From 215a7b93fdc46d460b7d6c4f97585420ff95dba4 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Tue, 1 Nov 2005 19:33:38 +0000
Subject: [PATCH] Permit nodes to be in multiple partitions simultaneously.
---
NEWS | 1 +
doc/html/quickstart_admin.html | 11 ++-
doc/man/man5/slurm.conf.5 | 18 +++--
etc/slurm.conf.example | 2 +-
slurm/slurm.h.in | 1 -
src/api/node_info.c | 13 +---
src/common/slurm_protocol_defs.c | 1 -
src/common/slurm_protocol_pack.c | 2 -
src/plugins/sched/backfill/backfill.c | 11 ++-
src/sinfo/sinfo.c | 104 +++++++++++++-------------
src/slurmctld/node_mgr.c | 50 +++++++++----
src/slurmctld/partition_mgr.c | 70 ++++++++++++-----
src/slurmctld/read_config.c | 37 ++++-----
src/slurmctld/sched_upcalls.c | 11 ++-
src/slurmctld/slurmctld.h | 4 +-
testsuite/expect/test3.5 | 72 +++++++++++++++++-
16 files changed, 268 insertions(+), 140 deletions(-)
diff --git a/NEWS b/NEWS
index 1659de504c2..188e01d6e1f 100644
--- a/NEWS
+++ b/NEWS
@@ -7,6 +7,7 @@ documents those changes that are of interest to users and admins.
with a DRAIN flag. COMPLETING state is changed to a COMPLETING flag.
-- Test suite moved into testsuite/expect from separate repository.
-- Added new document describing slurm APIs (doc/html/api.html).
+ -- Permit nodes to be in multiple partitions simultaneously.
* Changes in SLURM 0.7.0-pre2
=============================
diff --git a/doc/html/quickstart_admin.html b/doc/html/quickstart_admin.html
index d2a2d652f22..6275ba6243d 100644
--- a/doc/html/quickstart_admin.html
+++ b/doc/html/quickstart_admin.html
@@ -9,7 +9,7 @@
<meta http-equiv="keywords" content="Simple Linux Utility for Resource Management, SLURM, resource management,
Linux clusters, high-performance computing, Livermore Computing">
<meta name="LLNLRandR" content="UCRL-WEB-213976">
-<meta name="LLNLRandRdate" content="25 October 2005">
+<meta name="LLNLRandRdate" content="1 November 2005">
<meta name="distribution" content="global">
<meta name="description" content="Simple Linux Utility for Resource Management">
<meta name="copyright"
@@ -315,8 +315,11 @@ In this case "emcri" is the private management network interface
for the host "mcri". Port numbers to be used for
communications are specified as well as various timer values.</p>
-<p>A description of the nodes and their grouping into non-overlapping partitions
-is required. Partition and node specifications use node range expressions to identify
+<p>A description of the nodes and their grouping into partitions is required.
+Nodes can be in more than one partition and each partition can have different
+constraints (permitted users, time limits, job size limits, etc.).
+Each partition can thus be considered a separate queue.
+Partition and node specifications use node range expressions to identify
nodes in a concise fashion. This configuration file defines a 1154-node cluster
for SLURM, but it might be used for a much larger cluster by just changing a few
node range expressions. Specify the minimum processor count (Procs), real memory
@@ -571,7 +574,7 @@ in the NEWS file.
<td colspan="3"><hr> <p>For information about this page, contact <a href="mailto:slurm-dev@lists.llnl.gov">slurm-dev@lists.llnl.gov</a>.</p>
<p><a href="http://www.llnl.gov/"><img align=middle src="lll.gif" width="32" height="32" border="0"></a></p>
<p class="footer">UCRL-WEB-213976<br>
-Last modified 25 October 2005</p></td>
+Last modified 1 November 2005</p></td>
</tr>
</table>
</td>
diff --git a/doc/man/man5/slurm.conf.5 b/doc/man/man5/slurm.conf.5
index 47db07119d9..98dd95b7e4e 100644
--- a/doc/man/man5/slurm.conf.5
+++ b/doc/man/man5/slurm.conf.5
@@ -666,9 +666,13 @@ Weight is an integer value with a default value of 1.
.LP
The partition configuration permits you to establish different job
limits or access controls for various groups (or partitions) of nodes.
-Nodes may be in only one partition. Jobs are allocated resources
-within a single partition. The partition configuration
-file contains the following information:
+Nodes may be in more than one partition, making partitions serve
+as general purpose queues.
