diff --git a/NEWS b/NEWS
index defc3827f365063e1985c8da5c98ac5ae6325e05..a72f14d192c33e87451edc0f0dd4bc1524097b34 100644
--- a/NEWS
+++ b/NEWS
@@ -329,6 +329,11 @@ documents those changes that are of interest to users and administrators.
-- Fix sinfo to display mixed nodes as allocated in '%F' output.
-- Sview - Fix cpu and node counts for partitions.
-- Ignore NO_VAL in SLURMDB_PURGE_* macros.
+ -- ALPS - Don't drain nodes if epilog fails. It leaves them in drain state
+ with no way to get them out.
+ -- Fix issue with task/affinity oversubscribing cpus erroneously when
+ using --ntasks-per-node.
+ -- MYSQL - Fix load of archive files.
* Changes in Slurm 14.03.8
==========================
diff --git a/src/plugins/accounting_storage/mysql/as_mysql_archive.c b/src/plugins/accounting_storage/mysql/as_mysql_archive.c
index 63d2944a319a0564eb17f82a959a8a7c134e13bd..fa044e9d28e2c7e2ce083f52cbe501eeebf8f81b 100644
--- a/src/plugins/accounting_storage/mysql/as_mysql_archive.c
+++ b/src/plugins/accounting_storage/mysql/as_mysql_archive.c
@@ -1541,11 +1541,11 @@ _load_events(uint16_t rpc_version, Buf buffer, char *cluster_name,
xstrcat(insert, ", ");
xstrfmtcat(insert, format,
+ object.period_start,
+ object.period_end,
+ object.node_name,
object.cluster_nodes,
object.cpu_count,
- object.node_name,
- object.period_end,
- object.period_start,
object.reason,
object.reason_uid,
object.state);
@@ -1852,15 +1852,15 @@ static char *_load_resvs(uint16_t rpc_version, Buf buffer,
xstrcat(insert, ", ");
xstrfmtcat(insert, format,
+ object.id,
object.assocs,
object.cpus,
object.flags,
- object.id,
- object.name,
object.nodes,
object.node_inx,
- object.time_end,
- object.time_start);
+ object.name,
+ object.time_start,
+ object.time_end);
}
// END_TIMER2("step query");
// info("resv query took %s", TIME_STR);
@@ -2018,52 +2018,52 @@ static char *_load_steps(uint16_t rpc_version, Buf buffer,
xstrcat(insert, ", ");
xstrfmtcat(insert, format,
- object.ave_cpu,
- object.act_cpufreq,
- object.consumed_energy,
- object.ave_disk_read,
- object.ave_disk_write,
- object.ave_pages,
- object.ave_rss,
- object.ave_vsize,
- object.exit_code,
- object.cpus,
object.id,
- object.kill_requid,
- object.max_disk_read,
- object.max_disk_read_node,
- object.max_disk_read_task,
- object.max_disk_write,
- object.max_disk_write_node,
- object.max_disk_write_task,
- object.max_pages,
- object.max_pages_node,
- object.max_pages_task,
- object.max_rss,
- object.max_rss_node,
- object.max_rss_task,
- object.max_vsize,
- object.max_vsize_node,
- object.max_vsize_task,
- object.min_cpu,
- object.min_cpu_node,
- object.min_cpu_task,
+ object.stepid,
+ object.period_start,
+ object.period_end,
+ object.period_suspended,
object.name,
object.nodelist,
- object.nodes,
object.node_inx,
- object.period_end,
- object.period_start,
- object.period_suspended,
- object.req_cpufreq,
object.state,
- object.stepid,
- object.sys_sec,
- object.sys_usec,
+ object.kill_requid,
+ object.exit_code,
+ object.nodes,
+ object.cpus,
object.tasks,
object.task_dist,
object.user_sec,
- object.user_usec);
+ object.user_usec,
+ object.sys_sec,
+ object.sys_usec,
+ object.max_vsize,
+ object.max_vsize_task,
+ object.max_vsize_node,
+ object.ave_vsize,
+ object.max_rss,
+ object.max_rss_task,
+ object.max_rss_node,
+ object.ave_rss,
+ object.max_pages,
+ object.max_pages_task,
+ object.max_pages_node,
+ object.ave_pages,
+ object.min_cpu,
+ object.min_cpu_task,
+ object.min_cpu_node,
+ object.ave_cpu,
+ object.act_cpufreq,
+ object.consumed_energy,
+ object.req_cpufreq,
+ object.max_disk_read,
+ object.max_disk_read_task,
+ object.max_disk_read_node,
+ object.ave_disk_read,
+ object.max_disk_write,
+ object.max_disk_write_task,
+ object.max_disk_write_node,
+ object.ave_disk_write);
}
// END_TIMER2("step query");
@@ -2180,10 +2180,10 @@ static char *_load_suspend(uint16_t rpc_version, Buf buffer,
xstrcat(insert, ", ");
xstrfmtcat(insert, format,
- object.associd,
object.id,
- object.period_end,
- object.period_start);
+ object.associd,
+ object.period_start,
+ object.period_end);
}
// END_TIMER2("suspend query");
diff --git a/src/plugins/task/affinity/dist_tasks.c b/src/plugins/task/affinity/dist_tasks.c
index 7ab578d0d9ba185150ed7c18cd33f97ec37b2c46..0bc4b3a7b95fba8c746defd6aa094fd27c088576 100644
--- a/src/plugins/task/affinity/dist_tasks.c
+++ b/src/plugins/task/affinity/dist_tasks.c
@@ -213,7 +213,7 @@ void batch_bind(batch_job_launch_msg_t *req)
}
str = (char *)bit_fmt_hexmask(req_map);
- debug3("task/affinity: job %u CPU mask from slurmctld: %s",
+ debug3("task/affinity: job %u core mask from slurmctld: %s",
req->job_id, str);
xfree(str);
@@ -690,7 +690,7 @@ static bitstr_t *_get_avail_map(launch_tasks_request_msg_t *req,
}
str = (char *)bit_fmt_hexmask(req_map);
- debug3("task/affinity: job %u.%u CPU mask from slurmctld: %s",
+ debug3("task/affinity: job %u.%u core mask from slurmctld: %s",
req->job_id, req->job_step_id, str);
xfree(str);
@@ -878,7 +878,32 @@ static int _task_layout_lllp_cyclic(launch_tasks_request_msg_t *req,
last_taskcount = taskcount;
for (i = 0; i < size; i++) {
bool already_switched = false;
- uint16_t bit = socket_last_pu[s] + (s * offset);
+ uint16_t bit;
+ uint16_t orig_s = s;
+
+ while (socket_last_pu[s] >= offset) {
+ /* Switch to the next socket we have
+ * ran out here. */
+
+ /* This only happens if the slurmctld
+ * gave us an allocation that made a
+ * task split sockets. Or if the
+ * entire allocation is on one socket.
