From 05f38e08bcf3ed063f9d9efd4221c82582bfecc8 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Mon, 10 Mar 2008 16:12:50 +0000
Subject: [PATCH] svn merge -r13515:13533
https://eris.llnl.gov/svn/slurm/branches/slurm-1.2
---
NEWS | 5 +-
slurm/slurm.h.in | 95 +++++++++++++--------------
src/api/job_info.c | 4 +-
src/common/slurm_protocol_pack.c | 2 +-
src/scontrol/update_job.c | 10 +++
src/slurmctld/job_mgr.c | 31 +++++++--
src/slurmctld/slurmctld.h | 107 ++++++++++++++++---------------
7 files changed, 146 insertions(+), 108 deletions(-)
diff --git a/NEWS b/NEWS
index 36b9c098ae1..a0f61ba81c8 100644
--- a/NEWS
+++ b/NEWS
@@ -6,6 +6,8 @@ documents those changes that are of interest to users and admins.
-- Restructure the sbcast RPC to take advantage of larger buffers available
in Slurm v1.3 RPCs.
-- Fix several memory leaks.
+ -- In scontrol, show job's Requeue value, permit change of Requeue and COmment
+ values.
* Changes in SLURM 1.3.0-pre10
==============================
@@ -204,7 +206,8 @@ documents those changes that are of interest to users and admins.
to control message timeout.
-- Add threaded agent to manage a queue of Gold update requests for
performance reasons.
- -- Add slloc options --chdir and --get-user-env.
+ -- Add salloc options --chdir and --get-user-env (for Moab).
+ -- Modify scontrol update to support job comment changes.
* Changes in SLURM 1.2.24
=========================
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index dedb2dfdcd8..8eee044c451 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -577,27 +577,62 @@ typedef struct job_descriptor { /* For submit, allocate, and update requests */
} job_desc_msg_t;
typedef struct job_info {
- uint32_t job_id; /* job ID */
- char *name; /* name of the job */
- uint16_t batch_flag; /* 1 if batch: queued job with script */
- uint32_t alloc_sid; /* local sid making resource alloc */
+ char *account; /* charge to specified account */
char *alloc_node; /* local node making resource alloc */
- uint32_t user_id; /* user the job runs as */
+ uint32_t alloc_sid; /* local sid making resource alloc */
+ uint16_t batch_flag; /* 1 if batch: queued job with script */
+ char *command; /* command to be executed */
+ char *comment; /* arbitrary comment (used by Moab scheduler) */
+ uint16_t contiguous; /* 1 if job requires contiguous nodes */
+ uint16_t cpus_per_task; /* number of processors required for each task */
+ char *dependency; /* syncrhonize job execution with other jobs */
+ time_t end_time; /* time of termination, actual or expected */
+ char *exc_nodes; /* comma separated list of excluded nodes */
+ int *exc_node_inx; /* excluded list index pairs into node_table:
+ * start_range_1, end_range_1,
+ * start_range_2, .., -1 */
+ uint32_t exit_code; /* exit code for job (status from wait call) */
+ char *features; /* comma separated list of required features */
uint32_t group_id; /* group job sumitted as */
+ uint32_t job_id; /* job ID */
+ uint16_t job_min_cores; /* minimum cores per processor, default=0 */
+ uint32_t job_min_memory; /* minimum real memory per node, default=0 */
+ uint16_t job_min_procs; /* minimum processors per node, default=0 */
+ uint16_t job_min_sockets; /* minimum sockets per node, default=0 */
+ uint16_t job_min_threads; /* minimum threads per core, default=0 */
+ uint32_t job_min_tmp_disk; /* minimum tmp disk per node, default=0 */
uint16_t job_state; /* state of the job, see enum job_states */
- uint32_t time_limit; /* maximum run time in minutes or INFINITE */
- time_t submit_time; /* time of job submission */
- time_t start_time; /* time execution begins, actual or expected */
- time_t end_time; /* time of termination, actual or expected */