+For example one may put the same set of nodes into two different
+partitions, each with different constraints (time limit, job sizes,
+groups allowed to use the partition, etc.).
+Jobs are allocated resources within a single partition.
+The partition configuration file contains the following information:
.TP
\fBAllowGroups\fR
Comma separated list of group IDs which may execute jobs in the partition.
@@ -843,11 +847,13 @@ NodeName=dev20 State=DOWN Reason="power,ETA=Dec25"
.br
#
.br
-PartitionName=DEFAULT MaxTime=30 MaxNodes=10
+PartitionName=DEFAULT MaxTime=30 MaxNodes=10 State=UP
.br
-PartitionName=debug Nodes=dev[0-8,18-25] State=UP Default=YES
+PartitionName=debug Nodes=dev[0-8,18-25] Default=YES
.br
-PartitionName=batch Nodes=dev[9-17] State=UP MinNodes=4
+PartitionName=batch Nodes=dev[9-17] MinNodes=4
+.br
+PartitionName=long Nodes=dev[9-17] MaxTime=120 AllowGroups=admin
.SH "COPYING"
Copyright (C) 2002 The Regents of the University of California.
diff --git a/etc/slurm.conf.example b/etc/slurm.conf.example
index 6e0830d231a..53b1d4e32e3 100644
--- a/etc/slurm.conf.example
+++ b/etc/slurm.conf.example
@@ -549,7 +549,7 @@ JobAcctType=jobacct/none
# o Partition Configuration
#
# Paritions are groups of nodes which (possibly) have different limits
-# and access controls. Nodes may only be in one partition and jobs will
+# and access controls. Nodes may be in multiple partitions. Jobs will
# not be allowed to span partitions. The following partition configuration
# parameters are recognized:
#
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index fcbefd2db86..c8fef5607b3 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -448,7 +448,6 @@ typedef struct node_info {
uint32_t tmp_disk; /* configured MB of total disk in TMP_FS */
uint32_t weight; /* arbitrary priority of node for scheduling */
char *features; /* arbitrary list of features for node */
- char *partition; /* name of partition node configured to */
char *reason; /* reason for node being DOWN or DRAINING */
} node_info_t;
diff --git a/src/api/node_info.c b/src/api/node_info.c
index 3a59be12eca..fe3c1b87b0e 100644
--- a/src/api/node_info.c
+++ b/src/api/node_info.c
@@ -102,20 +102,13 @@ slurm_print_node_table ( FILE * out, node_info_t * node_ptr, int one_liner )
fprintf ( out, "\n ");
/****** Line 2 ******/
- fprintf ( out, "Weight=%u Partition=%s Features=%s",
- node_ptr->weight, node_ptr->partition, node_ptr->features);
+ fprintf ( out, "Weight=%u Features=%s " ,
+ node_ptr->weight, node_ptr->features);
+ fprintf ( out, "Reason=%s", node_ptr->reason);
if (one_liner)
fprintf ( out, " ");
else
fprintf ( out, "\n ");
-
-
- /****** Line 3 ******/
- fprintf ( out, "Reason=%s", node_ptr->reason);
- if (one_liner)
- fprintf ( out, "\n");
- else
- fprintf ( out, "\n\n");
}
diff --git a/src/common/slurm_protocol_defs.c b/src/common/slurm_protocol_defs.c
index 9ffd53dbd9d..18fe226a9d3 100644
--- a/src/common/slurm_protocol_defs.c
+++ b/src/common/slurm_protocol_defs.c
@@ -866,7 +866,6 @@ static void _slurm_free_node_info_members(node_info_t * node)
if (node) {
xfree(node->name);
xfree(node->features);
- xfree(node->partition);
xfree(node->reason);
}
}
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 54bd1865a07..e3e6619036b 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -1168,7 +1168,6 @@ _unpack_node_info_members(node_info_t * node, Buf buffer)
safe_unpack32(&node->tmp_disk, buffer);
safe_unpack32(&node->weight, buffer);
safe_unpackstr_xmalloc(&node->features, &uint16_tmp, buffer);
- safe_unpackstr_xmalloc(&node->partition, &uint16_tmp, buffer);
safe_unpackstr_xmalloc(&node->reason, &uint16_tmp, buffer);
return SLURM_SUCCESS;
