From 1e38eeb4dc1c3ab85ba1baecf559c2db99abaf8d Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Mon, 8 Sep 2008 21:59:42 +0000
Subject: [PATCH] svn merge -r14958:15006
https://eris.llnl.gov/svn/slurm/branches/slurm-1.3
---
NEWS | 10 +
configure | 2 +-
doc/html/faq.shtml | 9 +-
doc/html/team.shtml | 4 +-
doc/man/man5/slurm.conf.5 | 16 +-
etc/init.d.slurmdbd | 34 +-
src/api/allocate.c | 2 +-
src/common/assoc_mgr.c | 27 +-
src/common/assoc_mgr.h | 9 +-
src/common/read_config.c | 12 +-
.../filetxt/accounting_storage_filetxt.c | 10 +-
.../mysql/accounting_storage_mysql.c | 130 +-
.../mysql/mysql_jobacct_process.c | 16 +-
.../pgsql/accounting_storage_pgsql.c | 39 +-
src/plugins/sched/wiki/job_modify.c | 9 +-
src/plugins/sched/wiki2/job_modify.c | 8 +-
.../block_allocator/block_allocator.c | 1064 ++++++-----------
.../select/bluegene/plugin/bg_job_place.c | 11 +-
src/plugins/select/bluegene/plugin/bluegene.c | 1 -
.../select/bluegene/plugin/slurm_prolog.c | 5 +-
src/sacct/options.c | 18 +-
src/sacctmgr/user_functions.c | 19 +
src/salloc/Makefile.am | 2 +-
src/salloc/Makefile.in | 2 +-
src/salloc/salloc.c | 115 ++
src/slurmctld/controller.c | 5 +
src/slurmctld/job_mgr.c | 113 +-
src/slurmctld/job_scheduler.c | 4 +
src/slurmctld/proc_req.c | 13 +-
src/slurmctld/slurmctld.h | 12 +-
src/slurmctld/step_mgr.c | 2 +-
src/slurmd/slurmd/slurmd.c | 6 +-
src/slurmd/slurmstepd/mgr.c | 39 +-
src/slurmd/slurmstepd/req.c | 14 +-
src/slurmdbd/slurmdbd.c | 6 +-
src/smap/configure_functions.c | 2 +-
src/srun/allocate.c | 2 +-
src/srun/srun.c | 12 +-
testsuite/expect/globals | 51 +
testsuite/expect/test21.10 | 129 +-
testsuite/expect/test21.11 | 8 +
testsuite/expect/test21.12 | 8 +
testsuite/expect/test21.13 | 7 +
testsuite/expect/test21.14 | 5 +
testsuite/expect/test21.15 | 4 +
testsuite/expect/test21.16 | 6 +-
testsuite/expect/test21.17 | 5 +
testsuite/expect/test21.18 | 5 +
testsuite/expect/test21.19 | 5 +
testsuite/expect/test21.5 | 5 +
testsuite/expect/test21.6 | 71 +-
testsuite/expect/test21.7 | 10 +-
testsuite/expect/test21.8 | 8 +
testsuite/expect/test21.9 | 8 +
testsuite/expect/test7.3 | 2 +-
testsuite/expect/test7.7 | 2 +-
testsuite/expect/test7.7.prog.c | 94 +-
testsuite/expect/test7.8 | 2 +-
58 files changed, 1300 insertions(+), 939 deletions(-)
diff --git a/NEWS b/NEWS
index ad68f0d961c..d4c7f49d00f 100644
--- a/NEWS
+++ b/NEWS
@@ -39,6 +39,13 @@ documents those changes that are of interest to users and admins.
are created in SLURM tables for future use without a reboot of the SLURM
daemons, but are not reported by any SLURM commands or APIs.
+* Changes in SLURM 1.3.9
+========================
+ -- Fix jobs being cancelled by ctrl-C to have correct cancelled state in
+ accounting.
+ -- Slurmdbd will only cache user data, made for faster start up
+ -- Improved support for job steps in FRONT_END systems
+
* Changes in SLURM 1.3.8
========================
-- Added PrivateData flags for Users, Usage, and Accounts to Accounting.
@@ -56,6 +63,7 @@ documents those changes that are of interest to users and admins.
-- BLUEGENE - Set MPI environment variables from salloc.
-- BLUEGENE - Fix threading issue for overlap mode
-- Reject batch scripts containing DOS linebreaks.
+ -- BLUEGENE - Added wait for block boot to salloc
* Changes in SLURM 1.3.7
========================
@@ -512,6 +520,8 @@ documents those changes that are of interest to users and admins.
configuration or other runtime checks.
-- Add "include" keywork to SPANK plugstack.conf file to optionally include
other configuration files or directories of configuration files.
+ -- Srun to wait indefinitely for resource allocation to be made. Used to
+ abort after two minutes.
* Changes in SLURM 1.2.34
=========================
diff --git a/configure b/configure
index 00ef5f35adf..c052058b64a 100755
--- a/configure
+++ b/configure
@@ -25271,7 +25271,7 @@ echo "$as_me: WARNING: *** pkg-config not found. Cannot probe for libglade-2.0 o
# fi
-### Check for gtk2.8 package
+### Check for gtk2.7.1 package
if test "$ac_have_gtk" == "yes" ; then
$HAVEPKGCONFIG --exists gtk+-2.0
if ! test $? -eq 0 ; then
diff --git a/doc/html/faq.shtml b/doc/html/faq.shtml
index b9711d0b1be..6ca32bfc0aa 100644
--- a/doc/html/faq.shtml
+++ b/doc/html/faq.shtml
@@ -890,7 +890,12 @@ about these options.
clocks on the cluster?</b></a><br>
In general, yes. Having inconsistent clocks may cause nodes to
be unusable. SLURM log files should contain references to
-expired credentials.
+expired credentials. For example:
+<pre>
+error: Munge decode failed: Expired credential
+ENCODED: Wed May 12 12:34:56 2008
+DECODED: Wed May 12 12:01:12 2008
+</pre>
<p><a name="cred_invalid"><b>21. Why are "Invalid job credential"
errors generated?</b></a><br>
@@ -1001,6 +1006,6 @@ sinfo -t drain -h -o "scontrol update nodename='%N' state=drain reason='%E'"
<p class="footer"><a href="#top">top</a></p>
-<p style="text-align:center;">Last modified 2 September 2008</p>
+<p style="text-align:center;">Last modified 3 September 2008</p>
<!--#include virtual="footer.txt"-->
diff --git a/doc/html/team.shtml b/doc/html/team.shtml
index f82230ee631..df6f0fdfca3 100644
--- a/doc/html/team.shtml
+++ b/doc/html/team.shtml
@@ -29,6 +29,7 @@ and a host of others.
<li>Chuck Clouston (Bull)</li>
<li>Chris Dunlap (LLNL)</li>
<li>Joey Ekstrom (LLNL/Bringham Young University)</li>
+<li>Josh England (TGSMC)</li>
<li>Kent Engström (National Supercomputer Centre, Sweden)</li>
<li>Jim Garlick (LLNL)</li>
<li>Didier Gazen (Laboratoire d'Aerologie, France)</li>
@@ -59,11 +60,12 @@ Networking, Italy)</li>
<li>Federico Sacerdoti (D.E. Shaw)<li>
<li>Jeff Squyres (LAM MPI)</li>
<li>Prashanth Tamraparni (HP, India)</li>
+<li>Adam Todorski (Rensselaer Polytechnic Institute)</li
<li>Kevin Tew (LLNL/Bringham Young University)</li>
<li>Jay Windley (Linux NetworX)</li>
<li>Anne-Marie Wunderlin (Bull)</li>
</ul>
-<p style="text-align:center;">Last modified 28 July 2008</p>
+<p style="text-align:center;">Last modified 5 September 2008</p>
<!--#include virtual="footer.txt"-->
diff --git a/doc/man/man5/slurm.conf.5 b/doc/man/man5/slurm.conf.5
index dce6e249340..1c961e2830b 100644
--- a/doc/man/man5/slurm.conf.5
+++ b/doc/man/man5/slurm.conf.5
@@ -1661,11 +1661,19 @@ is very important.
\fBSelectTypeParameters\fR should be configured to treat
memory as a consumable resource and the \fB\-\-mem\fR option
should be used for job allocations.
-For more information see the following web page:
-\fIhttps://computing.llnl.gov/linux/slurm/cons_res_share.html\fR.
-.na
Possible values for \fBShared\fR are "EXCLUSIVE", "FORCE", "YES", and "NO".
-.ad
+Sharing of resources is typically useful only when using
+\fBSchedulerType=sched/gang\fR.
+For more information see the following web pages:
+.br
+\fIhttps://computing.llnl.gov/linux/slurm/cons_res.html\fR,
+.br
+\fIhttps://computing.llnl.gov/linux/slurm/cons_res_share.html\fR,
+.br
+\fIhttps://computing.llnl.gov/linux/slurm/gang_scheduling.html\fR, and
+.br
+\fIhttps://computing.llnl.gov/linux/slurm/preempt.html\fR.
+
.RS
.TP 12
\fBEXCLUSIVE\fR
diff --git a/etc/init.d.slurmdbd b/etc/init.d.slurmdbd
index 55473e4383d..45e4471e8e7 100755
--- a/etc/init.d.slurmdbd
+++ b/etc/init.d.slurmdbd
@@ -44,6 +44,12 @@ else
RETVAL=0
fi
+# We can not use a starter program without losing environment
+# variables that are critical on Blue Gene systems
+if [ -d /bgl/BlueLight/ppcfloor ]; then
+ STARTPROC=""
+fi
+
# Source slurm specific configuration
if [ -f /etc/sysconfig/slurm ] ; then
. /etc/sysconfig/slurm
@@ -53,18 +59,23 @@ fi
[ -f $CONFDIR/slurmdbd.conf ] || exit 1
+# setup library paths for slurm and munge support
+export LD_LIBRARY_PATH="$LIBDIR:$LD_LIBRARY_PATH"
+
start() {
- echo -n "starting slurmdbd: "
+ prog=$1
+ shift
+ echo -n "starting $prog: "
unset HOME MAIL USER USERNAME
- $STARTPROC $SBINDIR/slurmdbd $SLURMDBD_OPTIONS
+ $STARTPROC $SBINDIR/$prog $SLURMDBD_OPTIONS
rc_status -v
echo
touch /var/lock/subsys/slurmdbd
}
stop() {
- echo -n "stopping slurmdbd: "
- killproc slurmdbd -TERM
+ echo -n "stopping $1: "
+ killproc $1 -TERM
rc_status -v
echo
rm -f /var/lock/subsys/slurmdbd
@@ -76,7 +87,7 @@ slurmstatus() {
local rpid
local pidfile
- pidfile=`grep -i PidFile $CONFDIR/slurmdbd.conf | grep -v '^ *#'`
+ pidfile=`grep -i ${base}pid $CONFDIR/slurmdbd.conf | grep -v '^ *#'`
if [ $? = 0 ]; then
pidfile=${pidfile##*=}
pidfile=${pidfile%#*}
@@ -84,25 +95,26 @@ slurmstatus() {
pidfile=/var/run/slurmdbd.pid
fi
- pid=`pidof -o $$ -o $$PPID -o %PPID -x slurmdbd`
+ pid=`pidof -o $$ -o $$PPID -o %PPID -x $1 || \
+ pidof -o $$ -o $$PPID -o %PPID -x ${base}`
if [ -f $pidfile ]; then
read rpid < $pidfile
if [ "$rpid" != "" -a "$pid" != "" ]; then
for i in $pid ; do
if [ "$i" = "$rpid" ]; then
- echo $"slurmdbd (pid $pid) is running..."
+ echo $"${base} (pid $pid) is running..."
return 0
fi
done
elif [ "$rpid" != "" -a "$pid" = "" ]; then
- echo $"slurmdbd is stopped"
+ echo $"${base} is stopped"
return 1
fi
fi
- echo $"slurmdbd is stopped"
+ echo $"${base} is stopped"
return 3
}
@@ -125,8 +137,8 @@ case "$1" in
slurmstatus slurmdbd
;;
restart)
- stop slurmdbd
- start slurmdbd
+ $0 stop
+ $0 start
;;
condrestart)
if [ -f /var/lock/subsys/slurm ]; then
diff --git a/src/api/allocate.c b/src/api/allocate.c
index 41f7a96caa3..d87842f5e2a 100644
--- a/src/api/allocate.c
+++ b/src/api/allocate.c
@@ -262,7 +262,7 @@ slurm_allocate_resources_blocking (const job_desc_msg_t *user_req,
timeout);
/* If NULL, we didn't get the allocation in
the time desired, so just free the job id */
- if (resp == NULL) {
+ if (resp == NULL && errno != ESLURM_ALREADY_DONE) {
errnum = errno;
slurm_complete_job(job_id, -1);
}
diff --git a/src/common/assoc_mgr.c b/src/common/assoc_mgr.c
index 17f09dacf01..3dee13d3113 100644
--- a/src/common/assoc_mgr.c
+++ b/src/common/assoc_mgr.c
@@ -202,7 +202,7 @@ static int _get_local_user_list(void *db_conn, int enforce)
while((user = list_next(itr))) {
uid_t pw_uid = uid_from_string(user->name);
if(pw_uid == (uid_t) -1) {
- error("couldn't get a uid for user %s",
+ debug("couldn't get a uid for user %s",
user->name);
user->uid = (uint32_t)NO_VAL;
} else
@@ -218,26 +218,35 @@ static int _get_local_user_list(void *db_conn, int enforce)
extern int assoc_mgr_init(void *db_conn, assoc_init_args_t *args)
{
- int enforce = 0;
+ uint16_t enforce = 0;
+ uint16_t refresh = 0;
+ uint16_t cache_level = ASSOC_MGR_CACHE_ALL;
if(args) {
enforce = args->enforce;
if(args->remove_assoc_notify)
remove_assoc_notify = args->remove_assoc_notify;
+ refresh = args->refresh;
+ cache_level = args->cache_level;
}
- if(!local_cluster_name && !slurmdbd_conf)
+ if((!local_cluster_name || refresh) && !slurmdbd_conf) {
+ xfree(local_cluster_name);
local_cluster_name = slurm_get_cluster_name();
+ }
- if(!local_association_list)
+ if((!local_association_list || refresh)
+ && (cache_level & ASSOC_MGR_CACHE_ASSOC))
if(_get_local_association_list(db_conn, enforce) == SLURM_ERROR)
return SLURM_ERROR;
-
- if(!local_qos_list)
+
+ if((!local_qos_list || refresh)
+ && (cache_level & ASSOC_MGR_CACHE_QOS))
if(_get_local_qos_list(db_conn, enforce) == SLURM_ERROR)
return SLURM_ERROR;
-
- if(!local_user_list)
+
+ if((!local_user_list || refresh)
+ && (cache_level & ASSOC_MGR_CACHE_USER))
if(_get_local_user_list(db_conn, enforce) == SLURM_ERROR)
return SLURM_ERROR;
@@ -709,7 +718,7 @@ extern int assoc_mgr_update_local_users(acct_update_object_t *update)
}
pw_uid = uid_from_string(object->name);
if(pw_uid == (uid_t) -1) {
- error("couldn't get a uid for user %s",
+ debug("couldn't get a uid for user %s",
object->name);
object->uid = NO_VAL;
} else
diff --git a/src/common/assoc_mgr.h b/src/common/assoc_mgr.h
index df9b26f9b76..2ca1750013b 100644
--- a/src/common/assoc_mgr.h
+++ b/src/common/assoc_mgr.h
@@ -49,8 +49,15 @@
#include <slurm/slurm.h>
#include <slurm/slurm_errno.h>
+#define ASSOC_MGR_CACHE_ASSOC 0x0001
+#define ASSOC_MGR_CACHE_QOS 0x0002
+#define ASSOC_MGR_CACHE_USER 0x0004
+#define ASSOC_MGR_CACHE_ALL 0xffff
+
typedef struct {
- int enforce;
+ uint16_t cache_level;
+ uint16_t enforce;
+ uint16_t refresh;
void (*remove_assoc_notify) (acct_association_rec_t *rec);
} assoc_init_args_t;
diff --git a/src/common/read_config.c b/src/common/read_config.c
index 58ad7042c37..dbe52700f4e 100644
--- a/src/common/read_config.c
+++ b/src/common/read_config.c
@@ -810,6 +810,7 @@ static int _register_conf_node_aliases(slurm_conf_node_t *node_ptr)
"in FRONT_END mode");
goto cleanup;
}
+
hostname = node_ptr->hostnames;
address = node_ptr->addresses;
#else
@@ -825,15 +826,20 @@ static int _register_conf_node_aliases(slurm_conf_node_t *node_ptr)
#endif
/* now build the individual node structures */
+#ifdef HAVE_FRONT_END
+ /* we always want the first on in the list to be the one
+ * returned when looking for localhost
+ */
+ while ((alias = hostlist_pop(alias_list))) {
+#else
while ((alias = hostlist_shift(alias_list))) {
-#ifndef HAVE_FRONT_END
hostname = hostlist_shift(hostname_list);
address = hostlist_shift(address_list);
#endif
_push_to_hashtbls(alias, hostname, address, node_ptr->port,
- node_ptr->cpus, node_ptr->sockets,
- node_ptr->cores, node_ptr->threads);
+ node_ptr->cpus, node_ptr->sockets,
+ node_ptr->cores, node_ptr->threads);
free(alias);
#ifndef HAVE_FRONT_END
diff --git a/src/plugins/accounting_storage/filetxt/accounting_storage_filetxt.c b/src/plugins/accounting_storage/filetxt/accounting_storage_filetxt.c
index a37ad84ee4d..136d9d694f9 100644
--- a/src/plugins/accounting_storage/filetxt/accounting_storage_filetxt.c
+++ b/src/plugins/accounting_storage/filetxt/accounting_storage_filetxt.c
@@ -668,6 +668,7 @@ extern int jobacct_storage_p_step_complete(void *db_conn,
float ave_vsize = 0, ave_rss = 0, ave_pages = 0;
float ave_cpu = 0, ave_cpu2 = 0;
char *account;
+ uint32_t exit_code;
if(!storage_init) {
debug("jobacct init was not called or it failed");
@@ -684,7 +685,12 @@ extern int jobacct_storage_p_step_complete(void *db_conn,
if ((elapsed=now-step_ptr->start_time)<0)
elapsed=0; /* For *very* short jobs, if clock is wrong */
- if (step_ptr->exit_code)
+
+ exit_code = step_ptr->exit_code;
+ if (exit_code == NO_VAL) {
+ comp_status = JOB_CANCELLED;
+ exit_code = 0;
+ } else if (exit_code)
comp_status = JOB_FAILED;
else
comp_status = JOB_COMPLETE;
@@ -740,7 +746,7 @@ extern int jobacct_storage_p_step_complete(void *db_conn,
JOB_STEP,
step_ptr->step_id, /* stepid */
comp_status, /* completion status */
- step_ptr->exit_code, /* completion code */
