From 64f1f691484f99a732dfd1a49a0e53c5e76a6d56 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@schedmd.com>
Date: Mon, 29 Feb 2016 14:34:27 -0800
Subject: [PATCH] Add profile request to the batch step, continuation of commit
ff2c5b887db
---
src/common/slurm_protocol_defs.h | 1 +
src/common/slurm_protocol_pack.c | 156 ++++++++++++++++++++++++-
src/slurmctld/job_scheduler.c | 1 +
src/slurmctld/proc_req.c | 6 +
src/slurmd/slurmstepd/slurmstepd_job.c | 1 +
5 files changed, 163 insertions(+), 2 deletions(-)
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index 62e0d9057c9..91b58a79d9d 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -973,6 +973,7 @@ typedef struct batch_job_launch_msg {
uint16_t job_core_spec; /* Count of specialized cores */
char *alias_list; /* node name/address/hostnamne aliases */
char *nodes; /* list of nodes allocated to job_step */
+ uint32_t profile; /* what to profile for the batch step */
char *script; /* the actual job script, default NONE */
char *std_err; /* pathname of stderr */
char *std_in; /* pathname of stdin */
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index c54e13c6b09..f056c59bc72 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -12317,7 +12317,68 @@ _pack_batch_job_launch_msg(batch_job_launch_msg_t * msg, Buf buffer,
{
xassert(msg != NULL);
- if (protocol_version >= SLURM_15_08_PROTOCOL_VERSION) {
+ if (protocol_version >= SLURM_16_05_PROTOCOL_VERSION) {
+ pack32(msg->job_id, buffer);
+ pack32(msg->step_id, buffer);
+ pack32(msg->uid, buffer);
+ packstr(msg->partition, buffer);
+ packstr(msg->user_name, buffer);
+ pack32(msg->gid, buffer);
+ pack32(msg->ntasks, buffer);
+ pack32(msg->pn_min_memory, buffer);
+
+ pack8(msg->open_mode, buffer);
+ pack8(msg->overcommit, buffer);
+
+ pack32(msg->array_job_id, buffer);
+ pack32(msg->array_task_id, buffer);
+
+ packstr(msg->acctg_freq, buffer);
+ pack16(msg->cpu_bind_type, buffer);
+ pack16(msg->cpus_per_task, buffer);
+ pack16(msg->restart_cnt, buffer);
+ pack16(msg->job_core_spec, buffer);
+
+ pack32(msg->num_cpu_groups, buffer);
+ if (msg->num_cpu_groups) {
+ pack16_array(msg->cpus_per_node, msg->num_cpu_groups,
+ buffer);
+ pack32_array(msg->cpu_count_reps, msg->num_cpu_groups,
+ buffer);
+ }
+
+ packstr(msg->alias_list, buffer);
+ packstr(msg->cpu_bind, buffer);
+ packstr(msg->nodes, buffer);
+ packstr(msg->script, buffer);
+ packstr(msg->work_dir, buffer);
+ packstr(msg->ckpt_dir, buffer);
+ packstr(msg->restart_dir, buffer);
+
+ packstr(msg->std_err, buffer);
+ packstr(msg->std_in, buffer);
+ packstr(msg->std_out, buffer);
+
+ pack32(msg->argc, buffer);
+ packstr_array(msg->argv, msg->argc, buffer);
+ packstr_array(msg->spank_job_env, msg->spank_job_env_size,
+ buffer);
+
+ pack32(msg->envc, buffer);
+ packstr_array(msg->environment, msg->envc, buffer);
+
+ pack32(msg->job_mem, buffer);
+
+ slurm_cred_pack(msg->cred, buffer, protocol_version);
+
+ select_g_select_jobinfo_pack(msg->select_jobinfo, buffer,
+ protocol_version);
+
+ packstr(msg->account, buffer);
+ packstr(msg->qos, buffer);
+ packstr(msg->resv_name, buffer);
+ pack32(msg->profile, buffer);
+ } else if (protocol_version >= SLURM_15_08_PROTOCOL_VERSION) {
pack32(msg->job_id, buffer);
pack32(msg->step_id, buffer);
pack32(msg->uid, buffer);
@@ -12451,7 +12512,98 @@ _unpack_batch_job_launch_msg(batch_job_launch_msg_t ** msg, Buf buffer,
launch_msg_ptr = xmalloc(sizeof(batch_job_launch_msg_t));
*msg = launch_msg_ptr;
- if (protocol_version >= SLURM_15_08_PROTOCOL_VERSION) {
+ if (protocol_version >= SLURM_16_05_PROTOCOL_VERSION) {
+ safe_unpack32(&launch_msg_ptr->job_id, buffer);
+ safe_unpack32(&launch_msg_ptr->step_id, buffer);
+ safe_unpack32(&launch_msg_ptr->uid, buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->partition,
+ &uint32_tmp, buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->user_name,
+ &uint32_tmp, buffer);
+ safe_unpack32(&launch_msg_ptr->gid, buffer);
+ safe_unpack32(&launch_msg_ptr->ntasks, buffer);
+ safe_unpack32(&launch_msg_ptr->pn_min_memory, buffer);
+
+ safe_unpack8(&launch_msg_ptr->open_mode, buffer);
+ safe_unpack8(&launch_msg_ptr->overcommit, buffer);
+
+ safe_unpack32(&launch_msg_ptr->array_job_id, buffer);
+ safe_unpack32(&launch_msg_ptr->array_task_id, buffer);
+
+ safe_unpackstr_xmalloc(&launch_msg_ptr->acctg_freq,
+ &uint32_tmp, buffer);
+ safe_unpack16(&launch_msg_ptr->cpu_bind_type, buffer);
+ safe_unpack16(&launch_msg_ptr->cpus_per_task, buffer);