@@ -1176,7 +1175,6 @@ _unpack_node_info_members(node_info_t * node, Buf buffer)
unpack_error:
xfree(node->name);
xfree(node->features);
- xfree(node->partition);
xfree(node->reason);
return SLURM_ERROR;
}
diff --git a/src/plugins/sched/backfill/backfill.c b/src/plugins/sched/backfill/backfill.c
index 57781a9c182..a07e7c0aeb6 100644
--- a/src/plugins/sched/backfill/backfill.c
+++ b/src/plugins/sched/backfill/backfill.c
@@ -317,7 +317,7 @@ _attempt_backfill(struct part_record *part_ptr)
static void
_get_part_specs(struct part_record *part_ptr, part_specs_t *part_specs)
{
- int i;
+ int i, j;
part_specs->idle_node_cnt = 0;
part_specs->max_cpus = 0;
@@ -327,8 +327,15 @@ _get_part_specs(struct part_record *part_ptr, part_specs_t *part_specs)
for (i=0; i<node_record_count; i++) {
struct node_record *node_ptr = &node_record_table_ptr[i];
+ bool found_part = false;
- if (node_ptr->partition_ptr != part_ptr)
+ for (j=0; j<node_ptr->part_cnt; j++) {
+ if (node_ptr->part_pptr[j] != part_ptr)
+ continue;
+ found_part = true;
+ break;
+ }
+ if (found_part == false)
continue; /* different partition */
if (node_ptr->node_state == NODE_STATE_IDLE)
part_specs->idle_node_cnt++;
diff --git a/src/sinfo/sinfo.c b/src/sinfo/sinfo.c
index 6168f66a489..af1930322dd 100644
--- a/src/sinfo/sinfo.c
+++ b/src/sinfo/sinfo.c
@@ -57,8 +57,7 @@ static void _create_sinfo(List sinfo_list, partition_info_t* part_ptr,
uint16_t part_inx, node_info_t *node_ptr);
static bool _filter_out(node_info_t *node_ptr);
static void _sinfo_list_delete(void *data);
-static partition_info_t *_find_part(char *part_name,
- partition_info_msg_t *partition_msg, uint16_t *part_inx);
+static node_info_t *_find_node(char *node_name, node_info_msg_t *node_msg);
static bool _match_node_data(sinfo_data_t *sinfo_ptr,
node_info_t *node_ptr);
static bool _match_part_data(sinfo_data_t *sinfo_ptr,
@@ -67,8 +66,7 @@ static int _query_server(partition_info_msg_t ** part_pptr,
node_info_msg_t ** node_pptr);
static void _sort_hostlist(List sinfo_list);
static int _strcmp(char *data1, char *data2);
-static void _update_sinfo(sinfo_data_t *sinfo_ptr,
- partition_info_t* part_ptr, node_info_t *node_ptr);
+static void _update_sinfo(sinfo_data_t *sinfo_ptr, node_info_t *node_ptr);
int main(int argc, char *argv[])
{
@@ -277,7 +275,9 @@ static int _build_sinfo_data(List sinfo_list,
partition_info_t* part_ptr;
ListIterator i;
int j;
- uint16_t part_inx;
+ hostlist_t hl;
+ sinfo_data_t *sinfo_ptr;
+ char *node_name = NULL;
/* by default every partition is shown, even if no nodes */
if ((!params.node_flag) && params.match_flags.partition_flag) {
@@ -290,37 +290,42 @@ static int _build_sinfo_data(List sinfo_list,
}
}
- /* make sinfo_list entries for each node */
- for (j=0; j<node_msg->record_count; j++) {
- sinfo_data_t *sinfo_ptr;
- node_ptr = &(node_msg->node_array[j]);
-
- if (params.filtering && _filter_out(node_ptr))
- continue;
-
- part_ptr = _find_part(node_ptr->partition, partition_msg,
- &part_inx);
- if ( ! part_ptr )
+ /* make sinfo_list entries for every node in every partition */
+ for (j=0; j<partition_msg->record_count; j++, part_ptr++) {
+ part_ptr = &(partition_msg->partition_array[j]);
+ if (params.filtering
+ && _strcmp(part_ptr->name, params.partition))
continue;
- i = list_iterator_create(sinfo_list);
- /* test if node can be added to existing sinfo_data entry */
- while ((sinfo_ptr = list_next(i))) {
- if (!_match_part_data(sinfo_ptr, part_ptr))
+ hl = hostlist_create(part_ptr->nodes);
+ while (1) {
+ if (node_name)
+ free(node_name);
+ node_name = hostlist_shift(hl);
+ if (!node_name)
+ break;
+ node_ptr = _find_node(node_name, node_msg);
