From 25227d02af5919c7f90ee626ceff2c70d2fd013d Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Mon, 20 Dec 2004 17:28:45 +0000
Subject: [PATCH] Fix how job kill RPC is handled, include node_select data if
available, don't seg fault if data is missing, free the data when the agent
is finished with it.
---
src/common/node_select.c | 18 ++++++++++--------
src/common/slurm_protocol_defs.h | 3 ++-
src/slurmctld/agent.c | 3 +++
src/slurmctld/job_mgr.c | 19 +++++++++++++------
src/slurmctld/node_mgr.c | 8 ++++----
src/slurmctld/slurmctld.h | 2 ++
6 files changed, 34 insertions(+), 19 deletions(-)
diff --git a/src/common/node_select.c b/src/common/node_select.c
index 24380651d7e..92267a6c65f 100644
--- a/src/common/node_select.c
+++ b/src/common/node_select.c
@@ -10,7 +10,7 @@
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
- * Written by Morris Jette <jette@llnl.gov>.
+ * Written by Morris Jette <jette1@llnl.gov>.
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -409,7 +409,7 @@ extern int select_g_set_jobinfo (select_jobinfo_t jobinfo,
char * tmp_char = (char *) data;
if (jobinfo->magic != JOBINFO_MAGIC) {
- error("select_p_set_jobinfo: jobinfo magic bad");
+ error("select_g_set_jobinfo: jobinfo magic bad");
return SLURM_ERROR;
}
@@ -452,7 +452,7 @@ extern int select_g_get_jobinfo (select_jobinfo_t jobinfo,
char **tmp_char = (char **) data;
if (jobinfo->magic != JOBINFO_MAGIC) {
- error("select_p_set_jobinfo: jobinfo magic bad");
+ error("select_g_get_jobinfo: jobinfo magic bad");
return SLURM_ERROR;
}
@@ -494,8 +494,10 @@ extern select_jobinfo_t select_g_copy_jobinfo(select_jobinfo_t jobinfo)
{
struct select_jobinfo *rc = NULL;
- if (jobinfo->magic != JOBINFO_MAGIC)
- error("select_p_copy_jobinfo: jobinfo magic bad");
+ if (jobinfo == NULL)
+ ;
+ else if (jobinfo->magic != JOBINFO_MAGIC)
+ error("select_g_copy_jobinfo: jobinfo magic bad");
else {
int i;
rc = xmalloc(sizeof(struct select_jobinfo));
@@ -523,7 +525,7 @@ extern int select_g_free_jobinfo (select_jobinfo_t *jobinfo)
if (*jobinfo == NULL) /* never set, treat as not an error */
;
else if ((*jobinfo)->magic != JOBINFO_MAGIC) {
- error("select_p_set_jobinfo: jobinfo magic bad");
+ error("select_g_free_jobinfo: jobinfo magic bad");
rc = EINVAL;
} else {
(*jobinfo)->magic = 0;
@@ -595,13 +597,13 @@ extern char *select_g_sprint_jobinfo(select_jobinfo_t jobinfo,
int i;
if (buf == NULL) {
- error("select_p_sprint_jobinfo: buf is null");
+ error("select_g_sprint_jobinfo: buf is null");
return NULL;
}
if ((mode != SELECT_PRINT_DATA)
&& jobinfo && (jobinfo->magic != JOBINFO_MAGIC)) {
- error("select_p_copy_jobinfo: jobinfo magic bad");
+ error("select_g_sprint_jobinfo: jobinfo magic bad");
return NULL;
}
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index 480ceb211cd..7cae18d1a95 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -319,7 +319,8 @@ typedef struct return_code_msg {
/* Note: We include select_jobinfo here in addition to the job launch
* RPC in order to insure reliable clean-up of a BlueGene partition in
* the event of some launch failure or race condition preventing slurmd
- * from getting the BGL_PARTITION_ID at that time */
+ * from getting the BGL_PARTITION_ID at that time. It is needed for
+ * the job epilog. */
typedef struct kill_job_msg {
uint32_t job_id;
uint32_t job_uid;
diff --git a/src/slurmctld/agent.c b/src/slurmctld/agent.c
index 7e34c9605e8..9de112fe525 100644
--- a/src/slurmctld/agent.c
+++ b/src/slurmctld/agent.c
@@ -935,6 +935,9 @@ static void _purge_agent_args(agent_arg_t *agent_arg_ptr)
RESPONSE_RESOURCE_ALLOCATION)
slurm_free_resource_allocation_response_msg(
agent_arg_ptr->msg_args);
+ else if (agent_arg_ptr->msg_type ==
+ REQUEST_KILL_JOB)
+ slurm_free_kill_job_msg(agent_arg_ptr->msg_args);
else
xfree(agent_arg_ptr->msg_args);
}
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 8c6a74885ce..1ca40dd9573 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -3177,7 +3177,7 @@ validate_jobs_on_node(char *node_name, uint32_t * job_count,
if (job_ptr == NULL) {
error("Orphan job %u.%u reported on node %s",
job_id_ptr[i], step_id_ptr[i], node_name);
