diff --git a/testsuite/expect/test18.29 b/testsuite/expect/test18.29
index 7de8986afa298c255a5bbffb35e6adbedd7b5ca0..d756f7c60e0703c4dde8c063b54302270f1d8495 100755
--- a/testsuite/expect/test18.29
+++ b/testsuite/expect/test18.29
@@ -1,17 +1,12 @@
#!/usr/bin/expect
############################################################################
# Purpose: Test of SLURM functionality
-# Test of cpus-per-task option on a single node (--cpus-per-task
-# option).
+# Test of slaunch --cpus-per-task option.
#
# Output: "TEST: #.#" followed by "SUCCESS" if test was successful, OR
# "WARNING: ..." with an explanation of why the test can't be made, OR
# "FAILURE: ..." otherwise with an explanation of the failure, OR
# anything else indicates a failure mode that must be investigated.
-#
-# NOTE: This assumes node names are of the form <alpha><number>, where
-# the value of <number> indicates the nodes relative location.
-# Change tha node name parsing logic as needed for other formats.
############################################################################
# Copyright (C) 2002-2006 The Regents of the University of California.
# Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -56,115 +51,143 @@ make_bash_script $file_in "
env | grep SLURM_JOB_CPUS_PER_NODE
$bin_hostname"
-#
-# Submit a 1 node job to determine the node's CPU count
-#
set cpu_cnt 0
-set fat_nodes 0
-set host ""
set job_id 0
set timeout $max_job_delay
-set task_cnt 0
+set node_cnt 2
+
+set available [available_nodes [default_partition]]
+if {$available < 2} {
+ send_user "\nWARNING: not enough nodes currently available"
+ send_user " ($available avail, 2)\n"
+ exit $exit_code
+}
-spawn $salloc -N1 -t1 $slaunch --cpus-per-task=1 -l $file_in
+# Script will print SLURM_JOB_CPUS_PER_NODE, then hold the allocation
+# for a long time.
+set script_name test18.28.sh
+make_bash_script $script_name {
+ printenv SLURM_JOB_CPUS_PER_NODE
+ sleep 600
+}
+set salloc_pid [spawn $salloc -N $node_cnt $slaunch -n 1 $script_name]
+set salloc_spawn_id $spawn_id
expect {
- -re "Granted job allocation ($number)" {
- set job_id $expect_out(1,string)
- exp_continue
- }
- -re "SLURM_JOB_CPUS_PER_NODE=($number)" {
- set cpu_cnt $expect_out(1,string)
- exp_continue
- }
- -re "($number): ($alpha_numeric)" {
- set task $expect_out(1,string)
- set host $expect_out(2,string)
- incr task_cnt
- exp_continue
- }
- -re "Task count specification invalid" {
- send_user "\nWARNING: nodes have too many CPUs for test\n"
- set fat_nodes 1
- exp_continue
- }
- timeout {
- send_user "\nFAILURE: salloc not responding\n"
- if {$job_id == 0} {
- kill_salloc
- } else {
- cancel_job $job_id
- }
- set exit_code 1
- exp_continue
- }
- eof {
- wait
- }
-}
-# On Blue Gene/L each node has 1024 CPUs, more than slurmd can launch
-if {$fat_nodes == 1} {
- exit 0
+ -re "Granted job allocation ($number)" {
+ set jobid $expect_out(1,string)
+ }
+ timeout {
+ slow_kill $salloc_pid
+ set exit_code 1
+ }
}
+expect {
+ -re "($number)" {
+ set cpu_cnt $expect_out(1,string)
+ }
+ timeout {
+ slow_kill $salloc_pid
+ set exit_code 1
+ }
+}
+file delete $script_name
+
+send_user "cpu count is $cpu_cnt\n"
-#
# Check CPU count
+if {$cpu_cnt < 2} {
+ send_user "\nWARNING: The node only has one CPU\n"
+ exec $scancel $jobid
+ exit $exit_code
+}
+
+#
+# Now start the real testing of --cpus-per-task.
