<node_record_count;
- i++, node_ptr++) {
- info("%s: %u", node_ptr->name, node_ptr->node_rank);
- }
-#endif
}
-
diff --git a/src/plugins/topology/3d_torus/topology_3d_torus.c b/src/plugins/topology/3d_torus/topology_3d_torus.c
index c2d37a8a603..5220be4630a 100644
--- a/src/plugins/topology/3d_torus/topology_3d_torus.c
+++ b/src/plugins/topology/3d_torus/topology_3d_torus.c
@@ -79,7 +79,7 @@
*/
const char plugin_name[] = "topology 3d_torus plugin";
const char plugin_type[] = "topology/3d_torus";
-const uint32_t plugin_version = 100;
+const uint32_t plugin_version = 101;
extern void nodes_to_hilbert_curve(void);
@@ -107,19 +107,29 @@ extern int fini(void)
* after a system startup or reconfiguration.
*/
extern int topo_build_config(void)
-{ static bool first_run = true;
+{
+ return SLURM_SUCCESS;
+}
+
+/*
+ * topo_generate_node_ranking - populate node_rank fields
+ */
+extern bool topo_generate_node_ranking(void)
+{
+#ifdef HAVE_BG
+ return false;
+#else
+ static bool first_run = true;
/* We can only re-order the nodes once at slurmctld startup.
* After that time, many bitmaps are created based upon the
* index of each node name in the array. */
if (!first_run)
- return SLURM_SUCCESS;
- first_run = false;
+ return false;
-#ifndef HAVE_BG
nodes_to_hilbert_curve();
+ return true;
#endif
- return SLURM_SUCCESS;
}
/*
diff --git a/src/plugins/topology/node_rank/topology_node_rank.c b/src/plugins/topology/node_rank/topology_node_rank.c
index 3c4802948a8..7abaaca27a0 100644
--- a/src/plugins/topology/node_rank/topology_node_rank.c
+++ b/src/plugins/topology/node_rank/topology_node_rank.c
@@ -84,7 +84,7 @@
*/
const char plugin_name[] = "topology node_rank plugin";
const char plugin_type[] = "topology/node_rank";
-const uint32_t plugin_version = 100;
+const uint32_t plugin_version = 101;
/*
* init() is called when the plugin is loaded, before any other functions
@@ -110,79 +110,25 @@ extern int fini(void)
* after a system startup or reconfiguration.
*/
extern int topo_build_config(void)
+{
+ return SLURM_SUCCESS;
+}
+
+/*
+ * topo_generate_node_ranking - populate node_rank fields
+ */
+extern bool topo_generate_node_ranking(void)
{
static bool first_run = true;
- struct node_record *node_ptr, *node_ptr2;
- int i, j, min_inx;
- uint32_t min_val;
/* We can only re-order the nodes once at slurmctld startup.
* After that time, many bitmaps are created based upon the
* index of each node name in the array. */
if (!first_run)
- return SLURM_SUCCESS;
+ return false;
first_run = false;
- /* Now we need to sort the node records. We only need to move a few
- * fields since the others were all initialized to identical values.
- * The fields needing to be copied are those set by the function
- * _build_single_nodeline_info() in src/common/read_conf.c */
- for (i=0; i<node_record_count; i++) {
- min_val = node_record_table_ptr[i].node_rank;
- min_inx = i;
- for (j=(i+1); j<node_record_count; j++) {
- if (node_record_table_ptr[j].node_rank < min_val) {
- min_val = node_record_table_ptr[j].node_rank;
- min_inx = j;
- }
- }
- if (min_inx != i) { /* swap records */
- char *tmp_str;
- uint16_t tmp_uint16;
- uint32_t tmp_uint32;
-
- node_ptr = node_record_table_ptr + i;
- node_ptr2 = node_record_table_ptr + min_inx;
-
- tmp_str = node_ptr->name;
- node_ptr->name = node_ptr2->name;
- node_ptr2->name = tmp_str;
-
- tmp_str = node_ptr->comm_name;
- node_ptr->comm_name = node_ptr2->comm_name;
- node_ptr2->comm_name = tmp_str;
-
- tmp_uint32 = node_ptr->node_rank;
- node_ptr->node_rank = node_ptr2->node_rank;
- node_ptr2->node_rank = tmp_uint32;
-
- tmp_str = node_ptr->features;
- node_ptr->features = node_ptr2->features;
- node_ptr2->features = tmp_str;
-
- tmp_uint16 = node_ptr->port;
- node_ptr->port = node_ptr2->port;
- node_ptr2->port = tmp_uint16;
-
- tmp_str = node_ptr->reason;
- node_ptr->reason = node_ptr2->reason;
- node_ptr2->reason = tmp_str;
-
- tmp_uint32 = node_ptr->weight;
- node_ptr->weight = node_ptr2->weight;
- node_ptr2->weight = tmp_uint32;
- }
- }
-
-#if _DEBUG
- /* Log the results */
- for (i=0, node_ptr=node_record_table_ptr; i<node_record_count;
- i++, node_ptr++) {
- info("%s: %u", node_ptr->name, node_ptr->node_rank);
- }
-#endif
-
- return SLURM_SUCCESS;
+ return true;
}
/*
diff --git a/src/plugins/topology/none/topology_none.c b/src/plugins/topology/none/topology_none.c
index 20b0cb51739..806ac300809 100644
--- a/src/plugins/topology/none/topology_none.c
+++ b/src/plugins/topology/none/topology_none.c
@@ -77,7 +77,7 @@
*/
const char plugin_name[] = "topology NONE plugin";
const char plugin_type[] = "topology/none";
