From 6e0a4c1c76805f0d8b765a31b6ddda470cecfd4f Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 26 Jun 2009 17:49:50 +0000
Subject: [PATCH] svn merge -r17976:17985
https://eris.llnl.gov/svn/slurm/branches/slurm-2.0
---
NEWS | 7 +++++++
.../multifactor/priority_multifactor.c | 14 ++++++++++++--
src/plugins/select/bluegene/plugin/bluegene.c | 19 ++-----------------
src/sacctmgr/sacctmgr.c | 3 +++
src/salloc/salloc.c | 13 +++++++------
src/srun/allocate.c | 13 +++++++------
6 files changed, 38 insertions(+), 31 deletions(-)
diff --git a/NEWS b/NEWS
index dee3d412526..67e4f5f4247 100644
--- a/NEWS
+++ b/NEWS
@@ -68,6 +68,13 @@ documents those changes that are of interest to users and admins.
-- Better logging for when job's request bad output file locations.
-- Fix issue where if user specified non-existant file to write to slurmstepd
will regain privileges before sending batch script ended to the controller.
+ -- Fix bug when using the priority_multifactor plugin with no associations yet.
+ -- BLUEGENE - we no longer check for the images to sync state. This was
+ needed long ago when rebooting blocks wasn't a possibility and should
+ had been removed when that functionality was available.
+ -- Added message about no connection with the database for sacctmgr.
+ -- On BlueGene, let srun or salloc exit on SIGINT if slurmctld dies while
+ booting its block.
* Changes in SLURM 2.0.3
========================
diff --git a/src/plugins/priority/multifactor/priority_multifactor.c b/src/plugins/priority/multifactor/priority_multifactor.c
index b7940509650..1f5ee0b56cf 100644
--- a/src/plugins/priority/multifactor/priority_multifactor.c
+++ b/src/plugins/priority/multifactor/priority_multifactor.c
@@ -138,6 +138,7 @@ static int _apply_decay(double decay_factor)
return SLURM_SUCCESS;
xassert(assoc_mgr_association_list);
+ xassert(assoc_mgr_qos_list);
slurm_mutex_lock(&assoc_mgr_association_lock);
itr = list_iterator_create(assoc_mgr_association_list);
@@ -938,7 +939,14 @@ int init ( void )
temp);
calc_fairshare = 0;
weight_fs = 0;
- } else {
+ } else if(weight_fs) {
+ if(!assoc_mgr_root_assoc)
+ fatal("It appears you don't have any association "
+ "data from your database. "
+ "The priority/multifactor plugin requires "
+ "this information to run correctly. Please "
+ "check your database connection and try again.");
+
if(!cluster_procs)
fatal("We need to have a cluster cpu count "
"before we can init the priority/multifactor "
@@ -960,7 +968,9 @@ int init ( void )
fatal("pthread_create error %m");
slurm_attr_destroy(&thread_attr);
- }
+ } else
+ calc_fairshare = 0;
+
xfree(temp);
verbose("%s loaded", plugin_name);
diff --git a/src/plugins/select/bluegene/plugin/bluegene.c b/src/plugins/select/bluegene/plugin/bluegene.c
index 074805a2111..1eedebf3614 100644
--- a/src/plugins/select/bluegene/plugin/bluegene.c
+++ b/src/plugins/select/bluegene/plugin/bluegene.c
@@ -1416,7 +1416,8 @@ static int _validate_config_nodes(List curr_block_list,
while ((bg_record = list_next(itr_conf))) {
list_iterator_reset(itr_curr);
while ((init_bg_record = list_next(itr_curr))) {
- if (strcasecmp(bg_record->nodes, init_bg_record->nodes))
+ if (strcasecmp(bg_record->nodes,
+ init_bg_record->nodes))
continue; /* wrong nodes */
if(!bit_equal(bg_record->ionode_bitmap,
init_bg_record->ionode_bitmap))
@@ -1424,28 +1425,12 @@ static int _validate_config_nodes(List curr_block_list,
#ifdef HAVE_BGL
if (bg_record->conn_type != init_bg_record->conn_type)
continue; /* wrong conn_type */
- if(bg_record->blrtsimage &&
- strcasecmp(bg_record->blrtsimage,
- init_bg_record->blrtsimage))
- continue;
#else
if ((bg_record->conn_type != init_bg_record->conn_type)
&& ((bg_record->conn_type < SELECT_SMALL)
&& (init_bg_record->conn_type < SELECT_SMALL)))
continue; /* wrong conn_type */
#endif
- if(bg_record->linuximage &&
- strcasecmp(bg_record->linuximage,
- init_bg_record->linuximage))
- continue;
- if(bg_record->mloaderimage &&
- strcasecmp(bg_record->mloaderimage,
