From 39ca450e35d26786c07fa49f35c2d6f88daa9d44 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 5 Sep 2008 21:04:59 +0000
Subject: [PATCH] Add configuration parameter CompleteTime to control how long
to wait for a job's completion before allocating already released
resources to pending jobs. This can be used to reduce fragmentation of
resources.
---
NEWS | 3 +++
doc/html/configurator.html.in | 1 +
doc/man/man5/slurm.conf.5 | 16 ++++++++++++-
slurm/slurm.h.in | 2 ++
src/api/config_info.c | 2 ++
src/common/read_config.c | 14 ++++++++---
src/common/read_config.h | 1 +
src/common/slurm_protocol_pack.c | 2 ++
src/slurmctld/proc_req.c | 1 +
src/sview/popups.c | 40 +++++++++++++++++++++++---------
10 files changed, 67 insertions(+), 15 deletions(-)
diff --git a/NEWS b/NEWS
index 6eee814bcd3..03bed7dd7de 100644
--- a/NEWS
+++ b/NEWS
@@ -12,6 +12,9 @@ documents those changes that are of interest to users and admins.
-- Added configuration parameter PrologSlurmctld, which can be used to boot
nodes into a particular state for each job. See "man slurm.conf" for
details.
+ -- Add configuration parameter CompleteTime to control how long to wait for
+ a job's completion before allocating already released resources to pending
+ jobs. This can be used to reduce fragmentation of resources.
* Changes in SLURM 1.4.0-pre1
=============================
diff --git a/doc/html/configurator.html.in b/doc/html/configurator.html.in
index 613a844d9b5..01fdda1e9c6 100644
--- a/doc/html/configurator.html.in
+++ b/doc/html/configurator.html.in
@@ -197,6 +197,7 @@ function displayfile()
"#HealthCheckProgram= <br>" +
"InactiveLimit=" + document.config.inactive_limit.value + "<br>" +
"MinJobAge=" + document.config.min_job_age.value + "<br>" +
+ "#CompleteWait=0 <br>" +
"KillWait=" + document.config.kill_wait.value + "<br>" +
"#MessageTimeout=10 <br>" +
"SlurmctldTimeout=" + document.config.slurmctld_timeout.value + "<br>" +
diff --git a/doc/man/man5/slurm.conf.5 b/doc/man/man5/slurm.conf.5
index 4aad0ae536b..dfc2cb84e15 100644
--- a/doc/man/man5/slurm.conf.5
+++ b/doc/man/man5/slurm.conf.5
@@ -1,4 +1,4 @@
-.TH "slurm.conf" "5" "September 2008" "slurm.conf 1.3" "Slurm configuration file"
+.TH "slurm.conf" "5" "September 2008" "slurm.conf 1.4" "Slurm configuration file"
.SH "NAME"
slurm.conf \- Slurm configuration file
@@ -176,6 +176,20 @@ The name by which this SLURM managed cluster is known for accounting
purposes. This is needed distinguish between accounting data from
multiple clusters being recorded in a single database.
+.TP
+\fBCompleteWait\fR
+The time, in seconds, given for a job to remain in COMPLETING state
+before any any additional jobs are scheduled.
+If set to zero, pending jobs will be started as soon as possible.
+Since a COMPLETING job's resources are released for use by other
+jobs as soon as the \fBEpilog\fR completes on each individual node,
+this can result in very fragmented resource allocations.
+In order to minimize fragmentation of resources without substantial
+performance degradation, a value equal to \fBKillWait\fR plus two
+is recommended.
+The default value is 0 seconds for fastest job scheduling.
+The value may not exceed 65533.