- time_t suspend_time; /* time job last suspended or resumed */
- time_t pre_sus_time; /* time job ran prior to last suspend */
- uint32_t priority; /* relative priority of the job,
- * 0=held, 1=required nodes DOWN/DRAINED */
+ char *licenses; /* licenses required by the job */
+ char *name; /* name of the job */
+ char *network; /* network specification */
char *nodes; /* list of nodes allocated to job */
int *node_inx; /* list index pairs into node_table for *nodes:
* start_range_1, end_range_1,
* start_range_2, .., -1 */
+ uint16_t ntasks_per_core;/* number of tasks to invoke on each core */
+ uint16_t ntasks_per_node;/* number of tasks to invoke on each node */
+ uint16_t ntasks_per_socket;/* number of tasks to invoke on each socket */
char *partition; /* name of assigned partition */
+ time_t pre_sus_time; /* time job ran prior to last suspend */
+ uint32_t priority; /* relative priority of the job,
+ * 0=held, 1=required nodes DOWN/DRAINED */
+ char *req_nodes; /* comma separated list of required nodes */
+ int *req_node_inx; /* required list index pairs into node_table:
+ * start_range_1, end_range_1,
+ * start_range_2, .., -1 */
+ uint16_t requeue; /* enable or disable job requeue option */
+ select_jobinfo_t select_jobinfo; /* opaque data type,
+ * process using select_g_get_jobinfo() */
+ uint16_t shared; /* 1 if job can share nodes with other jobs */
+ time_t start_time; /* time execution begins, actual or expected */
+ uint16_t state_reason; /* reason job still pending or failed, see
+ * slurm.h:enum job_state_reason */
+ time_t submit_time; /* time of job submission */
+ time_t suspend_time; /* time job last suspended or resumed */
+ uint32_t time_limit; /* maximum run time in minutes or INFINITE */
+ uint32_t user_id; /* user the job runs as */
+ char *work_dir; /* pathname of working directory */
+
uint16_t num_cpu_groups;/* elements in below cpu arrays */
uint32_t *cpus_per_node;/* cpus per node */
uint32_t *cpu_count_reps;/* how many nodes have same cpu count */
@@ -610,40 +645,6 @@ typedef struct job_info {
uint16_t max_cores; /* maximum number of cores per cpu */
uint16_t min_threads; /* minimum number of threads per core */
uint16_t max_threads; /* maximum number of threads per core */
- uint16_t shared; /* 1 if job can share nodes with other jobs */
- uint16_t contiguous; /* 1 if job requires contiguous nodes */
- uint16_t cpus_per_task; /* number of processors required for each task */
- uint16_t ntasks_per_node;/* number of tasks to invoke on each node */
- uint16_t ntasks_per_socket;/* number of tasks to invoke on each socket */
- uint16_t ntasks_per_core;/* number of tasks to invoke on each core */
- /* job constraints: */
- uint16_t job_min_procs; /* minimum processors per node, default=0 */
- uint16_t job_min_sockets; /* minimum sockets per node, default=0 */
- uint16_t job_min_cores; /* minimum cores per processor, default=0 */
- uint16_t job_min_threads; /* minimum threads per core, default=0 */
- uint32_t job_min_memory; /* minimum real memory per node, default=0 */
- uint32_t job_min_tmp_disk; /* minimum tmp disk per node, default=0 */
- char *req_nodes; /* comma separated list of required nodes */
- int *req_node_inx; /* required list index pairs into node_table:
- * start_range_1, end_range_1,
- * start_range_2, .., -1 */
- char *exc_nodes; /* comma separated list of excluded nodes */
- int *exc_node_inx; /* excluded list index pairs into node_table:
- * start_range_1, end_range_1,
- * start_range_2, .., -1 */
- char *features; /* comma separated list of required features */
- char *dependency; /* syncrhonize job execution with other jobs */
- uint32_t exit_code; /* exit code for job (status from wait call) */
