From 3a6c788fc867b6a61a90f1d083af6dee3fbadcf5 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 21 May 2010 17:02:11 +0000
Subject: [PATCH] add logic to save/restore/log a job step's gres details (i.e.
specific bits)
---
src/common/gres.c | 216 +++++++++++++++++++++++++++++++-
src/common/gres.h | 26 ++++
src/plugins/gres/gpu/gres_gpu.c | 97 ++++++++++++++
src/plugins/gres/nic/gres_nic.c | 97 ++++++++++++++
src/slurmctld/job_mgr.c | 2 +-
src/slurmctld/slurmctld.h | 7 +-
src/slurmctld/step_mgr.c | 17 ++-
7 files changed, 456 insertions(+), 6 deletions(-)
diff --git a/src/common/gres.c b/src/common/gres.c
index d2aa93b92a5..67dfa0f19c1 100644
--- a/src/common/gres.c
+++ b/src/common/gres.c
@@ -130,6 +130,12 @@ typedef struct slurm_gres_ops {
int (*job_dealloc) ( void *job_gres_data,
void *node_gres_data,
int node_offset );
+ int (*step_state_pack) ( void *gres_data,
+ Buf buffer );
+ int (*step_state_unpack) ( void **gres_data,
+ Buf buffer );
+ void (*step_state_log) ( void *gres_data,
+ uint32_t job_id );
} slurm_gres_ops_t;
typedef struct slurm_gres_context {
@@ -194,7 +200,10 @@ static int _load_gres_plugin(char *plugin_name,
"job_state_log",
"job_test",
"job_alloc",
- "job_dealloc"
+ "job_dealloc",
+ "step_state_pack",
+ "step_state_unpack",
+ "step_state_log"
};
int n_syms = sizeof(syms) / sizeof(char *);
@@ -1341,7 +1350,7 @@ extern int gres_plugin_job_state_unpack(List *gres_list, Buf buffer,
}
if (i >= gres_context_cnt) {
error("gres_plugin_job_state_unpack: no plugin "
- "configured to unpack data type %u from job %j",
+ "configured to unpack data type %u from job %u",
plugin_id, job_id);
/* A likely sign that GresPlugins has changed.
* Not a fatal error, skip over the data. */
@@ -1597,3 +1606,206 @@ extern void gres_plugin_job_state_log(List gres_list, uint32_t job_id)
list_iterator_destroy(gres_iter);
slurm_mutex_unlock(&gres_context_lock);
}
+
+/*
+ * Pack a step's current gres status, called from slurmctld for save/restore
+ * IN gres_list - generated by gres_plugin_step_allocate()
+ * IN/OUT buffer - location to write state to
+ * IN job_id, step_id - job and step ID for logging
+ */
+extern int gres_plugin_step_state_pack(List gres_list, Buf buffer,
+ uint32_t job_id, uint32_t step_id)
+{
+ int i, rc = SLURM_SUCCESS, rc2;
+ uint32_t top_offset, gres_size = 0;
+ uint32_t header_offset, size_offset, data_offset, tail_offset;
+ uint32_t magic = GRES_MAGIC;
+ uint16_t rec_cnt = 0;
+ ListIterator gres_iter;
+ gres_state_t *gres_ptr;
+
+ top_offset = get_buf_offset(buffer);
+ pack16(rec_cnt, buffer); /* placeholder if data */
+
+ if (gres_list == NULL)
+ return rc;
+
+ (void) gres_plugin_init();
+
+ slurm_mutex_lock(&gres_context_lock);
+ gres_iter = list_iterator_create(gres_list);
