From ce370ca371a81c7bcdcebab0dbeae10474c30a91 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 14 May 2010 21:07:28 +0000
Subject: [PATCH] Added limited gres support to select/linear. It prevents
overallocation of gres resources including a map of specific allocated
resources.
Still need to add save/restore of that information,
propagate detains to slurmd, consider topology, etc.
---
src/common/gres.c | 65 +++++++----
src/common/gres.h | 20 ++--
src/plugins/gres/gpu/gres_gpu.c | 125 +++++++++++++++++++---
src/plugins/gres/nic/gres_nic.c | 125 +++++++++++++++++++---
src/plugins/select/linear/select_linear.c | 31 ++++--
5 files changed, 300 insertions(+), 66 deletions(-)
diff --git a/src/common/gres.c b/src/common/gres.c
index a6c09f8225a..c0f4f394bc8 100644
--- a/src/common/gres.c
+++ b/src/common/gres.c
@@ -112,11 +112,14 @@ typedef struct slurm_gres_ops {
uint32_t (*job_test) ( void *job_gres_data,
void *node_gres_data,
bool use_total_gres );
- void (*job_alloc) ( void *job_gres_data,
+ int (*job_alloc) ( void *job_gres_data,
void *node_gres_data,
+ int node_cnt,
+ int node_offset,
uint32_t cpu_cnt );
- void (*job_dealloc) ( void *job_gres_data,
+ int (*job_dealloc) ( void *job_gres_data,
void *node_gres_data,
+ int node_offset,
uint32_t cpu_cnt );
void (*job_state_log) ( void *gres_data,
uint32_t job_id );
@@ -1259,20 +1262,27 @@ extern uint32_t gres_plugin_job_test(List job_gres_list, List node_gres_list,
/*
* Allocate resource to a job and update node and job gres information
* IN job_gres_list - job's gres_list built by gres_plugin_job_gres_validate()
- * IN node_gres_list - node's gres_list built by gres_plugin_node_config_validate()
+ * IN node_gres_list - node's gres_list built by
+ * gres_plugin_node_config_validate()
+ * IN node_cnt - total number of nodes originally allocated to the job
+ * IN node_offset - zero-origin index to the node of interest
* IN cpu_cnt - number of CPUs allocated to this job on this node
+ * RET SLURM_SUCCESS or error code
*/
-extern void gres_plugin_job_alloc(List job_gres_list, List node_gres_list,
- uint32_t cpu_cnt)
+extern int gres_plugin_job_alloc(List job_gres_list, List node_gres_list,
+ int node_cnt, int node_offset,
+ uint32_t cpu_cnt)
{
- int i;
+ int i, rc, rc2;
ListIterator job_gres_iter, node_gres_iter;
gres_state_t *job_gres_ptr, *node_gres_ptr;
- if ((job_gres_list == NULL) || (node_gres_list == NULL))
- return;
+ if (job_gres_list == NULL)
+ return SLURM_SUCCESS;
+ if (node_gres_list == NULL)
+ return SLURM_ERROR;
- (void) gres_plugin_init();
+ rc = gres_plugin_init();
slurm_mutex_lock(&gres_context_lock);
job_gres_iter = list_iterator_create(job_gres_list);
@@ -1291,33 +1301,43 @@ extern void gres_plugin_job_alloc(List job_gres_list, List node_gres_list,
if (job_gres_ptr->plugin_id !=
*(gres_context[i].ops.plugin_id))
continue;
- (*(gres_context[i].ops.job_alloc))
+ rc2 = (*(gres_context[i].ops.job_alloc))
(job_gres_ptr->gres_data,
- node_gres_ptr->gres_data, cpu_cnt);
+ node_gres_ptr->gres_data, node_cnt,
+ node_offset, cpu_cnt);