- char *account; /* charge to specified account */
- uint16_t state_reason; /* reason job still pending or failed, see
- * slurm.h:enum job_state_reason */
- char *network; /* network specification */
- char *comment; /* arbitrary comment (used by Moab scheduler) */
- char *work_dir; /* pathname of working directory */
- char *command; /* command to be executed */
- select_jobinfo_t select_jobinfo; /* opaque data type,
- * process using select_g_get_jobinfo() */
- char *licenses; /* licenses required by the job */
} job_info_t;
typedef struct job_info_msg {
diff --git a/src/api/job_info.c b/src/api/job_info.c
index e8a7ffa53d7..2df156bc8ce 100644
--- a/src/api/job_info.c
+++ b/src/api/job_info.c
@@ -381,8 +381,8 @@ slurm_sprint_job_info ( job_info_t * job_ptr, int one_liner )
/****** Line 11 ******/
snprintf(tmp_line, sizeof(tmp_line),
- "Dependency=%s Account=%s",
- job_ptr->dependency, job_ptr->account);
+ "Dependency=%s Account=%s Requeue=%u",
+ job_ptr->dependency, job_ptr->account, job_ptr->requeue);
xstrcat(out, tmp_line);
if (one_liner)
xstrcat(out, " ");
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index e8b87494276..f1446dfcaea 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -2122,7 +2122,7 @@ _unpack_job_info_members(job_info_t * job, Buf buffer)
safe_unpack32(&job->num_nodes, buffer);
safe_unpack32(&job->max_nodes, buffer);
-
+ safe_unpack16(&job->requeue, buffer);
/*** unpack pending job details ***/
safe_unpack16(&job->shared, buffer);
diff --git a/src/scontrol/update_job.c b/src/scontrol/update_job.c
index 3813fdeb44b..540c216f617 100644
--- a/src/scontrol/update_job.c
+++ b/src/scontrol/update_job.c
@@ -209,6 +209,10 @@ scontrol_update_job (int argc, char *argv[])
job_msg.job_id =
(uint32_t) strtol(&argv[i][6],
(char **) NULL, 10);
+ else if (strncasecmp(argv[i], "Comment=", 8) == 0) {
+ job_msg.comment = &argv[i][8];
+ update_cnt++;
+ }
else if (strncasecmp(argv[i], "TimeLimit=", 10) == 0) {
int time_limit = time_str2mins(&argv[i][10]);
if ((time_limit < 0) && (time_limit != INFINITE)) {
@@ -247,6 +251,12 @@ scontrol_update_job (int argc, char *argv[])
(char **) NULL, 10);
update_cnt++;
}
+ else if (strncasecmp(argv[i], "Requeue=", 8) == 0) {
+ job_msg.requeue =
+ (uint16_t) strtol(&argv[i][8],
+ (char **) NULL, 10);
+ update_cnt++;
+ }
else if ((strncasecmp(argv[i], "MinNodes=", 9) == 0) ||
(strncasecmp(argv[i], "ReqNodes=", 9) == 0)) {
char *tmp;
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index cce759191f9..c969f049825 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -3098,6 +3098,7 @@ static void _pack_default_job_details(struct job_details *detail_ptr,
pack32(detail_ptr->min_nodes, buffer);
pack32(detail_ptr->max_nodes, buffer);
+ pack16(detail_ptr->requeue, buffer);
} else {
packnull(buffer);
packnull(buffer);
@@ -3106,6 +3107,7 @@ static void _pack_default_job_details(struct job_details *detail_ptr,
pack32((uint32_t) 0, buffer);
pack32((uint32_t) 0, buffer);
+ pack16((uint16_t) 0, buffer);
}
}
@@ -3785,6 +3787,9 @@ int update_job(job_desc_msg_t * job_specs, uid_t uid)
info("update_job: setting features to %s for "
"job_id %u", job_specs->features,
job_specs->job_id);
+ } else {
+ info("update_job: cleared features for job %u",
+ job_specs->job_id);
}
} else {
error("Attempt to change features for job %u",
@@ -3793,11 +3798,26 @@ int update_job(job_desc_msg_t * job_specs, uid_t uid)
}
}
+ if (job_specs->comment) {
+ xfree(job_ptr->comment);
+ job_ptr->comment = job_specs->comment;
+ job_specs->comment = NULL; /* Nothing left to free */
+ info("update_job: setting comment to %s for job_id %u",
+ job_ptr->comment, job_specs->job_id);
+ }
+
if (job_specs->name) {
