From d9640972940118f21b1823c116768c83e70da97a Mon Sep 17 00:00:00 2001
From: Danny Auble <da@schedmd.com>
Date: Fri, 8 Jul 2016 13:54:25 -0700
Subject: [PATCH] Rename LostJobs RunawayJobs in sacctmgr.
---
doc/man/man1/sacctmgr.1 | 7 ++-
src/common/slurm_accounting_storage.c | 9 ++--
src/common/slurm_accounting_storage.h | 7 +--
src/common/slurmdbd_defs.c | 6 +--
src/common/slurmdbd_defs.h | 2 +-
.../filetxt/accounting_storage_filetxt.c | 4 +-
.../accounting_storage/mysql/Makefile.am | 2 +-
.../accounting_storage/mysql/Makefile.in | 20 +++----
.../mysql/accounting_storage_mysql.c | 6 +--
...ost_jobs.c => as_mysql_fix_runaway_jobs.c} | 20 +++----
...ost_jobs.h => as_mysql_fix_runaway_jobs.h} | 10 ++--
.../none/accounting_storage_none.c | 4 +-
.../slurmdbd/accounting_storage_slurmdbd.c | 6 +--
src/sacctmgr/Makefile.am | 2 +-
src/sacctmgr/Makefile.in | 6 +--
...bs_functions.c => runaway_job_functions.c} | 53 ++++++++++---------
src/sacctmgr/sacctmgr.c | 15 ++++--
src/sacctmgr/sacctmgr.h | 4 +-
src/slurmdbd/proc_req.c | 24 ++++-----
19 files changed, 109 insertions(+), 98 deletions(-)
rename src/plugins/accounting_storage/mysql/{as_mysql_fix_lost_jobs.c => as_mysql_fix_runaway_jobs.c} (91%)
rename src/plugins/accounting_storage/mysql/{as_mysql_fix_lost_jobs.h => as_mysql_fix_runaway_jobs.h} (88%)
rename src/sacctmgr/{lost_jobs_functions.c => runaway_job_functions.c} (84%)
diff --git a/doc/man/man1/sacctmgr.1 b/doc/man/man1/sacctmgr.1
index 2a708a643bd..543ad1ce936 100644
--- a/doc/man/man1/sacctmgr.1
+++ b/doc/man/man1/sacctmgr.1
@@ -208,8 +208,11 @@ account and user associations inside their realm.
Events like downed or draining nodes on clusters.
.TP
-\fILostJobs\fR
-Jobs that have been orphanded on the local cluster.
+\fIRunawayJobs\fR
+Used only with the \fIlist\fR or \fIshow\fR command to report current
+jobs that have been orphanded on the local cluster and are now
+runaway. If there are jobs in this state it will also give you an
+option to "fix" them.
.TP
\fIjob\fR
diff --git a/src/common/slurm_accounting_storage.c b/src/common/slurm_accounting_storage.c
index 9387862020a..b3e6d25a8e8 100644
--- a/src/common/slurm_accounting_storage.c
+++ b/src/common/slurm_accounting_storage.c
@@ -170,7 +170,7 @@ typedef struct slurm_acct_storage_ops {
int (*roll_usage) (void *db_conn,
time_t sent_start, time_t sent_end,
uint16_t archive_data);
- int (*fix_lost_jobs) (void *db_conn, uint32_t uid, List jobs);
+ int (*fix_runaway_jobs) (void *db_conn, uint32_t uid, List jobs);
int (*node_down) (void *db_conn,
struct node_record *node_ptr,
time_t event_time,
@@ -257,7 +257,7 @@ static const char *syms[] = {
"acct_storage_p_get_txn",
"acct_storage_p_get_usage",
"acct_storage_p_roll_usage",
- "acct_storage_p_fix_lost_jobs",
+ "acct_storage_p_fix_runaway_jobs",
"clusteracct_storage_p_node_down",
"clusteracct_storage_p_node_up",
"clusteracct_storage_p_cluster_tres",
@@ -733,11 +733,12 @@ extern int acct_storage_g_roll_usage(void *db_conn,
return (*(ops.roll_usage))(db_conn, sent_start, sent_end, archive_data);
}
-extern int acct_storage_g_fix_lost_jobs(void *db_conn, uint32_t uid, List jobs)
+extern int acct_storage_g_fix_runaway_jobs(void *db_conn,
+ uint32_t uid, List jobs)
{
if (slurm_acct_storage_init(NULL) < 0)
return SLURM_ERROR;
- return (*(ops.fix_lost_jobs))(db_conn, uid, jobs);
+ return (*(ops.fix_runaway_jobs))(db_conn, uid, jobs);
}
diff --git a/src/common/slurm_accounting_storage.h b/src/common/slurm_accounting_storage.h
index f968bca6140..6e6d1390f1d 100644
--- a/src/common/slurm_accounting_storage.h
+++ b/src/common/slurm_accounting_storage.h
@@ -472,11 +472,12 @@ extern int acct_storage_g_roll_usage(void *db_conn,