+ exit_code, /* completion code */
cpus, /* number of tasks */
cpus, /* number of cpus */
elapsed, /* elapsed seconds */
diff --git a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
index d77c5ea2f37..44817f3cf45 100644
--- a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
+++ b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
@@ -2011,7 +2011,8 @@ extern int acct_storage_p_add_coord(mysql_conn_t *mysql_conn, uint32_t uid,
"insert into %s "
"(timestamp, action, name, "
"actor, info) "
- "values (%d, %u, '%s', '%s', '%s')",
+ "values (%d, %u, '%s', "
+ "'%s', \"%s\")",
txn_table,
now, DBD_ADD_ACCOUNT_COORDS, user,
user_name, acct);
@@ -6705,10 +6706,11 @@ extern int clusteracct_storage_p_node_down(mysql_conn_t *mysql_conn,
xstrfmtcat(query,
"insert into %s "
"(node_name, cluster, cpu_count, period_start, reason) "
- "values ('%s', '%s', %u, %d, '%s') on duplicate key "
+ "values ('%s', '%s', %u, %d, \"%s\") on duplicate key "
"update period_end=0;",
event_table, node_ptr->name, cluster,
cpus, event_time, my_reason);
+ debug3("%d(%d) query\n%s", mysql_conn->conn, __LINE__, query);
rc = mysql_db_query(mysql_conn->db_conn, query);
xfree(query);
@@ -6733,6 +6735,7 @@ extern int clusteracct_storage_p_node_up(mysql_conn_t *mysql_conn,
"update %s set period_end=%d where cluster='%s' "
"and period_end=0 and node_name='%s';",
event_table, event_time, cluster, node_ptr->name);
+ debug3("%d(%d) query\n%s", mysql_conn->conn, __LINE__, query);
rc = mysql_db_query(mysql_conn->db_conn, query);
xfree(query);
return rc;
@@ -6964,7 +6967,7 @@ extern int jobacct_storage_p_job_start(mysql_conn_t *mysql_conn,
{
#ifdef HAVE_MYSQL
int rc=SLURM_SUCCESS;
- char *jname, *nodes;
+ char *jname = NULL, *nodes = NULL;
long priority;
int track_steps = 0;
char *block_id = NULL;
@@ -7001,7 +7004,7 @@ extern int jobacct_storage_p_job_start(mysql_conn_t *mysql_conn,
if (job_ptr->nodes && job_ptr->nodes[0])
nodes = job_ptr->nodes;
else
- nodes = "(null)";
+ nodes = "None assigned";
if(job_ptr->batch_flag)
track_steps = 1;
@@ -7026,27 +7029,52 @@ extern int jobacct_storage_p_job_start(mysql_conn_t *mysql_conn,
if(!job_ptr->db_index) {
query = xstrdup_printf(
"insert into %s "
- "(jobid, account, associd, uid, gid, partition, "
- "blockid, eligible, submit, start, name, track_steps, "
- "state, priority, req_cpus, alloc_cpus, nodelist) "
- "values (%u, '%s', %u, %u, %u, '%s', '%s', "
- "%d, %d, %d, '%s', %u, "
- "%u, %u, %u, %u, '%s') "
- "on duplicate key update id=LAST_INSERT_ID(id), "
- "end=0, state=%u",
- job_table, job_ptr->job_id, job_ptr->account,
- job_ptr->assoc_id,
- job_ptr->user_id, job_ptr->group_id,
- job_ptr->partition, block_id,
- (int)job_ptr->details->begin_time,
- (int)job_ptr->details->submit_time,
- (int)job_ptr->start_time,
- jname, track_steps,
- job_ptr->job_state & (~JOB_COMPLETING),
- priority, job_ptr->num_procs,
- job_ptr->total_procs, nodes,
- job_ptr->job_state & (~JOB_COMPLETING));
-
+ "(jobid, associd, uid, gid, nodelist, ",
+ job_table);
+
+ if(job_ptr->account)
+ xstrcat(query, "account, ");
+ if(job_ptr->partition)
+ xstrcat(query, "partition, ");
+ if(block_id)
+ xstrcat(query, "blockid, ");
+
+ xstrfmtcat(query,
+ "eligible, submit, start, name, track_steps, "
+ "state, priority, req_cpus, alloc_cpus) "
+ "values (%u, %u, %u, %u, '%s', ",
+ job_ptr->job_id, job_ptr->assoc_id,
+ job_ptr->user_id, job_ptr->group_id, nodes);
+
+ if(job_ptr->account)
+ xstrfmtcat(query, "'%s', ", job_ptr->account);
+ if(job_ptr->partition)
+ xstrfmtcat(query, "'%s', ", job_ptr->partition);
+ if(block_id)
+ xstrfmtcat(query, "'%s', ", block_id);
+
+ xstrfmtcat(query,
+ "%d, %d, %d, '%s', %u, %u, %u, %u, %u) "
+ "on duplicate key update "
+ "id=LAST_INSERT_ID(id), state=%u, associd=%u",
+ (int)job_ptr->details->begin_time,
+ (int)job_ptr->details->submit_time,
+ (int)job_ptr->start_time,
+ jname, track_steps,
+ job_ptr->job_state & (~JOB_COMPLETING),
+ priority, job_ptr->num_procs,
+ job_ptr->total_procs,
+ job_ptr->job_state & (~JOB_COMPLETING),
+ job_ptr->assoc_id);
+
+ if(job_ptr->account)
+ xstrfmtcat(query, ", account='%s'", job_ptr->account);
+ if(job_ptr->partition)
+ xstrfmtcat(query, ", partition='%s'",
+ job_ptr->partition);
+ if(block_id)
+ xstrfmtcat(query, ", blockid='%s'", block_id);
+
debug3("%d(%d) query\n%s", mysql_conn->conn, __LINE__, query);
try_again:
if(!(job_ptr->db_index = mysql_insert_ret_id(
@@ -7065,16 +7093,25 @@ extern int jobacct_storage_p_job_start(mysql_conn_t *mysql_conn,
rc = SLURM_ERROR;
}
} else {
- query = xstrdup_printf(
- "update %s set partition='%s', blockid='%s', start=%d, "
- "name='%s', state=%u, alloc_cpus=%u, nodelist='%s', "
- "account='%s', end=0 where id=%d",
- job_table, job_ptr->partition, block_id,
- (int)job_ptr->start_time,
- jname,
- job_ptr->job_state & (~JOB_COMPLETING),
- job_ptr->total_procs, nodes,
- job_ptr->account, job_ptr->db_index);
+ query = xstrdup_printf("update %s set nodelist='%s', ",
+ job_table, nodes);
+
+ if(job_ptr->account)
+ xstrfmtcat(query, "account='%s', ",
+ job_ptr->account);
+ if(job_ptr->partition)
+ xstrfmtcat(query, "partition='%s', ",
+ job_ptr->partition);
+ if(block_id)
+ xstrfmtcat(query, "blockid='%s', ", block_id);
+
+ xstrfmtcat(query, "start=%d, name='%s', state=%u, "
+ "alloc_cpus=%u, associd=%u where id=%d",
+ (int)job_ptr->start_time,
+ jname, job_ptr->job_state & (~JOB_COMPLETING),
+ job_ptr->total_procs, nodes,
+ job_ptr->assoc_id,
+ job_ptr->db_index);
debug3("%d(%d) query\n%s", mysql_conn->conn, __LINE__, query);
rc = mysql_db_query(mysql_conn->db_conn, query);
}
@@ -7099,7 +7136,7 @@ extern int jobacct_storage_p_job_complete(mysql_conn_t *mysql_conn,
#ifdef HAVE_MYSQL
char *query = NULL, *nodes = NULL;
int rc=SLURM_SUCCESS;
-
+
if (!job_ptr->db_index
&& (!job_ptr->details || !job_ptr->details->submit_time)) {
error("jobacct_storage_p_job_complete: "
@@ -7110,15 +7147,19 @@ extern int jobacct_storage_p_job_complete(mysql_conn_t *mysql_conn,
if(_check_connection(mysql_conn) != SLURM_SUCCESS)
return SLURM_ERROR;
debug2("mysql_jobacct_job_complete() called");
+
+ /* If we get an error with this just fall through to avoid an
+ * infinite loop
+ */
if (job_ptr->end_time == 0) {
debug("mysql_jobacct: job %u never started", job_ptr->job_id);
- return SLURM_ERROR;
+ return SLURM_SUCCESS;
}
if (job_ptr->nodes && job_ptr->nodes[0])
nodes = job_ptr->nodes;
else
- nodes = "(null)";
+ nodes = "None assigned";
if(!job_ptr->db_index) {
if(!(job_ptr->db_index =
@@ -7135,7 +7176,6 @@ extern int jobacct_storage_p_job_complete(mysql_conn_t *mysql_conn,
job_ptr->job_id);
return SLURM_SUCCESS;
}
- jobacct_storage_p_job_start(mysql_conn, job_ptr);
}
}
@@ -7272,7 +7312,8 @@ extern int jobacct_storage_p_step_complete(mysql_conn_t *mysql_conn,
float ave_cpu = 0, ave_cpu2 = 0;
char *query = NULL;
int rc =SLURM_SUCCESS;
-
+ uint32_t exit_code = 0;
+
if (!step_ptr->job_ptr->db_index
&& (!step_ptr->job_ptr->details
|| !step_ptr->job_ptr->details->submit_time)) {
@@ -7309,7 +7350,12 @@ extern int jobacct_storage_p_step_complete(mysql_conn_t *mysql_conn,
if ((elapsed=now-step_ptr->start_time)<0)
elapsed=0; /* For *very* short jobs, if clock is wrong */
- if (step_ptr->exit_code)
+
+ exit_code = step_ptr->exit_code;
+ if (exit_code == NO_VAL) {
+ comp_status = JOB_CANCELLED;
+ exit_code = 0;
+ } else if (exit_code)
comp_status = JOB_FAILED;
else
comp_status = JOB_COMPLETE;
@@ -7369,7 +7415,7 @@ extern int jobacct_storage_p_step_complete(mysql_conn_t *mysql_conn,
step_table, (int)now,
comp_status,
step_ptr->job_ptr->requid,
- step_ptr->exit_code,
+ exit_code,
/* user seconds */
jobacct->user_cpu_sec,
/* user microseconds */
diff --git a/src/plugins/accounting_storage/mysql/mysql_jobacct_process.c b/src/plugins/accounting_storage/mysql/mysql_jobacct_process.c
index 318daf6dc28..ace4c6f4cd0 100644
--- a/src/plugins/accounting_storage/mysql/mysql_jobacct_process.c
+++ b/src/plugins/accounting_storage/mysql/mysql_jobacct_process.c
@@ -500,6 +500,7 @@ no_cond:
if(row[JOB_REQ_ACCOUNT])
job->account = xstrdup(row[JOB_REQ_ACCOUNT]);
+
if(row[JOB_REQ_BLOCKID])
job->blockid = xstrdup(row[JOB_REQ_BLOCKID]);
@@ -569,11 +570,15 @@ no_cond:
}
} else {
job->suspended = atoi(row[JOB_REQ_SUSPENDED]);
- if(!job->end) {
+
+ if(!job->start) {
+ job->elapsed = 0;
+ } else if(!job->end) {
job->elapsed = now - job->start;
} else {
job->elapsed = job->end - job->start;
}
+
job->elapsed -= job->suspended;
}
@@ -581,8 +586,13 @@ no_cond:
job->jobname = xstrdup(row[JOB_REQ_NAME]);
job->gid = atoi(row[JOB_REQ_GID]);
job->exitcode = atoi(row[JOB_REQ_COMP_CODE]);
- job->partition = xstrdup(row[JOB_REQ_PARTITION]);
- job->nodes = xstrdup(row[JOB_REQ_NODELIST]);
+
+ if(row[JOB_REQ_PARTITION])
+ job->partition = xstrdup(row[JOB_REQ_PARTITION]);
+
+ if(row[JOB_REQ_NODELIST])
+ job->nodes = xstrdup(row[JOB_REQ_NODELIST]);
+
if (!job->nodes || !strcmp(job->nodes, "(null)")) {
xfree(job->nodes);
job->nodes = xstrdup("(unknown)");
diff --git a/src/plugins/accounting_storage/pgsql/accounting_storage_pgsql.c b/src/plugins/accounting_storage/pgsql/accounting_storage_pgsql.c
index 26c6e05350d..41d4a280848 100644
--- a/src/plugins/accounting_storage/pgsql/accounting_storage_pgsql.c
+++ b/src/plugins/accounting_storage/pgsql/accounting_storage_pgsql.c
@@ -1143,7 +1143,7 @@ extern int jobacct_storage_p_job_start(PGconn *acct_pgsql_db,
if (job_ptr->nodes && job_ptr->nodes[0])
nodes = job_ptr->nodes;
else
- nodes = "(null)";
+ nodes = "None assigned";
if(job_ptr->batch_flag)
track_steps = 1;
@@ -1250,15 +1250,24 @@ extern int jobacct_storage_p_job_complete(PGconn *acct_pgsql_db,
if (job_ptr->nodes && job_ptr->nodes[0])
nodes = job_ptr->nodes;
else
- nodes = "(null)";
-
+ nodes = "None assigned";
+
if(!job_ptr->db_index) {
- job_ptr->db_index = _get_db_index(acct_pgsql_db,
- job_ptr->details->submit_time,
- job_ptr->job_id,
- job_ptr->assoc_id);
- if(job_ptr->db_index == -1)
- return SLURM_ERROR;
+ if(!(job_ptr->db_index =
+ _get_db_index(acct_pgsql_db,
+ job_ptr->details->submit_time,
+ job_ptr->job_id,
+ job_ptr->assoc_id))) {
+ /* If we get an error with this just fall
+ * through to avoid an infinite loop
+ */
+ if(jobacct_storage_p_job_start(acct_pgsql_db, job_ptr)
+ == SLURM_ERROR) {
+ error("couldn't add job %u at job completion",
+ job_ptr->job_id);
+ return SLURM_SUCCESS;
+ }
+ }
}
query = xstrdup_printf("update %s set start=%u, endtime=%u, state=%d, "
"nodelist='%s', comp_code=%u, "
@@ -1385,7 +1394,8 @@ extern int jobacct_storage_p_step_complete(PGconn *acct_pgsql_db,
float ave_cpu = 0, ave_cpu2 = 0;
char *query = NULL;
int rc =SLURM_SUCCESS;
-
+ uint32_t exit_code;
+
if (!step_ptr->job_ptr->db_index
&& (!step_ptr->job_ptr->details
|| !step_ptr->job_ptr->details->submit_time)) {
@@ -1425,7 +1435,12 @@ extern int jobacct_storage_p_step_complete(PGconn *acct_pgsql_db,
if ((elapsed=now-step_ptr->start_time)<0)
elapsed=0; /* For *very* short jobs, if clock is wrong */
- if (step_ptr->exit_code)
+
+ exit_code = step_ptr->exit_code;
+ if (exit_code == NO_VAL) {
+ comp_status = JOB_CANCELLED;
+ exit_code = 0;
+ } else if (exit_code)
comp_status = JOB_FAILED;
else
comp_status = JOB_COMPLETE;
@@ -1475,7 +1490,7 @@ extern int jobacct_storage_p_step_complete(PGconn *acct_pgsql_db,
step_table, (int)now,
comp_status,
step_ptr->job_ptr->requid,
- step_ptr->exit_code,
+ exit_code,
/* user seconds */
jobacct->user_cpu_sec,
/* user microseconds */
diff --git a/src/plugins/sched/wiki/job_modify.c b/src/plugins/sched/wiki/job_modify.c
index 4ac123c9ca7..7cad50eadc1 100644
--- a/src/plugins/sched/wiki/job_modify.c
+++ b/src/plugins/sched/wiki/job_modify.c
@@ -95,11 +95,10 @@ static int _job_modify(uint32_t jobid, char *bank_ptr,
old_time) * 60);
last_job_update = time(NULL);
}
- if (bank_ptr) {
- info("wiki: change job %u bank %s", jobid, bank_ptr);
- xfree(job_ptr->account);
- job_ptr->account = xstrdup(bank_ptr);
- last_job_update = time(NULL);
+
+ if (bank_ptr &&
+ (update_job_account("wiki", job_ptr, bank_ptr) != SLURM_SUCCESS)) {
+ return EINVAL;
}
if (new_hostlist) {
diff --git a/src/plugins/sched/wiki2/job_modify.c b/src/plugins/sched/wiki2/job_modify.c
index 9e5f7aec178..714a38976b9 100644
--- a/src/plugins/sched/wiki2/job_modify.c
+++ b/src/plugins/sched/wiki2/job_modify.c
@@ -99,11 +99,9 @@ static int _job_modify(uint32_t jobid, char *bank_ptr,
last_job_update = now;
}
- if (bank_ptr) {
- info("wiki: change job %u bank %s", jobid, bank_ptr);
- xfree(job_ptr->account);
- job_ptr->account = xstrdup(bank_ptr);
- last_job_update = now;
+ if (bank_ptr &&
+ (update_job_account("wiki", job_ptr, bank_ptr) != SLURM_SUCCESS)) {
+ return EINVAL;
}
if (feature_ptr) {
diff --git a/src/plugins/select/bluegene/block_allocator/block_allocator.c b/src/plugins/select/bluegene/block_allocator/block_allocator.c
index 6001c2ac07f..c55a91280fa 100644
--- a/src/plugins/select/bluegene/block_allocator/block_allocator.c
+++ b/src/plugins/select/bluegene/block_allocator/block_allocator.c
@@ -96,6 +96,11 @@ s_p_options_t bg_conf_file_options[] = {
{NULL}
};
+typedef enum {
+ BLOCK_ALGO_FIRST,
+ BLOCK_ALGO_SECOND
+} block_algo_t;
+
#ifdef HAVE_BG
/** internal helper functions */
#ifdef HAVE_BG_FILES
@@ -153,7 +158,7 @@ static void _delete_path_list(void *object);
static int _find_match(ba_request_t* ba_request, List results);
/** */
-static bool _node_used(ba_node_t* ba_node, int *geometry);
+static bool _node_used(ba_node_t* ba_node, int x_size);
/* */
static void _switch_config(ba_node_t* source, ba_node_t* target, int dim,
@@ -167,14 +172,9 @@ static int _set_external_wires(int dim, int count, ba_node_t* source,
static char *_set_internal_wires(List nodes, int size, int conn_type);
/* */
-static int _find_x_path(List results, ba_node_t *ba_node,
- int *start, int *first,
- int *geometry, int found, int conn_type);
-
-/* */
-static int _find_x_path2(List results, ba_node_t *ba_node,
- int *start, int *first,
- int *geometry, int found, int conn_type);
+static int _find_x_path(List results, ba_node_t *ba_node, int *start,
+ int x_size, int found, int conn_type,
+ block_algo_t algo);
/* */
static int _remove_node(List results, int *node_tar);
@@ -190,7 +190,7 @@ static int _find_next_free_using_port_2(ba_switch_t *curr_switch,
/* int count, int highest_phys_x); */
/* */
static int _finish_torus(ba_switch_t *curr_switch, int source_port,
- List nodes, int dim, int count, int *start);
+ int dim, int count, int *start);
/* */
static int *_set_best_path();
@@ -1432,6 +1432,7 @@ end_it:
* IN start - where to start the allocation.
