From 5d4db9cdcbbd7ff82158d8c57b2605b6c4ba0e71 Mon Sep 17 00:00:00 2001
From: Brian Christiansen <brian@schedmd.com>
Date: Tue, 4 Apr 2017 23:27:39 -0600
Subject: [PATCH] Add scancel --sibling=<sib_name>
to remove active sibling from federated job.
---
doc/man/man1/scancel.1 | 4 ++++
src/scancel/opt.c | 16 +++++++++++----
src/scancel/scancel.c | 1 +
src/scancel/scancel.h | 1 +
src/slurmctld/fed_mgr.c | 42 ++++++++++++++++++++++++++++++++++++++++
src/slurmctld/fed_mgr.h | 1 +
src/slurmctld/proc_req.c | 15 +++++++++-----
7 files changed, 71 insertions(+), 9 deletions(-)
diff --git a/doc/man/man1/scancel.1 b/doc/man/man1/scancel.1
index d54d51f870f..6bb80179c43 100644
--- a/doc/man/man1/scancel.1
+++ b/doc/man/man1/scancel.1
@@ -90,6 +90,10 @@ This option is incompatible with the \fB\-\-verbose\fR option.
\fB\-R\fR, \fB\-\-reservation\fR=\fIreservation_name\fR
Restrict the scancel operation to jobs with this reservation name.
+.TP
+\fB\-\-sibling\fR=\fIcluster_name\fR
+Remove an active sibling job from a federated job.
+
.TP
\fB\-s\fR, \fB\-\-signal\fR=\fIsignal_name\fR
The name or number of the signal to send. If this option is not used
diff --git a/src/scancel/opt.c b/src/scancel/opt.c
index 3de55ddc460..bece7b8b00b 100644
--- a/src/scancel/opt.c
+++ b/src/scancel/opt.c
@@ -61,10 +61,11 @@
#include "src/scancel/scancel.h"
/* getopt_long options, integers but not characters */
-#define OPT_LONG_HELP 0x100
-#define OPT_LONG_USAGE 0x101
-#define OPT_LONG_CTLD 0x102
-#define OPT_LONG_WCKEY 0x103
+#define OPT_LONG_HELP 0x100
+#define OPT_LONG_USAGE 0x101
+#define OPT_LONG_CTLD 0x102
+#define OPT_LONG_WCKEY 0x103
+#define OPT_LONG_SIBLING 0x104
#define SIZE(a) (sizeof(a)/sizeof(a[0]))
@@ -255,6 +256,7 @@ static void _opt_default(void)
opt.partition = NULL;
opt.qos = NULL;
opt.reservation = NULL;
+ opt.sibling = NULL;
opt.signal = (uint16_t) NO_VAL;
opt.state = JOB_END;
opt.user_id = 0;
@@ -385,6 +387,7 @@ static void _opt_args(int argc, char **argv)
{"qos", required_argument, 0, 'q'},
{"quiet", no_argument, 0, 'Q'},
{"reservation", required_argument, 0, 'R'},
+ {"sibling", required_argument, 0, OPT_LONG_SIBLING},
{"signal", required_argument, 0, 's'},
{"state", required_argument, 0, 't'},
{"usage", no_argument, 0, OPT_LONG_USAGE},
@@ -465,6 +468,9 @@ static void _opt_args(int argc, char **argv)
case (int)'w':
opt.nodelist = xstrdup(optarg);
break;
+ case OPT_LONG_SIBLING:
+ opt.sibling = xstrdup(optarg);
+ break;
case OPT_LONG_WCKEY:
opt.wckey = xstrdup(optarg);
break;
@@ -652,6 +658,7 @@ static void _opt_list(void)
info("partition : %s", opt.partition);
info("qos : %s", opt.qos);
info("reservation : %s", opt.reservation);
+ info("sibling : %s", opt.sibling);
if (opt.signal != (uint16_t) NO_VAL)
info("signal : %u", opt.signal);
info("state : %s", job_state_string(opt.state));
@@ -713,6 +720,7 @@ static void _help(void)
printf(" -Q, --quiet disable warnings\n");
printf(" -q, --qos=qos act only on jobs with this quality of service\n");
printf(" -R, --reservation=reservation act only on jobs with this reservation\n");
+ printf(" --sibling=cluster_name remove an active sibling job from a federated job\n");
printf(" -s, --signal=name | integer signal to send to job, default is SIGKILL\n");
printf(" -t, --state=states act only on jobs in this state. Valid job\n");
printf(" states are PENDING, RUNNING and SUSPENDED\n");
diff --git a/src/scancel/scancel.c b/src/scancel/scancel.c
index 5e0125378f0..7113849fcdd 100644
--- a/src/scancel/scancel.c
+++ b/src/scancel/scancel.c
@@ -755,6 +755,7 @@ _cancel_job_id (void *ci)
kill_msg.flags = flags;
kill_msg.job_id = NO_VAL;
kill_msg.job_step_id = NO_VAL;
+ kill_msg.sibling = opt.sibling;
kill_msg.signal = cancel_info->sig;
kill_msg.sjob_id = cancel_info->job_id_str;
diff --git a/src/scancel/scancel.h b/src/scancel/scancel.h
index 2aa03ce79bd..7e4a8a3dd5b 100644
--- a/src/scancel/scancel.h
+++ b/src/scancel/scancel.h
@@ -46,6 +46,7 @@
typedef struct scancel_options {
char *account; /* --account=n, -a */
bool batch; /* --batch, -b */
+ char *sibling; /* --sibling=<sib_name> */
bool ctld; /* --ctld */
List clusters; /* --cluster=cluster_name -Mcluster-name */
bool full; /* --full, -f */
diff --git a/src/slurmctld/fed_mgr.c b/src/slurmctld/fed_mgr.c
index 3732085530f..5a02a624313 100644
--- a/src/slurmctld/fed_mgr.c
+++ b/src/slurmctld/fed_mgr.c
@@ -3828,3 +3828,45 @@ end_it:
return SLURM_SUCCESS;
}
+/*
+ * Remove active sibling from the given job.