+ if (!node_ptr)
continue;
- if (sinfo_ptr->nodes_tot &&
- (!_match_node_data(sinfo_ptr, node_ptr)))
+ if (params.filtering && _filter_out(node_ptr))
continue;
-
- /* This node has the same configuration as this
- * sinfo_data, just add to this record */
- _update_sinfo(sinfo_ptr, part_ptr, node_ptr);
- break;
+ i = list_iterator_create(sinfo_list);
+ while ((sinfo_ptr = list_next(i))) {
+ if (!_match_part_data(sinfo_ptr, part_ptr))
+ continue;
+ if (sinfo_ptr->nodes_tot
+ && (!_match_node_data(sinfo_ptr, node_ptr)))
+ continue;
+ _update_sinfo(sinfo_ptr, node_ptr);
+ break;
+ }
+ /* if no match, create new sinfo_data entry */
+ if (sinfo_ptr == NULL) {
+ _create_sinfo(sinfo_list, part_ptr,
+ (uint16_t) j, node_ptr);
+ }
+ list_iterator_destroy(i);
}
-
- /* no match, create new sinfo_data entry */
- if (sinfo_ptr == NULL)
- _create_sinfo(sinfo_list, part_ptr, part_inx, node_ptr);
- list_iterator_destroy(i);
+ hostlist_destroy(hl);
}
_sort_hostlist(sinfo_list);
@@ -337,10 +342,6 @@ static bool _filter_out(node_info_t *node_ptr)
{
static hostlist_t host_list = NULL;
- if (params.partition &&
- _strcmp(node_ptr->partition, params.partition))
- return true;
-
if (params.nodes) {
if (host_list == NULL)
host_list = hostlist_create(params.nodes);
@@ -477,8 +478,7 @@ static bool _match_part_data(sinfo_data_t *sinfo_ptr,
return true;
}
-static void _update_sinfo(sinfo_data_t *sinfo_ptr, partition_info_t* part_ptr,
- node_info_t *node_ptr)
+static void _update_sinfo(sinfo_data_t *sinfo_ptr, node_info_t *node_ptr)
{
uint16_t base_state;
@@ -494,6 +494,10 @@ static void _update_sinfo(sinfo_data_t *sinfo_ptr, partition_info_t* part_ptr,
sinfo_ptr->max_mem = node_ptr->real_memory;
sinfo_ptr->min_weight = node_ptr->weight;
sinfo_ptr->max_weight = node_ptr->weight;
+ } else if (hostlist_find(sinfo_ptr->nodes, node_ptr->name) != -1) {
+ /* we already have this node in this record,
+ * just return, don't duplicate */
+ return;
} else {
if (sinfo_ptr->min_cpus > node_ptr->cpus)
sinfo_ptr->min_cpus = node_ptr->cpus;
@@ -585,25 +589,23 @@ static void _create_sinfo(List sinfo_list, partition_info_t* part_ptr,
}
/*
- * _find_part - find a partition by name
- * part_name IN - name of partition to locate
- * partition_msg IN - partition information message from API
- * part_inx OUT - index of the partition within the table (0-origin)
+ * _find_node - find a node by name
+ * node_name IN - name of node to locate
+ * node_msg IN - node information message from API
*/
-static partition_info_t *_find_part(char *part_name,
- partition_info_msg_t *partition_msg,
- uint16_t *part_inx)
+static node_info_t *_find_node(char *node_name, node_info_msg_t *node_msg)
{
int i;
- for (i=0; i<partition_msg->record_count; i++) {
- if (_strcmp(part_name,
- partition_msg->partition_array[i].name))
+ if (node_name == NULL)
+ return NULL;
+
+ for (i=0; i<node_msg->record_count; i++) {
+ if (_strcmp(node_name, node_msg->node_array[i].name))
continue;
- *part_inx = i;
- return &(partition_msg->partition_array[i]);
+ return &(node_msg->node_array[i]);
}
- *part_inx = 0; /* not correct, but better than random data */
+ /* not found */
return NULL;
}
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index acf3df8ab62..6c9481fdc14 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -80,6 +80,7 @@ static void _list_delete_config (void *config_entry);
static int _list_find_config (void *config_entry, void *key);
static void _make_node_down(struct node_record *node_ptr);
static void _node_did_resp(struct node_record *node_ptr);
+static bool _node_is_hidden(struct node_record *node_ptr);
static void _node_not_resp (struct node_record *node_ptr, time_t msg_time);