+ while ((gres_ptr = (gres_state_t *) list_next(gres_iter))) {
+ for (i=0; i<gres_context_cnt; i++) {
+ if (gres_ptr->plugin_id !=
+ *(gres_context[i].ops.plugin_id))
+ continue;
+ header_offset = get_buf_offset(buffer);
+ pack32(magic, buffer);
+ pack32(gres_ptr->plugin_id, buffer);
+ size_offset = get_buf_offset(buffer);
+ pack32(gres_size, buffer); /* placeholder */
+ data_offset = get_buf_offset(buffer);
+ rc2 = (*(gres_context[i].ops.step_state_pack))
+ (gres_ptr->gres_data, buffer);
+ if (rc2 != SLURM_SUCCESS) {
+ rc = rc2;
+ set_buf_offset(buffer, header_offset);
+ continue;
+ }
+ tail_offset = get_buf_offset(buffer);
+ set_buf_offset(buffer, size_offset);
+ gres_size = tail_offset - data_offset;
+ pack32(gres_size, buffer);
+ set_buf_offset(buffer, tail_offset);
+ rec_cnt++;
+ break;
+ }
+ if (i >= gres_context_cnt) {
+ error("Could not find plugin id %u to pack record for "
+ "step %u.%u",
+ gres_ptr->plugin_id, job_id, step_id);
+ }
+ }
+ list_iterator_destroy(gres_iter);
+ slurm_mutex_unlock(&gres_context_lock);
+
+ tail_offset = get_buf_offset(buffer);
+ set_buf_offset(buffer, top_offset);
+ pack16(rec_cnt, buffer);
+ set_buf_offset(buffer, tail_offset);
+
+ return rc;
+}
+
+/*
+ * Unpack a step's current gres status, called from slurmctld for save/restore
+ * OUT gres_list - restored state stored by gres_plugin_step_state_pack()
+ * IN/OUT buffer - location to read state from
+ * IN job_id, step_id - job and step ID for logging
+ */
+extern int gres_plugin_step_state_unpack(List *gres_list, Buf buffer,
+ uint32_t job_id, uint32_t step_id)
+{
+ int i, rc, rc2;
+ uint32_t gres_size, magic, tail_offset, plugin_id;
+ uint16_t rec_cnt;
+ gres_state_t *gres_ptr;
+ void *gres_data;
+
+ safe_unpack16(&rec_cnt, buffer);
+ if (rec_cnt == 0)
+ return SLURM_SUCCESS;
+
+ rc = gres_plugin_init();
+
+ slurm_mutex_lock(&gres_context_lock);
+ if ((gres_context_cnt > 0) && (*gres_list == NULL)) {
+ *gres_list = list_create(_gres_job_list_delete);
+ if (*gres_list == NULL)
+ fatal("list_create malloc failure");
+ }
+
+ for (i=0; i<gres_context_cnt; i++)
+ gres_context[i].unpacked_info = false;
+
+ while ((rc == SLURM_SUCCESS) && (rec_cnt)) {
+ if ((buffer == NULL) || (remaining_buf(buffer) == 0))
+ break;
+ rec_cnt--;
+ safe_unpack32(&magic, buffer);
+ if (magic != GRES_MAGIC)
+ goto unpack_error;
+ safe_unpack32(&plugin_id, buffer);
+ safe_unpack32(&gres_size, buffer);
+ for (i=0; i<gres_context_cnt; i++) {
+ if (*(gres_context[i].ops.plugin_id) == plugin_id)
+ break;
+ }
+ if (i >= gres_context_cnt) {
+ error("gres_plugin_step_state_unpack: no plugin "
+ "configured to unpack data type %u from "
+ "step %u.%u",
+ plugin_id, job_id, step_id);
+ /* A likely sign that GresPlugins has changed.