+ *
+ * IN job_id - job_id of job to remove active sibling from.
+ * IN sib_name - name of sibling job to remove from active siblings.
+ * RET SLURM_SUCCESS on sucess, error code on error.
+ */
+extern int fed_mgr_remove_active_sibling(uint32_t job_id, char *sib_name)
+{
+ uint32_t origin_id;
+ struct job_record *job_ptr = NULL;
+ slurmdb_cluster_rec_t *sibling;
+
+ if (!(job_ptr = find_job_record(job_id)))
+ return ESLURM_INVALID_JOB_ID;
+
+ if (!_is_fed_job(job_ptr, &origin_id))
+ return ESLURM_JOB_NOT_FEDERATED;
+
+ if (job_ptr->fed_details->cluster_lock)
+ return ESLURM_JOB_NOT_PENDING;
+
+ if (!(sibling = fed_mgr_get_cluster_by_name(sib_name)))
+ return ESLURM_INVALID_CLUSTER_NAME;
+
+ if (job_ptr->fed_details->siblings_active &
+ FED_SIBLING_BIT(sibling->fed.id)) {
+ if (fed_mgr_cluster_rec == sibling)
+ fed_mgr_job_revoke(job_ptr, false, 0, 0);
+ else
+ _revoke_sibling_jobs(job_ptr->job_id,
+ fed_mgr_cluster_rec->fed.id,
+ FED_SIBLING_BIT(sibling->fed.id),
+ 0);
+ job_ptr->fed_details->siblings_active &=
+ ~(FED_SIBLING_BIT(sibling->fed.id));
+ update_job_fed_details(job_ptr);
+ }
+
+ return SLURM_SUCCESS;
+}
+
diff --git a/src/slurmctld/fed_mgr.h b/src/slurmctld/fed_mgr.h
index 9a5fa2b8da3..a9b22222ab3 100644
--- a/src/slurmctld/fed_mgr.h
+++ b/src/slurmctld/fed_mgr.h
@@ -78,6 +78,7 @@ extern int fed_mgr_job_revoke(struct job_record *job_ptr,
extern int fed_mgr_job_revoke_sibs(struct job_record *job_ptr);
extern int fed_mgr_job_start(struct job_record *job_ptr,
uint32_t cluster_id, time_t start_time);
+extern int fed_mgr_remove_active_sibling(uint32_t job_id, char *sib_name);
extern int fed_mgr_sib_will_run(slurm_msg_t *msg,
job_desc_msg_t *job_desc, uid_t uid,
will_run_response_msg_t **resp);
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 771dc521434..7cad5d329f3 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -5582,11 +5582,16 @@ _slurm_rpc_kill_job2(slurm_msg_t *msg)
_throttle_start(&active_rpc_cnt);
lock_slurmctld(lock);
- cc = job_str_signal(kill->sjob_id,
- kill->signal,
- kill->flags,
- uid,
- 0);
+ if (kill->sibling) {
+ uint32_t job_id = strtol(kill->sjob_id, NULL, 10);
+ cc = fed_mgr_remove_active_sibling(job_id, kill->sibling);
+ } else {
+ cc = job_str_signal(kill->sjob_id,
+ kill->signal,
+ kill->flags,
+ uid,
+ 0);
+ }
unlock_slurmctld(lock);
_throttle_fini(&active_rpc_cnt);
--
GitLab