+ if (rc2 != SLURM_SUCCESS)
+ rc = rc2;
break;
}
}
list_iterator_destroy(job_gres_iter);
slurm_mutex_unlock(&gres_context_lock);
+
+ return rc;
}
/*
* Deallocate resource from a job and update node and job gres information
* IN job_gres_list - job's gres_list built by gres_plugin_job_gres_validate()
- * IN node_gres_list - node's gres_list built by gres_plugin_node_config_validate()
+ * IN node_gres_list - node's gres_list built by
+ * gres_plugin_node_config_validate()
+ * IN node_offset - zero-origin index to the node of interest
* IN cpu_cnt - number of CPUs allocated to this job on this node
+ * RET SLURM_SUCCESS or error code
*/
-extern void gres_plugin_job_dealloc(List job_gres_list, List node_gres_list,
- uint32_t cpu_cnt)
+extern int gres_plugin_job_dealloc(List job_gres_list, List node_gres_list,
+ int node_offset, uint32_t cpu_cnt)
{
- int i;
+ int i, rc, rc2;
ListIterator job_gres_iter, node_gres_iter;
gres_state_t *job_gres_ptr, *node_gres_ptr;
- if ((job_gres_list == NULL) || (node_gres_list == NULL))
- return;
+ if (job_gres_list == NULL)
+ return SLURM_SUCCESS;
+ if (node_gres_list == NULL)
+ return SLURM_ERROR;
- (void) gres_plugin_init();
+ rc = gres_plugin_init();
slurm_mutex_lock(&gres_context_lock);
job_gres_iter = list_iterator_create(job_gres_list);
@@ -1336,14 +1356,19 @@ extern void gres_plugin_job_dealloc(List job_gres_list, List node_gres_list,
if (job_gres_ptr->plugin_id !=
*(gres_context[i].ops.plugin_id))
continue;
- (*(gres_context[i].ops.job_dealloc))
+ rc2 = (*(gres_context[i].ops.job_dealloc))
(job_gres_ptr->gres_data,
- node_gres_ptr->gres_data, cpu_cnt);
+ node_gres_ptr->gres_data, node_offset,
+ cpu_cnt);
+ if (rc2 != SLURM_SUCCESS)
+ rc = rc2;
break;
}
}
list_iterator_destroy(job_gres_iter);
slurm_mutex_unlock(&gres_context_lock);
+
+ return rc;
}
/*
diff --git a/src/common/gres.h b/src/common/gres.h
index baaf96acaed..4581ee4e922 100644
--- a/src/common/gres.h
+++ b/src/common/gres.h
@@ -208,20 +208,28 @@ extern uint32_t gres_plugin_job_test(List job_gres_list, List node_gres_list,
/*
* Allocate resource to a job and update node and job gres information
* IN job_gres_list - job's gres_list built by gres_plugin_job_gres_validate()
- * IN node_gres_list - node's gres_list built by gres_plugin_node_config_validate()
+ * IN node_gres_list - node's gres_list built by
+ * gres_plugin_node_config_validate()
+ * IN node_cnt - total number of nodes originally allocated to the job
+ * IN node_offset - zero-origin index to the node of interest
* IN cpu_cnt - number of CPUs allocated to this job on this node
+ * RET SLURM_SUCCESS or error code
*/
-extern void gres_plugin_job_alloc(List job_gres_list, List node_gres_list,
- uint32_t cpu_cnt);
+extern int gres_plugin_job_alloc(List job_gres_list, List node_gres_list,
+ int node_cnt, int node_offset,
+ uint32_t cpu_cnt);
/*
* Deallocate resource from a job and update node and job gres information
* IN job_gres_list - job's gres_list built by gres_plugin_job_gres_validate()
- * IN node_gres_list - node's gres_list built by gres_plugin_node_config_validate()