+
.TP
\fBControlAddr\fR
Name that \fBControlMachine\fR should be referred to in
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index c752d1c0bc8..510a6fe2647 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -961,6 +961,8 @@ typedef struct slurm_ctl_conf {
uint16_t cache_groups; /* cache /etc/groups to avoid initgroups(2) */
char *checkpoint_type; /* checkpoint plugin type */
char *cluster_name; /* general name of the entire cluster */
+ uint16_t complete_wait; /* seconds to wait for job completion before
+ * scheduling another job */
char *control_addr; /* comm path of slurmctld primary server */
char *control_machine; /* name of slurmctld primary server */
char *crypto_type; /* cryptographic signature plugin */
diff --git a/src/api/config_info.c b/src/api/config_info.c
index b147f822472..9b21664013c 100644
--- a/src/api/config_info.c
+++ b/src/api/config_info.c
@@ -148,6 +148,8 @@ void slurm_print_ctl_conf ( FILE* out,
slurm_ctl_conf_ptr->checkpoint_type);
fprintf(out, "ClusterName = %s\n",
slurm_ctl_conf_ptr->cluster_name);
+ fprintf(out, "CompleteWait = %u\n",
+ slurm_ctl_conf_ptr->complete_wait);
fprintf(out, "ControlAddr = %s\n",
slurm_ctl_conf_ptr->control_addr);
fprintf(out, "ControlMachine = %s\n",
diff --git a/src/common/read_config.c b/src/common/read_config.c
index a07d99c00df..58ad7042c37 100644
--- a/src/common/read_config.c
+++ b/src/common/read_config.c
@@ -138,6 +138,7 @@ s_p_options_t slurm_conf_options[] = {
{"CheckpointType", S_P_STRING},
{"CacheGroups", S_P_UINT16},
{"ClusterName", S_P_STRING},
+ {"CompleteWait", S_P_UINT16},
{"ControlAddr", S_P_STRING},
{"ControlMachine", S_P_STRING},
{"CryptoType", S_P_STRING},
@@ -1213,6 +1214,7 @@ init_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr)
ctl_conf_ptr->cache_groups = 0;
xfree (ctl_conf_ptr->checkpoint_type);
xfree (ctl_conf_ptr->cluster_name);
+ ctl_conf_ptr->complete_wait = (uint16_t) NO_VAL;
xfree (ctl_conf_ptr->control_addr);
xfree (ctl_conf_ptr->control_machine);
xfree (ctl_conf_ptr->crypto_type);
@@ -1523,6 +1525,9 @@ validate_and_set_defaults(slurm_ctl_conf_t *conf, s_p_hashtbl_t *hashtbl)
s_p_get_string(&conf->cluster_name, "ClusterName", hashtbl);
+ if (!s_p_get_uint16(&conf->complete_wait, "CompleteWait", hashtbl))
+ conf->complete_wait = DEFAULT_COMPLETE_WAIT;
+
if (!s_p_get_string(&conf->control_machine, "ControlMachine", hashtbl))
fatal ("validate_and_set_defaults: "
"ControlMachine not specified.");
@@ -1580,7 +1585,8 @@ validate_and_set_defaults(slurm_ctl_conf_t *conf, s_p_hashtbl_t *hashtbl)
if (s_p_get_uint32(&conf->def_mem_per_task, "DefMemPerCPU", hashtbl))
conf->def_mem_per_task |= MEM_PER_CPU;
- else if (!s_p_get_uint32(&conf->def_mem_per_task, "DefMemPerNode", hashtbl))
+ else if (!s_p_get_uint32(&conf->def_mem_per_task, "DefMemPerNode",
+ hashtbl))
conf->def_mem_per_task = DEFAULT_MEM_PER_CPU;
if (!s_p_get_boolean((bool *) &conf->disable_root_jobs,
@@ -1707,8 +1713,10 @@ validate_and_set_defaults(slurm_ctl_conf_t *conf, s_p_hashtbl_t *hashtbl)
if (!s_p_get_uint16(&conf->get_env_timeout, "GetEnvTimeout", hashtbl))
conf->get_env_timeout = DEFAULT_GET_ENV_TIMEOUT;
- s_p_get_uint16(&conf->health_check_interval, "HealthCheckInterval", hashtbl);
- s_p_get_string(&conf->health_check_program, "HealthCheckProgram", hashtbl);
+ s_p_get_uint16(&conf->health_check_interval, "HealthCheckInterval",
+ hashtbl);
+ s_p_get_string(&conf->health_check_program, "HealthCheckProgram",
+ hashtbl);
if (!s_p_get_uint16(&conf->kill_wait, "KillWait", hashtbl))
conf->kill_wait = DEFAULT_KILL_WAIT;
diff --git a/src/common/read_config.h b/src/common/read_config.h
index 987294c2ebf..39915150911 100644
--- a/src/common/read_config.h
+++ b/src/common/read_config.h
@@ -54,6 +54,7 @@ extern char *default_plugstack;
#define DEFAULT_ACCOUNTING_STORAGE_TYPE "accounting_storage/none"
#define DEFAULT_AUTH_TYPE "auth/none"
#define DEFAULT_CACHE_GROUPS 0
+#define DEFAULT_COMPLETE_WAIT 0
#define DEFAULT_CRYPTO_TYPE "crypto/openssl"
#define DEFAULT_EPILOG_MSG_TIME 2000
#define DEFAULT_FAST_SCHEDULE 1