+ * Not a fatal error, skip over the data. */
+ tail_offset = get_buf_offset(buffer);
+ tail_offset += gres_size;
+ set_buf_offset(buffer, tail_offset);
+ continue;
+ }
+ gres_context[i].unpacked_info = true;
+ rc2 = (*(gres_context[i].ops.step_state_unpack))
+ (&gres_data, buffer);
+ if (rc2 != SLURM_SUCCESS) {
+ rc = rc2;
+ } else {
+ gres_ptr = xmalloc(sizeof(gres_state_t));
+ gres_ptr->plugin_id = *(gres_context[i].ops.plugin_id);
+ gres_ptr->gres_data = gres_data;
+ list_append(*gres_list, gres_ptr);
+ }
+ }
+
+fini: /* Insure that every gres plugin is called for unpack, even if no data
+ * was packed by the job. A likely sign that GresPlugins is
+ * inconsistently configured. */
+ for (i=0; i<gres_context_cnt; i++) {
+ if (gres_context[i].unpacked_info)
+ continue;
+ debug("gres_plugin_job_state_unpack: no info packed for %s "
+ "by step %u.%u",
+ gres_context[i].gres_type, job_id, step_id);
+ rc2 = (*(gres_context[i].ops.job_state_unpack))
+ (&gres_data, NULL);
+ if (rc2 != SLURM_SUCCESS) {
+ rc = rc2;
+ } else {
+ gres_ptr = xmalloc(sizeof(gres_state_t));
+ gres_ptr->plugin_id = *(gres_context[i].ops.plugin_id);
+ gres_ptr->gres_data = gres_data;
+ list_append(*gres_list, gres_ptr);
+ }
+ }
+ slurm_mutex_unlock(&gres_context_lock);
+
+ return rc;
+
+unpack_error:
+ error("gres_plugin_job_state_unpack: unpack error from step %u.%u",
+ job_id, step_id);
+ rc = SLURM_ERROR;
+ goto fini;
+}
+
+/*
+ * Log a step's current gres state
+ * IN gres_list - generated by gres_plugin_step_allocate()
+ * IN job_id - job's ID
+ */
+extern void gres_plugin_step_state_log(List gres_list, uint32_t job_id,
+ uint32_t step_id)
+{
+ int i;
+ ListIterator gres_iter;
+ gres_state_t *gres_ptr;
+
+ if (!gres_debug || (gres_list == NULL))
+ return;
+
+ (void) gres_plugin_init();
+
+ slurm_mutex_lock(&gres_context_lock);
+ gres_iter = list_iterator_create(gres_list);
+ while ((gres_ptr = (gres_state_t *) list_next(gres_iter))) {
+ for (i=0; i<gres_context_cnt; i++) {
+ if (gres_ptr->plugin_id !=
+ *(gres_context[i].ops.plugin_id))
+ continue;
+ (*(gres_context[i].ops.step_state_log))
+ (gres_ptr->gres_data, job_id);
+ break;
+ }
+ }
+ list_iterator_destroy(gres_iter);
+ slurm_mutex_unlock(&gres_context_lock);
+}
diff --git a/src/common/gres.h b/src/common/gres.h
index a6f079f5b10..727cf7fa089 100644
--- a/src/common/gres.h
+++ b/src/common/gres.h
@@ -275,4 +275,30 @@ extern int gres_plugin_job_dealloc(List job_gres_list, List node_gres_list,
*/
extern void gres_plugin_job_state_log(List gres_list, uint32_t job_id);
+/*
+ * Pack a step's current gres status, called from slurmctld for save/restore
+ * IN gres_list - generated by gres_plugin_step_allocate()
+ * IN/OUT buffer - location to write state to
+ * IN job_id, step_id - job and step ID for logging
+ */
+extern int gres_plugin_step_state_pack(List gres_list, Buf buffer,
+ uint32_t job_id, uint32_t step_id);
+
+/*
+ * Unpack a step's current gres status, called from slurmctld for save/restore
+ * OUT gres_list - restored state stored by gres_plugin_step_state_pack()
+ * IN/OUT buffer - location to read state from
+ * IN job_id, step_id - job and step ID for logging
+ */
+extern int gres_plugin_step_state_unpack(List *gres_list, Buf buffer,
+ uint32_t job_id, uint32_t step_id);
+
+/*
+ * Log a step's current gres state
+ * IN gres_list - generated by gres_plugin_step_allocate()
+ * IN job_id - job's ID
+ */
+extern void gres_plugin_step_state_log(List gres_list, uint32_t job_id,
+ uint32_t step_id);
+
#endif /* !_GRES_H */
diff --git a/src/plugins/gres/gpu/gres_gpu.c b/src/plugins/gres/gpu/gres_gpu.c
index 6859fd76696..c48a0b07c37 100644
--- a/src/plugins/gres/gpu/gres_gpu.c
+++ b/src/plugins/gres/gpu/gres_gpu.c
@@ -140,8 +140,29 @@ typedef struct gpu_job_state {
/* Resources currently allocated to job on each node */
uint32_t node_cnt;
bitstr_t **gpu_bit_alloc;
+#if 0
+Need to modify state save, allocation, free, etc.