+ * IN node_gres_list - node's gres_list built by
+ * gres_plugin_node_config_validate()
+ * IN node_offset - zero-origin index to the node of interest
* IN cpu_cnt - number of CPUs allocated to this job on this node
+ * RET SLURM_SUCCESS or error code
*/
-extern void gres_plugin_job_dealloc(List job_gres_list, List node_gres_list,
- uint32_t cpu_cnt);
+extern int gres_plugin_job_dealloc(List job_gres_list, List node_gres_list,
+ int node_offset, uint32_t cpu_cnt);
/*
* Log a job's current gres state
diff --git a/src/plugins/gres/gpu/gres_gpu.c b/src/plugins/gres/gpu/gres_gpu.c
index 9b4c1317d34..55182e93213 100644
--- a/src/plugins/gres/gpu/gres_gpu.c
+++ b/src/plugins/gres/gpu/gres_gpu.c
@@ -136,6 +136,10 @@ typedef struct gpu_job_state {
/* If 0 then gpu_cnt_alloc is per node,
* if 1 then gpu_cnt_alloc is per CPU */
uint8_t gpu_cnt_mult;
+
+ /* Resources currently allocated to job on each node */
+ uint32_t node_cnt;
+ bitstr_t **gpu_bit_alloc;
} gpu_job_state_t;
/*
@@ -628,38 +632,129 @@ extern uint32_t job_test(void *job_gres_data, void *node_gres_data,
}
}
-extern void job_alloc(void *job_gres_data, void *node_gres_data, int cpu_cnt)
+extern int job_alloc(void *job_gres_data, void *node_gres_data,
+ int node_cnt, int node_offset, uint32_t cpu_cnt)
{
+ int i;
uint32_t gres_cnt;
gpu_job_state_t *job_gres_ptr = (gpu_job_state_t *) job_gres_data;
gpu_node_state_t *node_gres_ptr = (gpu_node_state_t *) node_gres_data;
+ /*
+ * Validate data structures. Either job_gres_data->node_cnt and
+ * job_gres_data->gpu_bit_alloc are both set or both zero/NULL.
+ */
+ xassert(node_cnt);
+ xassert(node_offset >= 0);
+ xassert(job_gres_ptr);
+ xassert(node_gres_ptr);
+ xassert(node_gres_ptr->gpu_bit_alloc);
+ if (job_gres_ptr->node_cnt == 0) {
+ job_gres_ptr->node_cnt = node_cnt;
+ if (job_gres_ptr->gpu_bit_alloc) {
+ error("%s: node_cnt==0 and bit_alloc is set",
+ plugin_name);
+ xfree(job_gres_ptr->gpu_bit_alloc);
+ }
+ job_gres_ptr->gpu_bit_alloc =
+ xmalloc(sizeof(bitstr_t *) * node_cnt);
+ } else if (job_gres_ptr->node_cnt < node_cnt) {
+ error("%s: node_cnt increase from %u to %d",
+ plugin_name, job_gres_ptr->node_cnt, node_cnt);
+ if (node_offset >= job_gres_ptr->node_cnt)
+ return SLURM_ERROR;
+ } else if (job_gres_ptr->node_cnt > node_cnt) {
+ error("%s: node_cnt decrease from %u to %d",
+ plugin_name, job_gres_ptr->node_cnt, node_cnt);
+ }
+
+ /*
+ * Check that sufficient resources exist on this node
+ */
if (job_gres_ptr->gpu_cnt_mult == 0)
gres_cnt = job_gres_ptr->gpu_cnt_alloc;
else
gres_cnt = (job_gres_ptr->gpu_cnt_alloc * cpu_cnt);
+ i = node_gres_ptr->gpu_cnt_alloc + gres_cnt;
+ i -= node_gres_ptr->gpu_cnt_avail;
+ if (i > 0) {
+ error("%s: overallocated resources by %d", plugin_name, i);
+ /* proceed with request, give job what's available */
+ }
- node_gres_ptr->gpu_cnt_alloc += gres_cnt;
- if (node_gres_ptr->gpu_cnt_alloc > node_gres_ptr->gpu_cnt_avail)
- error("%s: overallocated resources", plugin_name);
+ /*
+ * Select the specific resources to use for this job.