uint16_t archive_data);
/*
- * Fix lost jobs
- * IN: jobs, a list of all the lost jobs
+ * Fix runaway jobs
+ * IN: jobs, a list of all the runaway jobs
* RET: SLURM_SUCCESS on success SLURM_ERROR else
*/
-extern int acct_storage_g_fix_lost_jobs(void *db_conn, uint32_t uid, List jobs);
+extern int acct_storage_g_fix_runaway_jobs(void *db_conn,
+ uint32_t uid, List jobs);
/*
* record shares used information for backup in case slurmctld restarts
diff --git a/src/common/slurmdbd_defs.c b/src/common/slurmdbd_defs.c
index da48ed0e65d..441e57adcf9 100644
--- a/src/common/slurmdbd_defs.c
+++ b/src/common/slurmdbd_defs.c
@@ -521,7 +521,7 @@ extern Buf pack_slurmdbd_msg(slurmdbd_msg_t *req, uint16_t rpc_version)
case DBD_GOT_MULT_JOB_START:
case DBD_SEND_MULT_MSG:
case DBD_GOT_MULT_MSG:
- case DBD_FIX_LOST_JOB:
+ case DBD_FIX_RUNAWAY_JOB:
slurmdbd_pack_list_msg(
(dbd_list_msg_t *)req->data, rpc_version,
req->msg_type, buffer);
@@ -3671,7 +3671,7 @@ extern void slurmdbd_pack_list_msg(dbd_list_msg_t *msg,
my_function = pack_config_key_pair;
break;
case DBD_GOT_JOBS:
- case DBD_FIX_LOST_JOB:
+ case DBD_FIX_RUNAWAY_JOB:
my_function = slurmdb_pack_job_rec;
break;
case DBD_GOT_LIST:
@@ -3773,7 +3773,7 @@ extern int slurmdbd_unpack_list_msg(dbd_list_msg_t **msg, uint16_t rpc_version,
my_destroy = destroy_config_key_pair;
break;
case DBD_GOT_JOBS:
- case DBD_FIX_LOST_JOB:
+ case DBD_FIX_RUNAWAY_JOB:
my_function = slurmdb_unpack_job_rec;
my_destroy = slurmdb_destroy_job_rec;
break;
diff --git a/src/common/slurmdbd_defs.h b/src/common/slurmdbd_defs.h
index a177d60d849..65fb720f227 100644
--- a/src/common/slurmdbd_defs.h
+++ b/src/common/slurmdbd_defs.h
@@ -156,7 +156,7 @@ typedef enum {
DBD_ADD_TRES, /* Add tres to the database */
DBD_GET_TRES, /* Get tres from the database */
DBD_GOT_TRES, /* Got tres from the database */
- DBD_FIX_LOST_JOB, /* Fix the lost jobs */
+ DBD_FIX_RUNAWAY_JOB, /* Fix any runaway jobs */
} slurmdbd_msg_type_t;
/*****************************************************************************\
diff --git a/src/plugins/accounting_storage/filetxt/accounting_storage_filetxt.c b/src/plugins/accounting_storage/filetxt/accounting_storage_filetxt.c
index 49c9478fabe..d7db21aaaaf 100644
--- a/src/plugins/accounting_storage/filetxt/accounting_storage_filetxt.c
+++ b/src/plugins/accounting_storage/filetxt/accounting_storage_filetxt.c
@@ -547,8 +547,8 @@ extern int acct_storage_p_roll_usage(void *db_conn,
return rc;
}
-extern int acct_storage_p_fix_lost_jobs(void *db_conn, uint32_t uid,
- List jobs)
+extern int acct_storage_p_fix_runaway_jobs(void *db_conn, uint32_t uid,
+ List jobs)
{
return SLURM_SUCCESS;
}
diff --git a/src/plugins/accounting_storage/mysql/Makefile.am b/src/plugins/accounting_storage/mysql/Makefile.am
index 178459e25a1..6ebd26e78eb 100644
--- a/src/plugins/accounting_storage/mysql/Makefile.am
+++ b/src/plugins/accounting_storage/mysql/Makefile.am
@@ -13,7 +13,7 @@ AS_MYSQL_SOURCES = accounting_storage_mysql.c accounting_storage_mysql.h \
as_mysql_assoc.c as_mysql_assoc.h \
as_mysql_cluster.c as_mysql_cluster.h \
as_mysql_convert.c as_mysql_convert.h \
- as_mysql_fix_lost_jobs.c as_mysql_fix_lost_jobs.h \
+ as_mysql_fix_runaway_jobs.c as_mysql_fix_runaway_jobs.h \
as_mysql_job.c as_mysql_job.h \
as_mysql_jobacct_process.c as_mysql_jobacct_process.h \
as_mysql_problems.c as_mysql_problems.h \
diff --git a/src/plugins/accounting_storage/mysql/Makefile.in b/src/plugins/accounting_storage/mysql/Makefile.in
index 98427145a35..018dad24784 100644
--- a/src/plugins/accounting_storage/mysql/Makefile.in
+++ b/src/plugins/accounting_storage/mysql/Makefile.in