static void _pack_node (struct node_record *dump_node_ptr, Buf buffer);
static void _sync_bitmaps(struct node_record *node_ptr, int job_count);
@@ -197,7 +198,8 @@ create_node_record (struct config_record *config_ptr, char *node_name)
node_ptr->node_state = default_node_record.node_state;
node_ptr->last_response = default_node_record.last_response;
node_ptr->config_ptr = config_ptr;
- node_ptr->partition_ptr = NULL;
+ node_ptr->part_cnt = 0;
+ node_ptr->part_pptr = NULL;
/* these values will be overwritten when the node actually registers */
node_ptr->cpus = config_ptr->cpus;
node_ptr->real_memory = config_ptr->real_memory;
@@ -414,6 +416,8 @@ extern int load_all_node_state ( bool state_only )
node_ptr->node_state = node_state;
xfree(node_ptr->reason);
node_ptr->reason = reason;
+ node_ptr->part_cnt = 0;
+ xfree(node_ptr->part_pptr);
node_ptr->cpus = cpus;
node_ptr->real_memory = real_memory;
node_ptr->tmp_disk = tmp_disk;
@@ -530,7 +534,8 @@ int init_node_conf (void)
default_node_record.real_memory = 1;
default_node_record.tmp_disk = 1;
default_node_record.config_ptr = NULL;
- default_node_record.partition_ptr = NULL;
+ default_node_record.part_cnt = 0;
+ default_node_record.part_pptr = NULL;
default_config_record.cpus = 1;
default_config_record.real_memory = 1;
default_config_record.tmp_disk = 1;
@@ -650,6 +655,22 @@ extern int node_name2bitmap (char *node_names, bool best_effort,
return rc;
}
+static bool _node_is_hidden(struct node_record *node_ptr)
+{
+ int i;
+ bool shown = false;
+
+ for (i=0; i<node_ptr->part_cnt; i++) {
+ if (node_ptr->part_pptr[i]->hidden == 0) {
+ shown = true;
+ break;
+ }
+ }
+
+ if (shown || (node_ptr->part_cnt == 0))
+ return false;
+ return true;
+}
/*
* pack_all_node - dump all configuration and node information for all nodes
@@ -689,9 +710,8 @@ extern void pack_all_node (char **buffer_ptr, int *buffer_size,
xassert (node_ptr->config_ptr->magic ==
CONFIG_MAGIC);
- if (((show_flags & SHOW_ALL) == 0) &&
- (node_ptr->partition_ptr) &&
- (node_ptr->partition_ptr->hidden))
+ if (((show_flags & SHOW_ALL) == 0)
+ && (_node_is_hidden(node_ptr)))
continue;
_pack_node(node_ptr, buffer);
@@ -735,10 +755,6 @@ static void _pack_node (struct node_record *dump_node_ptr, Buf buffer)
}
pack32 (dump_node_ptr->config_ptr->weight, buffer);
packstr (dump_node_ptr->config_ptr->feature, buffer);
- if (dump_node_ptr->partition_ptr)
- packstr (dump_node_ptr->partition_ptr->name, buffer);
- else
- packstr (NULL, buffer);
packstr (dump_node_ptr->reason, buffer);
}
@@ -1037,7 +1053,7 @@ validate_node_specs (char *node_name, uint32_t cpus,
uint32_t real_memory, uint32_t tmp_disk,
uint32_t job_count, uint32_t status)
{
- int error_code;
+ int error_code, i;
struct config_record *config_ptr;
struct node_record *node_ptr;
char *reason_down = NULL;
@@ -1056,9 +1072,13 @@ validate_node_specs (char *node_name, uint32_t cpus,
error_code = EINVAL;
reason_down = "Low CPUs";
}
- if ((node_ptr->cpus != cpus) && (node_ptr->partition_ptr) &&
- (slurmctld_conf.fast_schedule == 0))
- node_ptr->partition_ptr->total_cpus += (cpus - node_ptr->cpus);
+ if ((node_ptr->cpus != cpus)
+ && (slurmctld_conf.fast_schedule == 0)) {
+ for (i=0; i<node_ptr->part_cnt; i++) {
+ node_ptr->part_pptr[i]->total_cpus +=
+ (cpus - node_ptr->cpus);
+ }
+ }
node_ptr->cpus = cpus;
if (real_memory < config_ptr->real_memory) {
@@ -1849,8 +1869,10 @@ void node_fini(void)
config_list = NULL;
}
- for (i=0; i< node_record_count; i++)
+ for (i=0; i< node_record_count; i++) {
+ xfree(node_record_table_ptr[i].part_pptr);
xfree(node_record_table_ptr[i].reason);
+ }
FREE_NULL_BITMAP(idle_node_bitmap);