+ * We'll need to add topology information in the future
+ */
+ if (job_gres_ptr->gpu_bit_alloc[node_offset]) {
+ error("%s: job's bit_alloc is set for node %d",
+ plugin_name, node_offset);
+ bit_free(job_gres_ptr->gpu_bit_alloc[node_offset]);
+ }
+ job_gres_ptr->gpu_bit_alloc[node_offset] = bit_alloc(node_gres_ptr->
+ gpu_cnt_avail);
+ if (job_gres_ptr->gpu_bit_alloc[node_offset] == NULL)
+ fatal("bit_copy: malloc failure");
+ for (i=0; i<node_gres_ptr->gpu_cnt_avail && gres_cnt>0; i++) {
+ if (bit_test(node_gres_ptr->gpu_bit_alloc, i))
+ continue;
+ bit_set(node_gres_ptr->gpu_bit_alloc, i);
+ bit_set(job_gres_ptr->gpu_bit_alloc[node_offset], i);
+ node_gres_ptr->gpu_cnt_alloc++;
+ gres_cnt--;
+ }
+
+ return SLURM_SUCCESS;
}
-extern void job_dealloc(void *job_gres_data, void *node_gres_data, int cpu_cnt)
+extern int job_dealloc(void *job_gres_data, void *node_gres_data,
+ int node_offset, uint32_t cpu_cnt)
{
- uint32_t gres_cnt;
+ int i, len;
gpu_job_state_t *job_gres_ptr = (gpu_job_state_t *) job_gres_data;
gpu_node_state_t *node_gres_ptr = (gpu_node_state_t *) node_gres_data;
- if (job_gres_ptr->gpu_cnt_mult == 0)
- gres_cnt = job_gres_ptr->gpu_cnt_alloc;
- else
- gres_cnt = (job_gres_ptr->gpu_cnt_alloc * cpu_cnt);
+ /*
+ * Validate data structures. Either job_gres_data->node_cnt and
+ * job_gres_data->gpu_bit_alloc are both set or both zero/NULL.
+ */
+ xassert(node_offset >= 0);
+ xassert(job_gres_ptr);
+ xassert(node_gres_ptr);
+ xassert(node_gres_ptr->gpu_bit_alloc);
+ if (job_gres_ptr->node_cnt <= node_offset) {
+ error("%s: bad node_offset %d count is %u",
+ plugin_name, node_offset, job_gres_ptr->node_cnt);
+ return SLURM_ERROR;
+ }
+ if (job_gres_ptr->gpu_bit_alloc == NULL) {
+ error("%s: job's bitmap is NULL", plugin_name);
+ return SLURM_ERROR;
+ }
+ if (job_gres_ptr->gpu_bit_alloc[node_offset] == NULL) {
+ error("%s: job's bitmap is empty", plugin_name);
+ return SLURM_ERROR;
+ }
- if (gres_cnt > node_gres_ptr->gpu_cnt_alloc) {
- error("%s: resource count underflow", plugin_name);
- node_gres_ptr->gpu_cnt_alloc = 0;
- } else
- node_gres_ptr->gpu_cnt_alloc -= gres_cnt;
+ len = bit_size(job_gres_ptr->gpu_bit_alloc[node_offset]);
+ i = bit_size(node_gres_ptr->gpu_bit_alloc);
+ if (i != len) {
+ error("%s: job and node bitmap sizes differ (%d != %d)",
+ plugin_name, len, i);
+ len = MIN(len, i);
+ /* proceed with request, make best effort */
+ }
+ for (i=0; i<len; i++) {
+ if (!bit_test(job_gres_ptr->gpu_bit_alloc[node_offset], i))
+ continue;
+ bit_clear(node_gres_ptr->gpu_bit_alloc, i);
+ /* NOTE: Do not clear bit from
+ * job_gres_ptr->gpu_bit_alloc[node_offset]
+ * since this may only be an emulated deallocate */
+ node_gres_ptr->gpu_cnt_alloc--;
+ }
+
+ return SLURM_SUCCESS;
}
extern void job_state_log(void *gres_data, uint32_t job_id)
diff --git a/src/plugins/gres/nic/gres_nic.c b/src/plugins/gres/nic/gres_nic.c
index 40cbf97c7ea..148d19cee63 100644
--- a/src/plugins/gres/nic/gres_nic.c
+++ b/src/plugins/gres/nic/gres_nic.c
@@ -136,6 +136,10 @@ typedef struct nic_job_state {
/* If 0 then nic_cnt_alloc is per node,
* if 1 then nic_cnt_alloc is per CPU */
uint8_t nic_cnt_mult;
+
+ /* Resources currently allocated to job on each node */
+ uint32_t node_cnt;
+ bitstr_t **nic_bit_alloc;
} nic_job_state_t;
/*
@@ -628,38 +632,129 @@ extern uint32_t job_test(void *job_gres_data, void *node_gres_data,
}
}
-extern void job_alloc(void *job_gres_data, void *node_gres_data, int cpu_cnt)
+extern int job_alloc(void *job_gres_data, void *node_gres_data,
+ int node_cnt, int node_offset, uint32_t cpu_cnt)
{
+ int i;
uint32_t gres_cnt;
nic_job_state_t *job_gres_ptr = (nic_job_state_t *) job_gres_data;
nic_node_state_t *node_gres_ptr = (nic_node_state_t *) node_gres_data;
+ /*
+ * Validate data structures. Either job_gres_data->node_cnt and
+ * job_gres_data->nic_bit_alloc are both set or both zero/NULL.