-const uint32_t plugin_version = 100;
+const uint32_t plugin_version = 101;
/*
* init() is called when the plugin is loaded, before any other functions
@@ -107,6 +107,14 @@ extern int topo_build_config(void)
return SLURM_SUCCESS;
}
+/*
+ * topo_generate_node_ranking - this plugin does not set any node_rank fields
+ */
+extern bool topo_generate_node_ranking(void)
+{
+ return false;
+}
+
/*
* topo_get_node_addr - build node address and the associated pattern
* based on the topology information
diff --git a/src/plugins/topology/tree/topology_tree.c b/src/plugins/topology/tree/topology_tree.c
index 07ea2bab125..12939a16bce 100644
--- a/src/plugins/topology/tree/topology_tree.c
+++ b/src/plugins/topology/tree/topology_tree.c
@@ -81,7 +81,7 @@
*/
const char plugin_name[] = "topology tree plugin";
const char plugin_type[] = "topology/tree";
-const uint32_t plugin_version = 100;
+const uint32_t plugin_version = 101;
typedef struct slurm_conf_switches {
uint32_t link_speed; /* link speed, arbitrary units */
@@ -139,6 +139,13 @@ extern int topo_build_config(void)
return SLURM_SUCCESS;
}
+/*
+ * topo_generate_node_ranking - this plugin does not set any node_rank fields
+ */
+extern bool topo_generate_node_ranking(void)
+{
+ return false;
+}
/*
* topo_get_node_addr - build node address and the associated pattern
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 5e891aabba4..4c3e25a3e71 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -121,6 +121,77 @@ static int _update_preempt(uint16_t old_enable_preempt);
static void _validate_node_proc_count(void);
#endif
+/*
+ * _reorder_node_record_table - order node table in ascending order of node_rank
+ * This depends on the TopologyPlugin, which may generate such a ranking.
+ */
+static void _reorder_node_record_table(void)
+{
+ struct node_record *node_ptr, *node_ptr2;
+ int i, j, min_inx;
+ uint32_t min_val;
+
+ /* Now we need to sort the node records. We only need to move a few
+ * fields since the others were all initialized to identical values.
+ * The fields needing to be copied are those set by the function
+ * _build_single_nodeline_info() in src/common/read_conf.c */
+ for (i = 0; i < node_record_count; i++) {
+ min_val = node_record_table_ptr[i].node_rank;
+ min_inx = i;
+ for (j = i + 1; j < node_record_count; j++) {
+ if (node_record_table_ptr[j].node_rank < min_val) {
+ min_val = node_record_table_ptr[j].node_rank;
+ min_inx = j;
+ }
+ }
+ if (min_inx != i) { /* swap records */
+ char *tmp_str;
+ uint16_t tmp_uint16;
+ uint32_t tmp_uint32;
+
+ node_ptr = node_record_table_ptr + i;
+ node_ptr2 = node_record_table_ptr + min_inx;
+
+ tmp_str = node_ptr->name;
+ node_ptr->name = node_ptr2->name;
+ node_ptr2->name = tmp_str;
+
+ tmp_str = node_ptr->comm_name;
+ node_ptr->comm_name = node_ptr2->comm_name;
+ node_ptr2->comm_name = tmp_str;
+
+ tmp_uint32 = node_ptr->node_rank;
+ node_ptr->node_rank = node_ptr2->node_rank;
+ node_ptr2->node_rank = tmp_uint32;
+
+ tmp_str = node_ptr->features;
+ node_ptr->features = node_ptr2->features;
+ node_ptr2->features = tmp_str;
+
+ tmp_uint16 = node_ptr->port;
+ node_ptr->port = node_ptr2->port;
+ node_ptr2->port = tmp_uint16;
+
+ tmp_str = node_ptr->reason;
+ node_ptr->reason = node_ptr2->reason;
+ node_ptr2->reason = tmp_str;
+
+ tmp_uint32 = node_ptr->weight;
+ node_ptr->weight = node_ptr2->weight;
+ node_ptr2->weight = tmp_uint32;
+ }
+ }
+
+#if _DEBUG
+ /* Log the results */
+ for (i=0, node_ptr = node_record_table_ptr; i < node_record_count;
+ i++, node_ptr++) {
+ info("%s: %u", node_ptr->name, node_ptr->node_rank);
+ }
+#endif
+}
+
+
/*
* _build_bitmaps_pre_select - recover some state for jobs and nodes prior to
* calling the select_* functions
@@ -633,6 +704,7 @@ int read_slurm_conf(int recover, bool reconfig)
int error_code, i, rc, load_job_ret = SLURM_SUCCESS;
int old_node_record_count = 0;
struct node_record *old_node_table_ptr = NULL, *node_ptr;
+ bool do_reorder_nodes = false;
List old_part_list = NULL;
char *old_def_part_name = NULL;
char *old_auth_type = xstrdup(slurmctld_conf.authtype);
@@ -713,6 +785,17 @@ int read_slurm_conf(int recover, bool reconfig)
return EINVAL;
}
+ /*
+ * Node reordering needs to be done by the topology and/or select
+ * plugin. Reordering the table must be done before hashing the
+ * nodes, and before any position-relative bitmaps are created.
+ */
+ do_reorder_nodes |= slurm_topo_generate_node_ranking();
+ do_reorder_nodes |= select_g_node_ranking(node_record_table_ptr,
+ node_record_count);
+ if (do_reorder_nodes)
+ _reorder_node_record_table();
+
rehash_node();
rehash_jobs();
set_slurmd_addr();
--
GitLab