+ /* Resources currently allocated to job steps on each node.
+ * This will be a subset of resources allocated to the job.
+ * gpu_bit_step_alloc is a subset of gpu_bit_alloc */
+ bitstr_t **gpu_bit_step_alloc;
+#endif
} gpu_job_state_t;
+/* Gres job step state as used by slurmctld. */
+typedef struct gpu_step_state {
+ /* Count of resources needed */
+ uint32_t gpu_cnt_alloc;
+
+ /* 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 the job step on each node */
+ uint32_t node_cnt;
+ bitstr_t **gpu_bit_alloc;
+} gpu_step_state_t;
+
/*
* This will be the output for "--gres=help" option.
* Called only by salloc, sbatch and srun.
@@ -988,3 +1009,79 @@ extern void job_state_log(void *gres_data, uint32_t job_id)
info(" gpu_bit_alloc:NULL");
}
}
+
+extern int step_state_pack(void *gres_data, Buf buffer)
+{
+ int i;
+ gpu_step_state_t *gres_ptr = (gpu_step_state_t *) gres_data;
+
+ pack32(gres_ptr->gpu_cnt_alloc, buffer);
+ pack8 (gres_ptr->gpu_cnt_mult, buffer);
+
+ pack32(gres_ptr->node_cnt, buffer);
+ for (i=0; i<gres_ptr->node_cnt; i++)
+ pack_bit_str(gres_ptr->gpu_bit_alloc[i], buffer);
+
+ return SLURM_SUCCESS;
+}
+
+extern int step_state_unpack(void **gres_data, Buf buffer)
+{
+ int i;
+ gpu_step_state_t *gres_ptr;
+
+ gres_ptr = xmalloc(sizeof(gpu_step_state_t));
+
+ if (buffer) {
+ safe_unpack32(&gres_ptr->gpu_cnt_alloc, buffer);
+ safe_unpack8 (&gres_ptr->gpu_cnt_mult, buffer);
+
+ safe_unpack32(&gres_ptr->node_cnt, buffer);
+ gres_ptr->gpu_bit_alloc = xmalloc(sizeof(bitstr_t *) *
+ (gres_ptr->node_cnt + 1));
+ for (i=0; i<gres_ptr->node_cnt; i++)
+ unpack_bit_str(&gres_ptr->gpu_bit_alloc[i], buffer);
+ }
+
+ *gres_data = gres_ptr;
+ return SLURM_SUCCESS;
+
+unpack_error:
+ error("Unpacking %s step state info", plugin_name);
+ if (gres_ptr->gpu_bit_alloc) {
+ for (i=0; i<gres_ptr->node_cnt; i++) {
+ if (gres_ptr->gpu_bit_alloc[i])
+ bit_free(gres_ptr->gpu_bit_alloc[i]);
+ }
+ xfree(gres_ptr->gpu_bit_alloc);
+ }
+ xfree(gres_ptr);
+ *gres_data = NULL;
+ return SLURM_ERROR;
+}
+
+extern void step_state_log(void *gres_data, uint32_t job_id, uint32_t step_id)
+{
+ gpu_step_state_t *gres_ptr;
+ char *mult, tmp_str[128];
+ int i;
+
+ xassert(gres_data);
+ gres_ptr = (gpu_step_state_t *) gres_data;
+ info("%s state for step %u.%u", plugin_name, job_id, step_id);
+ if (gres_ptr->gpu_cnt_mult)
+ mult = "cpu";
+ else
+ mult = "node";
+ info(" gpu_cnt %u per %s", gres_ptr->gpu_cnt_alloc, mult);
+
+ if (gres_ptr->node_cnt && gres_ptr->gpu_bit_alloc) {
+ for (i=0; i<gres_ptr->node_cnt; i++) {
+ bit_fmt(tmp_str, sizeof(tmp_str),
+ gres_ptr->gpu_bit_alloc[i]);
+ info(" gpu_bit_alloc[%d]:%s", i, tmp_str);
+ }
+ } else {
+ info(" gpu_bit_alloc:NULL");
+ }
+}