+ */
+ xassert(node_cnt);
+ xassert(node_offset >= 0);
+ xassert(job_gres_ptr);
+ xassert(node_gres_ptr);
+ xassert(node_gres_ptr->nic_bit_alloc);
+ if (job_gres_ptr->node_cnt == 0) {
+ job_gres_ptr->node_cnt = node_cnt;
+ if (job_gres_ptr->nic_bit_alloc) {
+ error("%s: node_cnt==0 and bit_alloc is set",
+ plugin_name);
+ xfree(job_gres_ptr->nic_bit_alloc);
+ }
+ job_gres_ptr->nic_bit_alloc =
+ xmalloc(sizeof(bitstr_t *) * node_cnt);
+ } else if (job_gres_ptr->node_cnt < node_cnt) {
+ error("%s: node_cnt increase from %u to %d",
+ plugin_name, job_gres_ptr->node_cnt, node_cnt);
+ if (node_offset >= job_gres_ptr->node_cnt)
+ return SLURM_ERROR;
+ } else if (job_gres_ptr->node_cnt > node_cnt) {
+ error("%s: node_cnt decrease from %u to %d",
+ plugin_name, job_gres_ptr->node_cnt, node_cnt);
+ }
+
+ /*
+ * Check that sufficient resources exist on this node
+ */
if (job_gres_ptr->nic_cnt_mult == 0)
gres_cnt = job_gres_ptr->nic_cnt_alloc;
else
gres_cnt = (job_gres_ptr->nic_cnt_alloc * cpu_cnt);
+ i = node_gres_ptr->nic_cnt_alloc + gres_cnt;
+ i -= node_gres_ptr->nic_cnt_avail;
+ if (i > 0) {
+ error("%s: overallocated resources by %d", plugin_name, i);
+ /* proceed with request, give job what's available */
+ }
- node_gres_ptr->nic_cnt_alloc += gres_cnt;
- if (node_gres_ptr->nic_cnt_alloc > node_gres_ptr->nic_cnt_avail)
- error("%s: overallocated resources", plugin_name);
+ /*
+ * Select the specific resources to use for this job.