* IN geometry - the requested geometry of the block.
* IN conn_type - mesh, torus, or small.
+ *
* RET char * - hostlist of midplanes results represent must be
* xfreed. NULL on failure
*/
@@ -1450,24 +1451,22 @@ extern char *set_bg_block(List results, int *start,
|| start[Y]>=DIM_SIZE[Y]
|| start[Z]>=DIM_SIZE[Z])
return NULL;
- if(geometry[X]<=0
- || geometry[Y]<=0
- || geometry[Z]<=0) {
+
+ if(geometry[X] <= 0 || geometry[Y] <= 0 || geometry[Z] <= 0) {
error("problem with geometry %c%c%c, needs to be at least 111",
alpha_num[geometry[X]],
alpha_num[geometry[Y]],
alpha_num[geometry[Z]]);
return NULL;
}
+
size = geometry[X] * geometry[Y] * geometry[Z];
- ba_node = &ba_system_ptr->
- grid[start[X]][start[Y]][start[Z]];
+ ba_node = &ba_system_ptr->grid[start[X]][start[Y]][start[Z]];
#else
if(start[X]>=DIM_SIZE[X])
return NULL;
size = geometry[X];
- ba_node = &ba_system_ptr->
- grid[start[X]];
+ ba_node = &ba_system_ptr->grid[start[X]];
#endif
@@ -1478,36 +1477,34 @@ extern char *set_bg_block(List results, int *start,
results = list_create(NULL);
else
send_results = 1;
-
+ /* This midplane should have already been checked if it was in
+ use or not */
list_append(results, ba_node);
if(conn_type == SELECT_SMALL) {
/* adding the ba_node and ending */
ba_node->used = true;
- name = xmalloc(4);
- snprintf(name, 4, "%c%c%c",
- alpha_num[ba_node->coord[X]],
- alpha_num[ba_node->coord[Y]],
- alpha_num[ba_node->coord[Z]]);
+ name = xstrdup_printf("%c%c%c",
+ alpha_num[ba_node->coord[X]],
+ alpha_num[ba_node->coord[Y]],
+ alpha_num[ba_node->coord[Z]]);
goto end_it;
}
found = _find_x_path(results, ba_node,
ba_node->coord,
- ba_node->coord,
- geometry,
+ geometry[X],
1,
- conn_type);
+ conn_type, BLOCK_ALGO_FIRST);
if(!found) {
debug2("trying less efficient code");
remove_block(results, color_count);
list_delete_all(results, &empty_null_destroy_list, "");
list_append(results, ba_node);
- found = _find_x_path2(results, ba_node,
- ba_node->coord,
- ba_node->coord,
- geometry,
- 1,
- conn_type);
+ found = _find_x_path(results, ba_node,
+ ba_node->coord,
+ geometry[X],
+ 1,
+ conn_type, BLOCK_ALGO_SECOND);
}
if(found) {
#ifdef HAVE_BG
@@ -2709,10 +2706,25 @@ static int _append_geo(int *geometry, List geos, int rotate)
}
/*
+ * Fill in the paths and extra midplanes we need for the block.
+ * Basically copy the x path sent in with the start_list in each Y anx
+ * Z dimension filling in every midplane for the block and then
+ * completing the Y and Z wiring, tying the whole block together.
*
+ * IN/OUT results - total list of midplanes after this function
+ * returns successfully. Should be
+ * an exact copy of the start_list at first.
+ * IN start_list - exact copy of results at first, This should only be
+ * a list of midplanes on the X dim. We will work off this and
+ * the geometry to fill in this wiring for the X dim in all the
+ * Y and Z coords.
+ * IN geometry - What the block looks like
+ * IN conn_type - Mesh or Torus
+ *
+ * RET: 0 on failure 1 on success
*/
static int _fill_in_coords(List results, List start_list,
- int *geometry, int conn_type)
+ int *geometry, int conn_type)
{
ba_node_t *ba_node = NULL;
ba_node_t *check_node = NULL;
@@ -2722,8 +2734,9 @@ static int _fill_in_coords(List results, List start_list,
ba_switch_t *curr_switch = NULL;
ba_switch_t *next_switch = NULL;
- if(!start_list)
+ if(!start_list || !results)
return 0;
+ /* go through the start_list and add all the midplanes */
itr = list_iterator_create(start_list);
while((check_node = (ba_node_t*) list_next(itr))) {
curr_switch = &check_node->axis_switch[X];
@@ -2744,18 +2757,23 @@ static int _fill_in_coords(List results, List start_list,
[check_node->coord[X]]
[check_node->coord[Y]+y]
[check_node->coord[Z]+z];
- if(ba_node->coord[Y]
- == check_node->coord[Y]
- && ba_node->coord[Z]
- == check_node->coord[Z])
+
+ if(ba_node->coord[Y] == check_node->coord[Y]
+ && ba_node->coord[Z] == check_node->coord[Z])
continue;
- if (!_node_used(ba_node,geometry)) {
+
+ if (!_node_used(ba_node, geometry[X])) {
debug3("here Adding %c%c%c",
alpha_num[ba_node->coord[X]],
alpha_num[ba_node->coord[Y]],
alpha_num[ba_node->coord[Z]]);
list_append(results, ba_node);
next_switch = &ba_node->axis_switch[X];
+
+ /* since we are going off the
+ * main system we can send NULL
+ * here
+ */
_copy_the_path(NULL, curr_switch,
next_switch,
0, X);
@@ -2788,6 +2806,24 @@ failed:
return rc;
}
+/*
+ * Copy a path through the wiring of a switch to another switch on a
+ * starting port on a dimension.
+ *
+ * IN/OUT: nodes - Local list of midplanes you are keeping track of. If
+ * you visit any new midplanes a copy from ba_system_grid
+ * will be added to the list. If NULL the path will be
+ * set in mark_switch of the main virtual system (ba_system_grid).
+ * IN: curr_switch - The switch you want to copy the path of
+ * IN/OUT: mark_switch - The switch you want to fill in. On success
+ * this switch will contain a complete path from the curr_switch
+ * starting from the source port.
+ * IN: source - source port number (If calling for the first time
+ * should be 0 since we are looking for 1 at the end)
+ * IN: dim - Dimension XYZ
+ *
+ * RET: on success 1, on error 0
+ */
static int _copy_the_path(List nodes, ba_switch_t *curr_switch,
ba_switch_t *mark_switch,
int source, int dim)
@@ -2798,7 +2834,8 @@ static int _copy_the_path(List nodes, ba_switch_t *curr_switch,
int port_tar, port_tar1;
ba_switch_t *next_switch = NULL;
ba_switch_t *next_mark_switch = NULL;
- /*set the switch to not be used */
+
+ /* Copy the source used and port_tar */
mark_switch->int_wire[source].used =
curr_switch->int_wire[source].used;
mark_switch->int_wire[source].port_tar =
@@ -2806,6 +2843,7 @@ static int _copy_the_path(List nodes, ba_switch_t *curr_switch,
port_tar = curr_switch->int_wire[source].port_tar;
+ /* Now to the same thing from the other end */
mark_switch->int_wire[port_tar].used =
curr_switch->int_wire[port_tar].used;
mark_switch->int_wire[port_tar].port_tar =
@@ -2828,6 +2866,7 @@ static int _copy_the_path(List nodes, ba_switch_t *curr_switch,
port_tar);
if(port_tar == 1) {
+ /* found the end of the line */
mark_switch->int_wire[1].used =
curr_switch->int_wire[1].used;
mark_switch->int_wire[1].port_tar =
@@ -2841,12 +2880,18 @@ static int _copy_the_path(List nodes, ba_switch_t *curr_switch,
if(node_curr[X] == node_tar[X]
&& node_curr[Y] == node_tar[Y]
&& node_curr[Z] == node_tar[Z]) {
+ /* We are going to the same node! this should never
+ happen */
debug4("something bad happened!!");
return 0;
}
+
+ /* see what the next switch is going to be */
next_switch = &ba_system_ptr->
grid[node_tar[X]][node_tar[Y]][node_tar[Z]].axis_switch[dim];
if(!nodes) {
+ /* If no nodes then just get the next switch to fill
+ in from the main system */
next_mark_switch = &ba_system_ptr->
grid[mark_node_tar[X]]
[mark_node_tar[Y]]
@@ -2855,6 +2900,7 @@ static int _copy_the_path(List nodes, ba_switch_t *curr_switch,
} else {
ba_node_t *ba_node = NULL;
ListIterator itr = list_iterator_create(nodes);
+ /* see if we have already been to this node */
while((ba_node = list_next(itr))) {
if (ba_node->coord[X] == mark_node_tar[X] &&
ba_node->coord[Y] == mark_node_tar[Y] &&
@@ -2863,6 +2909,7 @@ static int _copy_the_path(List nodes, ba_switch_t *curr_switch,
}
list_iterator_destroy(itr);
if(!ba_node) {
+ /* If node grab a copy and add it to the list */
ba_node = ba_copy_node(&ba_system_ptr->
grid[mark_node_tar[X]]
[mark_node_tar[Y]]
@@ -2877,8 +2924,10 @@ static int _copy_the_path(List nodes, ba_switch_t *curr_switch,
next_mark_switch = &ba_node->axis_switch[dim];
}
+
+ /* Keep going until we reach the end of the line */
return _copy_the_path(nodes, next_switch, next_mark_switch,
- port_tar, dim);
+ port_tar, dim);
}
static int _find_yz_path(ba_node_t *ba_node, int *first,
@@ -3350,7 +3399,7 @@ start_again:
#endif
;
- if (!_node_used(ba_node, ba_request->geometry)) {
+ if (!_node_used(ba_node, ba_request->geometry[X])) {
debug3("trying this node %c%c%c %c%c%c %d",
alpha_num[start[X]],
alpha_num[start[Y]],
@@ -3419,10 +3468,15 @@ requested_end:
return 0;
}
-/* bool _node_used(ba_node_t* ba_node, int geometry, */
-static bool _node_used(ba_node_t* ba_node, int *geometry)
+/*
+ * Used to check if midplane is usable in the block we are creating
+ *
+ * IN: ba_node - node to check if is used
+ * IN: x_size - How big is the block in the X dim used to see if the
+ * wires are full hence making this midplane unusable.
+ */
+static bool _node_used(ba_node_t* ba_node, int x_size)
{
- int i=0;
ba_switch_t* ba_switch = NULL;
/* if we've used this node in another block already */
@@ -3433,17 +3487,27 @@ static bool _node_used(ba_node_t* ba_node, int *geometry)
alpha_num[ba_node->coord[Z]]);
return true;
}
- /* if we've used this nodes switches completely in another
- block already */
- for(i=0;i<1;i++) {
- if(geometry[i]>1) {
- ba_switch = &ba_node->axis_switch[i];
-
- if(ba_switch->int_wire[3].used
- && ba_switch->int_wire[5].used) {
- debug3("switch in use dim %d!",i);
- return true;
- }
+ /* Check If we've used this node's switches completely in another
+ block already. Right now we are only needing to look at
+ the X dim since it is the only one with extra wires. This
+ can be set up to do all the dim's if in the future if it is
+ needed. We only need to check this if we are planning on
+ using more than 1 midplane in the block creation */
+ if(x_size > 1) {
+ /* get the switch of the X Dimension */
+ ba_switch = &ba_node->axis_switch[X];
+
+ /* If both of these ports are used then the node
+ is in use since there are no more wires we
+ can use since these can not connect to each
+ other they must be connected to the other ports.
+ */
+ if(ba_switch->int_wire[3].used && ba_switch->int_wire[5].used) {
+ debug3("switch full in the X dim on node %c%c%c!",
+ alpha_num[ba_node->coord[X]],
+ alpha_num[ba_node->coord[Y]],
+ alpha_num[ba_node->coord[Z]]);
+ return true;
}
}
@@ -3844,7 +3908,7 @@ static char *_set_internal_wires(List nodes, int size, int conn_type)
name = xmalloc(BUFSIZE);
hostlist = hostlist_create(NULL);
itr = list_iterator_create(nodes);
- while((ba_node[count] = (ba_node_t*) list_next(itr))) {
+ while((ba_node[count] = list_next(itr))) {
snprintf(temp_name, sizeof(temp_name), "%c%c%c",
alpha_num[ba_node[count]->coord[X]],
alpha_num[ba_node[count]->coord[Y]],
@@ -3894,41 +3958,67 @@ static char *_set_internal_wires(List nodes, int size, int conn_type)
return name;
}
+/*
+ * Used to find a complete path based on the conn_type for an x dim.
+ * When starting to wire a block together this should be called first.
+ *
+ * IN/OUT: results - contains the number of midplanes we are
+ * potentially going to use in the X dim.
+ * IN: ba_node - current node we are looking at and have already added
+ * to results.