@@ -183,7 +183,7 @@ am__accounting_storage_mysql_la_SOURCES_DIST = \
as_mysql_tres.h as_mysql_archive.c as_mysql_archive.h \
as_mysql_assoc.c as_mysql_assoc.h as_mysql_cluster.c \
as_mysql_cluster.h as_mysql_convert.c as_mysql_convert.h \
- as_mysql_fix_lost_jobs.c as_mysql_fix_lost_jobs.h \
+ as_mysql_fix_runaway_jobs.c as_mysql_fix_runaway_jobs.h \
as_mysql_job.c as_mysql_job.h as_mysql_jobacct_process.c \
as_mysql_jobacct_process.h as_mysql_problems.c \
as_mysql_problems.h as_mysql_qos.c as_mysql_qos.h \
@@ -200,7 +200,7 @@ am__objects_1 = \
accounting_storage_mysql_la-as_mysql_assoc.lo \
accounting_storage_mysql_la-as_mysql_cluster.lo \
accounting_storage_mysql_la-as_mysql_convert.lo \
- accounting_storage_mysql_la-as_mysql_fix_lost_jobs.lo \
+ accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.lo \
accounting_storage_mysql_la-as_mysql_job.lo \
accounting_storage_mysql_la-as_mysql_jobacct_process.lo \
accounting_storage_mysql_la-as_mysql_problems.lo \
@@ -220,7 +220,7 @@ am__EXTRA_accounting_storage_mysql_la_SOURCES_DIST = \
as_mysql_tres.h as_mysql_archive.c as_mysql_archive.h \
as_mysql_assoc.c as_mysql_assoc.h as_mysql_cluster.c \
as_mysql_cluster.h as_mysql_convert.c as_mysql_convert.h \
- as_mysql_fix_lost_jobs.c as_mysql_fix_lost_jobs.h \
+ as_mysql_fix_runaway_jobs.c as_mysql_fix_runaway_jobs.h \
as_mysql_job.c as_mysql_job.h as_mysql_jobacct_process.c \
as_mysql_jobacct_process.h as_mysql_problems.c \
as_mysql_problems.h as_mysql_qos.c as_mysql_qos.h \
@@ -572,7 +572,7 @@ AS_MYSQL_SOURCES = accounting_storage_mysql.c accounting_storage_mysql.h \
as_mysql_assoc.c as_mysql_assoc.h \
as_mysql_cluster.c as_mysql_cluster.h \
as_mysql_convert.c as_mysql_convert.h \
- as_mysql_fix_lost_jobs.c as_mysql_fix_lost_jobs.h \
+ as_mysql_fix_runaway_jobs.c as_mysql_fix_runaway_jobs.h \
as_mysql_job.c as_mysql_job.h \
as_mysql_jobacct_process.c as_mysql_jobacct_process.h \
as_mysql_problems.c as_mysql_problems.h \
@@ -680,7 +680,7 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/accounting_storage_mysql_la-as_mysql_assoc.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/accounting_storage_mysql_la-as_mysql_cluster.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/accounting_storage_mysql_la-as_mysql_convert.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/accounting_storage_mysql_la-as_mysql_fix_lost_jobs.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/accounting_storage_mysql_la-as_mysql_job.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/accounting_storage_mysql_la-as_mysql_jobacct_process.Plo@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/accounting_storage_mysql_la-as_mysql_problems.Plo@am__quote@
@@ -764,12 +764,12 @@ accounting_storage_mysql_la-as_mysql_convert.lo: as_mysql_convert.c
@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(accounting_storage_mysql_la_CFLAGS) $(CFLAGS) -c -o accounting_storage_mysql_la-as_mysql_convert.lo `test -f 'as_mysql_convert.c' || echo '$(srcdir)/'`as_mysql_convert.c
-accounting_storage_mysql_la-as_mysql_fix_lost_jobs.lo: as_mysql_fix_lost_jobs.c
-@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(accounting_storage_mysql_la_CFLAGS) $(CFLAGS) -MT accounting_storage_mysql_la-as_mysql_fix_lost_jobs.lo -MD -MP -MF $(DEPDIR)/accounting_storage_mysql_la-as_mysql_fix_lost_jobs.Tpo -c -o accounting_storage_mysql_la-as_mysql_fix_lost_jobs.lo `test -f 'as_mysql_fix_lost_jobs.c' || echo '$(srcdir)/'`as_mysql_fix_lost_jobs.c
-@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/accounting_storage_mysql_la-as_mysql_fix_lost_jobs.Tpo $(DEPDIR)/accounting_storage_mysql_la-as_mysql_fix_lost_jobs.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='as_mysql_fix_lost_jobs.c' object='accounting_storage_mysql_la-as_mysql_fix_lost_jobs.lo' libtool=yes @AMDEPBACKSLASH@
+accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.lo: as_mysql_fix_runaway_jobs.c
+@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(accounting_storage_mysql_la_CFLAGS) $(CFLAGS) -MT accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.lo -MD -MP -MF $(DEPDIR)/accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.Tpo -c -o accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.lo `test -f 'as_mysql_fix_runaway_jobs.c' || echo '$(srcdir)/'`as_mysql_fix_runaway_jobs.c
+@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) $(DEPDIR)/accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.Tpo $(DEPDIR)/accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='as_mysql_fix_runaway_jobs.c' object='accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.lo' libtool=yes @AMDEPBACKSLASH@
@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(accounting_storage_mysql_la_CFLAGS) $(CFLAGS) -c -o accounting_storage_mysql_la-as_mysql_fix_lost_jobs.lo `test -f 'as_mysql_fix_lost_jobs.c' || echo '$(srcdir)/'`as_mysql_fix_lost_jobs.c
+@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(accounting_storage_mysql_la_CFLAGS) $(CFLAGS) -c -o accounting_storage_mysql_la-as_mysql_fix_runaway_jobs.lo `test -f 'as_mysql_fix_runaway_jobs.c' || echo '$(srcdir)/'`as_mysql_fix_runaway_jobs.c
accounting_storage_mysql_la-as_mysql_job.lo: as_mysql_job.c
@am__fastdepCC_TRUE@ $(AM_V_CC)$(LIBTOOL) $(AM_V_lt) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(accounting_storage_mysql_la_CFLAGS) $(CFLAGS) -MT accounting_storage_mysql_la-as_mysql_job.lo -MD -MP -MF $(DEPDIR)/accounting_storage_mysql_la-as_mysql_job.Tpo -c -o accounting_storage_mysql_la-as_mysql_job.lo `test -f 'as_mysql_job.c' || echo '$(srcdir)/'`as_mysql_job.c
diff --git a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
index 4c6774cde92..e039eb98b72 100644
--- a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
+++ b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
@@ -51,7 +51,7 @@
#include "as_mysql_assoc.h"
#include "as_mysql_cluster.h"
#include "as_mysql_convert.h"
-#include "as_mysql_fix_lost_jobs.h"
+#include "as_mysql_fix_runaway_jobs.h"
#include "as_mysql_job.h"
#include "as_mysql_jobacct_process.h"
#include "as_mysql_problems.h"
@@ -2875,10 +2875,10 @@ extern int acct_storage_p_roll_usage(mysql_conn_t *mysql_conn,
sent_end, archive_data);
}
-extern int acct_storage_p_fix_lost_jobs(void *db_conn, uint32_t uid,
+extern int acct_storage_p_fix_runaway_jobs(void *db_conn, uint32_t uid,
List jobs)
{
- return as_mysql_fix_lost_jobs(db_conn, uid, jobs);
+ return as_mysql_fix_runaway_jobs(db_conn, uid, jobs);
}
extern int clusteracct_storage_p_node_down(mysql_conn_t *mysql_conn,
diff --git a/src/plugins/accounting_storage/mysql/as_mysql_fix_lost_jobs.c b/src/plugins/accounting_storage/mysql/as_mysql_fix_runaway_jobs.c
similarity index 91%
rename from src/plugins/accounting_storage/mysql/as_mysql_fix_lost_jobs.c
rename to src/plugins/accounting_storage/mysql/as_mysql_fix_runaway_jobs.c
index b07dfc56fab..6e06aed3a31 100644
--- a/src/plugins/accounting_storage/mysql/as_mysql_fix_lost_jobs.c
+++ b/src/plugins/accounting_storage/mysql/as_mysql_fix_runaway_jobs.c
@@ -1,5 +1,5 @@
/*****************************************************************************\
- * as_mysql_fix_lost_jobs.c - functions dealing with lost jobs.
+ * as_mysql_fix_runaway_jobs.c - functions dealing with runaway jobs.
*****************************************************************************
* Copyright (C) 2016 SchedMD LLC.