- init_bg_record->mloaderimage))
- continue;
- if(bg_record->ramdiskimage &&
- strcasecmp(bg_record->ramdiskimage,
- init_bg_record->ramdiskimage))
- continue;
copy_bg_record(init_bg_record, bg_record);
/* remove from the curr list since we just
diff --git a/src/sacctmgr/sacctmgr.c b/src/sacctmgr/sacctmgr.c
index b62c43033dc..32841d82a7d 100644
--- a/src/sacctmgr/sacctmgr.c
+++ b/src/sacctmgr/sacctmgr.c
@@ -193,11 +193,14 @@ main (int argc, char *argv[])
errno = 0;
db_conn = acct_storage_g_get_connection(false, 0, 1);
if(errno != SLURM_SUCCESS) {
+ int tmp_errno = errno;
if((input_field_count == 2) &&
(!strncasecmp(argv[2], "Configuration", strlen(argv[1]))) &&
((!strncasecmp(argv[1], "list", strlen(argv[0]))) ||
(!strncasecmp(argv[1], "show", strlen(argv[0])))))
sacctmgr_list_config(false);
+ errno = tmp_errno;
+ fprintf(stderr, "Problem talking to the database: %m\n");
exit(1);
}
my_uid = getuid();
diff --git a/src/salloc/salloc.c b/src/salloc/salloc.c
index 68fee323429..5f3a77243df 100644
--- a/src/salloc/salloc.c
+++ b/src/salloc/salloc.c
@@ -102,7 +102,7 @@ static void _node_fail_handler(srun_node_fail_msg_t *msg);
#define POLL_SLEEP 3 /* retry interval in seconds */
static int _wait_bluegene_block_ready(
resource_allocation_response_msg_t *alloc);
-static int _blocks_dealloc();
+static int _blocks_dealloc(void);
#endif
#ifdef HAVE_CRAY_XT
@@ -689,7 +689,7 @@ static int _wait_bluegene_block_ready(resource_allocation_response_msg_t *alloc)
&block_id);
for (i=0; (cur_delay < max_delay); i++) {
- if(i == 1)
+ if (i == 1)
info("Waiting for block %s to become ready for job",
block_id);
if (i) {
@@ -712,13 +712,14 @@ static int _wait_bluegene_block_ready(resource_allocation_response_msg_t *alloc)
is_ready = 1;
break;
}
+ if (allocation_interrupted)
+ break;
}
if (is_ready)
info("Block %s is ready for job", block_id);
- else if(!allocation_interrupted)
+ else if (!allocation_interrupted)
error("Block %s still not ready", block_id);
- else /* this should never happen, but if allocation_intrrupted
- send back not ready */
+ else /* allocation_interrupted and slurmctld not responing */
is_ready = 0;
xfree(block_id);
@@ -734,7 +735,7 @@ static int _wait_bluegene_block_ready(resource_allocation_response_msg_t *alloc)
* 0: no deallocate in progress
* -1: error occurred
*/
-static int _blocks_dealloc()
+static int _blocks_dealloc(void)
{
static node_select_info_msg_t *bg_info_ptr = NULL, *new_bg_ptr = NULL;
int rc = 0, error_code = 0, i;
diff --git a/src/srun/allocate.c b/src/srun/allocate.c
index a52a56b9725..0b57e6fcf3a 100644
--- a/src/srun/allocate.c
+++ b/src/srun/allocate.c
@@ -102,7 +102,7 @@ static void _intr_handler(int signo);
#define POLL_SLEEP 3 /* retry interval in seconds */
static int _wait_bluegene_block_ready(
resource_allocation_response_msg_t *alloc);
-static int _blocks_dealloc();
+static int _blocks_dealloc(void);
#endif
#ifdef HAVE_CRAY_XT
@@ -237,7 +237,7 @@ static int _wait_bluegene_block_ready(resource_allocation_response_msg_t *alloc)
&block_id);
for (i=0; (cur_delay < max_delay); i++) {
- if(i == 1)
+ if (i == 1)
debug("Waiting for block %s to become ready for job",
block_id);
if (i) {
@@ -260,13 +260,14 @@ static int _wait_bluegene_block_ready(resource_allocation_response_msg_t *alloc)
is_ready = 1;
break;
}
+ if (destroy_job)
+ break;
}
if (is_ready)
debug("Block %s is ready for job", block_id);
- else if(!destroy_job)
+ else if (!destroy_job)
error("Block %s still not ready", block_id);
- else /* this should never happen, but if destroy_job
- send back not ready */
+ else /* destroy_job set and slurmctld not responing */
is_ready = 0;
xfree(block_id);
@@ -282,7 +283,7 @@ static int _wait_bluegene_block_ready(resource_allocation_response_msg_t *alloc)
* 0: no deallocate in progress
* -1: error occurred
*/
-static int _blocks_dealloc()
+static int _blocks_dealloc(void)
{
static node_select_info_msg_t *bg_info_ptr = NULL, *new_bg_ptr = NULL;
int rc = 0, error_code = 0, i;
--
GitLab