+ * IN: start - coordinates of the first midplane (so we know when when
+ * to end with a torus)
+ * IN: x_size - How many midplanes are we looking for in the X dim
+ * IN: found - count of how many midplanes we have found in the x dim
+ * IN: conn_type - MESH or TORUS
+ * IN: algo - algorythm to try an allocation by
+ *
+ * RET: 0 on failure, 1 on success
+ */
static int _find_x_path(List results, ba_node_t *ba_node,
- int *start, int *first, int *geometry,
- int found, int conn_type)
+ int *start, int x_size,
+ int found, int conn_type, block_algo_t algo)
{
ba_switch_t *curr_switch = NULL;
ba_switch_t *next_switch = NULL;
int port_tar = 0;
int source_port=0;
- int target_port=0;
+ int target_port=1;
int broke = 0, not_first = 0;
- int ports_to_try[2] = {3,5};
+ int ports_to_try[2] = {4, 2};
int *node_tar = NULL;
int i = 0;
ba_node_t *next_node = NULL;
ba_node_t *check_node = NULL;
-/* int highest_phys_x = geometry[X] - start[X]; */
+/* int highest_phys_x = x_size - start[X]; */
/* info("highest_phys_x is %d", highest_phys_x); */
- ListIterator itr;
+ ListIterator itr = NULL;
- if(!ba_node)
+ if(!ba_node || !results || !start)
return 0;
- if(!source_port) {
- target_port=1;
+ /* we don't need to go any further */
+ if(x_size == 1)
+ return 1;
+
+ if(algo == BLOCK_ALGO_FIRST) {
ports_to_try[0] = 4;
ports_to_try[1] = 2;
-
- }
+ } else if(algo == BLOCK_ALGO_SECOND) {
+ ports_to_try[0] = 2;
+ ports_to_try[1] = 4;
+ } else {
+ error("Unknown algo %d", algo);
+ return 0;
+ }
+
curr_switch = &ba_node->axis_switch[X];
- if(geometry[X] == 1) {
- goto found_one;
- }
- debug3("found - %d",found);
+
+ debug3("Algo(%d) found - %d", algo, found);
+
+ /* Check the 2 ports we can leave though in ports_to_try */
for(i=0;i<2;i++) {
/* info("trying port %d", ports_to_try[i]); */
/* check to make sure it isn't used */
@@ -3950,53 +4040,48 @@ static int _find_x_path(List results, ba_node_t *ba_node,
/* port_tar); */
/* check to see if we are back at the start of the
block */
- if((node_tar[X] ==
- start[X] &&
- node_tar[Y] ==
- start[Y] &&
- node_tar[Z] ==
- start[Z])) {
+ if((node_tar[X] == start[X]
+ && node_tar[Y] == start[Y]
+ && node_tar[Z] == start[Z])) {
broke = 1;
goto broke_it;
}
/* check to see if the port points to itself */
- if((node_tar[X] ==
- ba_node->coord[X] &&
- node_tar[Y] ==
- ba_node->coord[Y] &&
- node_tar[Z] ==
- ba_node->coord[Z])) {
+ if((node_tar[X] == ba_node->coord[X]
+ && node_tar[Y] == ba_node->coord[Y]
+ && node_tar[Z] == ba_node->coord[Z])) {
continue;
}
/* check to see if I am going to a place I have
already been before */
itr = list_iterator_create(results);
- while((next_node = (ba_node_t*) list_next(itr))) {
- debug3("looking at %c%c%c and %c%c%c",
+ while((next_node = list_next(itr))) {
+ debug3("Algo(%d) looking at %c%c%c and %c%c%c",
+ algo,
alpha_num[next_node->coord[X]],
alpha_num[next_node->coord[Y]],
alpha_num[next_node->coord[Z]],
alpha_num[node_tar[X]],
alpha_num[node_tar[Y]],
alpha_num[node_tar[Z]]);
- if((node_tar[X] == next_node->coord[X] &&
- node_tar[Y] == next_node->coord[Y] &&
- node_tar[Z] == next_node->coord[Z])) {
+ if((node_tar[X] == next_node->coord[X]
+ && node_tar[Y] == next_node->coord[Y]
+ && node_tar[Z] == next_node->coord[Z])) {
not_first = 1;
break;
}
}
list_iterator_destroy(itr);
- if(not_first && found<DIM_SIZE[X]) {
- debug2("already been there before");
+ if(not_first && found < DIM_SIZE[X]) {
+ debug2("Algo(%d) already been there before",
+ algo);
not_first = 0;
continue;
}
not_first = 0;
broke_it:
- next_node = &ba_system_ptr->
- grid[node_tar[X]]
+ next_node = &ba_system_ptr->grid[node_tar[X]]
#ifdef HAVE_BG
[node_tar[Y]]
[node_tar[Z]]
@@ -4004,97 +4089,36 @@ static int _find_x_path(List results, ba_node_t *ba_node,
;
next_switch = &next_node->axis_switch[X];
- if((conn_type == SELECT_MESH)
- && (found == (geometry[X]))) {
- debug2("we found the end of the mesh");
+ if((conn_type == SELECT_MESH) && (found == (x_size))) {
+ debug2("Algo(%d) we found the end of the mesh",
+ algo);
return 1;
}
- debug3("Broke = %d Found = %d geometry[X] = %d",
- broke, found, geometry[X]);
-
-/* This doesnt' appear to be of any use since we are doing a circular
- * system not a linear one. Kept just to make sure.
- */
+ debug3("Algo(%d) Broke = %d Found = %d x_size = %d",
+ algo, broke, found, x_size);
-/* debug3("Next Phys X %d Highest X %d", */
-/* next_node->phys_x, highest_phys_x); */
-/* if(next_node->phys_x >= highest_phys_x) { */
-/* debug3("looking for a passthrough"); */
-/* if(best_path) */
-/* list_destroy(best_path); */
-/* best_path = list_create(_delete_path_list); */
-/* if(path) */
-/* list_destroy(path); */
-/* path = list_create(_delete_path_list); */
-
-/* _find_passthrough(curr_switch, */
-/* 0, */
-/* results, */
-/* X, */
-/* 0, */
-/* highest_phys_x); */
-/* if(best_count < BEST_COUNT_INIT) { */
-/* debug2("yes found next free %d", */
-/* best_count); */
-/* node_tar = _set_best_path(); */
-/* next_node = &ba_system_ptr-> */
-/* grid[node_tar[X]] */
-/* #ifdef HAVE_BG */
-/* [node_tar[Y]] */
-/* [node_tar[Z]] */
-/* #endif */
-/* ; */
-/* next_switch = */
-/* &next_node->axis_switch[X]; */
-
-/* #ifdef HAVE_BG */
-/* debug2("found %d looking at " */
-/* "%c%c%c going to %c%c%c %d", */
-/* found, */
-/* alpha_num[ba_node->coord[X]], */
-/* alpha_num[ba_node->coord[Y]], */
-/* alpha_num[ba_node->coord[Z]], */
-/* alpha_num[node_tar[X]], */
-/* alpha_num[node_tar[Y]], */
-/* alpha_num[node_tar[Z]], */
-/* port_tar); */
-/* #endif */
-/* list_append(results, next_node); */
-/* found++; */
-/* if(_find_x_path(results, next_node, */
-/* start, first, geometry, */
-/* found, conn_type)) { */
-/* return 1; */
-/* } else { */
-/* found--; */
-/* _reset_the_path(curr_switch, 0, */
-/* 1, X); */
-/* _remove_node(results, */
-/* next_node->coord); */
-/* return 0; */
-/* } */
-/* } */
-/* } */
-
- if(broke && (found == geometry[X])) {
+ if(broke && (found == x_size)) {
goto found_path;
- } else if(found == geometry[X]) {
- debug2("finishing the torus!");
+ } else if(found == x_size) {
+ debug2("Algo(%d) finishing the torus!", algo);
+
if(best_path)
- list_destroy(best_path);
- best_path = list_create(_delete_path_list);
+ list_flush(best_path);
+ else
+ best_path =
+ list_create(_delete_path_list);
+
if(path)
- list_destroy(path);
- path = list_create(_delete_path_list);
- _finish_torus(curr_switch,
- 0,
- results,
- X,
- 0,
- start);
+ list_flush(path);
+ else
+ path = list_create(_delete_path_list);
+
+ _finish_torus(curr_switch, 0, X, 0, start);
+
if(best_count < BEST_COUNT_INIT) {
- debug2("Found a best path with %d "
- "steps.", best_count);
+ debug2("Algo(%d) Found a best path "
+ "with %d steps.",
+ algo, best_count);
_set_best_path();
return 1;
} else {
@@ -4105,10 +4129,11 @@ static int _find_x_path(List results, ba_node_t *ba_node,
continue;
}
- if (!_node_used(next_node, geometry)) {
+ if (!_node_used(next_node, x_size)) {
#ifdef HAVE_BG
- debug2("found %d looking at %c%c%c "
+ debug2("Algo(%d) found %d looking at %c%c%c "
"%d going to %c%c%c %d",
+ algo,
found,
alpha_num[ba_node->coord[X]],
alpha_num[ba_node->coord[Y]],
@@ -4120,13 +4145,11 @@ static int _find_x_path(List results, ba_node_t *ba_node,
port_tar);
#endif
itr = list_iterator_create(results);
- while((check_node =
- (ba_node_t*) list_next(itr))) {
- if((node_tar[X] ==
- check_node->coord[X] &&
- node_tar[Y] ==
- check_node->coord[Y] &&
- node_tar[Z] ==
+ while((check_node = list_next(itr))) {
+ if((node_tar[X] == check_node->coord[X]
+ && node_tar[Y] ==
+ check_node->coord[Y]
+ && node_tar[Z] ==
check_node->coord[Z])) {
break;
}
@@ -4134,7 +4157,8 @@ static int _find_x_path(List results, ba_node_t *ba_node,
list_iterator_destroy(itr);
if(!check_node) {
#ifdef HAVE_BG
- debug2("add %c%c%c",
+ debug2("Algo(%d) add %c%c%c",
+ algo,
alpha_num[next_node->coord[X]],
alpha_num[next_node->coord[Y]],
alpha_num[next_node->coord[Z]]);
@@ -4142,8 +4166,9 @@ static int _find_x_path(List results, ba_node_t *ba_node,
list_append(results, next_node);
} else {
#ifdef HAVE_BG
- debug2("Hey this is already added "
- "%c%c%c",
+ debug2("Algo(%d) Hey this is already "
+ "added %c%c%c",
+ algo,
alpha_num[node_tar[X]],
alpha_num[node_tar[Y]],
alpha_num[node_tar[Z]]);
@@ -4151,19 +4176,20 @@ static int _find_x_path(List results, ba_node_t *ba_node,
continue;
}
found++;
-
+
+ /* look for the next closest midplane */
if(!_find_x_path(results, next_node,
- start, first, geometry,
- found, conn_type)) {
- _remove_node(results,
- next_node->coord);
+ start, x_size,
+ found, conn_type, algo)) {
+ _remove_node(results, next_node->coord);
found--;
continue;
} else {
found_path:
#ifdef HAVE_BG
- debug2("added node %c%c%c %d %d -> "
- "%c%c%c %d %d",
+ debug2("Algo(%d) added node %c%c%c "
+ "%d %d -> %c%c%c %d %d",
+ algo,
alpha_num[ba_node->coord[X]],
alpha_num[ba_node->coord[Y]],
alpha_num[ba_node->coord[Z]],
@@ -4175,341 +4201,104 @@ static int _find_x_path(List results, ba_node_t *ba_node,
port_tar,
target_port);
#endif
- found_one:
- if(geometry[X] != 1) {
- curr_switch->
- int_wire
- [source_port].used = 1;
- curr_switch->
- int_wire
- [source_port].port_tar
- = ports_to_try[i];
- curr_switch->
- int_wire
- [ports_to_try[i]].used
- = 1;
- curr_switch->
- int_wire
- [ports_to_try[i]].
- port_tar = source_port;
+ curr_switch->int_wire[source_port].used
+ = 1;
+ curr_switch->int_wire
+ [source_port].port_tar
+ = ports_to_try[i];
+ curr_switch->int_wire
+ [ports_to_try[i]].used = 1;
+ curr_switch->int_wire
+ [ports_to_try[i]].port_tar
+ = source_port;
- next_switch->
- int_wire[port_tar].used
- = 1;
- next_switch->
- int_wire
- [port_tar].port_tar
- = target_port;
- next_switch->
- int_wire
- [target_port].used = 1;
- next_switch->
- int_wire
- [target_port].port_tar
- = port_tar;
- }
+ next_switch->int_wire[port_tar].used
+ = 1;
+ next_switch->int_wire[port_tar].port_tar
+ = target_port;
+ next_switch->int_wire[target_port].used
+ = 1;
+ next_switch->int_wire
+ [target_port].port_tar
+ = port_tar;
return 1;
-
}
}
}
}
- debug2("couldn't find path");
- return 0;
-}
-
-static int _find_x_path2(List results, ba_node_t *ba_node,
- int *start, int *first, int *geometry,
- int found, int conn_type)
-{
- ba_switch_t *curr_switch = NULL;
- ba_switch_t *next_switch = NULL;
-
- int port_tar = 0;
- int source_port=0;
- int target_port=0;
- int broke = 0, not_first = 0;
- int ports_to_try[2] = {3,5};
- int *node_tar = NULL;
- int i = 0;
- ba_node_t *next_node = NULL;
- ba_node_t *check_node = NULL;
-
- ListIterator itr;
-
- if(!ba_node)
+ if(algo == BLOCK_ALGO_FIRST) {
+ debug2("Algo(%d) couldn't find path", algo);
return 0;
-
- if(!source_port) {
- target_port=1;
- ports_to_try[0] = 2;
- ports_to_try[1] = 4;
+ } else if(algo == BLOCK_ALGO_SECOND) {
+#ifdef HAVE_BG
+ debug2("Algo(%d) looking for the next free node "
+ "starting at %c%c%c",
+ algo,
+ alpha_num[ba_node->coord[X]],
+ alpha_num[ba_node->coord[Y]],
+ alpha_num[ba_node->coord[Z]]);
+#endif
+
+ if(best_path)
+ list_flush(best_path);
+ else
+ best_path = list_create(_delete_path_list);
+
+ if(path)
+ list_flush(path);
+ else
+ path = list_create(_delete_path_list);
+
+ _find_next_free_using_port_2(curr_switch, 0, results, X, 0);
+
+ if(best_count < BEST_COUNT_INIT) {
+ debug2("Algo(%d) yes found next free %d", algo,
+ best_count);
+ node_tar = _set_best_path();
- }
- curr_switch = &ba_node->axis_switch[X];
- if(geometry[X] == 1) {
- goto found_one;
- }
- debug2("found - %d",found);
- for(i=0;i<2;i++) {
- /* check to make sure it isn't used */
- if(!curr_switch->int_wire[ports_to_try[i]].used) {
- node_tar = curr_switch->
- ext_wire[ports_to_try[i]].node_tar;
- port_tar = curr_switch->
- ext_wire[ports_to_try[i]].port_tar;
- if((node_tar[X] ==
- start[X] &&
- node_tar[Y] ==
- start[Y] &&
- node_tar[Z] ==
- start[Z])) {
- broke = 1;
- goto broke_it;
- }
- if((node_tar[X] ==
- ba_node->coord[X] &&
- node_tar[Y] ==
- ba_node->coord[Y] &&
- node_tar[Z] ==
- ba_node->coord[Z])) {
- continue;
- }
- itr = list_iterator_create(results);
- while((next_node = (ba_node_t*) list_next(itr))) {
- if((node_tar[X] ==
- next_node->coord[X] &&
- node_tar[Y] ==
- next_node->coord[Y] &&
- node_tar[Z] ==
- next_node->coord[Z])) {
- not_first = 1;
- break;
- }
-
- }
- list_iterator_destroy(itr);
- if(not_first && found<DIM_SIZE[X]) {
- not_first = 0;
- continue;
- }
- not_first = 0;
-
- broke_it:
- next_node = &ba_system_ptr->
- grid[node_tar[X]]
+ next_node = &ba_system_ptr->grid[node_tar[X]]
#ifdef HAVE_BG
[node_tar[Y]]
[node_tar[Z]]
#endif
;
-
+
next_switch = &next_node->axis_switch[X];
-
- if((conn_type == SELECT_MESH)
- && (found == (geometry[X]))) {
- debug2("we found the end of the mesh");
+#ifdef HAVE_BG
+ debug2("Algo(%d) found %d looking at %c%c%c "
+ "going to %c%c%c %d",
+ algo, found,
+ alpha_num[ba_node->coord[X]],
+ alpha_num[ba_node->coord[Y]],
+ alpha_num[ba_node->coord[Z]],
+ alpha_num[node_tar[X]],
+ alpha_num[node_tar[Y]],
+ alpha_num[node_tar[Z]],
+ port_tar);
+#endif
+ list_append(results, next_node);
+ found++;
+ if(_find_x_path(results, next_node,
+ start, x_size, found,
+ conn_type, algo)) {
return 1;
+ } else {
+ found--;
+ _reset_the_path(curr_switch, 0, 1, X);
+ _remove_node(results, next_node->coord);
+ debug2("Algo(%d) couldn't finish "
+ "the path off this one", algo);
}
- debug3("Broke = %d Found = %d geometry[X] = %d",
- broke, found, geometry[X]);
- if(broke && (found == geometry[X])) {
- goto found_path;
- } else if(found == geometry[X]) {
- debug2("finishing the torus!");
- if(best_path)
- list_destroy(best_path);
- best_path = list_create(_delete_path_list);
- if(path)
- list_destroy(path);
- path = list_create(_delete_path_list);
- _finish_torus(curr_switch,
- 0,
- results,
- X,
- 0,
- start);
- if(best_count < BEST_COUNT_INIT) {
- debug2("Found a best path with %d "
- "steps.", best_count);
- _set_best_path();
- return 1;
- } else {
- return 0;
- }
- } else if(broke) {
- broke = 0;
- continue;
- }
-
- if (!_node_used(next_node, geometry)) {
-#ifdef HAVE_BG
- debug2("found %d looking at %c%c%c "
- "%d going to %c%c%c %d",
- found,
- alpha_num[ba_node->coord[X]],
- alpha_num[ba_node->coord[Y]],
- alpha_num[ba_node->coord[Z]],
- ports_to_try[i],
- alpha_num[node_tar[X]],
- alpha_num[node_tar[Y]],
- alpha_num[node_tar[Z]],
- port_tar);
-#endif
- itr = list_iterator_create(results);
- while((check_node =
- (ba_node_t*) list_next(itr))) {
- if((node_tar[X] ==
- check_node->coord[X] &&
- node_tar[Y] ==
- check_node->coord[Y] &&
- node_tar[Z] ==
- check_node->coord[Z])) {
- break;
- }
- }
- list_iterator_destroy(itr);
- if(!check_node) {
-#ifdef HAVE_BG
- debug2("add %c%c%c",
- alpha_num[next_node->coord[X]],
- alpha_num[next_node->coord[Y]],
- alpha_num[next_node->coord[Z]]);
-#endif
- list_append(results, next_node);
- } else {
-#ifdef HAVE_BG
- debug2("Hey this is already added "
- "%c%c%c",
- alpha_num[node_tar[X]],
- alpha_num[node_tar[Y]],
- alpha_num[node_tar[Z]]);
-#endif
- continue;
- }
- found++;
-
- if(!_find_x_path2(results, next_node,
- start, first, geometry,
- found, conn_type)) {
- _remove_node(results,
- next_node->coord);
- found--;
- continue;
- } else {
- found_path:
-#ifdef HAVE_BG
- debug2("added node %c%c%c %d %d -> "
- "%c%c%c %d %d",
- alpha_num[ba_node->coord[X]],
- alpha_num[ba_node->coord[Y]],
- alpha_num[ba_node->coord[Z]],
- source_port,
- ports_to_try[i],
- alpha_num[node_tar[X]],
- alpha_num[node_tar[Y]],
- alpha_num[node_tar[Z]],
- port_tar,
- target_port);
-#endif
- found_one:
- if(geometry[X] != 1) {
- curr_switch->
- int_wire
- [source_port].used = 1;
- curr_switch->
- int_wire
- [source_port].port_tar
- = ports_to_try[i];
- curr_switch->
- int_wire
- [ports_to_try[i]].used
- = 1;
- curr_switch->
- int_wire
- [ports_to_try[i]].