* Written by Nathan Yee <nyee32@schedmd.com>
@@ -34,7 +34,7 @@
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
\*****************************************************************************/
-#include "as_mysql_fix_lost_jobs.h"
+#include "as_mysql_fix_runaway_jobs.h"
#include "src/common/list.h"
static int _job_sort_by_start_time(void *void1, void * void2)
@@ -87,8 +87,8 @@ static int _first_job_roll_up(mysql_conn_t *mysql_conn, time_t first_start)
return rc;
}
-extern int as_mysql_fix_lost_jobs(mysql_conn_t *mysql_conn, uint32_t uid,
- List lost_jobs)
+extern int as_mysql_fix_runaway_jobs(mysql_conn_t *mysql_conn, uint32_t uid,
+ List runaway_jobs)
{
char *query = NULL, *job_ids = NULL;
slurmdb_job_rec_t *job = NULL;
@@ -96,8 +96,8 @@ extern int as_mysql_fix_lost_jobs(mysql_conn_t *mysql_conn, uint32_t uid,
int rc = SLURM_SUCCESS;
slurmdb_job_rec_t *first_job;
- list_sort(lost_jobs, _job_sort_by_start_time);
- first_job = list_peek(lost_jobs);
+ list_sort(runaway_jobs, _job_sort_by_start_time);
+ first_job = list_peek(runaway_jobs);
if (check_connection(mysql_conn) != SLURM_SUCCESS)
return ESLURM_DB_CONNECTION;
@@ -110,12 +110,12 @@ extern int as_mysql_fix_lost_jobs(mysql_conn_t *mysql_conn, uint32_t uid,
if (!is_user_any_coord(mysql_conn, &user)) {
error("Only admins/operators/coordinators "
- "can fix lost jobs");
+ "can fix runaway jobs");
return ESLURM_ACCESS_DENIED;
}
}
- iter = list_iterator_create(lost_jobs);
+ iter = list_iterator_create(runaway_jobs);
while ((job = list_next(iter))) {
xstrfmtcat(job_ids, "%s%d", ((job_ids) ? "," : ""), job->jobid);
}
@@ -132,10 +132,10 @@ extern int as_mysql_fix_lost_jobs(mysql_conn_t *mysql_conn, uint32_t uid,
xfree(job_ids);
/* Set rollup to the the last day of the previous month of the first
- * lost job */
+ * runaway job */
rc = _first_job_roll_up(mysql_conn, first_job->start);
if (rc != SLURM_SUCCESS) {
- error("Failed to fix lost jobs");
+ error("Failed to fix runaway jobs");
return SLURM_ERROR;
}
diff --git a/src/plugins/accounting_storage/mysql/as_mysql_fix_lost_jobs.h b/src/plugins/accounting_storage/mysql/as_mysql_fix_runaway_jobs.h
similarity index 88%
rename from src/plugins/accounting_storage/mysql/as_mysql_fix_lost_jobs.h
rename to src/plugins/accounting_storage/mysql/as_mysql_fix_runaway_jobs.h
index fba0103a3b5..43cc665b60a 100644
--- a/src/plugins/accounting_storage/mysql/as_mysql_fix_lost_jobs.h
+++ b/src/plugins/accounting_storage/mysql/as_mysql_fix_runaway_jobs.h
@@ -1,5 +1,5 @@
/*****************************************************************************\
- * as_mysql_fix_lost_jobs.h - functions dealing with lost jobs.
+ * as_mysql_fix_runaway_jobs.h - functions dealing with runaway jobs.
*****************************************************************************
* Copyright (C) 2016 SchedMD LLC.
* Written by Nathan Yee <nyee32@schedmd.com>
@@ -34,13 +34,13 @@
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
\*****************************************************************************/
-#ifndef _HAVE_MYSQL_FIX_LOST_JOBS_H
-#define _HAVE_MYSQL_FIX_LOST_JOBS_H
+#ifndef _HAVE_MYSQL_FIX_RUNAWAY_JOBS_H
+#define _HAVE_MYSQL_FIX_RUNAWAY_JOBS_H
#include "accounting_storage_mysql.h"
#include "src/common/slurm_time.h"
-extern int as_mysql_fix_lost_jobs(mysql_conn_t *mysql_conn, uint32_t uid,
- List lost_jobs);
+extern int as_mysql_fix_runaway_jobs(mysql_conn_t *mysql_conn, uint32_t uid,
+ List runaway_jobs);
#endif
diff --git a/src/plugins/accounting_storage/none/accounting_storage_none.c b/src/plugins/accounting_storage/none/accounting_storage_none.c
index 2bd1b7b8cac..b5f3f7952e0 100644
--- a/src/plugins/accounting_storage/none/accounting_storage_none.c
+++ b/src/plugins/accounting_storage/none/accounting_storage_none.c
@@ -371,8 +371,8 @@ extern int acct_storage_p_roll_usage(void *db_conn,
return rc;
}
-extern int acct_storage_p_fix_lost_jobs(void *db_conn, uint32_t uid,
- List jobs)
+extern int acct_storage_p_fix_runaway_jobs(void *db_conn, uint32_t uid,