+ * We'll need to add topology information in the future
+ */
+ if (job_gres_ptr->nic_bit_alloc[node_offset]) {
+ error("%s: job's bit_alloc is set for node %d",
+ plugin_name, node_offset);
+ bit_free(job_gres_ptr->nic_bit_alloc[node_offset]);
+ }
+ job_gres_ptr->nic_bit_alloc[node_offset] = bit_alloc(node_gres_ptr->
+ nic_cnt_avail);
+ if (job_gres_ptr->nic_bit_alloc[node_offset] == NULL)
+ fatal("bit_copy: malloc failure");
+ for (i=0; i<node_gres_ptr->nic_cnt_avail && gres_cnt>0; i++) {
+ if (bit_test(node_gres_ptr->nic_bit_alloc, i))
+ continue;
+ bit_set(node_gres_ptr->nic_bit_alloc, i);
+ bit_set(job_gres_ptr->nic_bit_alloc[node_offset], i);
+ node_gres_ptr->nic_cnt_alloc++;
+ gres_cnt--;
+ }
+
+ return SLURM_SUCCESS;
}
-extern void job_dealloc(void *job_gres_data, void *node_gres_data, int cpu_cnt)
+extern int job_dealloc(void *job_gres_data, void *node_gres_data,
+ int node_offset, uint32_t cpu_cnt)
{
- uint32_t gres_cnt;
+ int i, len;
nic_job_state_t *job_gres_ptr = (nic_job_state_t *) job_gres_data;
nic_node_state_t *node_gres_ptr = (nic_node_state_t *) node_gres_data;
- if (job_gres_ptr->nic_cnt_mult == 0)
- gres_cnt = job_gres_ptr->nic_cnt_alloc;
- else
- gres_cnt = (job_gres_ptr->nic_cnt_alloc * cpu_cnt);
+ /*
+ * Validate data structures. Either job_gres_data->node_cnt and
+ * job_gres_data->nic_bit_alloc are both set or both zero/NULL.
+ */
+ xassert(node_offset >= 0);
+ xassert(job_gres_ptr);
+ xassert(node_gres_ptr);
+ xassert(node_gres_ptr->nic_bit_alloc);
+ if (job_gres_ptr->node_cnt <= node_offset) {
+ error("%s: bad node_offset %d count is %u",
+ plugin_name, node_offset, job_gres_ptr->node_cnt);
+ return SLURM_ERROR;
+ }
+ if (job_gres_ptr->nic_bit_alloc == NULL) {
+ error("%s: job's bitmap is NULL", plugin_name);
+ return SLURM_ERROR;
+ }
+ if (job_gres_ptr->nic_bit_alloc[node_offset] == NULL) {
+ error("%s: job's bitmap is empty", plugin_name);
+ return SLURM_ERROR;
+ }
- if (gres_cnt > node_gres_ptr->nic_cnt_alloc) {
- error("%s: resource count underflow", plugin_name);
- node_gres_ptr->nic_cnt_alloc = 0;
- } else
- node_gres_ptr->nic_cnt_alloc -= gres_cnt;
+ len = bit_size(job_gres_ptr->nic_bit_alloc[node_offset]);
+ i = bit_size(node_gres_ptr->nic_bit_alloc);
+ if (i != len) {
+ error("%s: job and node bitmap sizes differ (%d != %d)",
+ plugin_name, len, i);
+ len = MIN(len, i);
+ /* proceed with request, make best effort */
+ }
+ for (i=0; i<len; i++) {
+ if (!bit_test(job_gres_ptr->nic_bit_alloc[node_offset], i))
+ continue;
+ bit_clear(node_gres_ptr->nic_bit_alloc, i);
+ /* NOTE: Do not clear bit from
+ * job_gres_ptr->nic_bit_alloc[node_offset]
+ * since this may only be an emulated deallocate */
+ node_gres_ptr->nic_cnt_alloc--;
+ }
+
+ return SLURM_SUCCESS;
}
extern void job_state_log(void *gres_data, uint32_t job_id)
diff --git a/src/plugins/select/linear/select_linear.c b/src/plugins/select/linear/select_linear.c
index 30024f78be8..a60afafafc0 100644
--- a/src/plugins/select/linear/select_linear.c
+++ b/src/plugins/select/linear/select_linear.c
@@ -1348,7 +1348,7 @@ static int _rm_job_from_nodes(struct cr_record *cr_ptr,
struct job_record *job_ptr, char *pre_err,
bool remove_all)
{