- port_tar = source_port;
-
- next_switch->
- int_wire[port_tar].used
- = 1;
- next_switch->
- int_wire
- [port_tar].port_tar
- = target_port;
- next_switch->
- int_wire
- [target_port].used = 1;
- next_switch->
- int_wire
- [target_port].port_tar
- = port_tar;
- }
- return 1;
- }
- }
- }
+ }
+
+ debug2("Algo(%d) couldn't find path", algo);
+ return 0;
}
-#ifdef HAVE_BG
- debug2("looking for the next free node starting at %c%c%c",
- alpha_num[ba_node->coord[X]],
- alpha_num[ba_node->coord[Y]],
- alpha_num[ba_node->coord[Z]]);
-#endif
- if(best_path)
- list_destroy(best_path);
- best_path = list_create(_delete_path_list);
- if(path)
- list_destroy(path);
- path = list_create(_delete_path_list);
-
- _find_next_free_using_port_2(curr_switch,
- 0,
- results,
- X,
- 0);
- if(best_count < BEST_COUNT_INIT) {
- debug2("yes found next free %d", best_count);
- node_tar = _set_best_path();
-
- next_node = &ba_system_ptr->
- grid[node_tar[X]]
-#ifdef HAVE_BG
- [node_tar[Y]]
- [node_tar[Z]]
-#endif
- ;
-
- next_switch = &next_node->axis_switch[X];
-
-#ifdef HAVE_BG
- debug2("found %d looking at %c%c%c going to %c%c%c %d",
- found,
- alpha_num[ba_node->coord[X]],
- alpha_num[ba_node->coord[Y]],
- alpha_num[ba_node->coord[Z]],
- alpha_num[node_tar[X]],
- alpha_num[node_tar[Y]],
- alpha_num[node_tar[Z]],
- port_tar);
-#endif
- list_append(results, next_node);
- found++;
- if(_find_x_path2(results, next_node,
- start, first, geometry, found, conn_type)) {
- return 1;
- } else {
- found--;
- _reset_the_path(curr_switch, 0, 1, X);
- _remove_node(results, next_node->coord);
- debug2("couldn't finish the path off this one");
- }
- }
-
- debug2("couldn't find path 2");
+ error("We got here meaning there is a bad algo, "
+ "but this should never happen algo(%d)", algo);
return 0;
}
@@ -4614,8 +4403,8 @@ static int _find_next_free_using_port_2(ba_switch_t *curr_switch,
if((source_port%2))
target_port=1;
- list_destroy(best_path);
- best_path = list_create(_delete_path_list);
+ list_flush(best_path);
+
found = true;
path_add->out = target_port;
list_push(path, path_add);
@@ -4634,7 +4423,7 @@ static int _find_next_free_using_port_2(ba_switch_t *curr_switch,
temp_switch->dim = path_switch->dim;
temp_switch->in = path_switch->in;
temp_switch->out = path_switch->out;
- list_append(best_path,temp_switch);
+ list_append(best_path, temp_switch);
}
list_iterator_destroy(itr);
best_count = count;
@@ -4711,205 +4500,28 @@ return_0:
return 0;
}
-/* static int _find_passthrough(ba_switch_t *curr_switch, int source_port, */
-/* List nodes, int dim, int count, int highest_phys_x) */
-/* { */
-/* ba_switch_t *next_switch = NULL; */
-/* ba_path_switch_t *path_add = */
-/* (ba_path_switch_t *) xmalloc(sizeof(ba_path_switch_t)); */
-/* ba_path_switch_t *path_switch = NULL; */
-/* ba_path_switch_t *temp_switch = NULL; */
-/* int port_tar; */
-/* int target_port = 0; */
-/* int ports_to_try[2] = {3,5}; */
-/* int *node_tar= curr_switch->ext_wire[0].node_tar; */
-/* int *node_src = curr_switch->ext_wire[0].node_tar; */
-/* int i; */
-/* int used=0; */
-/* int broke = 0; */
-/* ba_node_t *ba_node = NULL; */
-
-/* ListIterator itr; */
-/* static bool found = false; */
-
-/* path_add->geometry[X] = node_src[X]; */
-/* #ifdef HAVE_BG */
-/* path_add->geometry[Y] = node_src[Y]; */
-/* path_add->geometry[Z] = node_src[Z]; */
-/* #endif */
-/* path_add->dim = dim; */
-/* path_add->in = source_port; */
-
-/* if(count>=best_count) { */
-/* xfree(path_add); */
-/* return 0; */
-/* } */
-
-/* itr = list_iterator_create(nodes); */
-/* while((ba_node = (ba_node_t*) list_next(itr))) { */
-
-/* #ifdef HAVE_BG */
-/* if(node_tar[X] == ba_node->coord[X] */
-/* && node_tar[Y] == ba_node->coord[Y] */
-/* && node_tar[Z] == ba_node->coord[Z]) { */
-/* broke = 1; */
-/* break; */
-/* } */
-/* #else */
-/* if(node_tar[X] == ba_node->coord[X]) { */
-/* broke = 1; */
-/* break; */
-/* } */
-/* #endif */
-
-/* } */
-/* list_iterator_destroy(itr); */
-/* ba_node = &ba_system_ptr-> */
-/* grid[node_tar[X]] */
-/* #ifdef HAVE_BG */
-/* [node_tar[Y]] */
-/* [node_tar[Z]] */
-/* #endif */
-/* ; */
-/* if(!broke && count>0 */
-/* && !ba_node->used */
-/* && (ba_node->phys_x < highest_phys_x)) { */
-
-/* debug3("this one not found %c%c%c", */
-/* alpha_num[node_tar[X]], */
-/* alpha_num[node_tar[Y]], */
-/* alpha_num[node_tar[Z]]); */
-
-/* broke = 0; */
-
-/* if((source_port%2)) */
-/* target_port=1; */
-
-/* list_destroy(best_path); */
-/* best_path = list_create(_delete_path_list); */
-/* found = true; */
-/* path_add->out = target_port; */
-/* list_push(path, path_add); */
-
-/* itr = list_iterator_create(path); */
-/* while((path_switch = (ba_path_switch_t*) list_next(itr))){ */
-
-/* temp_switch = (ba_path_switch_t *) */
-/* xmalloc(sizeof(ba_path_switch_t)); */
-
-/* temp_switch->geometry[X] = path_switch->geometry[X]; */
-/* #ifdef HAVE_BG */
-/* temp_switch->geometry[Y] = path_switch->geometry[Y]; */
-/* temp_switch->geometry[Z] = path_switch->geometry[Z]; */
-/* #endif */
-/* temp_switch->dim = path_switch->dim; */
-/* temp_switch->in = path_switch->in; */
-/* temp_switch->out = path_switch->out; */
-/* list_append(best_path,temp_switch); */
-/* } */
-/* list_iterator_destroy(itr); */
-/* best_count = count; */
-/* return 1; */
-/* } */
-
-/* if(source_port==0 || source_port==3 || source_port==5) { */
-/* if(count==0) { */
-/* ports_to_try[0] = 2; */
-/* ports_to_try[1] = 4; */
-/* } else { */
-/* ports_to_try[0] = 4; */
-/* ports_to_try[1] = 2; */
-/* } */
-/* } */
-
-/* for(i=0;i<2;i++) { */
-/* used=0; */
-/* if(!curr_switch->int_wire[ports_to_try[i]].used) { */
-/* itr = list_iterator_create(path); */
-/* while((path_switch = */
-/* (ba_path_switch_t*) list_next(itr))){ */
-
-/* if(((path_switch->geometry[X] == node_src[X]) */
-/* #ifdef HAVE_BG */
-/* && (path_switch->geometry[Y] */
-/* == node_src[Y]) */
-/* && (path_switch->geometry[Z] */
-/* == node_tar[Z]) */
-/* #endif */
-/* )) { */
-
-/* if( path_switch->out */
-/* == ports_to_try[i]) { */
-/* used = 1; */
-/* break; */
-/* } */
-/* } */
-/* } */
-/* list_iterator_destroy(itr); */
-
-/* if(curr_switch-> */
-/* ext_wire[ports_to_try[i]].node_tar[X] */
-/* == curr_switch->ext_wire[0].node_tar[X] */
-/* #ifdef HAVE_BG */
-/* && curr_switch-> */
-/* ext_wire[ports_to_try[i]].node_tar[Y] */
-/* == curr_switch->ext_wire[0].node_tar[Y] */
-/* && curr_switch-> */
-/* ext_wire[ports_to_try[i]].node_tar[Z] */
-/* == curr_switch->ext_wire[0].node_tar[Z] */
-/* #endif */
-/* ) { */
-/* continue; */
-/* } */
-
-/* if(!used) { */
-/* port_tar = curr_switch-> */
-/* ext_wire[ports_to_try[i]].port_tar; */
-/* node_tar = curr_switch-> */
-/* ext_wire[ports_to_try[i]].node_tar; */
-
-/* next_switch = &ba_system_ptr-> */
-/* grid[node_tar[X]] */
-/* #ifdef HAVE_BG */
-/* [node_tar[Y]] */
-/* [node_tar[Z]] */
-/* #endif */
-/* .axis_switch[X]; */
-
-/* count++; */
-/* path_add->out = ports_to_try[i]; */
-/* list_push(path, path_add); */
-/* debug3("looking at this one " */
-/* "%c%c%c %d -> %c%c%c %d", */
-/* alpha_num[ba_node->coord[X]], */
-/* alpha_num[ba_node->coord[Y]], */
-/* alpha_num[ba_node->coord[Z]], */
-/* ports_to_try[i], */
-/* alpha_num[node_tar[X]], */
-/* alpha_num[node_tar[Y]], */
-/* alpha_num[node_tar[Z]], */
-/* port_tar); */
-
-/* _find_passthrough(next_switch, port_tar, nodes, */
-/* dim, count, highest_phys_x); */
-/* while((temp_switch = list_pop(path)) */
-/* != path_add){ */
-/* xfree(temp_switch); */
-/* debug3("something here 2"); */
-/* } */
-/* } */
-/* } */
-/* } */
-/* xfree(path_add); */
-/* return 0; */
-/* } */
+/*
+ * Used to tie the end of the block to the start. best_path and path
+ * should both be set up before calling this function.
+ *
+ * IN: curr_switch -
+ * IN: source_port -
+ * IN: dim -
+ * IN: count -
+ * IN: start -
+ *
+ * RET: 0 on failure, 1 on success
+ *
+ * Sets up global variable best_path, and best_count. On success
+ * best_count will be >= BEST_COUNT_INIT you can call _set_best_path
+ * to apply this path to the main system (ba_system_ptr)
+ */
static int _finish_torus(ba_switch_t *curr_switch, int source_port,
- List nodes, int dim, int count, int *start)
+ int dim, int count, int *start)
{
ba_switch_t *next_switch = NULL;
- ba_path_switch_t *path_add =
- (ba_path_switch_t *) xmalloc(sizeof(ba_path_switch_t));
+ ba_path_switch_t *path_add = xmalloc(sizeof(ba_path_switch_t));
ba_path_switch_t *path_switch = NULL;
ba_path_switch_t *temp_switch = NULL;
int port_tar;
@@ -4945,18 +4557,16 @@ static int _finish_torus(ba_switch_t *curr_switch, int source_port,
target_port=1;
if(!curr_switch->int_wire[target_port].used) {
- list_destroy(best_path);
- best_path = list_create(_delete_path_list);
+ list_flush(best_path);
+
found = true;
path_add->out = target_port;
list_push(path, path_add);
itr = list_iterator_create(path);
- while((path_switch =
- (ba_path_switch_t*) list_next(itr))){
+ while((path_switch = list_next(itr))) {
- temp_switch = (ba_path_switch_t *)
- xmalloc(sizeof(ba_path_switch_t));
+ temp_switch = xmalloc(sizeof(ba_path_switch_t));
temp_switch->geometry[X] =
path_switch->geometry[X];
@@ -4986,8 +4596,7 @@ static int _finish_torus(ba_switch_t *curr_switch, int source_port,
used=0;
if(!curr_switch->int_wire[ports_to_try[i]].used) {
itr = list_iterator_create(path);
- while((path_switch =
- (ba_path_switch_t*) list_next(itr))){
+ while((path_switch = list_next(itr))){
if(((path_switch->geometry[X] == node_src[X])
#ifdef HAVE_BG
@@ -5022,8 +4631,7 @@ static int _finish_torus(ba_switch_t *curr_switch, int source_port,
node_tar = curr_switch->
ext_wire[ports_to_try[i]].node_tar;
- next_switch = &ba_system_ptr->
- grid[node_tar[X]]
+ next_switch = &ba_system_ptr->grid[node_tar[X]]
#ifdef HAVE_BG
[node_tar[Y]]
[node_tar[Z]]
@@ -5034,8 +4642,8 @@ static int _finish_torus(ba_switch_t *curr_switch, int source_port,
count++;
path_add->out = ports_to_try[i];
list_push(path, path_add);
- _finish_torus(next_switch, port_tar, nodes,
- dim, count, start);
+ _finish_torus(next_switch, port_tar,
+ dim, count, start);
while((temp_switch = list_pop(path))
!= path_add){
xfree(temp_switch);
@@ -5048,14 +4656,22 @@ static int _finish_torus(ba_switch_t *curr_switch, int source_port,
return 0;
}
+/*
+ * using best_path set up previously from _finish_torus or
+ * _find_next_free_using_port_2. Will set up the path contained there
+ * into the main virtual system. With will also set the passthrough
+ * flag if there was a passthrough used.
+ */
static int *_set_best_path()
{
ListIterator itr;
ba_path_switch_t *path_switch = NULL;
ba_switch_t *curr_switch = NULL;
int *geo = NULL;
+
if(!best_path)
return NULL;
+
itr = list_iterator_create(best_path);
while((path_switch = (ba_path_switch_t*) list_next(itr))) {
if(passthrough && path_switch->in > 1 && path_switch->out > 1) {
@@ -5070,15 +4686,13 @@ static int *_set_best_path()
path_switch->in, path_switch->out);
if(!geo)
geo = path_switch->geometry;
- curr_switch = &ba_system_ptr->
- grid
+ curr_switch = &ba_system_ptr->grid
[path_switch->geometry[X]]
[path_switch->geometry[Y]]
[path_switch->geometry[Z]].
axis_switch[path_switch->dim];
#else
- curr_switch = &ba_system_ptr->
- grid[path_switch->geometry[X]].
+ curr_switch = &ba_system_ptr->grid[path_switch->geometry[X]].
axis_switch[path_switch->dim];
#endif
diff --git a/src/plugins/select/bluegene/plugin/bg_job_place.c b/src/plugins/select/bluegene/plugin/bg_job_place.c
index a7435af07c4..f62594eede8 100644
--- a/src/plugins/select/bluegene/plugin/bg_job_place.c
+++ b/src/plugins/select/bluegene/plugin/bg_job_place.c
@@ -380,7 +380,7 @@ static bg_record_t *_find_matching_block(List block_list,
test_only);
itr = list_iterator_create(block_list);
- while ((bg_record = (bg_record_t*) list_next(itr))) {
+ while ((bg_record = list_next(itr))) {
/* If test_only we want to fall through to tell the
scheduler that it is runnable just not right now.
