From 26fa811ba4fd2cf7c795c50ad628a3d837feb43b Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 10 Jul 2009 18:03:29 +0000
Subject: [PATCH] svn merge -r18102:18114
https://eris.llnl.gov/svn/slurm/branches/slurm-2.0
---
NEWS | 1 +
.../mysql/accounting_storage_mysql.c | 18 ++++++------
src/slurmctld/job_mgr.c | 28 +++++++++++++------
3 files changed, 29 insertions(+), 18 deletions(-)
diff --git a/NEWS b/NEWS
index 8ff16356734..40ad95536ac 100644
--- a/NEWS
+++ b/NEWS
@@ -116,6 +116,7 @@ documents those changes that are of interest to users and admins.
to properly handle user names that contain all digits. Return error code
from uid_from_string() and gid_from_string() functions rather than a uid of
-1, which might be a valid uid or gid on some systems.
+ -- Fix in re-calcuation of job priorities do to DOWN or DRAINED nodes.
* Changes in SLURM 2.0.3
========================
diff --git a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
index 49b017d5122..4d3e55d72cd 100644
--- a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
+++ b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
@@ -3032,14 +3032,14 @@ static int _mysql_acct_check_tables(MYSQL *db_conn)
{ "user_usec", "int unsigned default 0 not null" },
{ "sys_sec", "int unsigned default 0 not null" },
{ "sys_usec", "int unsigned default 0 not null" },
- { "max_vsize", "int unsigned default 0 not null" },
+ { "max_vsize", "bigint unsigned default 0 not null" },
{ "max_vsize_task", "smallint unsigned default 0 not null" },
{ "max_vsize_node", "int unsigned default 0 not null" },
- { "ave_vsize", "float default 0.0 not null" },
- { "max_rss", "int unsigned default 0 not null" },
+ { "ave_vsize", "double default 0.0 not null" },
+ { "max_rss", "bigint unsigned default 0 not null" },
{ "max_rss_task", "smallint unsigned default 0 not null" },
{ "max_rss_node", "int unsigned default 0 not null" },
- { "ave_rss", "float default 0.0 not null" },
+ { "ave_rss", "double default 0.0 not null" },
{ "max_pages", "int unsigned default 0 not null" },
{ "max_pages_task", "smallint unsigned default 0 not null" },
{ "max_pages_node", "int unsigned default 0 not null" },
@@ -10972,13 +10972,13 @@ extern int jobacct_storage_p_step_complete(mysql_conn_t *mysql_conn,
"user_sec=%ld, user_usec=%ld, "
"sys_sec=%ld, sys_usec=%ld, "
"max_vsize=%u, max_vsize_task=%u, "
- "max_vsize_node=%u, ave_vsize=%.2f, "
+ "max_vsize_node=%u, ave_vsize=%f, "
"max_rss=%u, max_rss_task=%u, "
- "max_rss_node=%u, ave_rss=%.2f, "
+ "max_rss_node=%u, ave_rss=%f, "
"max_pages=%u, max_pages_task=%u, "
- "max_pages_node=%u, ave_pages=%.2f, "
- "min_cpu=%.2f, min_cpu_task=%u, "
- "min_cpu_node=%u, ave_cpu=%.2f "
+ "max_pages_node=%u, ave_pages=%f, "
+ "min_cpu=%f, min_cpu_task=%u, "
+ "min_cpu_node=%u, ave_cpu=%f "
"where id=%d and stepid=%u",
step_table, (int)now,
comp_status,
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index a35dee4ec66..986485edc7c 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -2592,8 +2592,8 @@ static int _job_create(job_desc_msg_t * job_desc, int allocate, int will_run,
/* already confirmed submit_uid==0 */
/* If the priority isn't given we will figure it out later
- after we see if the job is eligible or not. So we want
- NO_VAL if not set. */
+ * after we see if the job is eligible or not. So we want
+ * NO_VAL if not set. */
job_ptr->priority = job_desc->priority;
if (update_job_dependency(job_ptr, job_desc->dependency)) {
@@ -3541,7 +3541,18 @@ void job_time_limit(void)
}
}
+ /* This needs to be near the top of the loop, checks every
+ * running, suspended and pending job */
resv_status = job_resv_check(job_ptr);
+
+ if ((job_ptr->priority == 1) && (!IS_JOB_FINISHED(job_ptr))) {
+ /* Rather than resetting job priorities whenever a
+ * DOWN, DRAINED or non-responsive node is returned to
+ * service, we pick them up here. There will be a small
+ * delay in restting a job's priority, but the code is
+ * a lot cleaner this way. */
+ _set_job_prio(job_ptr);
+ }
if (!IS_JOB_RUNNING(job_ptr))
continue;
@@ -5260,6 +5271,9 @@ int update_job(job_desc_msg_t * job_specs, uid_t uid)
if (IS_JOB_PENDING(job_ptr) && detail_ptr) {
detail_ptr->begin_time = job_specs->begin_time;
update_accounting = true;
+ if ((job_ptr->priority == 1) &&
+ (detail_ptr->begin_time <= now))
+ _set_job_prio(job_ptr);
} else
error_code = ESLURM_DISABLED;
}
@@ -6905,9 +6919,9 @@ extern int update_job_account(char *module, struct job_record *job_ptr,
!job_ptr->assoc_ptr &&
!(accounting_enforce & ACCOUNTING_ENFORCE_ASSOCS)) {
/* if not enforcing associations we want to look for
- the default account and use it to avoid getting
- trash in the accounting records.
- */
+ * the default account and use it to avoid getting
+ * trash in the accounting records.
+ */
assoc_rec.acct = NULL;
assoc_mgr_fill_in_assoc(acct_db_conn, &assoc_rec,
accounting_enforce,
@@ -7308,10 +7322,6 @@ _copy_job_record_to_job_desc(struct job_record *job_ptr)
/* construct a job_desc_msg_t from job */
job_desc = xmalloc(sizeof(job_desc_msg_t));
- if (!job_desc) {
- error("_pack_job_for_ckpt: memory exhausted");
- return NULL;
- }
job_desc->account = xstrdup(job_ptr->account);
job_desc->acctg_freq = details->acctg_freq;
--
GitLab