+ List jobs)
{
return SLURM_SUCCESS;
}
diff --git a/src/plugins/accounting_storage/slurmdbd/accounting_storage_slurmdbd.c b/src/plugins/accounting_storage/slurmdbd/accounting_storage_slurmdbd.c
index fdad804484f..2d40dec10e3 100644
--- a/src/plugins/accounting_storage/slurmdbd/accounting_storage_slurmdbd.c
+++ b/src/plugins/accounting_storage/slurmdbd/accounting_storage_slurmdbd.c
@@ -2244,8 +2244,8 @@ extern int acct_storage_p_roll_usage(void *db_conn,
return rc;
}
-extern int acct_storage_p_fix_lost_jobs(void *db_conn, uint32_t uid,
- List jobs)
+extern int acct_storage_p_fix_runaway_jobs(void *db_conn, uint32_t uid,
+ List jobs)
{
slurmdbd_msg_t req;
dbd_list_msg_t get_msg;
@@ -2254,7 +2254,7 @@ extern int acct_storage_p_fix_lost_jobs(void *db_conn, uint32_t uid,
memset(&get_msg, 0, sizeof(dbd_list_msg_t));
get_msg.my_list = jobs;
- req.msg_type = DBD_FIX_LOST_JOB;
+ req.msg_type = DBD_FIX_RUNAWAY_JOB;
req.data = &get_msg;
rc = slurm_send_slurmdbd_recv_rc_msg(SLURM_PROTOCOL_VERSION,
diff --git a/src/sacctmgr/Makefile.am b/src/sacctmgr/Makefile.am
index 564f28ce333..80754dc36f5 100644
--- a/src/sacctmgr/Makefile.am
+++ b/src/sacctmgr/Makefile.am
@@ -19,7 +19,7 @@ sacctmgr_SOURCES = \
common.c \
event_functions.c \
file_functions.c \
- lost_jobs_functions.c \
+ runaway_job_functions.c \
job_functions.c \
reservation_functions.c \
resource_functions.c \
diff --git a/src/sacctmgr/Makefile.in b/src/sacctmgr/Makefile.in
index 7647f4d27a1..04a010e2592 100644
--- a/src/sacctmgr/Makefile.in
+++ b/src/sacctmgr/Makefile.in
@@ -151,7 +151,7 @@ am_sacctmgr_OBJECTS = account_functions.$(OBJEXT) \
archive_functions.$(OBJEXT) association_functions.$(OBJEXT) \
config_functions.$(OBJEXT) cluster_functions.$(OBJEXT) \
common.$(OBJEXT) event_functions.$(OBJEXT) \
- file_functions.$(OBJEXT) lost_jobs_functions.$(OBJEXT) \
+ file_functions.$(OBJEXT) runaway_job_functions.$(OBJEXT) \
job_functions.$(OBJEXT) reservation_functions.$(OBJEXT) \
resource_functions.$(OBJEXT) sacctmgr.$(OBJEXT) \
qos_functions.$(OBJEXT) txn_functions.$(OBJEXT) \
@@ -502,7 +502,7 @@ sacctmgr_SOURCES = \
common.c \
event_functions.c \
file_functions.c \
- lost_jobs_functions.c \
+ runaway_job_functions.c\
job_functions.c \
reservation_functions.c \
resource_functions.c \
@@ -617,7 +617,7 @@ distclean-compile:
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/event_functions.Po@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/file_functions.Po@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/job_functions.Po@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/lost_jobs_functions.Po@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/runaway_job_functions.Po@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/problem_functions.Po@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/qos_functions.Po@am__quote@
@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/reservation_functions.Po@am__quote@
diff --git a/src/sacctmgr/lost_jobs_functions.c b/src/sacctmgr/runaway_job_functions.c
similarity index 84%
rename from src/sacctmgr/lost_jobs_functions.c
rename to src/sacctmgr/runaway_job_functions.c
index c3ab13da32a..f680077b194 100644
--- a/src/sacctmgr/lost_jobs_functions.c
+++ b/src/sacctmgr/runaway_job_functions.c
@@ -1,5 +1,5 @@
/*****************************************************************************\
- * lost_jobs_functions.c - functions dealing with lost jobs
+ * runaway_jobs_functions.c - functions dealing with runaway/orphan jobs
*****************************************************************************
* Copyright (C) 2016 SchedMD LLC.
* Written by Nathan Yee <nyee32@schedmd.com>
@@ -49,7 +49,7 @@ static int _job_sort_by_start_time(void *void1, void * void2)
return 0;
}
-static void _print_lost_jobs(List jobs)
+static void _print_runaway_jobs(List jobs)
{
char outbuf[FORMAT_STRING_SIZE];
slurmdb_job_rec_t *job = NULL;
@@ -131,17 +131,17 @@ static void _print_lost_jobs(List jobs)