*/
@@ -636,6 +636,8 @@ static int _check_for_booted_overlapping_blocks(
* bg_record
*/
list_remove(bg_record_itr);
+ slurm_mutex_lock(&block_state_mutex);
+
if(bg_record->original) {
debug3("This was a copy");
found_record =
@@ -651,8 +653,10 @@ static int _check_for_booted_overlapping_blocks(
}
destroy_bg_record(bg_record);
if(!found_record) {
- error("1 this record wasn't "
- "found in the list!");
+ debug2("This record wasn't "
+ "found in the bg_list, "
+ "no big deal, it "
+ "probably wasn't added");
//rc = SLURM_ERROR;
} else {
List temp_list =
@@ -663,6 +667,7 @@ static int _check_for_booted_overlapping_blocks(
free_block_list(temp_list);
list_destroy(temp_list);
}
+ slurm_mutex_unlock(&block_state_mutex);
}
rc = 1;
diff --git a/src/plugins/select/bluegene/plugin/bluegene.c b/src/plugins/select/bluegene/plugin/bluegene.c
index d3ff7a7749d..8c2b9410510 100644
--- a/src/plugins/select/bluegene/plugin/bluegene.c
+++ b/src/plugins/select/bluegene/plugin/bluegene.c
@@ -488,7 +488,6 @@ extern bg_record_t *find_and_remove_org_from_bg_list(List my_list,
if(bit_equal(bg_record->bitmap, found_record->bitmap)
&& bit_equal(bg_record->ionode_bitmap,
found_record->ionode_bitmap)) {
-
if(!strcmp(bg_record->bg_block_id,
found_record->bg_block_id)) {
list_remove(itr);
diff --git a/src/plugins/select/bluegene/plugin/slurm_prolog.c b/src/plugins/select/bluegene/plugin/slurm_prolog.c
index af652b7879c..f4ad1d0201d 100644
--- a/src/plugins/select/bluegene/plugin/slurm_prolog.c
+++ b/src/plugins/select/bluegene/plugin/slurm_prolog.c
@@ -176,10 +176,11 @@ static int _get_job_size(uint32_t job_id)
}
/*
- * Test if any BG blocks are in deallocating state
+ * Test if any BG blocks are in deallocating state since they are
+ * probably related to this job we will want to sleep longer
* RET 1: deallocate in progress
* 0: no deallocate in progress
- * -1: error occurred
+ * -1: error occurred
*/
static int _partitions_dealloc()
{
diff --git a/src/sacct/options.c b/src/sacct/options.c
index b7c2737fc28..da6921117a1 100644
--- a/src/sacct/options.c
+++ b/src/sacct/options.c
@@ -414,7 +414,7 @@ void _help_msg(void)
"-c, --completion\n"
" Use job completion instead of accounting data.\n"
"-C, --cluster\n"
- " Only send data about this cluster.\n"
+ " Only send data about this cluster -1 for all clusters.\n"
"-d, --dump\n"
" Dump the raw data records\n"
"--duplicates\n"
@@ -622,7 +622,7 @@ void parse_command_line(int argc, char **argv)
char *dot = NULL;
bool brief_output = FALSE, long_output = FALSE;
bool all_users = 0;
-
+ bool all_clusters = 1;
static struct option long_options[] = {
{"all", 0,0, 'a'},
{"accounts", 1, 0, 'A'},
@@ -690,6 +690,10 @@ void parse_command_line(int argc, char **argv)
params.opt_completion = 1;
break;
case 'C':
+ if(!strcasecmp(optarg, "-1")) {
+ all_clusters = 1;
+ break;
+ }
if(!params.opt_cluster_list)
params.opt_cluster_list =
list_create(slurm_destroy_char);
@@ -934,7 +938,15 @@ void parse_command_line(int argc, char **argv)
}
/* specific clusters requested? */
- if (params.opt_verbose && params.opt_cluster_list
+ if(all_clusters) {
+ if(params.opt_cluster_list
+ && list_count(params.opt_cluster_list)) {
+ list_destroy(params.opt_cluster_list);
+ params.opt_cluster_list = NULL;
+ }
+ if(params.opt_verbose)
+ fprintf(stderr, "Clusters requested:\n\t: all\n");
+ } else if (params.opt_verbose && params.opt_cluster_list
&& list_count(params.opt_cluster_list)) {
fprintf(stderr, "Clusters requested:\n");
itr = list_iterator_create(params.opt_cluster_list);
diff --git a/src/sacctmgr/user_functions.c b/src/sacctmgr/user_functions.c
index d87cbb8f55d..66d4289610c 100644
--- a/src/sacctmgr/user_functions.c
+++ b/src/sacctmgr/user_functions.c
@@ -37,6 +37,7 @@
\*****************************************************************************/
#include "src/sacctmgr/sacctmgr.h"
+#include "src/common/uid.h"
static int _set_cond(int *start, int argc, char *argv[],
acct_user_cond_t *user_cond,
@@ -504,6 +505,7 @@ extern int sacctmgr_add_user(int argc, char *argv[])
while((name = list_next(itr))) {
user = NULL;
if(!sacctmgr_find_user_from_list(local_user_list, name)) {
+ uid_t pw_uid;
if(!default_acct) {
exit_code=1;
fprintf(stderr, " Need a default account for "
@@ -524,6 +526,22 @@ extern int sacctmgr_add_user(int argc, char *argv[])
}
first = 0;
}
+ pw_uid = uid_from_string(name);
+ if(pw_uid == (uid_t) -1) {
+ char *warning = xstrdup_printf(
+ "There is no uid for user '%s'"
+ "\nAre you sure you want to continue?",
+ name);
+
+ if(!commit_check(warning)) {
+ xfree(warning);
+ rc = SLURM_ERROR;
+ list_flush(user_list);
+ goto no_default;
+ }
+ xfree(warning);
+ }
+
user = xmalloc(sizeof(acct_user_rec_t));
user->assoc_list = list_create(NULL);
user->name = xstrdup(name);
@@ -543,6 +561,7 @@ extern int sacctmgr_add_user(int argc, char *argv[])
}
user->admin_level = admin_level;
+
xstrfmtcat(user_str, " %s\n", name);
list_append(user_list, user);
diff --git a/src/salloc/Makefile.am b/src/salloc/Makefile.am
index e2da3019f5c..05fb0f3eb54 100644
--- a/src/salloc/Makefile.am
+++ b/src/salloc/Makefile.am
@@ -3,7 +3,7 @@
AUTOMAKE_OPTIONS = foreign
CLEANFILES = core.*
-INCLUDES = -I$(top_srcdir)
+INCLUDES = -I$(top_srcdir) $(BG_INCLUDES)
bin_PROGRAMS = salloc
diff --git a/src/salloc/Makefile.in b/src/salloc/Makefile.in
index 91ae04a4039..03315a66ab0 100644
--- a/src/salloc/Makefile.in
+++ b/src/salloc/Makefile.in
@@ -263,7 +263,7 @@ top_builddir = @top_builddir@
top_srcdir = @top_srcdir@
AUTOMAKE_OPTIONS = foreign
CLEANFILES = core.*
-INCLUDES = -I$(top_srcdir)
+INCLUDES = -I$(top_srcdir) $(BG_INCLUDES)
salloc_SOURCES = salloc.c salloc.h opt.c opt.h
convenience_libs = $(top_builddir)/src/api/libslurm.o -ldl
salloc_LDADD = \
diff --git a/src/salloc/salloc.c b/src/salloc/salloc.c
index de913220ffd..c1a48258353 100644
--- a/src/salloc/salloc.c
+++ b/src/salloc/salloc.c
@@ -50,6 +50,14 @@
#include "src/salloc/salloc.h"
#include "src/salloc/opt.h"
+#ifdef HAVE_BG
+#include "src/api/job_info.h"
+#include "src/api/node_select_info.h"
+#include "src/common/node_select.h"
+#include "src/plugins/select/bluegene/plugin/bg_boot_time.h"
+#include "src/plugins/select/bluegene/wrap_rm_api.h"
+#endif
+
#define MAX_RETRIES 3
char **command_argv;
@@ -76,6 +84,16 @@ static void _user_msg_handler(srun_user_msg_t *msg);
static void _ping_handler(srun_ping_msg_t *msg);
static void _node_fail_handler(srun_node_fail_msg_t *msg);
+#ifdef HAVE_BG
+
+#define POLL_SLEEP 3 /* retry interval in seconds */
+
+static int _wait_bluegene_block_ready(
+ resource_allocation_response_msg_t *alloc);
+static int _blocks_dealloc();
+#endif
+
+
int main(int argc, char *argv[])
{
log_options_t logopt = LOG_OPTS_STDERR_ONLY;
@@ -187,6 +205,13 @@ int main(int argc, char *argv[])
* Allocation granted!
*/
info("Granted job allocation %d", alloc->job_id);
+#ifdef HAVE_BG
+ if (!_wait_bluegene_block_ready(alloc)) {
+ error("Something is wrong with the boot of the block.");
+ goto relinquish;
+ }
+
+#endif
if (opt.bell == BELL_ALWAYS
|| (opt.bell == BELL_AFTER_DELAY
&& ((after - before) > DEFAULT_BELL_DELAY))) {
@@ -520,3 +545,93 @@ static void _node_fail_handler(srun_node_fail_msg_t *msg)
{
error("Node failure on %s", msg->nodelist);
}
+
+#ifdef HAVE_BG
+/* returns 1 if job and nodes are ready for job to begin, 0 otherwise */
+static int _wait_bluegene_block_ready(resource_allocation_response_msg_t *alloc)
+{
+ int is_ready = 0, i, rc;
+ char *block_id = NULL;
+ int cur_delay = 0;
+ int max_delay = BG_FREE_PREVIOUS_BLOCK + BG_MIN_BLOCK_BOOT +
+ (BG_INCR_BLOCK_BOOT * alloc->node_cnt);
+
+ select_g_get_jobinfo(alloc->select_jobinfo, SELECT_DATA_BLOCK_ID,
+ &block_id);
+
+ for (i=0; (cur_delay < max_delay); i++) {
+ if(i == 1)
+ info("Waiting for block %s to become ready for job",
+ block_id);
+ if (i) {
+ sleep(POLL_SLEEP);
+ rc = _blocks_dealloc();
+ if ((rc == 0) || (rc == -1))
+ cur_delay += POLL_SLEEP;
+ debug("still waiting");
+ }
+
+ rc = slurm_job_node_ready(alloc->job_id);
+
+ if (rc == READY_JOB_FATAL)
+ break; /* fatal error */
+ if (rc == READY_JOB_ERROR) /* error */
+ continue; /* retry */
+ if ((rc & READY_JOB_STATE) == 0) /* job killed */
+ break;
+ if (rc & READY_NODE_STATE) { /* job and node ready */
+ is_ready = 1;
+ break;
+ }
+ }
+
+ if (is_ready)
+ info("Block %s is ready for job", block_id);
+ else
+ error("Block %s still not ready", block_id);
+ xfree(block_id);
+
+ return is_ready;
+}
+
+/*
+ * Test if any BG blocks are in deallocating state since they are
+ * probably related to this job we will want to sleep longer
+ * RET 1: deallocate in progress
+ * 0: no deallocate in progress
+ * -1: error occurred
+ */
+static int _blocks_dealloc()
+{
+ static node_select_info_msg_t *bg_info_ptr = NULL, *new_bg_ptr = NULL;
+ int rc = 0, error_code = 0, i;
+
+ if (bg_info_ptr) {
+ error_code = slurm_load_node_select(bg_info_ptr->last_update,
+ &new_bg_ptr);
+ if (error_code == SLURM_SUCCESS)
+ select_g_free_node_info(&bg_info_ptr);
+ else if (slurm_get_errno() == SLURM_NO_CHANGE_IN_DATA) {
+ error_code = SLURM_SUCCESS;
+ new_bg_ptr = bg_info_ptr;
+ }
+ } else {
+ error_code = slurm_load_node_select((time_t) NULL, &new_bg_ptr);
+ }
+
+ if (error_code) {
+ error("slurm_load_partitions: %s\n",
+ slurm_strerror(slurm_get_errno()));
+ return -1;
+ }
+ for (i=0; i<new_bg_ptr->record_count; i++) {
+ if(new_bg_ptr->bg_info_array[i].state
+ == RM_PARTITION_DEALLOCATING) {
+ rc = 1;
+ break;
+ }
+ }
+ bg_info_ptr = new_bg_ptr;
+ return rc;
+}
+#endif
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 165d92c96f2..ff9a8313348 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -304,8 +304,13 @@ int main(int argc, char *argv[])
slurmctld_cluster_name = xstrdup(slurmctld_conf.cluster_name);
accounting_enforce = slurmctld_conf.accounting_storage_enforce;
acct_db_conn = acct_storage_g_get_connection(true, false);
+
+ memset(&assoc_init_arg, 0, sizeof(assoc_init_args_t));
assoc_init_arg.enforce = accounting_enforce;
assoc_init_arg.remove_assoc_notify = _remove_assoc;
+ assoc_init_arg.refresh = 0;
+ assoc_init_arg.cache_level = ASSOC_MGR_CACHE_ALL;
+
if (assoc_mgr_init(acct_db_conn, &assoc_init_arg) &&
accounting_enforce) {
error("assoc_mgr_init failure");
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index b45b3d9e2e9..8dd391c7a75 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -1878,6 +1878,7 @@ extern int job_complete(uint32_t job_id, uid_t uid, bool requeue,
xfree(job_ptr->state_desc);
} else
job_ptr->job_state = JOB_COMPLETE | job_comp_flag;
+
if (suspended) {
job_ptr->end_time = job_ptr->suspend_time;
job_ptr->tot_sus_time +=
@@ -4320,55 +4321,24 @@ int update_job(job_desc_msg_t * job_specs, uid_t uid)
}
if (job_specs->account) {
- if ((!IS_JOB_PENDING(job_ptr)) || (detail_ptr == NULL)) {
- info("update_job: attempt to modify account for "
- "non-pending job_id %u", job_specs->job_id);
- error_code = ESLURM_DISABLED;
- } else {
- acct_association_rec_t assoc_rec, *assoc_ptr;
- bzero(&assoc_rec, sizeof(acct_association_rec_t));
-
- assoc_rec.uid = job_ptr->user_id;
- assoc_rec.partition = job_ptr->partition;
- assoc_rec.acct = job_specs->account;
- if (assoc_mgr_fill_in_assoc(acct_db_conn, &assoc_rec,
- accounting_enforce,
- &assoc_ptr)) {
- info("job_update: invalid account %s for "
- "job_id %u",
- job_specs->account, job_ptr->job_id);
- error_code = ESLURM_INVALID_ACCOUNT;
- } else {
- xfree(job_ptr->account);
- if (assoc_rec.acct[0] != '\0') {
- job_ptr->account =
- xstrdup(assoc_rec.acct);
- info("update_job: setting account to "
- "%s for job_id %u",
- assoc_rec.acct, job_ptr->job_id);
- } else {
- info("update_job: cleared account for "
- "job_id %u",
- job_specs->job_id);
- }
- job_ptr->assoc_id = assoc_rec.id;
- job_ptr->assoc_ptr = (void *) assoc_ptr;
- }
- }
+ int rc = update_job_account("update_job", job_ptr,
+ job_specs->account);
+ if (rc != SLURM_SUCCESS)
+ error_code = rc;
}
if (job_specs->ntasks_per_node != (uint16_t) NO_VAL) {
if ((!IS_JOB_PENDING(job_ptr)) || (detail_ptr == NULL))
error_code = ESLURM_DISABLED;
else if (super_user) {
- detail_ptr->ntasks_per_node =
- job_specs->ntasks_per_node;
+ detail_ptr->ntasks_per_node = job_specs->
+ ntasks_per_node;
info("update_job: setting ntasks_per_node to %u for "
"job_id %u", job_specs->ntasks_per_node,
job_specs->job_id);
} else {
- error("Not super user: setting ntasks_oper_node to job %u",
- job_specs->job_id);
+ error("Not super user: setting ntasks_oper_node to "
+ "job %u", job_specs->job_id);
error_code = ESLURM_ACCESS_DENIED;
}
}
@@ -5077,6 +5047,16 @@ extern void job_completion_logger(struct job_record *job_ptr)
}
g_slurm_jobcomp_write(job_ptr);
+
+ /*
+ * This means the job wasn't ever eligible, but we want to
+ * keep track of all jobs, so we will set the db_inx to
+ * INFINITE and the database will understand what happened.