diff --git a/src/plugins/gres/nic/gres_nic.c b/src/plugins/gres/nic/gres_nic.c
index adee32805ad..d123bb11800 100644
--- a/src/plugins/gres/nic/gres_nic.c
+++ b/src/plugins/gres/nic/gres_nic.c
@@ -140,8 +140,29 @@ typedef struct nic_job_state {
/* Resources currently allocated to job on each node */
uint32_t node_cnt;
bitstr_t **nic_bit_alloc;
+#if 0
+Need to modify state save, allocation, free, etc.
+ /* Resources currently allocated to job steps on each node.
+ * This will be a subset of resources allocated to the job.
+ * nic_bit_step_alloc is a subset of nic_bit_alloc */
+ bitstr_t **nic_bit_step_alloc;
+#endif
} nic_job_state_t;
+/* Gres job step state as used by slurmctld. */
+typedef struct nic_step_state {
+ /* Count of resources needed */
+ uint32_t nic_cnt_alloc;
+
+ /* 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 the job step on each node */
+ uint32_t node_cnt;
+ bitstr_t **nic_bit_alloc;
+} nic_step_state_t;
+
/*
* This will be the output for "--gres=help" option.
* Called only by salloc, sbatch and srun.
@@ -988,3 +1009,79 @@ extern void job_state_log(void *gres_data, uint32_t job_id)
info(" nic_bit_alloc:NULL");
}
}
+
+extern int step_state_pack(void *gres_data, Buf buffer)
+{
+ int i;
+ nic_step_state_t *gres_ptr = (nic_step_state_t *) gres_data;
+
+ pack32(gres_ptr->nic_cnt_alloc, buffer);
+ pack8 (gres_ptr->nic_cnt_mult, buffer);
+
+ pack32(gres_ptr->node_cnt, buffer);
+ for (i=0; i<gres_ptr->node_cnt; i++)
+ pack_bit_str(gres_ptr->nic_bit_alloc[i], buffer);
+
+ return SLURM_SUCCESS;
+}
+
+extern int step_state_unpack(void **gres_data, Buf buffer)
+{
+ int i;
+ nic_step_state_t *gres_ptr;
+
+ gres_ptr = xmalloc(sizeof(nic_step_state_t));
+
+ if (buffer) {
+ safe_unpack32(&gres_ptr->nic_cnt_alloc, buffer);
+ safe_unpack8 (&gres_ptr->nic_cnt_mult, buffer);
+
+ safe_unpack32(&gres_ptr->node_cnt, buffer);
+ gres_ptr->nic_bit_alloc = xmalloc(sizeof(bitstr_t *) *
+ (gres_ptr->node_cnt + 1));
+ for (i=0; i<gres_ptr->node_cnt; i++)
+ unpack_bit_str(&gres_ptr->nic_bit_alloc[i], buffer);
+ }
+
+ *gres_data = gres_ptr;
+ return SLURM_SUCCESS;
+
+unpack_error:
+ error("Unpacking %s step state info", plugin_name);
+ if (gres_ptr->nic_bit_alloc) {
+ for (i=0; i<gres_ptr->node_cnt; i++) {
+ if (gres_ptr->nic_bit_alloc[i])
+ bit_free(gres_ptr->nic_bit_alloc[i]);
+ }
+ xfree(gres_ptr->nic_bit_alloc);
+ }
+ xfree(gres_ptr);
+ *gres_data = NULL;
+ return SLURM_ERROR;
+}
+
+extern void step_state_log(void *gres_data, uint32_t job_id, uint32_t step_id)
+{
+ nic_step_state_t *gres_ptr;
+ char *mult, tmp_str[128];
+ int i;
+
+ xassert(gres_data);
+ gres_ptr = (nic_step_state_t *) gres_data;
+ info("%s state for step %u.%u", plugin_name, job_id, step_id);
+ if (gres_ptr->nic_cnt_mult)