FREE_NULL_BITMAP(avail_node_bitmap);
diff --git a/src/slurmctld/partition_mgr.c b/src/slurmctld/partition_mgr.c
index 44b69095393..c9680a5baf7 100644
--- a/src/slurmctld/partition_mgr.c
+++ b/src/slurmctld/partition_mgr.c
@@ -72,7 +72,8 @@ static uid_t *_get_group_members(char *group_name);
static time_t _get_group_tlm(void);
static void _list_delete_part(void *part_entry);
static int _uid_list_size(uid_t * uid_list_ptr);
-static void _unlink_free_nodes(bitstr_t *old_bitmap);
+static void _unlink_free_nodes(bitstr_t *old_bitmap,
+ struct part_record *part_ptr);
/*
* _build_part_bitmap - update the total_cpus, total_nodes, and node_bitmap
@@ -109,7 +110,7 @@ static int _build_part_bitmap(struct part_record *part_ptr)
}
if (part_ptr->nodes == NULL) { /* no nodes in partition */
- _unlink_free_nodes(old_bitmap);
+ _unlink_free_nodes(old_bitmap, part_ptr);
FREE_NULL_BITMAP(old_bitmap);
return 0;
}
@@ -136,7 +137,10 @@ static int _build_part_bitmap(struct part_record *part_ptr)
part_ptr->total_cpus += node_ptr->config_ptr->cpus;
else
part_ptr->total_cpus += node_ptr->cpus;
- node_ptr->partition_ptr = part_ptr;
+ node_ptr->part_cnt++;
+ xrealloc(node_ptr->part_pptr, (node_ptr->part_cnt *
+ sizeof(struct part_record *)));
+ node_ptr->part_pptr[node_ptr->part_cnt-1] = part_ptr;
if (old_bitmap)
bit_clear(old_bitmap,
(int) (node_ptr -
@@ -147,27 +151,41 @@ static int _build_part_bitmap(struct part_record *part_ptr)
}
hostlist_destroy(host_list);
- _unlink_free_nodes(old_bitmap);
+ _unlink_free_nodes(old_bitmap, part_ptr);
last_node_update = time(NULL);
FREE_NULL_BITMAP(old_bitmap);
return 0;
}
-/* unlink nodes removed from the partition */
-static void _unlink_free_nodes(bitstr_t *old_bitmap)
+/* unlink nodes removed from a partition */
+static void _unlink_free_nodes(bitstr_t *old_bitmap,
+ struct part_record *part_ptr)
{
- int i, update_nodes = 0;
+ int i, j, k, update_nodes = 0;
+ struct node_record *node_ptr;
- if (old_bitmap) {
- for (i = 0; i < node_record_count; i++) {
- if (bit_test(old_bitmap, i) == 0)
+ if (old_bitmap == NULL)
+ return;
+
+ node_ptr = &node_record_table_ptr[0];
+ for (i = 0; i < node_record_count; i++, node_ptr++) {
+ if (bit_test(old_bitmap, i) == 0)
+ continue;
+ for (j=0; j<node_ptr->part_cnt; j++) {
+ if (node_ptr->part_pptr[j] != part_ptr)
continue;
- node_record_table_ptr[i].partition_ptr = NULL;
- update_nodes = 1;
+ node_ptr->part_cnt--;
+ for (k=j; k<node_ptr->part_cnt; k++) {
+ node_ptr->part_pptr[k] =
+ node_ptr->part_pptr[k+1];
+ }
+ break;
}
- if (update_nodes)
- last_node_update = time(NULL);
+ update_nodes = 1;
}
+
+ if (update_nodes)
+ last_node_update = time(NULL);
}
@@ -537,14 +555,22 @@ int init_part_conf(void)
static void _list_delete_part(void *part_entry)
{
struct part_record *part_ptr;
- int i;
+ struct node_record *node_ptr;
+ int i, j, k;
part_ptr = (struct part_record *) part_entry;
- for (i = 0; i < node_record_count; i++) {
- if (node_record_table_ptr[i].partition_ptr !=
- part_ptr)
- continue;
- node_record_table_ptr[i].partition_ptr = NULL;
+ node_ptr = &node_record_table_ptr[0];
+ for (i = 0; i < node_record_count; i++, node_ptr++) {
+ for (j=0; j<node_ptr->part_cnt; j++) {
+ if (node_ptr->part_pptr[j] != part_ptr)
+ continue;
+ node_ptr->part_cnt--;
+ for (k=j; k<node_ptr->part_cnt; k++) {
+ node_ptr->part_pptr[k] =
+ node_ptr->part_pptr[k+1];
+ }
+ break;
+ }
}
xfree(part_ptr->allow_groups);
xfree(part_ptr->allow_uids);
@@ -1038,6 +1064,10 @@ extern int delete_partition(delete_part_msg_t *part_desc_ptr)
if (part_ptr == NULL) /* No such partition */
return ESLURM_INVALID_PARTITION_NAME;
+ if (default_part_loc == part_ptr) {