- kill_job_on_node(job_id_ptr[i], node_ptr);
+ kill_job_on_node(job_id_ptr[i], job_ptr, node_ptr);
}
else if (job_ptr->job_state == JOB_RUNNING) {
@@ -3196,14 +3196,15 @@ validate_jobs_on_node(char *node_name, uint32_t * job_count,
error
("Registered job %u.%u on wrong node %s ",
job_id_ptr[i], step_id_ptr[i], node_name);
- kill_job_on_node(job_id_ptr[i], node_ptr);
+ kill_job_on_node(job_id_ptr[i], job_ptr,
+ node_ptr);
}
}
else if (job_ptr->job_state & JOB_COMPLETING) {
/* Re-send kill request as needed,
* not necessarily an error */
- kill_job_on_node(job_id_ptr[i], node_ptr);
+ kill_job_on_node(job_id_ptr[i], job_ptr, node_ptr);
}
@@ -3215,7 +3216,7 @@ validate_jobs_on_node(char *node_name, uint32_t * job_count,
last_job_update = now;
job_ptr->start_time = job_ptr->end_time = now;
delete_job_details(job_ptr);
- kill_job_on_node(job_id_ptr[i], node_ptr);
+ kill_job_on_node(job_id_ptr[i], job_ptr, node_ptr);
job_completion_logger(job_ptr);
}
@@ -3225,7 +3226,7 @@ validate_jobs_on_node(char *node_name, uint32_t * job_count,
job_id_ptr[i], step_id_ptr[i],
job_state_string(job_ptr->job_state),
node_name);
- kill_job_on_node(job_id_ptr[i], node_ptr);
+ kill_job_on_node(job_id_ptr[i], job_ptr, node_ptr);
}
}
@@ -3277,10 +3278,12 @@ static void _purge_lost_batch_jobs(int node_inx, time_t now)
* these jobs and use this function to kill them - one
* agent request per node as they register.
* IN job_id - id of the job to be killed
+ * IN job_ptr - pointer to terminating job (NULL if unknown, e.g. orphaned)
* IN node_ptr - pointer to the node on which the job resides
*/
extern void
-kill_job_on_node(uint32_t job_id, struct node_record *node_ptr)
+kill_job_on_node(uint32_t job_id, struct job_record *job_ptr,
+ struct node_record *node_ptr)
{
agent_arg_t *agent_info;
kill_job_msg_t *kill_req;
@@ -3289,6 +3292,10 @@ kill_job_on_node(uint32_t job_id, struct node_record *node_ptr)
kill_req = xmalloc(sizeof(kill_job_msg_t));
kill_req->job_id = job_id;
+ if (job_ptr) { /* NULL if unknown */
+ kill_req->select_jobinfo = select_g_copy_jobinfo(
+ job_ptr->select_jobinfo);
+ }
agent_info = xmalloc(sizeof(agent_arg_t));
agent_info->node_count = 1;
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 04a31d29bdc..30e54b5b325 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -1182,7 +1182,7 @@ extern int validate_nodes_via_front_end(uint32_t job_count,
if (job_ptr == NULL) {
error("Orphan job %u.%u reported",
job_id_ptr[i], step_id_ptr[i]);
- kill_job_on_node(job_id_ptr[i], node_ptr);
+ kill_job_on_node(job_id_ptr[i], job_ptr, node_ptr);
}
else if (job_ptr->job_state == JOB_RUNNING) {
@@ -1197,7 +1197,7 @@ extern int validate_nodes_via_front_end(uint32_t job_count,
else if (job_ptr->job_state & JOB_COMPLETING) {
/* Re-send kill request as needed,
* not necessarily an error */
- kill_job_on_node(job_id_ptr[i], node_ptr);
+ kill_job_on_node(job_id_ptr[i], job_ptr, node_ptr);
}
@@ -1209,7 +1209,7 @@ extern int validate_nodes_via_front_end(uint32_t job_count,
last_job_update = now;
job_ptr->start_time = job_ptr->end_time = now;
delete_job_details(job_ptr);
- kill_job_on_node(job_id_ptr[i], node_ptr);
+ kill_job_on_node(job_id_ptr[i], job_ptr, node_ptr);
job_completion_logger(job_ptr);
}
@@ -1217,7 +1217,7 @@ extern int validate_nodes_via_front_end(uint32_t job_count,
error("Registered job %u.%u in state %s",
job_id_ptr[i], step_id_ptr[i],
job_state_string(job_ptr->job_state));
- kill_job_on_node(job_id_ptr[i], node_ptr);
+ kill_job_on_node(job_id_ptr[i], job_ptr, node_ptr);
}
}
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 9a23b1082c8..e4b3f0483b1 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -736,9 +736,11 @@ extern int kill_job_by_part_name(char *part_name);
* these jobs and use this function to kill them - one
* agent request per node as they register.
* IN job_id - id of the job to be killed
+ * IN job_ptr - pointer to terminating job (NULL if unknown, e.g. orphaned)
* IN node_ptr - pointer to the node on which the job resides
*/
extern void kill_job_on_node(uint32_t job_id,
+ struct job_record *job_ptr,
struct node_record *node_ptr);
/*
--
GitLab