diff --git a/src/common/slurm_protocol_defs.c b/src/common/slurm_protocol_defs.c
index 91f3ec1d0249f0fa32b042fe34d19534e89a201c..2f8acf872968c435636d2452d21364e1f5ddc2c2 100644
--- a/src/common/slurm_protocol_defs.c
+++ b/src/common/slurm_protocol_defs.c
@@ -268,7 +268,10 @@ void slurm_free_launch_tasks_response_msg(launch_tasks_response_msg_t *
void slurm_free_kill_job_msg(kill_job_msg_t * msg)
{
- xfree(msg);
+ if (msg) {
+ select_g_free_jobinfo(&msg->select_jobinfo);
+ xfree(msg);
+ }
}
void slurm_free_update_job_time_msg(job_time_msg_t * msg)
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index 805b903b51561a9c922853ecaeba74f83132d805..480ceb211cdb7520f26be1daf5e30927bc355992 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -316,9 +316,14 @@ typedef struct return_code_msg {
int32_t return_code;
} return_code_msg_t;
+/* 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 */
typedef struct kill_job_msg {
uint32_t job_id;
uint32_t job_uid;
+ select_jobinfo_t select_jobinfo; /* opaque data type */
} kill_job_msg_t;
typedef struct job_time_msg {
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index c2d47a1d2dc049b83626656f5ebb0976cd947fad..7da0911cbe80940482b2c41c14147fc52570227e 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -1267,6 +1267,7 @@ _pack_kill_job_msg(kill_job_msg_t * msg, Buf buffer)
pack32(msg->job_id, buffer);
pack32(msg->job_uid, buffer);
+ select_g_pack_jobinfo(msg->select_jobinfo, buffer);
}
static int
@@ -1282,6 +1283,10 @@ _unpack_kill_job_msg(kill_job_msg_t ** msg, Buf buffer)
safe_unpack32(&(tmp_ptr->job_id), buffer);
safe_unpack32(&(tmp_ptr->job_uid), buffer);
+ if (select_g_alloc_jobinfo (&tmp_ptr->select_jobinfo)
+ || select_g_unpack_jobinfo(tmp_ptr->select_jobinfo, buffer))
+ goto unpack_error;
+
return SLURM_SUCCESS;
unpack_error:
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 1e06fa74dcfe2670cfd1d99e7ce71ed19df02e97..6c4e1b2a663f3eca7ab8e3ff591182b90e6a1480 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -3,7 +3,7 @@
*****************************************************************************
* 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>, Kevin Tew <tew1@llnl.gov>, et. al.
+ * Written by Morris Jette <jette1@llnl.gov>, Kevin Tew <tew1@llnl.gov>
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -345,9 +345,9 @@ int main(int argc, char *argv[])
/* Plugins are needed to purge job/node data structures,
* unplug after other data structures are purged */
- slurm_select_fini();
g_slurm_jobcomp_fini();
slurm_sched_fini();
+ slurm_select_fini();
checkpoint_fini();
slurm_auth_fini();
switch_fini();
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 3c1631dfc7dc099471b2064b532f6038b2b71976..4d594ec68a0c9528d7738373e1377562132a0389 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -168,6 +168,8 @@ extern void deallocate_nodes(struct job_record *job_ptr, bool timeout)
last_node_update = time(NULL);
kill_job->job_id = job_ptr->job_id;
kill_job->job_uid = job_ptr->user_id;
+ kill_job->select_jobinfo = select_g_copy_jobinfo(
+ job_ptr->select_jobinfo);
for (i = 0; i < node_record_count; i++) {
struct node_record *node_ptr = &node_record_table_ptr[i];
@@ -1149,6 +1151,8 @@ extern void re_kill_job(struct job_record *job_ptr)
kill_job = xmalloc(sizeof(kill_job_msg_t));
kill_job->job_id = job_ptr->job_id;
kill_job->job_uid = job_ptr->user_id;
+ kill_job->select_jobinfo = select_g_copy_jobinfo(
+ job_ptr->select_jobinfo);
for (i = 0; i < node_record_count; i++) {
struct node_record *node_ptr = &node_record_table_ptr[i];
diff --git a/src/slurmd/mgr.c b/src/slurmd/mgr.c
index f995bb94cbd1c3769919b55a70e31dab4c41c941..b527092b2329bd9dc411be0cc5f7772eeb8b2546 100644
--- a/src/slurmd/mgr.c
+++ b/src/slurmd/mgr.c
@@ -256,12 +256,14 @@ mgr_spawn_task(spawn_task_request_msg_t *msg, slurm_addr *cli,
}
/*
- * Run a prolog or epilog script.