list_iterator_destroy(itr);
}
-static List _get_lost_jobs(char *cluster)
+static List _get_runaway_jobs(char *cluster)
{
int i = 0;
- bool job_lost = true;
+ bool job_runaway = true;
List db_jobs_list = NULL;
ListIterator db_jobs_itr = NULL;
job_info_t *clus_job = NULL;
job_info_msg_t *clus_jobs = NULL;
slurmdb_job_rec_t *db_job = NULL;
slurmdb_job_cond_t *job_cond = xmalloc(sizeof(slurmdb_job_cond_t));
- List lost_jobs = NULL;
+ List runaway_jobs = NULL;
job_cond->without_steps = 1;
job_cond->without_usage_truncation = 1;
@@ -158,62 +158,63 @@ static List _get_lost_jobs(char *cluster)
return NULL;
}
- lost_jobs = list_create(NULL);
+ runaway_jobs = list_create(NULL);
db_jobs_itr = list_iterator_create(db_jobs_list);
while ((db_job = list_next(db_jobs_itr))) {
- job_lost = true;
+ job_runaway = true;
for (i = 0, clus_job = clus_jobs->job_array;
i < clus_jobs->record_count; i++, clus_job++) {
if (db_job->jobid == clus_job->job_id) {
- job_lost = false;
+ job_runaway = false;
break;
}
}
- if (job_lost)
- list_append(lost_jobs, db_job);
+ if (job_runaway)
+ list_append(runaway_jobs, db_job);
}
list_iterator_destroy(db_jobs_itr);
- return lost_jobs;
+ return runaway_jobs;
}
-static void _report_lost_jobs(List lost_jobs)
+static void _report_runaway_jobs(List runaway_jobs)
{
- if (list_count(lost_jobs)) {
- printf("NOTE: Lost jobs are jobs that don't exist in the "
+ if (list_count(runaway_jobs)) {
+ printf("NOTE: Runaway jobs are jobs that don't exist in the "
"controller but are still considered running in the "
"datbase\n");
- _print_lost_jobs(lost_jobs);
+ _print_runaway_jobs(runaway_jobs);
}
}
/*
- * List and ask user if they wish to fix the lost jobs
+ * List and ask user if they wish to fix the runaway jobs
*/
-extern int sacctmgr_list_lost_jobs(int argc, char *argv[])
+extern int sacctmgr_list_runaway_jobs(int argc, char *argv[])
{
- List lost_jobs = NULL;
+ List runaway_jobs = NULL;
int rc = SLURM_SUCCESS;
uint32_t my_uid = getuid();
char *cluster = NULL;
- char *ask_msg = "\nWould you like to fix these lost jobs?\n"
+ char *ask_msg = "\nWould you like to fix these runaway jobs?\n"
"(This will set the end times to start times and "
"states to completed for these jobs and will trigger "
"the rollup to reroll usage from before the oldest "
- "lost job.)\n\n";
+ "runaway job.)\n\n";
if (!(cluster = slurm_get_cluster_name()))
return SLURM_ERROR;
- if (!(lost_jobs = _get_lost_jobs(cluster)))
+ if (!(runaway_jobs = _get_runaway_jobs(cluster)))
return SLURM_ERROR;
xfree(cluster);
- _report_lost_jobs(lost_jobs);
+ _report_runaway_jobs(runaway_jobs);
- if (list_count(lost_jobs)) {
- rc = acct_storage_g_fix_lost_jobs(db_conn, my_uid, lost_jobs);
+ if (list_count(runaway_jobs)) {
+ rc = acct_storage_g_fix_runaway_jobs(
+ db_conn, my_uid, runaway_jobs);
if (rc == SLURM_SUCCESS) {
if (commit_check(ask_msg)) {
@@ -224,12 +225,12 @@ extern int sacctmgr_list_lost_jobs(int argc, char *argv[])
}
}
else {
- error("Failed to fix lost job: %s\n",
+ error("Failed to fix runaway job: %s\n",
slurm_strerror(rc));
}
} else {
- printf("Lost Jobs: No lost jobs found\n");
+ printf("Runaway Jobs: No runaway jobs found\n");
}
return rc;
diff --git a/src/sacctmgr/sacctmgr.c b/src/sacctmgr/sacctmgr.c
index dd28ece3424..a8d21564aaf 100644
--- a/src/sacctmgr/sacctmgr.c
+++ b/src/sacctmgr/sacctmgr.c
@@ -651,8 +651,10 @@ static void _show_it (int argc, char *argv[])
} else if (strncasecmp(argv[0], "Problems",
MAX(command_len, 1)) == 0) {
error_code = sacctmgr_list_problem((argc - 1), &argv[1]);
- } else if (strncasecmp(argv[0], "LostJobs", MAX(command_len, 1)) == 0) {
- error_code = sacctmgr_list_lost_jobs((argc - 1), &argv[1]);
+ } else if (!strncasecmp(argv[0], "RunawayJobs", MAX(command_len, 2)) ||
+ !strncasecmp(argv[0], "OrphanJobs", MAX(command_len, 1)) ||
+ !strncasecmp(argv[0], "LostJobs", MAX(command_len, 1))) {
+ error_code = sacctmgr_list_runaway_jobs((argc - 1), &argv[1]);