+ */
+ if(!job_ptr->nodes && !job_ptr->db_index) {
+ jobacct_storage_g_job_start(acct_db_conn, job_ptr);
+ }
+
jobacct_storage_g_job_complete(acct_db_conn, job_ptr);
}
@@ -5686,7 +5666,8 @@ extern int job_cancel_by_assoc_id(uint32_t assoc_id)
if ((job_ptr->assoc_id != assoc_id) ||
IS_JOB_FINISHED(job_ptr))
continue;
- info("Association deleted, cancelling job %u", job_ptr->job_id);
+ info("Association deleted, cancelling job %u",
+ job_ptr->job_id);
job_signal(job_ptr->job_id, SIGKILL, 0, 0);
job_ptr->state_reason = FAIL_BANK_ACCOUNT;
xfree(job_ptr->state_desc);
@@ -5695,3 +5676,55 @@ extern int job_cancel_by_assoc_id(uint32_t assoc_id)
list_iterator_destroy(job_iterator);
return cnt;
}
+
+/*
+ * Modify the account associated with a pending job
+ * IN module - where this is called from
+ * IN job_ptr - pointer to job which should be modified
+ * IN new_account - desired account name
+ * RET SLURM_SUCCESS or error code
+ */
+extern int update_job_account(char *module, struct job_record *job_ptr,
+ char *new_account)
+{
+ acct_association_rec_t assoc_rec, *assoc_ptr;
+
+ if ((!IS_JOB_PENDING(job_ptr)) || (job_ptr->details == NULL)) {
+ info("%s: attempt to modify account for non-pending "
+ "job_id %u", module, job_ptr->job_id);
+ return ESLURM_DISABLED;
+ }
+
+
+ bzero(&assoc_rec, sizeof(acct_association_rec_t));
+ assoc_rec.uid = job_ptr->user_id;
+ assoc_rec.partition = job_ptr->partition;
+ assoc_rec.acct = new_account;
+ if (assoc_mgr_fill_in_assoc(acct_db_conn, &assoc_rec,
+ accounting_enforce, &assoc_ptr)) {
+ info("%s: invalid account %s for job_id %u",
+ module, new_account, job_ptr->job_id);
+ return ESLURM_INVALID_ACCOUNT;
+ }
+
+
+ xfree(job_ptr->account);
+ if (assoc_rec.acct[0] != '\0') {
+ job_ptr->account = xstrdup(assoc_rec.acct);
+ info("%s: setting account to %s for job_id %u",
+ module, assoc_rec.acct, job_ptr->job_id);
+ } else {
+ info("%s: cleared account for job_id %u",
+ module, job_ptr->job_id);
+ }
+ job_ptr->assoc_id = assoc_rec.id;
+ job_ptr->assoc_ptr = (void *) assoc_ptr;
+
+ if (job_ptr->details && job_ptr->details->begin_time) {
+ /* Update account associated with the eligible time */
+ jobacct_storage_g_job_start(acct_db_conn, job_ptr);
+ }
+ last_job_update = time(NULL);
+
+ return SLURM_SUCCESS;
+}
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index 713f98a0b34..7d36219eba6 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -548,7 +548,11 @@ extern int make_batch_job_cred(batch_job_launch_msg_t *launch_msg_ptr,
cred_arg.jobid = launch_msg_ptr->job_id;
cred_arg.stepid = launch_msg_ptr->step_id;
cred_arg.uid = launch_msg_ptr->uid;
+#ifdef HAVE_FRONT_END
+ cred_arg.hostlist = node_record_table_ptr[0].name;
+#else
cred_arg.hostlist = launch_msg_ptr->nodes;
+#endif
if (job_ptr->details == NULL)
cred_arg.job_mem = 0;
else if (job_ptr->details->job_min_memory & MEM_PER_CPU) {
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 6a977229a4d..6f4be965c34 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -519,8 +519,11 @@ static int _make_step_cred(struct step_record *step_rec,
cred_arg.stepid = step_rec->step_id;
cred_arg.uid = job_ptr->user_id;
cred_arg.job_mem = job_ptr->details->job_min_memory;
+#ifdef HAVE_FRONT_END
+ cred_arg.hostlist = node_record_table_ptr[0].name;
+#else
cred_arg.hostlist = step_rec->step_layout->node_list;
-
+#endif
cred_arg.alloc_lps_cnt = job_ptr->alloc_lps_cnt;
if ((cred_arg.alloc_lps_cnt > 0) &&
bit_equal(job_ptr->node_bitmap, step_rec->step_node_bitmap)) {
@@ -554,8 +557,7 @@ static int _make_step_cred(struct step_record *step_rec,
cred_arg.alloc_lps = NULL;
}
- *slurm_cred = slurm_cred_create(slurmctld_config.cred_ctx,
- &cred_arg);
+ *slurm_cred = slurm_cred_create(slurmctld_config.cred_ctx, &cred_arg);
xfree(cred_arg.alloc_lps);
if (*slurm_cred == NULL) {
error("slurm_cred_create error");
@@ -1742,7 +1744,8 @@ static void _slurm_rpc_shutdown_controller_immediate(slurm_msg_t * msg)
* represent the termination of an entire job */
static void _slurm_rpc_step_complete(slurm_msg_t *msg)
{
- int error_code = SLURM_SUCCESS, rc, rem, step_rc;
+ int error_code = SLURM_SUCCESS, rc, rem;
+ uint32_t step_rc;
DEF_TIMERS;
step_complete_msg_t *req = (step_complete_msg_t *)msg->data;
/* Locks: Write job, write node */
@@ -1801,7 +1804,7 @@ static void _slurm_rpc_step_complete(slurm_msg_t *msg)
}
} else {
error_code = job_step_complete(req->job_id, req->job_step_id,
- uid, job_requeue, step_rc);
+ uid, job_requeue, step_rc);
unlock_slurmctld(job_write_lock);
END_TIMER2("_slurm_rpc_step_complete");
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index dedbe812713..3895e6fc064 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -1416,7 +1416,7 @@ extern bool step_on_node(struct job_record *job_ptr,
* RET 0 on success, otherwise ESLURM error code
*/
extern int step_partial_comp(step_complete_msg_t *req, int *rem,
- int *max_rc);
+ uint32_t *max_rc);
/* Update time stamps for job step suspend */
extern void suspend_job_step(struct job_record *job_ptr);
@@ -1438,6 +1438,16 @@ extern int sync_job_files(void);
*/
extern int update_job (job_desc_msg_t * job_specs, uid_t uid);
+/*
+ * Modify the account associated with a pending job
+ * IN module - where this is called from
+ * IN job_ptr - pointer to job which should be modified
+ * IN new_account - desired account name
+ * RET SLURM_SUCCESS or error code
+ */
+extern int update_job_account(char *module, struct job_record *job_ptr,
+ char *new_account);
+
/* Reset nodes_completing field for all jobs */
extern void update_job_nodes_completing(void);
diff --git a/src/slurmctld/step_mgr.c b/src/slurmctld/step_mgr.c
index 5c2c459dbfb..952306bb0ac 100644
--- a/src/slurmctld/step_mgr.c
+++ b/src/slurmctld/step_mgr.c
@@ -1468,7 +1468,7 @@ extern int job_step_checkpoint_task_comp(checkpoint_task_comp_msg_t *ckpt_ptr,
* RET 0 on success, otherwise ESLURM error code
*/
extern int step_partial_comp(step_complete_msg_t *req, int *rem,
- int *max_rc)
+ uint32_t *max_rc)
{
struct job_record *job_ptr;
struct step_record *step_ptr;
diff --git a/src/slurmd/slurmd/slurmd.c b/src/slurmd/slurmd/slurmd.c
index e02697f002d..2863a5eeebf 100644
--- a/src/slurmd/slurmd/slurmd.c
+++ b/src/slurmd/slurmd/slurmd.c
@@ -589,9 +589,11 @@ _read_config()
* valid aliases */
if (conf->node_name == NULL)
conf->node_name = slurm_conf_get_aliased_nodename();
- if (conf->node_name == NULL)
+
+ if (conf->node_name == NULL)
conf->node_name = slurm_conf_get_nodename("localhost");
- if (conf->node_name == NULL)
+
+ if (conf->node_name == NULL)
fatal("Unable to determine this slurmd's NodeName");
_massage_pathname(&conf->logfile);
diff --git a/src/slurmd/slurmstepd/mgr.c b/src/slurmd/slurmstepd/mgr.c
index b3a9378be46..4198bdad34a 100644
--- a/src/slurmd/slurmstepd/mgr.c
+++ b/src/slurmd/slurmstepd/mgr.c
@@ -233,9 +233,25 @@ static void
_batch_finish(slurmd_job_t *job, int rc)
{
int i;
- for (i = 0; i < job->ntasks; i++)
+ for (i = 0; i < job->ntasks; i++) {
+ /* If signalled we only need to check one and then
+ break out of the loop */
+ if(WIFSIGNALED(job->task[i]->estatus)) {
+ switch(WTERMSIG(job->task[i]->estatus)) {
+ case SIGTERM:
+ case SIGKILL:
+ case SIGINT:
+ step_complete.step_rc = NO_VAL;
+ break;
+ default:
+ step_complete.step_rc = job->task[i]->estatus;
+ break;
+ }
+ break;
+ }
step_complete.step_rc = MAX(step_complete.step_rc,
WEXITSTATUS(job->task[i]->estatus));
+ }
if (job->argv[0] && (unlink(job->argv[0]) < 0))
error("unlink(%s): %m", job->argv[0]);
@@ -494,10 +510,25 @@ _wait_for_children_slurmstepd(slurmd_job_t *job)
}
/* Find the maximum task return code */
- for (i = 0; i < job->ntasks; i++)
+ for (i = 0; i < job->ntasks; i++) {
+ /* If signalled we only need to check one and then
+ break out of the loop */
+ if(WIFSIGNALED(job->task[i]->estatus)) {
+ switch(WTERMSIG(job->task[i]->estatus)) {
+ case SIGTERM:
+ case SIGKILL:
+ case SIGINT:
+ step_complete.step_rc = NO_VAL;
+ break;
+ default:
+ step_complete.step_rc = job->task[i]->estatus;
+ break;
+ }
+ break;
+ }
step_complete.step_rc = MAX(step_complete.step_rc,
- WEXITSTATUS(job->task[i]->estatus));
-
+ WEXITSTATUS(job->task[i]->estatus));
+ }
step_complete.wait_children = false;
pthread_mutex_unlock(&step_complete.lock);
diff --git a/src/slurmd/slurmstepd/req.c b/src/slurmd/slurmstepd/req.c
index ebfbd430ac8..65f324d457c 100644
--- a/src/slurmd/slurmstepd/req.c
+++ b/src/slurmd/slurmstepd/req.c
@@ -174,10 +174,16 @@ _domain_socket_create(const char *dir, const char *nodename,
* First check to see if the named socket already exists.
*/
if (stat(name, &stat_buf) == 0) {
- error("Socket %s already exists", name);
- xfree(name);
- errno = ESLURMD_STEP_EXISTS;
- return -1;
+ /* Vestigial from a slurmd crash or job requeue that did not
+ * happen properly (very rare conditions). Try another name */
+ xstrcat(name, ".ALT");
+ if (stat(name, &stat_buf) == 0) {
+ error("Socket %s already exists", name);
+ xfree(name);
+ errno = ESLURMD_STEP_EXISTS;
+ return -1;
+ }
+ error("Using alternate socket name %s", name);
}
fd = _create_socket(name);
diff --git a/src/slurmdbd/slurmdbd.c b/src/slurmdbd/slurmdbd.c
index ae2a361ca39..eb2580e48f7 100644
--- a/src/slurmdbd/slurmdbd.c
+++ b/src/slurmdbd/slurmdbd.c
@@ -97,6 +97,7 @@ int main(int argc, char *argv[])
pthread_attr_t thread_attr;
char node_name[128];
void *db_conn = NULL;
+ assoc_init_args_t assoc_init_arg;
_init_config();
log_init(argv[0], log_opts, LOG_DAEMON, NULL);
@@ -139,7 +140,10 @@ int main(int argc, char *argv[])
db_conn = acct_storage_g_get_connection(false, false);
- if(assoc_mgr_init(db_conn, NULL) == SLURM_ERROR) {
+ memset(&assoc_init_arg, 0, sizeof(assoc_init_args_t));
+ assoc_init_arg.cache_level = ASSOC_MGR_CACHE_USER;
+
+ if(assoc_mgr_init(db_conn, &assoc_init_arg) == SLURM_ERROR) {
error("Problem getting cache of data");
acct_storage_g_close_connection(&db_conn);
goto end_it;
diff --git a/src/smap/configure_functions.c b/src/smap/configure_functions.c
index f899b0c3d71..80735294979 100644
--- a/src/smap/configure_functions.c
+++ b/src/smap/configure_functions.c
@@ -1276,7 +1276,7 @@ static void _print_text_command(allocated_block_t *allocated_block)
main_xcord += 7;
mvwprintw(text_win, main_ycord,
- main_xcord, "%d",allocated_block->request->size);
+ main_xcord, "%d", allocated_block->request->size);
main_xcord += 10;
if(allocated_block->request->conn_type == SELECT_SMALL) {
diff --git a/src/srun/allocate.c b/src/srun/allocate.c
index 5b5c53bd55b..5faf12f08b3 100644
--- a/src/srun/allocate.c
+++ b/src/srun/allocate.c
@@ -96,7 +96,7 @@ static void _signal_while_allocating(int signo)
{
destroy_job = 1;
if (pending_job_id != 0) {
- slurm_complete_job(pending_job_id, 0);
+ slurm_complete_job(pending_job_id, NO_VAL);
}
}
diff --git a/src/srun/srun.c b/src/srun/srun.c
index 1f9247d8a9c..d82a554d358 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -110,7 +110,7 @@
mpi_plugin_client_info_t mpi_job_info[1];
static struct termios termdefaults;
-int global_rc;
+uint32_t global_rc = 0;
srun_job_t *job = NULL;
struct {
@@ -250,7 +250,6 @@ int srun(int ac, char **av)
if (!job || create_job_step(job) < 0)
exit(1);
} else {
- got_alloc = 1;
/* Combined job allocation and job step launch */
#ifdef HAVE_FRONT_END
uid_t my_uid = getuid();
@@ -263,9 +262,11 @@ int srun(int ac, char **av)
if ( !(resp = allocate_nodes()) )
exit(1);
+ got_alloc = 1;
_print_job_information(resp);
_set_cpu_env_var(resp);
job = job_create_allocation(resp);
+
opt.exclusive = false; /* not applicable for this step */
if (!job || create_job_step(job) < 0) {
slurm_complete_job(job->jobid, 1);
@@ -433,7 +434,7 @@ cleanup:
_task_state_struct_free();
log_fini();
- return global_rc;
+ return (int)global_rc;
}
static int _call_spank_local_user (srun_job_t *job)
@@ -953,7 +954,7 @@ _task_finish(task_exit_msg_t *msg)
char buf[2048], *core_str = "", *msg_str, *node_list = NULL;
static bool first_done = true;
static bool first_error = true;
- int rc = 0;
+ uint32_t rc = 0;
int i;
verbose("%u tasks finished (rc=%u)",
@@ -975,7 +976,6 @@ _task_finish(task_exit_msg_t *msg)
}
} else if (WIFSIGNALED(msg->return_code)) {
bit_or(task_state.finish_abnormal, tasks_exited);
- rc = 1;
msg_str = strsignal(WTERMSIG(msg->return_code));
#ifdef WCOREDUMP
if (WCOREDUMP(msg->return_code))
@@ -983,9 +983,11 @@ _task_finish(task_exit_msg_t *msg)
#endif
node_list = _taskids_to_nodelist(tasks_exited);
if (job->state >= SRUN_JOB_CANCELLED) {
+ rc = NO_VAL;
verbose("%s: task %s: %s%s",
node_list, buf, msg_str, core_str);
} else {
+ rc = msg->return_code;
error("%s: task %s: %s%s",
node_list, buf, msg_str, core_str);
}
diff --git a/testsuite/expect/globals b/testsuite/expect/globals
index 0126835ee00..a9003d3dfbc 100755
--- a/testsuite/expect/globals
+++ b/testsuite/expect/globals
@@ -1134,3 +1134,54 @@ proc check_acct_associations { } {
log_user 1
return $rc
}
+
+################################################################
+#
+#
+#
+################################################################
+proc check_accounting_admin_level { } {
+ global sacctmgr alpha alpha_numeric_under bin_id
+
+ set admin_level ""
+
+ log_user 0
+
+ spawn $bin_id -u -n
+ expect {
+ -re "($alpha_numeric_under)" {
+ set user_name $expect_out(1,string)
+ exp_continue
+ }
+ eof {
+ wait
+ }
+ }
+
+ if { ![string length $user_name] } {
+ send_user "FAILURE: No name returned from id\n"
+ return ""
+ }
+
+ #
+ # Use sacctmgr to check admin_level
+ #
+ set s_pid [spawn $sacctmgr -n -P list user $user_name format=admin]
+ expect {
+ -re "($alpha)" {
+ set admin_level $expect_out(1,string)
+ exp_continue
+ }
+ timeout {
+ send_user "FAILURE: sacctmgr add not responding\n"
+ slow_kill $s_pid
+ set exit_code 1
+ }
+ eof {
+ wait
+ }
+ }
+
+ log_user 1
+ return $admin_level
+}
diff --git a/testsuite/expect/test21.10 b/testsuite/expect/test21.10
index e40d6a52d9b..f74f24bb17b 100755
--- a/testsuite/expect/test21.10
+++ b/testsuite/expect/test21.10
@@ -70,11 +70,138 @@ set access_err 0
print_header $test_id
+#
+# Check accounting config and bail if not found.
+#
if { [test_account_storage] == 0 } {
send_user "\nWARNING: This test can't be run without a usable AccountStorageType\n"
exit 0
}
-
+
+if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
+#
+# Use sacctmgr to delete the test cluster
+#
+ set nothing 0
+ set matches 0
+
+set sadel_pid [spawn $sacctmgr -i $del $clu $tc1]
+
+ expect {
+ -re "privilege to perform this action" {
+ set access_err 1
+ exp_continue
+ }
+ -re "(There was a problem|Unknown condition|Bad format on|Bad MaxWall|Unknown option)" {
+ send_user "FAILURE: there was a problem with the sacctmgr command\n"
+ incr exit_code 1
+ }
+ -re "Problem getting" {
+ send_user "FAILURE: there was a problem getting information from the database\n"
+ incr exit_code 1
+ }
+ -re "Problem adding" {
+ send_user "FAILURE: there was an unknwon problem\n"
+ incr exit_code 1
+ }
+ -re "No associations" {
+ send_user "FAILURE: your command didn't return anything\n"
+ incr exit_code 1
+ }
+ -re "Deleting clusters" {
+ incr matches
+ exp_continue
+ }
+ -re " Nothing deleted" {
+ incr matches
+ set nothing 1
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: sacctmgr delete not responding\n"
+ slow_kill $my_pid
+ incr exit_code 1
+ }
+ eof {
+ wait
+ }
+ }
+ if {$access_err != 0} {
+ return 1
+ }
+ if {$matches != 1} {
+ send_user "\nFAILURE: sacctmgr had a problem deleting cluster got $matches\n"
+ incr exit_code 1
+ }
+ if { !$nothing } {
+ if { ![check_acct_associations] } {
+ send_user "\nFAILURE: Our associations don't line up\n"
+ incr exit_code 1
+ }
+ }
+
+#
+# Use sacctmgr to remove an account
+#
+
+ set matches 0
+ set nothing 1
+ set check "Deleting account"
+
+ set my_pid [eval spawn $sacctmgr -i delete account $nm1]
+ expect {
+ -re "(There was a problem|Unknown condition|Bad format on|Bad MaxWall|Unknown option)" {
+ send_user "FAILURE: there was a problem with the sacctmgr command\n"
+ incr exit_code 1
+ }
+ -re "Problem getting" {
+ send_user "FAILURE: there was a problem getting information from the database\n"
+ incr exit_code 1
+ }
+ -re "Problem adding" {
+ send_user "FAILURE: there was an unknwon problem\n"
+ incr exit_code 1
+ }
+ -re "No associations" {
+ send_user "FAILURE: your command didn't return anything\n"
+ incr exit_code 1
+ }
+ -re "$check" {
+ incr matches
+ exp_continue
+ }
+ -re " Nothing deleted" {
+ incr matches
+ set nothing 1
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: sacctmgr add not responding\n"
+ slow_kill $my_pid
+ incr exit_code 1
+ }
+ eof {
+ wait
+ }
+ }
+
+ if {$matches != 1} {
+ send_user "\nFAILURE: sacctmgr had a problem deleting account.