- * returns -1 on failure.
- *
+ * Run a prolog or epilog script. Sets environment variables:
+ * SLURM_JOBID = jobid, SLURM_UID=uid, and
+ * BGL_PARTITION_ID=bgl_part_id (if not NULL)
+ * Returns -1 on failure.
*/
-int
-run_script(bool prolog, const char *path, uint32_t jobid, uid_t uid)
+extern int
+run_script(bool prolog, const char *path, uint32_t jobid, uid_t uid,
+ char *bgl_part_id)
{
int status;
pid_t cpid;
@@ -293,6 +295,8 @@ run_script(bool prolog, const char *path, uint32_t jobid, uid_t uid)
env[0] = NULL;
setenvpf(&env, "SLURM_JOBID", "%u", jobid);
setenvpf(&env, "SLURM_UID", "%u", uid);
+ if (bgl_part_id)
+ setenvpf(&env, "BGL_PARTITION_ID", "%s", bgl_part_id);
execve(path, argv, env);
error("help! %m");
diff --git a/src/slurmd/mgr.h b/src/slurmd/mgr.h
index e012676b9d8659c814b80777370933b2bace4169..d7243b33bbc4b5166cb8e3455b9086e0ecd4e7a6 100644
--- a/src/slurmd/mgr.h
+++ b/src/slurmd/mgr.h
@@ -50,8 +50,12 @@ int mgr_launch_tasks(launch_tasks_request_msg_t *msg, slurm_addr *client,
int mgr_launch_batch_job(batch_job_launch_msg_t *msg, slurm_addr *client);
/*
- * Run epilog or prolog on this node
+ * Run a prolog or epilog script. Sets environment variables:
+ * SLURM_JOBID = jobid, SLURM_UID=uid, and
+ * BGL_PARTITION_ID=bgl_part_id (if not NULL)
+ * Returns -1 on failure.
*/
-int run_script(bool prolog, const char *path, uint32_t jobid, uid_t uid);
+extern int run_script(bool prolog, const char *path, uint32_t jobid, uid_t uid,
+ char *bgl_part_id);
#endif
diff --git a/src/slurmd/req.c b/src/slurmd/req.c
index 27b790d9065723cde80c567951418cd314bc12ea..7857cc0b82942b06c0b4ab113db4dc3acabf996b 100644
--- a/src/slurmd/req.c
+++ b/src/slurmd/req.c
@@ -41,6 +41,7 @@
#include "src/common/hostlist.h"
#include "src/common/log.h"
#include "src/common/macros.h"
+#include "src/common/node_select.c"
#include "src/common/slurm_auth.h"
#include "src/common/slurm_cred.h"
#include "src/common/slurm_protocol_api.h"
@@ -77,8 +78,8 @@ static void _rpc_shutdown(slurm_msg_t *msg, slurm_addr *cli_addr);
static void _rpc_reconfig(slurm_msg_t *msg, slurm_addr *cli_addr);
static void _rpc_pid2jid(slurm_msg_t *msg, slurm_addr *);
static int _rpc_ping(slurm_msg_t *, slurm_addr *);
-static int _run_prolog(uint32_t jobid, uid_t uid);
-static int _run_epilog(uint32_t jobid, uid_t uid);
+static int _run_prolog(uint32_t jobid, uid_t uid, char *bgl_part_id);
+static int _run_epilog(uint32_t jobid, uid_t uid, char *bgl_part_id);
static int _spawn_task(spawn_task_request_msg_t *, slurm_addr *,
slurm_addr *);
@@ -395,7 +396,7 @@ _rpc_launch_tasks(slurm_msg_t *msg, slurm_addr *cli)
/* xassert(slurm_cred_jobid_cached(conf->vctx, req->job_id));*/
/* Run job prolog if necessary */
- if (run_prolog && (_run_prolog(req->job_id, req->uid) != 0)) {