+ error("Deleting default partition %s", part_ptr->name);
+ default_part_loc = NULL;
+ }
(void) kill_job_by_part_name(part_desc_ptr->name);
list_delete_all(part_list, list_find_part, part_desc_ptr->name);
last_part_update = time(NULL);
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index 0027756d4ff..4e3fa9e2fa1 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -106,7 +106,6 @@ static int _build_bitmaps(void)
struct node_record *node_ptr;
struct job_record *job_ptr;
ListIterator job_iterator;
- bitstr_t *all_part_node_bitmap;
hostlist_t host_list;
last_node_update = time(NULL);
@@ -187,9 +186,6 @@ static int _build_bitmaps(void)
}
/* scan partition table and identify nodes in each */
- all_part_node_bitmap = (bitstr_t *) bit_alloc(node_record_count);
- if (all_part_node_bitmap == NULL)
- fatal ("bit_alloc malloc failure");
part_iterator = list_iterator_create(part_list);
if (part_iterator == NULL)
fatal ("memory allocation failure");
@@ -222,29 +218,22 @@ static int _build_bitmaps(void)
continue;
}
j = node_ptr - node_record_table_ptr;
- if (bit_test(all_part_node_bitmap, j) == 1) {
- error("_build_bitmaps: node %s defined in "
- "more than one partition",
- this_node_name);
- error("_build_bitmaps: only the first "
- "specification is honored");
- } else {
- bit_set(part_ptr->node_bitmap, j);
- bit_set(all_part_node_bitmap, j);
- part_ptr->total_nodes++;
- if (slurmctld_conf.fast_schedule)
- part_ptr->total_cpus +=
- node_ptr->config_ptr->cpus;
- else
- part_ptr->total_cpus += node_ptr->cpus;
- node_ptr->partition_ptr = part_ptr;
- }
+ bit_set(part_ptr->node_bitmap, j);
+ part_ptr->total_nodes++;
+ if (slurmctld_conf.fast_schedule)
+ part_ptr->total_cpus +=
+ node_ptr->config_ptr->cpus;
+ else
+ part_ptr->total_cpus += node_ptr->cpus;
+ node_ptr->part_cnt++;
+ xrealloc(node_ptr->part_pptr, (node_ptr->part_cnt *
+ sizeof(struct part_record *)));
+ node_ptr->part_pptr[node_ptr->part_cnt-1] = part_ptr;
free(this_node_name);
}
hostlist_destroy(host_list);
}
list_iterator_destroy(part_iterator);
- bit_free(all_part_node_bitmap);
return error_code;
}
@@ -934,8 +923,10 @@ static void _purge_old_node_state(struct node_record *old_node_table_ptr,
{
int i;
- for (i = 0; i < old_node_record_count; i++)
+ for (i = 0; i < old_node_record_count; i++) {
+ xfree(old_node_table_ptr[i].part_pptr);
xfree(old_node_table_ptr[i].reason);
+ }
xfree(old_node_table_ptr);
}
diff --git a/src/slurmctld/sched_upcalls.c b/src/slurmctld/sched_upcalls.c
index 42ef602a875..93634514adb 100644
--- a/src/slurmctld/sched_upcalls.c
+++ b/src/slurmctld/sched_upcalls.c
@@ -1010,6 +1010,11 @@ sched_get_node_tmp_disk( sched_obj_list_t node_data,
/* ************************************************************************ */
/* TAG( sched_get_node_partition ) */
+/* NOTE: A SLURM node can be in multiple partitions/queues at the same time */
+/* We return only the first of these partition names here or NULL if there */
+/* are no associated partitions. There are 'part_cnt' partitions associated */
+/* with each node. There is an array of pointers to these partitions in the */
+/* array 'part_pptr'. We probably want to change this function accordingly. */
/* ************************************************************************ */
void *
sched_get_node_partition( sched_obj_list_t node_data,
@@ -1017,7 +1022,11 @@ sched_get_node_partition( sched_obj_list_t node_data,
char *type )
{
if ( type ) *type = 's';
- return ( (struct node_record *) node_data->data )[ idx ].partition_ptr->name;
+ if ( ((struct node_record *) node_data->data )[ idx ].part_cnt == 0 )
+ return NULL;
+
+ return ( (struct node_record *) node_data->data )[ idx ].