- int i, i_first, i_last, rc = SLURM_SUCCESS;
+ int i, i_first, i_last, node_offset, rc = SLURM_SUCCESS;
struct part_cr_record *part_cr_ptr;
job_resources_t *job_resrcs_ptr;
uint32_t job_memory, job_memory_cpu = 0, job_memory_node = 0;
@@ -1388,9 +1388,12 @@ static int _rm_job_from_nodes(struct cr_record *cr_ptr,
i_last = bit_fls(job_resrcs_ptr->node_bitmap);
if (i_first == -1) /* job has no nodes */
i_last = -2;
+ node_offset = -1;
for (i = i_first; i <= i_last; i++) {
- if (!bit_test(job_resrcs_ptr->node_bitmap, i) ||
- !bit_test(job_ptr->node_bitmap, i))
+ if (!bit_test(job_resrcs_ptr->node_bitmap, i))
+ continue;
+ node_offset++;
+ if (!bit_test(job_ptr->node_bitmap, i))
continue;
node_ptr = node_record_table_ptr + i;
@@ -1414,7 +1417,8 @@ static int _rm_job_from_nodes(struct cr_record *cr_ptr,
gres_list = cr_ptr->nodes[i].gres_list;
else
gres_list = node_ptr->gres_list;
- gres_plugin_job_dealloc(job_ptr->gres_list, gres_list, cpu_cnt);
+ gres_plugin_job_dealloc(job_ptr->gres_list, gres_list,
+ node_offset, cpu_cnt);
gres_plugin_node_state_log(gres_list, node_ptr->name);
if (exclusive) {
@@ -1526,7 +1530,8 @@ static int _rm_job_from_one_node(struct job_record *job_ptr,
}
first_bit = bit_ffs(job_resrcs_ptr->node_bitmap);
last_bit = node_inx;
- for (i = first_bit, node_offset = -1; i <= node_inx; i++) {
+ node_offset = -1;
+ for (i = first_bit; i <= node_inx; i++) {
if (!bit_test(job_resrcs_ptr->node_bitmap, i))
continue;
node_offset++;
@@ -1560,7 +1565,8 @@ static int _rm_job_from_one_node(struct job_record *job_ptr,
gres_list = cr_ptr->nodes[i].gres_list;
else
gres_list = node_ptr->gres_list;
- gres_plugin_job_dealloc(job_ptr->gres_list, gres_list, cpu_cnt);
+ gres_plugin_job_dealloc(job_ptr->gres_list, gres_list, node_offset,
+ cpu_cnt);
gres_plugin_node_state_log(gres_list, node_ptr->name);
exclusive = (job_ptr->details->shared == 0);
@@ -1624,7 +1630,7 @@ static int _add_job_to_nodes(struct cr_record *cr_ptr,
struct job_record *job_ptr, char *pre_err,
int alloc_all)
{
- int i, i_first, i_last, rc = SLURM_SUCCESS;
+ int i, i_first, i_last, node_cnt, node_offset, rc = SLURM_SUCCESS;
bool exclusive;
struct part_cr_record *part_cr_ptr;
job_resources_t *job_resrcs_ptr;
@@ -1658,11 +1664,15 @@ static int _add_job_to_nodes(struct cr_record *cr_ptr,
i_first = bit_ffs(job_resrcs_ptr->node_bitmap);
i_last = bit_fls(job_resrcs_ptr->node_bitmap);
+ node_cnt = bit_set_count(job_resrcs_ptr->node_bitmap);
if (i_first == -1) /* job has no nodes */
i_last = -2;
+ node_offset = -1;
for (i = i_first; i <= i_last; i++) {
- if (!bit_test(job_resrcs_ptr->node_bitmap, i) ||
- !bit_test(job_ptr->node_bitmap, i))
+ if (!bit_test(job_resrcs_ptr->node_bitmap, i))
+ continue;
+ node_offset++;
+ if (!bit_test(job_ptr->node_bitmap, i))
continue;
node_ptr = node_record_table_ptr + i;
@@ -1681,7 +1691,8 @@ static int _add_job_to_nodes(struct cr_record *cr_ptr,
gres_list = cr_ptr->nodes[i].gres_list;
else
gres_list = node_ptr->gres_list;
- gres_plugin_job_alloc(job_ptr->gres_list, gres_list, cpu_cnt);
+ gres_plugin_job_alloc(job_ptr->gres_list, gres_list,
+ node_cnt, node_offset, cpu_cnt);
gres_plugin_node_state_log(gres_list, node_ptr->name);
if (exclusive)
--
GitLab