+ if (run_prolog && (_run_prolog(req->job_id, req->uid, NULL) != 0)) {
error("[job %u] prolog failed", req->job_id);
errnum = ESLURMD_PROLOG_FAILED;
goto done;
@@ -469,7 +470,7 @@ _rpc_spawn_task(slurm_msg_t *msg, slurm_addr *cli)
/* xassert(slurm_cred_jobid_cached(conf->vctx, req->job_id));*/
/* Run job prolog if necessary */
- if (run_prolog && (_run_prolog(req->job_id, req->uid) != 0)) {
+ if (run_prolog && (_run_prolog(req->job_id, req->uid, NULL) != 0)) {
error("[job %u] prolog failed", req->job_id);
errnum = ESLURMD_PROLOG_FAILED;
goto done;
@@ -505,6 +506,7 @@ _rpc_batch_job(slurm_msg_t *msg, slurm_addr *cli)
batch_job_launch_msg_t *req = (batch_job_launch_msg_t *)msg->data;
int rc = SLURM_SUCCESS;
uid_t req_uid = g_slurm_auth_get_uid(msg->cred);
+ char *bgl_part_id = NULL;
if (!_slurm_authorized_user(req_uid)) {
error("Security violation, batch launch RPC from uid %u",
@@ -516,7 +518,11 @@ _rpc_batch_job(slurm_msg_t *msg, slurm_addr *cli)
/*
* Run job prolog on this node
*/
- if (_run_prolog(req->job_id, req->uid) != 0) {
+ select_g_get_jobinfo(req->select_jobinfo, SELECT_DATA_PART_ID,
+ &bgl_part_id);
+ rc = _run_prolog(req->job_id, req->uid, bgl_part_id);
+ xfree(bgl_part_id);
+ if (rc != 0) {
error("[job %u] prolog failed", req->job_id);
rc = ESLURMD_PROLOG_FAILED;
goto done;
@@ -966,6 +972,7 @@ _rpc_kill_job(slurm_msg_t *msg, slurm_addr *cli)
uid_t uid = g_slurm_auth_get_uid(msg->cred);
int nsteps = 0;
int delay;
+ char *bgl_part_id = NULL;
/*
* check that requesting user ID is the SLURM UID
@@ -1055,8 +1062,11 @@ _rpc_kill_job(slurm_msg_t *msg, slurm_addr *cli)
}
save_cred_state(conf->vctx);
-
- if (_run_epilog(req->job_id, req->job_uid) != 0) {
+ select_g_get_jobinfo(req->select_jobinfo, SELECT_DATA_PART_ID,
+ &bgl_part_id);
+ rc = _run_epilog(req->job_id, req->job_uid, bgl_part_id);
+ xfree(bgl_part_id);
+ if (rc != 0) {
error ("[job %u] epilog failed", req->job_id);
rc = ESLURMD_EPILOG_FAILED;
} else
@@ -1169,23 +1179,23 @@ _rpc_update_time(slurm_msg_t *msg, slurm_addr *cli)
}
static int
-_run_prolog(uint32_t jobid, uid_t uid)
+_run_prolog(uint32_t jobid, uid_t uid, char *bgl_part_id)
{
int error_code;
slurm_mutex_lock(&conf->config_mutex);
- error_code = run_script(true, conf->prolog, jobid, uid);
+ error_code = run_script(true, conf->prolog, jobid, uid, bgl_part_id);
slurm_mutex_unlock(&conf->config_mutex);
return error_code;
}
static int
-_run_epilog(uint32_t jobid, uid_t uid)
+_run_epilog(uint32_t jobid, uid_t uid, char *bgl_part_id)
{
int error_code;
slurm_mutex_lock(&conf->config_mutex);
- error_code = run_script(false, conf->epilog, jobid, uid);
+ error_code = run_script(false, conf->epilog, jobid, uid, bgl_part_id);
slurm_mutex_unlock(&conf->config_mutex);
return error_code;
}