+ mult = "cpu";
+ else
+ mult = "node";
+ info(" nic_cnt %u per %s", gres_ptr->nic_cnt_alloc, mult);
+
+ if (gres_ptr->node_cnt && gres_ptr->nic_bit_alloc) {
+ for (i=0; i<gres_ptr->node_cnt; i++) {
+ bit_fmt(tmp_str, sizeof(tmp_str),
+ gres_ptr->nic_bit_alloc[i]);
+ info(" nic_bit_alloc[%d]:%s", i, tmp_str);
+ }
+ } else {
+ info(" nic_bit_alloc:NULL");
+ }
+}
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 61139bb0bd4..47cbd75e040 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -797,7 +797,7 @@ static void _dump_job_state(struct job_record *dump_job_ptr, Buf buffer)
while ((step_ptr = (struct step_record *)
list_next(step_iterator))) {
pack16((uint16_t) STEP_FLAG, buffer);
- dump_job_step_state(step_ptr, buffer);
+ dump_job_step_state(dump_job_ptr, step_ptr, buffer);
}
list_iterator_destroy(step_iterator);
pack16((uint16_t) 0, buffer); /* no step flag */
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 38853f1e270..7d14b6fd86a 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -413,7 +413,7 @@ struct job_record {
uint32_t exit_code; /* exit code for job (status from
* wait call) */
char *gres; /* generic resources */
- List gres_list; /* generic resource allocation detaisl */
+ List gres_list; /* generic resource allocation detail */
uint32_t group_id; /* group submitted under */
uint32_t job_id; /* job ID */
struct job_record *job_next; /* next entry with same hash index */
@@ -525,6 +525,7 @@ struct step_record {
uint32_t exit_code; /* highest exit code from any task */
bitstr_t *exit_node_bitmap; /* bitmap of exited nodes */
char *gres; /* generic resources required */
+ List gres_list; /* generic resource allocation detail */
char *host; /* host for srun communications */
struct job_record* job_ptr; /* ptr to the job that owns the step */
jobacctinfo_t *jobacct; /* keep track of process info in the
@@ -686,10 +687,12 @@ extern void dump_job_desc(job_desc_msg_t * job_specs);
/*
* dump_job_step_state - dump the state of a specific job step to a buffer,
* load with load_step_state
+ * IN job_ptr - pointer to job for which information is to be dumpped
* IN step_ptr - pointer to job step for which information is to be dumpped
* IN/OUT buffer - location to store data, pointers automatically advanced
*/
-extern void dump_job_step_state(struct step_record *step_ptr, Buf buffer);
+extern void dump_job_step_state(struct job_record *job_ptr,
+ struct step_record *step_ptr, Buf buffer);
/*
* dump_step_desc - dump the incoming step initiate request message
diff --git a/src/slurmctld/step_mgr.c b/src/slurmctld/step_mgr.c
index 282ca8d0426..9e20be8528d 100644
--- a/src/slurmctld/step_mgr.c
+++ b/src/slurmctld/step_mgr.c
@@ -57,6 +57,7 @@