+ safe_unpack16(&launch_msg_ptr->restart_cnt, buffer);
+ safe_unpack16(&launch_msg_ptr->job_core_spec, buffer);
+
+ safe_unpack32(&launch_msg_ptr->num_cpu_groups, buffer);
+ if (launch_msg_ptr->num_cpu_groups) {
+ safe_unpack16_array(&(launch_msg_ptr->cpus_per_node),
+ &uint32_tmp, buffer);
+ if (launch_msg_ptr->num_cpu_groups != uint32_tmp)
+ goto unpack_error;
+ safe_unpack32_array(&(launch_msg_ptr->cpu_count_reps),
+ &uint32_tmp, buffer);
+ if (launch_msg_ptr->num_cpu_groups != uint32_tmp)
+ goto unpack_error;
+ }
+
+ safe_unpackstr_xmalloc(&launch_msg_ptr->alias_list,
+ &uint32_tmp, buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->cpu_bind, &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->nodes, &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->script, &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->work_dir, &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->ckpt_dir, &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->restart_dir,
+ &uint32_tmp, buffer);
+
+ safe_unpackstr_xmalloc(&launch_msg_ptr->std_err, &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->std_in, &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->std_out, &uint32_tmp,
+ buffer);
+
+ safe_unpack32(&launch_msg_ptr->argc, buffer);
+ safe_unpackstr_array(&launch_msg_ptr->argv,
+ &launch_msg_ptr->argc, buffer);
+ safe_unpackstr_array(&launch_msg_ptr->spank_job_env,
+ &launch_msg_ptr->spank_job_env_size,
+ buffer);
+
+ safe_unpack32(&launch_msg_ptr->envc, buffer);
+ safe_unpackstr_array(&launch_msg_ptr->environment,
+ &launch_msg_ptr->envc, buffer);
+
+ safe_unpack32(&launch_msg_ptr->job_mem, buffer);
+
+ if (!(launch_msg_ptr->cred = slurm_cred_unpack(
+ buffer, protocol_version)))
+ goto unpack_error;
+
+ if (select_g_select_jobinfo_unpack(&launch_msg_ptr->
+ select_jobinfo,
+ buffer, protocol_version))
+ goto unpack_error;
+
+ safe_unpackstr_xmalloc(&launch_msg_ptr->account,
+ &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->qos,
+ &uint32_tmp,
+ buffer);
+ safe_unpackstr_xmalloc(&launch_msg_ptr->resv_name,
+ &uint32_tmp,
+ buffer);
+ safe_unpack32(&launch_msg_ptr->profile, buffer);
+ } else if (protocol_version >= SLURM_15_08_PROTOCOL_VERSION) {
safe_unpack32(&launch_msg_ptr->job_id, buffer);
safe_unpack32(&launch_msg_ptr->step_id, buffer);
safe_unpack32(&launch_msg_ptr->uid, buffer);
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index e03f4816384..7bb5e50af70 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -2027,6 +2027,7 @@ extern batch_job_launch_msg_t *build_launch_job_msg(struct job_record *job_ptr,
launch_msg_ptr->cpus_per_task = job_ptr->details->cpus_per_task;
launch_msg_ptr->pn_min_memory = job_ptr->details->pn_min_memory;
launch_msg_ptr->restart_cnt = job_ptr->restart_cnt;
+ launch_msg_ptr->profile = job_ptr->profile;
if (make_batch_job_cred(launch_msg_ptr, job_ptr, protocol_version)) {
/* FIXME: This is a kludge, but this event indicates a serious
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index aa3dc4a00c3..54f222b91c5 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -4675,6 +4675,7 @@ static int _launch_batch_step(job_desc_msg_t *job_desc_msg, uid_t uid,
launch_msg_ptr->uid = uid;
launch_msg_ptr->nodes = xstrdup(job_ptr->alias_list);
launch_msg_ptr->partition = xstrdup(job_ptr->partition);
+
launch_msg_ptr->restart_cnt = job_ptr->restart_cnt;
if (job_ptr->details) {
launch_msg_ptr->pn_min_memory = job_ptr->details->
@@ -4735,6 +4736,11 @@ static int _launch_batch_step(job_desc_msg_t *job_desc_msg, uid_t uid,
launch_msg_ptr->select_jobinfo = select_g_select_jobinfo_copy(
job_ptr->select_jobinfo);
+ if (job_desc_msg->profile != ACCT_GATHER_PROFILE_NOT_SET)
+ launch_msg_ptr->profile = job_desc_msg->profile;
+ else
+ launch_msg_ptr->profile = job_ptr->profile;
+
/* FIXME: for some reason these CPU arrays total all the CPUs
* actually allocated, rather than totaling up to the requested
* CPU count for the allocation.
diff --git a/src/slurmd/slurmstepd/slurmstepd_job.c b/src/slurmd/slurmstepd/slurmstepd_job.c
index efedd5defd3..31b94eed3b6 100644
--- a/src/slurmd/slurmstepd/slurmstepd_job.c
+++ b/src/slurmd/slurmstepd/slurmstepd_job.c
@@ -457,6 +457,7 @@ batch_stepd_step_rec_create(batch_job_launch_msg_t *msg)
job->batch = true;
job->node_name = xstrdup(conf->node_name);
job->user_name = xstrdup(msg->user_name);
+ job->profile = msg->profile;
/* This needs to happen before acct_gather_profile_startpoll
and only really looks at the profile in the job.
*/
--
GitLab