+ */
+ s = (s + 1) % hw_sockets;
+ if (orig_s == s) {
+ /* This should rarely happen,
+ * but is here for sanity sake.
+ */
+ debug("allocation is full, "
+ "oversubscribing");
+ memset(socket_last_pu, 0,
+ sizeof(hw_sockets
+ * sizeof(int)));
+ }
+ }
+
+ bit = socket_last_pu[s] + (s * offset);
/* In case hardware and config differ */
bit %= size;
@@ -888,17 +913,6 @@ static int _task_layout_lllp_cyclic(launch_tasks_request_msg_t *req,
/* skip unrequested threads */
if (req->cpu_bind_type & CPU_BIND_ONE_THREAD_PER_CORE)
socket_last_pu[s] += hw_threads-1;
- if (socket_last_pu[s] >= offset) {
- /* Switch to the next socket we have
- * ran out here. */
-
- /* This only happens if the slurmctld
- gave us an allocation that made a
- task split sockets.
- */
- s = (s + 1) % hw_sockets;
- already_switched = true;
- }
if (!bit_test(avail_map, bit))
continue;
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index ce258b6af46cfe22d8ba7c12b7f6899c1cca68c8..6c5222162fb3913cb4a01b80c9610444b946027c 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -1817,6 +1817,12 @@ extern int drain_nodes ( char *nodes, char *reason, uint32_t reason_uid )
return ESLURM_INVALID_NODE_NAME;
}
+#ifdef HAVE_ALPS_CRAY
+ error("We cannot drain nodes on a Cray/ALPS system, "
+ "use native Cray tools such as xtprocadmin(8).");
+ return SLURM_SUCCESS;
+#endif
+
if ( (host_list = hostlist_create (nodes)) == NULL) {
error ("hostlist_create error on %s: %m", nodes);
return ESLURM_INVALID_NODE_NAME;
diff --git a/testsuite/expect/globals b/testsuite/expect/globals
index 85898c154773455fa9a10ae928a1e23114dbb92c..abdd7064a0e233bf54b250d31018e21f6ec62c22 100755
--- a/testsuite/expect/globals
+++ b/testsuite/expect/globals
@@ -778,6 +778,37 @@ proc wait_for_all_jobs { job_name incr_sleep } {
return $matches
}
+################################################################
+#
+# Proc: test_fast_schedule_2
+#
+# Purpose: Determine if we are runnning FastSchedule=2 usually meaning we are pretending we have more resources than reality.
+# This is based upon
+# the value of FastSchedule in the slurm.conf.
+#
+# Returns level of 1 if it is 2, 0 otherwise
+#
+################################################################
+proc test_fast_schedule_2 { } {
+ global scontrol number
+
+ log_user 0
+ set is_2 0
+ spawn $scontrol show config
+ expect {
+ -re "FastSchedule *= 2" {
+ set is_2 1
+ exp_continue
+ }
+ eof {
+ wait
+ }
+ }
+
+ log_user 1
+ return $is_2
+}
+
################################################################
#
diff --git a/testsuite/expect/test1.91 b/testsuite/expect/test1.91
index 36c660fe474bb7f04685a205092b8f90964db106..0de1d9cbddfebe26955a0330662684fc53ae68b9 100755
--- a/testsuite/expect/test1.91
+++ b/testsuite/expect/test1.91
@@ -520,10 +520,14 @@ expect {
if {$exit_code == 0} {
exec $bin_rm -f $file_prog
send_user "\nSUCCESS\n"
-} else {
+} elseif { [test_fast_schedule_2] } {
+ exec $bin_rm -f $file_prog
send_user "\nNOTE: This test can fail if the node configuration in slurm.conf\n"
send_user " (sockets, cores, threads) differs from the actual configuration\n"
send_user " or if using task/cgroup without task/affinity.\n"
+} else {
+ send_user "\nFAILURE\n"
+ set exit_code 1
}
exit $exit_code