+ part_pptr[0]->name;
}
/* ************************************************************************ */
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 7e17d20e5b5..c4e7245067d 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -167,7 +167,9 @@ struct node_record {
uint32_t real_memory; /* MB real memory on the node */
uint32_t tmp_disk; /* MB total disk in TMP_FS */
struct config_record *config_ptr; /* configuration spec ptr */
- struct part_record *partition_ptr; /* partition for this node */
+ uint16_t part_cnt; /* number of associated partitions */
+ 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 */
slurm_addr slurm_addr; /* network address */
uint16_t comp_job_cnt; /* count of jobs completing on node */
diff --git a/testsuite/expect/test3.5 b/testsuite/expect/test3.5
index 8776a52e0cd..b2fb109de51 100755
--- a/testsuite/expect/test3.5
+++ b/testsuite/expect/test3.5
@@ -70,10 +70,44 @@ if {$found == 1} {
exit 1
}
+#
+# Identify usable nodes in default partition
+#
+set def_name ""
+set def_node ""
+spawn $sinfo -h -o %32P
+expect {
+ -re "($alpha_numeric_under)(\\*)" {
+ set def_name $expect_out(1,string)
+ exp_continue
+ }
+ eof {
+ wait
+ }
+}
+if {[string compare $def_name ""] == 0} {
+ send_user "\nFAILURE: failed to find default partition\n"
+ exit 1
+}
+spawn $sinfo -h -o %N -p $def_name
+expect {
+ -re "(\[a-zA-Z0-9_-\]+)" {
+ set def_node $expect_out(1,string)
+ exp_continue
+ }
+ eof {
+ wait
+ }
+}
+if {[string compare $def_node ""] == 0} {
+ send_user "\nFAILURE:default partition seems to have no nodes\n"
+ exit 1
+}
+
#
# Create a new partition
#
-spawn $scontrol update PartitionName=$part_name
+spawn $scontrol update PartitionName=$part_name Nodes=$def_node
expect {
-re "slurm_update error: ($alpha_numeric) ($alpha_numeric)" {
set access_err 0
@@ -139,7 +173,7 @@ if {$allow != 1} {
}
#
-# Now set group to mine
+# Now set AllowGroups to mine and TimeLimit=1
#
spawn $bin_id -gn
expect {
@@ -155,7 +189,7 @@ expect {
wait
}
}
-spawn $scontrol update PartitionName=$part_name AllowGroups=$my_group
+spawn $scontrol update PartitionName=$part_name AllowGroups=$my_group MaxTime=1
expect {
timeout {
send_user "\nFAILURE: scontrol not responding\n"
@@ -185,6 +219,38 @@ if {$found != 1} {
set exit_code 1
}
+#
+# Run a job in this new partition and validate the time limit
+#
+set timed_out 0
+set sleep_time 300
+set timeout [expr $max_job_delay + $sleep_time]
+spawn $srun -t1 -p $part_name $bin_sleep $sleep_time
+expect {
+ -re "job exceeded timelimit" {
+ set timed_out 1
+ exp_continue
+ }
+ -re "Terminated" {
+ set timed_out 1
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: srun not responding\n"
+ kill_srun
+ exp_continue
+ }
+ eof {
+ wait
+ }
+}
+if {$timed_out == 1} {
+ send_user "Early termination is expected, no worries.\n"
+} else {
+ send_user "\nFAILURE: partition time limit not enforced\n"
+ set exit_code 1
+}
+
#
# Now reset AllowGroups to ALL
#
--
GitLab