+ got $matches\n"
+ incr exit_code 1
+ }
+
+ if { !$nothing } {
+ if { ![check_acct_associations] } {
+ send_user "\nFAILURE: Our associations don't line up\n"
+ incr exit_code 1
+ }
+ }
+
#
# Use sacctmgr to create a cluster
#
diff --git a/testsuite/expect/test21.11 b/testsuite/expect/test21.11
index 3411f0b80fa..99571a5925d 100755
--- a/testsuite/expect/test21.11
+++ b/testsuite/expect/test21.11
@@ -78,6 +78,14 @@ if { [test_account_storage] == 0 } {
exit 0
}
+#
+# Verify if Administrator privileges
+#
+ if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to create a cluster
#
diff --git a/testsuite/expect/test21.12 b/testsuite/expect/test21.12
index a7a363bbe37..ed897c9542b 100755
--- a/testsuite/expect/test21.12
+++ b/testsuite/expect/test21.12
@@ -79,6 +79,14 @@ if { [test_account_storage] == 0 } {
exit 0
}
+#
+# Verify if Administrator privileges
+#
+ if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to create a cluster
#
diff --git a/testsuite/expect/test21.13 b/testsuite/expect/test21.13
index c6b9074282d..4da00c44013 100755
--- a/testsuite/expect/test21.13
+++ b/testsuite/expect/test21.13
@@ -74,6 +74,13 @@ if { [test_account_storage] == 0 } {
exit 0
}
+#
+# Verify if Administrator privileges
+#
+ if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
#
# Use sacctmgr to create a cluster
diff --git a/testsuite/expect/test21.14 b/testsuite/expect/test21.14
index b73c438cc4b..c6b138ebeb0 100755
--- a/testsuite/expect/test21.14
+++ b/testsuite/expect/test21.14
@@ -75,6 +75,11 @@ if { [test_account_storage] == 0 } {
exit 0
}
+if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to create a cluster
diff --git a/testsuite/expect/test21.15 b/testsuite/expect/test21.15
index 529e24a4cda..c6a7f0eeb68 100755
--- a/testsuite/expect/test21.15
+++ b/testsuite/expect/test21.15
@@ -102,6 +102,10 @@ if { [test_account_storage] == 0 } {
exit 0
}
+if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
#
# Use sacctmgr to create a cluster
diff --git a/testsuite/expect/test21.16 b/testsuite/expect/test21.16
index 5d318f87e60..4c8dfc49d6d 100755
--- a/testsuite/expect/test21.16
+++ b/testsuite/expect/test21.16
@@ -88,7 +88,7 @@ set mn maxnode
set mw maxwall
set dbu debug
set access_err 0
-
+#set user_name "id -u -n"
print_header $test_id
@@ -102,6 +102,10 @@ if { [test_account_storage] == 0 } {
exit 0
}
+if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
#
# Use sacctmgr to create a cluster
diff --git a/testsuite/expect/test21.17 b/testsuite/expect/test21.17
index e7376f3debc..21d001a5d1e 100755
--- a/testsuite/expect/test21.17
+++ b/testsuite/expect/test21.17
@@ -102,6 +102,11 @@ if { [test_account_storage] == 0 } {
exit 0
}
+if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to create a cluster
diff --git a/testsuite/expect/test21.18 b/testsuite/expect/test21.18
index 84662a10b0d..20c848f138f 100755
--- a/testsuite/expect/test21.18
+++ b/testsuite/expect/test21.18
@@ -102,6 +102,11 @@ if { [test_account_storage] == 0 } {
exit 0
}
+if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to create a cluster
diff --git a/testsuite/expect/test21.19 b/testsuite/expect/test21.19
index 4ef4e677330..2d62d68384b 100755
--- a/testsuite/expect/test21.19
+++ b/testsuite/expect/test21.19
@@ -102,6 +102,11 @@ if { [test_account_storage] == 0 } {
exit 0
}
+if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to create a cluster
diff --git a/testsuite/expect/test21.5 b/testsuite/expect/test21.5
index 4110ad1184a..5daa909f2de 100755
--- a/testsuite/expect/test21.5
+++ b/testsuite/expect/test21.5
@@ -67,6 +67,11 @@ if { [test_account_storage] == 0 } {
exit 0
}
+if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to delete the test cluster
#
diff --git a/testsuite/expect/test21.6 b/testsuite/expect/test21.6
index 3bfb62ac521..ac83876bc07 100755
--- a/testsuite/expect/test21.6
+++ b/testsuite/expect/test21.6
@@ -61,11 +61,80 @@ set access_err 0
print_header $test_id
+#
+# Check accounting config and bail if not found.
+#
if { [test_account_storage] == 0 } {
send_user "\nWARNING: This test can't be run without a usable AccountStorageType\n"
exit 0
}
-
+
+if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
+#
+# Use sacctmgr to delete the test cluster
+#
+ set nothing 0
+ set matches 0
+
+set sadel_pid [spawn $sacctmgr -i $del $clu $tc1,$tc2,$tc3]
+
+ expect {
+ -re "privilege to perform this action" {
+ set access_err 1
+ exp_continue
+ }
+ -re "(There was a problem|Unknown condition|Bad format on|Bad MaxWall|Unknown option)" {
+ send_user "FAILURE: there was a problem with the sacctmgr command\n"
+ incr exit_code 1
+ }
+ -re "Problem getting" {
+ send_user "FAILURE: there was a problem getting information from the database\n"
+ incr exit_code 1
+ }
+ -re "Problem adding" {
+ send_user "FAILURE: there was an unknwon problem\n"
+ incr exit_code 1
+ }
+ -re "No associations" {
+ send_user "FAILURE: your command didn't return anything\n"
+ incr exit_code 1
+ }
+ -re "Deleting clusters" {
+ incr matches
+ exp_continue
+ }
+ -re " Nothing deleted" {
+ incr matches
+ set nothing 1
+ exp_continue
+ }
+ timeout {
+ send_user "\nFAILURE: sacctmgr delete not responding\n"
+ slow_kill $my_pid
+ incr exit_code 1
+ }
+ eof {
+ wait
+ }
+ }
+ if {$access_err != 0} {
+ return 1
+ }
+ if {$matches != 1} {
+ send_user "\nFAILURE: sacctmgr had a problem deleting cluster got $matches\n"
+ incr exit_code 1
+ }
+ if { !$nothing } {
+ if { ![check_acct_associations] } {
+ send_user "\nFAILURE: Our associations don't line up\n"
+ incr exit_code 1
+ }
+ }
+
#
# Use sacctmgr to create a cluster
#
diff --git a/testsuite/expect/test21.7 b/testsuite/expect/test21.7
index 0f464fdd195..ee0067db4de 100755
--- a/testsuite/expect/test21.7
+++ b/testsuite/expect/test21.7
@@ -65,7 +65,15 @@ if { [test_account_storage] == 0 } {
send_user "\nWARNING: This test can't be run without a usable AccountStorageType\n"
exit 0
}
-
+
+#
+# Verify if Administrator privileges
+#
+ if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to create a cluster
#
diff --git a/testsuite/expect/test21.8 b/testsuite/expect/test21.8
index 6d414415c6a..e90f04bace4 100755
--- a/testsuite/expect/test21.8
+++ b/testsuite/expect/test21.8
@@ -72,6 +72,14 @@ if { [test_account_storage] == 0 } {
exit 0
}
+#
+# Verify if Administrator privileges
+#
+ if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to create a cluster
#
diff --git a/testsuite/expect/test21.9 b/testsuite/expect/test21.9
index 4ec5194503a..bb5aad587da 100755
--- a/testsuite/expect/test21.9
+++ b/testsuite/expect/test21.9
@@ -72,6 +72,14 @@ if { [test_account_storage] == 0 } {
exit 0
}
+#
+# Verify if Administrator privileges
+#
+ if { [string compare [check_accounting_admin_level] "Administrator"] } {
+ send_user "\nWARNING: This test can't be run without being an Accounting administrator.\nUse sacctmgr mod user \$USER_NAME admin=admin.\n"
+ exit 0
+}
+
#
# Use sacctmgr to create a cluster
#
diff --git a/testsuite/expect/test7.3 b/testsuite/expect/test7.3
index 86fe053a947..07b72d80a21 100755
--- a/testsuite/expect/test7.3
+++ b/testsuite/expect/test7.3
@@ -57,7 +57,7 @@ send_user "slurm_dir is $slurm_dir\n"
if {[test_aix]} {
send_user "$bin_cc ${test_prog}.c -Wl,-brtl -g -pthread -o ${test_prog} -I${slurm_dir}/include -L${slurm_dir}/lib -lslurm -lntbl\n"
exec $bin_cc ${test_prog}.c -Wl,-brtl -g -pthread -o ${test_prog} -I${slurm_dir}/include -L${slurm_dir}/lib -lslurm -lntbl
-} elseif [file exists ${slurm_dir}/lib64] {
+} elseif [file exists ${slurm_dir}/lib64/libslurm.so] {
send_user "$bin_cc ${test_prog}.c -g -pthread -o ${test_prog} -I${slurm_dir}/include -Wl,--rpath=${slurm_dir}/lib64 -L${slurm_dir}/lib64 -lslurm\n"
exec $bin_cc ${test_prog}.c -g -pthread -o ${test_prog} -I${slurm_dir}/include -Wl,--rpath=${slurm_dir}/lib64 -L${slurm_dir}/lib64 -lslurm
} else {
diff --git a/testsuite/expect/test7.7 b/testsuite/expect/test7.7
index c55a44bc164..7a90ba1cfe5 100755
--- a/testsuite/expect/test7.7
+++ b/testsuite/expect/test7.7
@@ -150,7 +150,7 @@ make_bash_script $file_in "
echo BEGIN
$bin_sleep 20
echo FINI
- exit 123"
+ exit 0"
set job_id1 0
set job_id2 0
set sbatch_pid [spawn $sbatch -N1-1024 --output=$file_out --comment=test -t1 $file_in]
diff --git a/testsuite/expect/test7.7.prog.c b/testsuite/expect/test7.7.prog.c
index cc8a536858b..81ba2a5e36d 100644
--- a/testsuite/expect/test7.7.prog.c
+++ b/testsuite/expect/test7.7.prog.c
@@ -25,6 +25,7 @@
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
\*****************************************************************************/
+#include <errno.h>
#include <netdb.h>
#include <stdio.h>
#include <stdlib.h>
@@ -71,7 +72,7 @@ static int _conn_wiki_port(char *host, int port)
static int _conn_event_port(char *host, int port)
{
- int sock_fd;
+ int i, rc, sock_fd;
struct sockaddr_in wiki_addr;
struct hostent *hptr;
@@ -88,11 +89,19 @@ static int _conn_event_port(char *host, int port)
wiki_addr.sin_family = AF_INET;
wiki_addr.sin_port = htons(port);
memcpy(&wiki_addr.sin_addr.s_addr, hptr->h_addr, hptr->h_length);
- if (bind(sock_fd, (struct sockaddr *) &wiki_addr,
- sizeof(wiki_addr))) {
- printf("WARNING: bind to port %i failed, may not be real error\n",
- port);
- return -1;
+ for (i=0; ; i++) {
+ if (i)
+ sleep(5);
+ rc = bind(sock_fd, (struct sockaddr *) &wiki_addr,
+ sizeof(wiki_addr));
+ if (rc == 0)
+ break;
+ if ((errno != EINVAL) || (i > 5)) {
+ printf("WARNING: bind to port %i; %s\n",
+ port, strerror(errno));
+ return -1;
+ }
+ printf("WARNING: port %i in use, retrying\n", port);
}
listen(sock_fd, 1);
return sock_fd;
@@ -179,7 +188,7 @@ static char *_recv_msg(int fd)
return buf;
}
-static void _xmit(char *msg)
+static int _xmit(char *msg)
{
int msg_len = strlen(msg);
char *out_msg, *in_msg, sum[20], *sc_ptr;
@@ -199,12 +208,11 @@ static void _xmit(char *msg)
printf("recv:%s\n\n", in_msg);
sc_ptr = strstr(in_msg, "SC=");
sc = atoi(sc_ptr+3);
- if (sc != 0) {
+ if (sc != 0)
fprintf(stderr, "RPC failure\n");
- exit(1);
- }
free(in_msg);
close(wiki_fd);
+ return sc;
}
static void _event_mgr(void)
@@ -240,19 +248,22 @@ static void _get_jobs(void)
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=%s",
(uint32_t) now, "CMD=GETJOBS ARG=0:ALL");
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
/* Dump volitile data */
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=GETJOBS ARG=%u:ALL",
(uint32_t) now, (uint32_t) 1);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
/* Dump state only */
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=GETJOBS ARG=%u:ALL",
(uint32_t) now, (uint32_t) (now+2));
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _get_nodes(void)
@@ -264,19 +275,22 @@ static void _get_nodes(void)
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=%s",
(uint32_t) now, "CMD=GETNODES ARG=0:ALL");
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
/* Dump volitile data */
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=GETNODES ARG=%u:ALL",
(uint32_t) now, (uint32_t) 1);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
/* Dump state only */
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=GETNODES ARG=%u:ALL",
(uint32_t) now, (uint32_t) (now+2));
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _cancel_job(long my_job_id)
@@ -289,20 +303,32 @@ static void _cancel_job(long my_job_id)
"TYPE=ADMIN "
"COMMENT=\"cancel comment\" ",
(uint32_t) now, my_job_id);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _start_job(long my_job_id)
{
time_t now = time(NULL);
char out_msg[128];
+ int i, rc;
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=STARTJOB ARG=%ld "
"COMMENT=\'start comment\' "
"TASKLIST=", /* Empty TASKLIST means we don't care */
(uint32_t) now, my_job_id);
- _xmit(out_msg);
+
+ for (i=0; i<10; i++) {
+ if (i)
+ sleep(10);
+ rc = _xmit(out_msg);
+ if (rc == 0)
+ break;
+ /* Still completing after requeue */
+ }
+ if (rc != 0)
+ exit(1);
}
static void _suspend_job(long my_job_id)
@@ -313,7 +339,8 @@ static void _suspend_job(long my_job_id)
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=SUSPENDJOB ARG=%ld",
(uint32_t) now, my_job_id);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _signal_job(long my_job_id)
@@ -324,7 +351,8 @@ static void _signal_job(long my_job_id)
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=SIGNALJOB ARG=%ld VALUE=URG",
(uint32_t) now, my_job_id);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _modify_job(long my_job_id)
@@ -343,7 +371,8 @@ static void _modify_job(long my_job_id)
/* "INVALID=123 " */
"TIMELIMIT=10 BANK=test_bank",
(uint32_t) now, my_job_id);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _notify_job(long my_job_id)
@@ -355,7 +384,8 @@ static void _notify_job(long my_job_id)
"TS=%u AUTH=root DT=CMD=NOTIFYJOB ARG=%ld "
"MSG=this_is_a_test",
(uint32_t) now, my_job_id);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _resume_job(long my_job_id)
@@ -366,7 +396,8 @@ static void _resume_job(long my_job_id)
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=RESUMEJOB ARG=%ld",
(uint32_t) now, my_job_id);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _job_requeue(long my_job_id)
@@ -377,7 +408,8 @@ static void _job_requeue(long my_job_id)
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=REQUEUEJOB ARG=%ld",
(uint32_t) now, my_job_id);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _job_will_run(long my_job_id)
@@ -389,7 +421,8 @@ static void _job_will_run(long my_job_id)
"TS=%u AUTH=root DT=CMD=JOBWILLRUN ARG=JOBID=%ld,%s",
(uint32_t) now, my_job_id,
""); /* put available node list here */
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _initialize(void)
@@ -400,7 +433,8 @@ static void _initialize(void)
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=INITIALIZE ARG=USEHOSTEXP=N EPORT=%u",
(uint32_t) now, e_port);
- _xmit(out_msg);
+ if (_xmit(out_msg))
+ exit(1);
}
static void _single_msg(void)
@@ -411,8 +445,10 @@ static void _single_msg(void)
snprintf(out_msg, sizeof(out_msg),
"TS=%u AUTH=root DT=CMD=%s",
(uint32_t) now,
- "JOBWILLRUN ARG=JOBID=65537,bgl[000x733] JOBID=65539,bgl[000x733] JOBID=65538,bgl[000x733]");
- _xmit(out_msg);
+ "JOBWILLRUN ARG=JOBID=65537,bgl[000x733] "
+ "JOBID=65539,bgl[000x733] JOBID=65538,bgl[000x733]");
+ if (_xmit(out_msg))
+ exit(1);
}
int main(int argc, char * argv[])
@@ -458,7 +494,7 @@ int main(int argc, char * argv[])
}
_cancel_job(job_id+1);
_job_requeue(job_id); /* Put job back into HELD state */
- sleep(15);
+ sleep(10);
_start_job(job_id);
_get_jobs();
#endif
diff --git a/testsuite/expect/test7.8 b/testsuite/expect/test7.8
index 0713a576b7f..e170fa2f7ee 100755
--- a/testsuite/expect/test7.8
+++ b/testsuite/expect/test7.8
@@ -110,7 +110,7 @@ make_bash_script $file_in "
echo BEGIN
$bin_sleep 20
echo FINI
- exit 123"
+ exit 0"
set job_id1 0
set job_id2 0
--
GitLab