#
-if {[string compare $host ""] == 0} {
- send_user "\nFAILURE: Did not get hostname\n"
- set exit_code 1
+
+#
+# Test A
+# Should run: --cpus-per-task=1, -n=(node_count*cpu count)
+#
+set task_cnt 0
+set pid [spawn $slaunch --jobid $jobid -l --cpus-per-task 1 -n [expr $node_cnt * $cpu_cnt] $bin_printenv SLURMD_NODENAME]
+expect {
+ -re "($number): ($alpha_numeric)" {
+ incr task_cnt
+ exp_continue
+ }
+ eof {
+ wait
+ }
}
-if {$cpu_cnt != $task_cnt} {
- send_user "FAILURE: should have run $cpu_cnt tasks (one per CPU) "
- send_user "instead of $task_cnt tasks\n"
- set exit_code 1
+set expected_task_cnt [expr $node_cnt * $cpu_cnt]
+if {$task_cnt < $expected_task_cnt} {
+ send_user "\nFAILURE Test A: Fewer tasks ($task_cnt) then expected ($expected_task_cnt)\n"
+ set exit_code 1
}
-if {$cpu_cnt < 2} {
- send_user "\nWARNING: The node only has one CPU\n"
- exit $exit_code
+if {$task_cnt > $expected_task_cnt} {
+ send_user "\nFAILURE Test A: More tasks ($task_cnt) then expected ($expected_task_cnt)\n"
+ set exit_code 1
}
#
-# Check task count when using 2 cpus per task
+# Test B
+# Should NOT run: --cpus-per-task=2, -n=(2*cpu count)
#
-set job_id 0
-set task_cnt 0
-spawn $salloc -N1 -t1 --nodelist=$host $slaunch --cpus-per-task=2 -l $bin_hostname
+set task_cnt 0
+set pid [spawn $slaunch --jobid $jobid -l --cpus-per-task $cpu_cnt -n [expr $node_cnt * $cpu_cnt] $bin_printenv SLURMD_NODENAME]
expect {
- -re "Granted job allocation ($number)" {
- set job_id $expect_out(1,string)
- exp_continue
- }
- -re "Invalid node name specified" {
- send_user "\nWARNING: Appears you are using "
- send_user "multiple slurmd testing.\n"
- send_user "This test won't work in that fashion.\n"
- exit 0
- }
- -re "$number: ($alpha_numeric)" {
- incr task_cnt
- exp_continue
- }
- timeout {
- send_user "\nFAILURE: srun not responding\n"
- if {$job_id == 0} {
- kill_salloc
- } else {
- cancel_job $job_id
- }
- set exit_code 1
- exp_continue
- }
- eof {
- wait
- }
+ -re "($number): ($alpha_numeric)" {
+ incr task_cnt
+ exp_continue
+ }
+ "error: Failed creating job step context" {
+ send_user "This error was expected!\n"
+ }
+ eof {
+ wait
+ }
+}
+if {$task_cnt > 0} {
+ send_user "FAILURE Test B: Test should NOT have run.\n"
+ set exit_code 1
}
+
#
-# Check CPU count
+# Test C
+# Should run: --cpus-per-task=2, -n=(cpu count)
#
-if {$task_cnt != [expr $cpu_cnt / 2]} {
- send_user "\nFAILURE: Improper task count for given cpus-per-task\n"
- set exit_code 1
+set task_cnt 0
+set pid [spawn $slaunch --jobid $jobid -l --cpus-per-task $cpu_cnt -n $node_cnt $bin_printenv SLURMD_NODENAME]
+expect {
+ -re "($number): ($alpha_numeric)" {
+ incr task_cnt
+ exp_continue
+ }
+ eof {
+ wait
+ }
+}
+set expected_task_cnt $node_cnt
+if {$task_cnt < $expected_task_cnt} {
+ send_user "\nFAILURE Test C: Fewer tasks ($task_cnt) then expected ($expected_task_cnt)\n"
+ set exit_code 1
+}
+if {$task_cnt > $expected_task_cnt} {
+ send_user "\nFAILURE Test C: More tasks ($task_cnt) then expected ($expected_task_cnt)\n"
+ set exit_code 1
+}
+
+#
+# Clean up the job allocation.
+#
+exec $scancel $jobid
+set spawn_id $salloc_spawn_id
+expect {
+ eof {
+ wait
+ }
}