xfree(job_ptr->name);
- job_ptr->name = xstrdup(job_specs->name);
+ job_ptr->name = job_specs->name;
+ job_specs->name = NULL; /* Nothing left to free */
info("update_job: setting name to %s for job_id %u",
- job_specs->name, job_specs->job_id);
+ job_ptr->name, job_specs->job_id);
+ }
+
+ if (job_specs->requeue) {
+ detail_ptr->requeue = job_specs->requeue;
+ info("update_job: setting requeue to %u for job_id %u",
+ job_specs->requeue, job_specs->job_id);
}
if (job_specs->partition) {
@@ -3881,10 +3901,13 @@ int update_job(job_desc_msg_t * job_specs, uid_t uid)
if (job_specs->account) {
xfree(job_ptr->account);
if (job_specs->account[0] != '\0') {
- job_ptr->account = job_specs->account ;
+ job_ptr->account = job_specs->account;
+ job_specs->account = NULL; /* Nothing left to free */
info("update_job: setting account to %s for job_id %u",
job_ptr->account, job_specs->job_id);
- job_specs->account = NULL;
+ } else {
+ info("update_job: cleared account for job_id %u",
+ job_specs->job_id);
}
}
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 82938e618e8..7be782183b1 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -337,73 +337,83 @@ struct job_details {
};
struct job_record {
- uint32_t job_id; /* job ID */
- uint32_t magic; /* magic cookie for data integrity */
- char *name; /* name of the job */
- char *partition; /* name of the partition */
- struct part_record *part_ptr; /* pointer to the partition record */
+ char *account; /* account number to charge */
+ char *alloc_node; /* local node making resource alloc */
+ uint16_t alloc_resp_port; /* RESPONSE_RESOURCE_ALLOCATION port */
+ uint32_t alloc_sid; /* local sid making resource alloc */
uint16_t batch_flag; /* 1 or 2 if batch job (with script),
* 2 indicates retry mode (one retry) */
- uint32_t user_id; /* user the job runs as */
+ char *comment; /* arbitrary comment */
+ uint16_t cr_enabled; /* specify if if Consumable Resources
+ * is enabled. Needed since CR deals
+ * with a finer granularity in its
+ * node/cpu scheduling (available cpus
+ * instead of available nodes) than the
+ * bluegene and the linear plugins
+ * 0 if cr is NOT enabled,
+ * 1 if cr is enabled */
+ uint32_t db_index; /* used only for database
+ plugins */
+ struct job_details *details; /* job details */
+ time_t end_time; /* time of termination,
+ * actual or expected */
+ uint32_t exit_code; /* exit code for job (status from
+ * wait call) */
uint32_t group_id; /* group submitted under */
+ uint32_t job_id; /* job ID */
+ struct job_record *job_next; /* next entry with same hash index */
enum job_states job_state; /* state of the job */
uint16_t kill_on_node_fail; /* 1 if job should be killed on
* node failure */
uint16_t kill_on_step_done; /* 1 if job should be killed when
* the job step completes, 2 if kill
* in progress */
- select_jobinfo_t select_jobinfo;/* opaque data */
+ char *licenses; /* licenses required by the job */
+ uint16_t mail_type; /* see MAIL_JOB_* in slurm.h */
+ char *mail_user; /* user to get e-mail notification */
+ uint32_t magic; /* magic cookie for data integrity */
+ char *name; /* name of the job */
+ char *network; /* network/switch requirement spec */
+ uint16_t next_step_id; /* next step id to be used */
char *nodes; /* list of nodes allocated to job */
+ slurm_addr *node_addr; /* addresses of the nodes allocated to
+ * job */
bitstr_t *node_bitmap; /* bitmap of nodes allocated to job */
+ uint32_t node_cnt; /* count of nodes allocated to job */
char *nodes_completing; /* nodes still in completing state
* for this job, used to insure
* epilog is not re-run for job */
uint32_t num_procs; /* count of required processors */
- uint32_t total_procs; /* number of allocated processors,