#include "src/common/bitstring.h"
#include "src/common/checkpoint.h"
#include "src/common/forward.h"
+#include "src/common/gres.h"
#include "src/common/slurm_accounting_storage.h"
#include "src/common/slurm_jobacct_gather.h"
#include "src/common/slurm_protocol_interface.h"
@@ -2328,10 +2329,12 @@ resume_job_step(struct job_record *job_ptr)
/*
* dump_job_step_state - dump the state of a specific job step to a buffer,
* load with load_step_state
+ * IN job_ptr - pointer to job for which information is to be dumpped
* IN step_ptr - pointer to job step for which information is to be dumpped
* IN/OUT buffer - location to store data, pointers automatically advanced
*/
-extern void dump_job_step_state(struct step_record *step_ptr, Buf buffer)
+extern void dump_job_step_state(struct job_record *job_ptr,
+ struct step_record *step_ptr, Buf buffer)
{
pack32(step_ptr->step_id, buffer);
pack16(step_ptr->cyclic_alloc, buffer);
@@ -2368,7 +2371,11 @@ extern void dump_job_step_state(struct step_record *step_ptr, Buf buffer)
packstr(step_ptr->name, buffer);
packstr(step_ptr->network, buffer);
packstr(step_ptr->ckpt_dir, buffer);
+
packstr(step_ptr->gres, buffer);
+ (void) gres_plugin_step_state_pack(step_ptr->gres_list, buffer,
+ job_ptr->job_id, step_ptr->step_id);
+
pack16(step_ptr->batch_step, buffer);
if (!step_ptr->batch_step) {
pack_slurm_step_layout(step_ptr->step_layout, buffer,
@@ -2400,6 +2407,7 @@ extern int load_step_state(struct job_record *job_ptr, Buf buffer,
switch_jobinfo_t *switch_tmp = NULL;
check_jobinfo_t check_tmp = NULL;
slurm_step_layout_t *step_layout = NULL;
+ List gres_list = NULL;
if(protocol_version >= SLURM_2_2_PROTOCOL_VERSION) {
safe_unpack32(&step_id, buffer);
@@ -2433,7 +2441,11 @@ extern int load_step_state(struct job_record *job_ptr, Buf buffer,
safe_unpackstr_xmalloc(&name, &name_len, buffer);
safe_unpackstr_xmalloc(&network, &name_len, buffer);
safe_unpackstr_xmalloc(&ckpt_dir, &name_len, buffer);
+
safe_unpackstr_xmalloc(&gres, &name_len, buffer);
+ if (gres_plugin_step_state_unpack(&gres_list, buffer,
+ job_ptr->job_id, step_id) != SLURM_SUCCESS)
+ goto unpack_error;
safe_unpack16(&batch_step, buffer);
if (!batch_step) {
@@ -2534,6 +2546,7 @@ extern int load_step_state(struct job_record *job_ptr, Buf buffer,
step_ptr->no_kill = no_kill;
step_ptr->ckpt_dir = ckpt_dir;
step_ptr->gres = gres;
+ step_ptr->gres_list = gres_list;
step_ptr->port = port;
step_ptr->ckpt_interval= ckpt_interval;
step_ptr->mem_per_cpu = mem_per_cpu;
@@ -2591,6 +2604,8 @@ extern int load_step_state(struct job_record *job_ptr, Buf buffer,
xfree(network);
xfree(ckpt_dir);
xfree(gres);
+ if (gres_list)
+ list_destroy(gres_list);
xfree(bit_fmt);
xfree(core_job);
if (switch_tmp)
--
GitLab