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 85832294dbd..d1aa1988450 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -2241,6 +2241,7 @@ _pack_slurm_ctl_conf_msg(slurm_ctl_conf_info_msg_t * build_ptr, Buf buffer)
pack16(build_ptr->cache_groups, buffer);
packstr(build_ptr->checkpoint_type, buffer);
packstr(build_ptr->cluster_name, buffer);
+ pack16(build_ptr->complete_wait, buffer);
packstr(build_ptr->control_addr, buffer);
packstr(build_ptr->control_machine, buffer);
packstr(build_ptr->crypto_type, buffer);
@@ -2394,6 +2395,7 @@ _unpack_slurm_ctl_conf_msg(slurm_ctl_conf_info_msg_t **
safe_unpackstr_xmalloc(&build_ptr->checkpoint_type, &uint32_tmp,
buffer);
safe_unpackstr_xmalloc(&build_ptr->cluster_name, &uint32_tmp, buffer);
+ safe_unpack16(&build_ptr->complete_wait, buffer);
safe_unpackstr_xmalloc(&build_ptr->control_addr, &uint32_tmp, buffer);
safe_unpackstr_xmalloc(&build_ptr->control_machine, &uint32_tmp,
buffer);
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 65ae919a1bb..6a977229a4d 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -358,6 +358,7 @@ void _fill_ctld_conf(slurm_ctl_conf_t * conf_ptr)
conf_ptr->cache_groups = conf->cache_groups;
conf_ptr->checkpoint_type = xstrdup(conf->checkpoint_type);
conf_ptr->cluster_name = xstrdup(conf->cluster_name);
+ conf_ptr->complete_wait = conf->complete_wait;
conf_ptr->control_addr = xstrdup(conf->control_addr);
conf_ptr->control_machine = xstrdup(conf->control_machine);
conf_ptr->crypto_type = xstrdup(conf->crypto_type);
diff --git a/src/sview/popups.c b/src/sview/popups.c
index 6415cc5c0b4..9ffb523359e 100644
--- a/src/sview/popups.c
+++ b/src/sview/popups.c
@@ -258,6 +258,11 @@ static void _layout_ctl_conf(GtkTreeStore *treestore,
add_display_treestore_line(update, treestore, &iter,
"CheckpointType",
slurm_ctl_conf_ptr->checkpoint_type);
+ snprintf(temp_str, sizeof(temp_str), "%u",
+ slurm_ctl_conf_ptr->complete_wait);
+ add_display_treestore_line(update, treestore, &iter,
+ "CompleteWait",
+ temp_str);
add_display_treestore_line(update, treestore, &iter,
"ControlAddr",
slurm_ctl_conf_ptr->control_addr);
@@ -267,11 +272,17 @@ static void _layout_ctl_conf(GtkTreeStore *treestore,
add_display_treestore_line(update, treestore, &iter,
"CryptoType",
slurm_ctl_conf_ptr->crypto_type);
- snprintf(temp_str, sizeof(temp_str), "%u",
- slurm_ctl_conf_ptr->def_mem_per_task);
- add_display_treestore_line(update, treestore, &iter,
- "DefMemPerTask",
- temp_str);
+ if (slurm_ctl_conf_ptr->def_mem_per_task & MEM_PER_CPU) {
+ snprintf(temp_str, sizeof(temp_str), "%u",
+ slurm_ctl_conf_ptr->def_mem_per_task & (~MEM_PER_CPU));
+ add_display_treestore_line(update, treestore, &iter,
+ "DefMemPerCPU", temp_str);
+ } else {
+ snprintf(temp_str, sizeof(temp_str), "%u",
+ slurm_ctl_conf_ptr->def_mem_per_task);
+ add_display_treestore_line(update, treestore, &iter,
+ "DefMemPerNode", temp_str);
+ }
add_display_treestore_line(update, treestore, &iter,
"Epilog",
slurm_ctl_conf_ptr->epilog);
@@ -324,7 +335,8 @@ static void _layout_ctl_conf(GtkTreeStore *treestore,
add_display_treestore_line(update, treestore, &iter,
"JobCredentialPrivateKey",
- slurm_ctl_conf_ptr->job_credential_private_key);
+ slurm_ctl_conf_ptr->
+ job_credential_private_key);
add_display_treestore_line(update, treestore, &iter,
"JobCredentialPublicCertificate",
slurm_ctl_conf_ptr->
@@ -346,11 +358,17 @@ static void _layout_ctl_conf(GtkTreeStore *treestore,
add_display_treestore_line(update, treestore, &iter,
"MaxJobCount",
temp_str);
- snprintf(temp_str, sizeof(temp_str), "%u",
- slurm_ctl_conf_ptr->max_mem_per_task);
- add_display_treestore_line(update, treestore, &iter,
- "MaxMemPerTask",
- temp_str);
+ if (slurm_ctl_conf_ptr->max_mem_per_task & MEM_PER_CPU) {
+ snprintf(temp_str, sizeof(temp_str), "%u",
+ slurm_ctl_conf_ptr->max_mem_per_task & (~MEM_PER_CPU));
+ add_display_treestore_line(update, treestore, &iter,
+ "MaxMemPerCPU", temp_str);
+ } else {
+ snprintf(temp_str, sizeof(temp_str), "%u",
+ slurm_ctl_conf_ptr->max_mem_per_task);
+ add_display_treestore_line(update, treestore, &iter,
+ "MaxMemPerNode", temp_str);
+ }
snprintf(temp_str, sizeof(temp_str), "%u",
slurm_ctl_conf_ptr->msg_timeout);
add_display_treestore_line(update, treestore, &iter,
--
GitLab