} else if (strncasecmp(argv[0], "QOS", MAX(command_len, 1)) == 0) {
error_code = sacctmgr_list_qos((argc - 1), &argv[1]);
} else if (!strncasecmp(argv[0], "Resource", MAX(command_len, 4))) {
@@ -677,7 +679,8 @@ static void _show_it (int argc, char *argv[])
fprintf(stderr, "\"Account\", \"Association\", "
"\"Cluster\", \"Configuration\",\n\"Event\", "
"\"Problem\", \"QOS\", \"Resource\", \"Reservation\", "
- "\"Transaction\", \"TRES\", \"User\", or \"WCKey\"\n");
+ "\"RunAwayJobs\", \"Transaction\", \"TRES\", "
+ "\"User\", or \"WCKey\"\n");
}
if (error_code != SLURM_SUCCESS) {
@@ -855,8 +858,8 @@ sacctmgr [<OPTION>] [<COMMAND>] \n\
<ENTITY> may be \"account\", \"association\", \"cluster\", \n\
\"configuration\", \"coordinator\", \"event\", \"job\", \n\
\"problem\", \"qos\", \"resource\", \"reservation\", \n\
- \"transaction\", \"tres\", \n\
- \"user\" or \"wckey\" \n\
+ \"runawayjobs\", \"transaction\", \"tres\", \n\
+ \"user\" or \"wckey\" \n\
\n\
<SPECS> are different for each command entity pair. \n\
list account - Clusters=, Descriptions=, Format=, \n\
@@ -944,6 +947,8 @@ sacctmgr [<OPTION>] [<COMMAND>] \n\
\n\
list reservation - Clusters=, End=, ID=, Names=, Nodes=, Start= \n\
\n\
+ list runawayjobs \n\
+ \n\
list transactions - Accounts=, Action=, Actor=, Clusters=, End=, \n\
Format=, ID=, Start=, User=, and WithAssoc \n\
\n\
diff --git a/src/sacctmgr/sacctmgr.h b/src/sacctmgr/sacctmgr.h
index c01da83e733..ab4e2c80bca 100644
--- a/src/sacctmgr/sacctmgr.h
+++ b/src/sacctmgr/sacctmgr.h
@@ -358,7 +358,7 @@ extern void load_sacctmgr_cfg_file (int argc, char *argv[]);
/* txn_functions.c */
extern int sacctmgr_list_txn(int argc, char *argv[]);
-/* lost_jobs_functions.c */
-extern int sacctmgr_list_lost_jobs(int argc, char *argv[]);
+/* runaway_jobs_functions.c */
+extern int sacctmgr_list_runaway_jobs(int argc, char *argv[]);
#endif
diff --git a/src/slurmdbd/proc_req.c b/src/slurmdbd/proc_req.c
index bfa7d020eed..535c309da31 100644
--- a/src/slurmdbd/proc_req.c
+++ b/src/slurmdbd/proc_req.c
@@ -180,8 +180,8 @@ static int _step_complete(slurmdbd_conn_t *slurmdbd_conn,
Buf in_buffer, Buf *out_buffer, uint32_t *uid);
static int _step_start(slurmdbd_conn_t *slurmdbd_conn,
Buf in_buffer, Buf *out_buffer, uint32_t *uid);
-static int _fix_lost_jobs(slurmdbd_conn_t *slurmdbd_conn, Buf in_buffer,
- Buf *out_buffer, uint32_t *uid);
+static int _fix_runaway_jobs(slurmdbd_conn_t *slurmdbd_conn, Buf in_buffer,
+ Buf *out_buffer, uint32_t *uid);
/* Process an incoming RPC
* slurmdbd_conn IN/OUT - in will that the newsockfd set before
@@ -466,9 +466,9 @@ proc_req(slurmdbd_conn_t *slurmdbd_conn,
rc = _step_start(slurmdbd_conn,
in_buffer, out_buffer, uid);
break;
- case DBD_FIX_LOST_JOB:
- rc = _fix_lost_jobs(slurmdbd_conn,
- in_buffer, out_buffer, uid);
+ case DBD_FIX_RUNAWAY_JOB:
+ rc = _fix_runaway_jobs(slurmdbd_conn,
+ in_buffer, out_buffer, uid);
break;
default:
comment = "Invalid RPC";
@@ -569,29 +569,29 @@ end_it:
return rc;
}
-static int _fix_lost_jobs(slurmdbd_conn_t *slurmdbd_conn, Buf in_buffer,
- Buf *out_buffer, uint32_t *uid)
+static int _fix_runaway_jobs(slurmdbd_conn_t *slurmdbd_conn, Buf in_buffer,
+ Buf *out_buffer, uint32_t *uid)
{
int rc = SLURM_SUCCESS;
dbd_list_msg_t *get_msg = NULL;
char *comment = NULL;
if (slurmdbd_unpack_list_msg(&get_msg, slurmdbd_conn->rpc_version,
- DBD_FIX_LOST_JOB, in_buffer) !=
+ DBD_FIX_RUNAWAY_JOB, in_buffer) !=
SLURM_SUCCESS) {
- comment = "Failed to unpack DBD_LOST_JOBS message";
+ comment = "Failed to unpack DBD_RUNAWAY_JOBS message";
error("CONN:%u %s", slurmdbd_conn->newsockfd, comment);
rc = SLURM_ERROR;
goto end_it;
}
- rc = acct_storage_g_fix_lost_jobs(slurmdbd_conn->db_conn, *uid,
- get_msg->my_list);
+ rc = acct_storage_g_fix_runaway_jobs(slurmdbd_conn->db_conn, *uid,
+ get_msg->my_list);
end_it:
slurmdbd_free_list_msg(get_msg);
*out_buffer = make_dbd_rc_msg(slurmdbd_conn->rpc_version,
- rc, comment, DBD_FIX_LOST_JOB);
+ rc, comment, DBD_FIX_RUNAWAY_JOB);
return rc;
}
--
GitLab