- for accounting */
- uint32_t time_limit; /* time_limit minutes or INFINITE,
- * NO_VAL implies partition max_time */
+ uint16_t other_port; /* port for client communications */
+ char *partition; /* name of the partition */
+ time_t pre_sus_time; /* time job ran prior to last suspend */
+ uint32_t priority; /* relative priority of the job,
+ * zero == held (don't initiate) */
+ uint32_t requid; /* requester user ID */
+ char *resp_host; /* host for srun communications */
+ select_jobinfo_t select_jobinfo;/* opaque data */
time_t start_time; /* time execution begins,
* actual or expected */
- time_t end_time; /* time of termination,
- * actual or expected */
+ uint16_t state_reason; /* reason job still pending or failed
+ * see slurm.h:enum job_wait_reason */
+ List step_list; /* list of job's steps */
time_t suspend_time; /* time job last suspended or resumed */
- time_t pre_sus_time; /* time job ran prior to last suspend */
- time_t tot_sus_time; /* total time in suspend state */
time_t time_last_active; /* time of last job activity */
- uint32_t priority; /* relative priority of the job,
- * zero == held (don't initiate) */
- struct job_details *details; /* job details */
+ uint32_t time_limit; /* time_limit minutes or INFINITE,
+ * NO_VAL implies partition max_time */
+ time_t tot_sus_time; /* total time in suspend state */
+ uint32_t total_procs; /* number of allocated processors,
+ for accounting */
+ struct part_record *part_ptr; /* pointer to the partition record */
+ uint32_t user_id; /* user the job runs as */
+
+ /* Per node allocation details */
uint16_t num_cpu_groups; /* record count in cpus_per_node and
* cpu_count_reps */
uint32_t *cpus_per_node; /* array of cpus per node allocated */
uint32_t *cpu_count_reps; /* array of consecutive nodes with
* same cpu count */
- uint32_t alloc_sid; /* local sid making resource alloc */
- char *alloc_node; /* local node making resource alloc */
- uint16_t next_step_id; /* next step id to be used */
- uint32_t node_cnt; /* count of nodes allocated to job */
- slurm_addr *node_addr; /* addresses of the nodes allocated to
- * job */
- List step_list; /* list of job's steps */
- char *resp_host; /* host for srun communications */
- uint16_t alloc_resp_port; /* RESPONSE_RESOURCE_ALLOCATION port */
- uint16_t other_port; /* port for client communications */
- char *account; /* account number to charge */
- char *comment; /* arbitrary comment */
- char *network; /* network/switch requirement spec */
- struct job_record *job_next; /* next entry with same hash index */
- uint16_t cr_enabled; /* specify if if Consumable
- * Resources is
- * enabled. Needed since CR
- * deals with a finer
- * granularity in its node/cpu
- * scheduling (available cpus
- * instead of available nodes)
- * than the bluegene and the
- * linear plugins
- * 0 if cr is NOT enabled,
- * 1 if cr is enabled */
+
uint32_t alloc_lps_cnt; /* number of hosts in alloc_lps
* or 0 if alloc_lps is not needed
* for the credentials */
@@ -411,16 +421,7 @@ struct job_record {
* allocated for this job */
uint32_t *used_lps; /* number of logical processors
* already allocated to job steps */
- char *licenses; /* licenses required by the job */
- uint16_t mail_type; /* see MAIL_JOB_* in slurm.h */
- char *mail_user; /* user to get e-mail notification */
- uint32_t requid; /* requester user ID */
- uint32_t exit_code; /* exit code for job (status from
- * wait call) */
- uint16_t state_reason; /* reason job still pending or failed
- * see slurm.h:enum job_wait_reason */
- uint32_t db_index; /* used only for database
- plugins */
+
};
/* Job dependency specification, used in "depend_list" within job_record */
--
GitLab