#!/usr/bin/env expect
############################################################################
# Purpose: Test of Slurm functionality
#          Test some valid combinations of srun --gpu and non-GPU GRES options
#
# Output:  "TEST: #.#" followed by "SUCCESS" if test was successful, OR
#          "FAILURE: ..." otherwise with an explanation of the failure, OR
#          anything else indicates a failure mode that must be investigated.
############################################################################
# Copyright (C) 2018 SchedMD LLC
# Written by Morris Jette
#
# This file is part of Slurm, a resource management program.
# For details, see <https://slurm.schedmd.com/>.
# Please also read the included file: DISCLAIMER.
#
# Slurm is free software; you can redistribute it and/or modify it under
# the terms of the GNU General Public License as published by the Free
# Software Foundation; either version 2 of the License, or (at your option)
# any later version.
#
# Slurm is distributed in the hope that it will be useful, but WITHOUT ANY
# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more
# details.
#
# You should have received a copy of the GNU General Public License along
# with Slurm; if not, write to the Free Software Foundation, Inc.,
# 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA.
############################################################################
source ./globals

set test_id     "39.12"
set exit_code   0
set file_in     "test$test_id.input"
set number_commas  "\[0-9_,\]+"

print_header $test_id

if {[test_cons_tres]} {
	send_user "\nValid configuration, using select/cons_tres\n"
} else {
	send_user "\nWARNING: This test is only compatible with select/cons_tres\n"
	exit 0
}
if {[test_front_end]} {
	send_user "\nWARNING: This test is incompatible with front-end systems\n"
	exit $exit_code
}

set def_part_name [default_partition]
set nb_nodes [get_node_cnt_in_part $def_part_name]
if {$nb_nodes > 1} {
	set nb_nodes 2
}
set gpu_cnt [get_gpu_count $nb_nodes]
if {$gpu_cnt < 0} {
	send_user "\nFAILURE: Error getting GPU count\n"
	exit 1
}
if {$gpu_cnt < 1} {
	send_user "\nWARNING: This test requires 1 or more GPUs per node for at least $nb_nodes nodes in the default partition\n"
	exit 0
}

set craynetwork_count [get_craynetwork_count $nb_nodes]
if {$craynetwork_count < 0} {
	send_user "\nFAILURE: Error getting craynetwork GRES count\n"
	exit 1
}
if {$craynetwork_count < 1} {
	send_user "\nWARNING: This test requires 1 or more craynetwork GRES per node for at least $nb_nodes nodes in the default partition\n"
	exit 0
}

set one_task_per_core [test_select_type_params "CR_ONE_TASK_PER_CORE"]
get_node_config
set sockets_with_gpus [get_gpu_socket_count $gpu_cnt $sockets_per_node]

send_user "\nGPU count is $gpu_cnt\n"
send_user "Sockets with GPUs $sockets_with_gpus\n"
send_user "Sockets per node is $sockets_per_node\n"
send_user "Cores per socket is $cores_per_socket\n"
send_user "CPUs per socket is $cpus_per_socket\n"
send_user "gres/craynetwork count is $craynetwork_count\n"
send_user "CR_ONE_TASK_PER_CORE is $one_task_per_core\n"

set cores_per_node [expr $sockets_per_node * $cores_per_socket]
set cpus_per_node  [expr $sockets_per_node * $cpus_per_socket]
if {$one_task_per_core > 0} {
	if {$gpu_cnt > $cores_per_node} {
		set gpu_cnt $cores_per_node
	}
} else {
	if {$gpu_cnt > $cpus_per_node} {
		set gpu_cnt $cpus_per_node
	}
}

set tot_gpus $gpu_cnt
if {$nb_nodes > 1} {
	incr tot_gpus $gpu_cnt
}
set gpus_per_node $gpu_cnt
if {$gpus_per_node > 1 && $sockets_per_node > 1} {
	set sockets_per_node 2
	set gpus_per_socket [expr $gpus_per_node / $sockets_per_node]
} else {
	set gpus_per_socket $gpus_per_node
}
set sockets_per_node [expr $gpus_per_node / $gpus_per_socket]

#
# Build input script file
#
make_bash_script $file_in "echo HOST:\$SLURMD_NODENAME CUDA_VISIBLE_DEVICES:\$CUDA_VISIBLE_DEVICES
if \[ \$SLURM_PROCID -eq 0 \]; then
sleep 1
$scontrol -dd show job \$SLURM_JOB_ID | grep \"GRES=\"
fi
exit 0"

#
# Test --gpus options using a subset of GPUs actually available on the node
#
send_user "\n\nTEST: --gpus option\n"
set match_craynetwork 0
set match_gpu 0
set timeout $max_job_delay
if {$tot_gpus > 1} {
	set use_gpus_per_job [expr $tot_gpus - 1]
} else {
	set use_gpus_per_job $tot_gpus
}
set srun_pid [spawn $srun --gres=craynetwork --cpus-per-gpu=1 --gpus=$use_gpus_per_job --nodes=$nb_nodes -t1 -J "test$test_id" -l ./$file_in]
expect {
	-re "CUDA_VISIBLE_DEVICES:($number_commas)" {
		incr match_gpu [cuda_count $expect_out(1,string)]
		exp_continue
	}
	-re "craynetwork.CNT:($number)" {
		if {$expect_out(1,string) == 1} {
			incr match_craynetwork
		} else {
			set match_craynetwork -9999
		}
		exp_continue
	}
	timeout {
		send_user "\nFAILURE: srun not responding\n"
		slow_kill $srun_pid
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$match_craynetwork < 1} {
	send_user "\nFAILURE: srun --gres=craynetwork failure ($match_craynetwork < 1)\n"
	set exit_code 1
}
set expected_gpus $use_gpus_per_job
if {$match_gpu != $expected_gpus} {
	send_user "\nFAILURE: srun --gpus failure ($match_gpu != $expected_gpus)\n"
	set exit_code 1
}

if {[expr $use_gpus_per_job - 2] > $nb_nodes} {
	send_user "\n\nTEST: --gpus option, part 2\n"
	set match_craynetwork 0
	set match_gpu 0
	incr use_gpus_per_job -2
	set srun_pid [spawn $srun --gres=craynetwork:1 --cpus-per-gpu=1 --gpus=$use_gpus_per_job --nodes=$nb_nodes -t1 -J "test$test_id" -l ./$file_in]
	expect {
		-re "CUDA_VISIBLE_DEVICES:($number_commas)" {
			incr match_gpu [cuda_count $expect_out(1,string)]
			exp_continue
		}
		-re "craynetwork.CNT:($number)" {
			if {$expect_out(1,string) == 1} {
				incr match_craynetwork
			} else {
				set match_craynetwork -9999
			}
			exp_continue
		}
		timeout {
			send_user "\nFAILURE: srun not responding\n"
			slow_kill $srun_pid
			set exit_code 1
		}
		eof {
			wait
		}
	}
	if {$match_craynetwork < 1} {
		send_user "\nFAILURE: srun --gres=craynetwork failure ($match_craynetwork < 1)\n"
		set exit_code 1
	}
	set expected_gpus $use_gpus_per_job
	if {$match_gpu != $expected_gpus} {
		send_user "\nFAILURE: srun --gpus failure ($match_gpu != $expected_gpus)\n"
		set exit_code 1
	}
}

#
# Test --gpus-per-node options using a subset of GPUs actually available on the node
#
send_user "\n\nTEST: --gpus-per-node option\n"
set match_craynetwork 0
set match_gpu 0
if {$gpus_per_node > 1} {
	set use_gpus_per_node [expr $gpus_per_node - 1]
} else {
	set use_gpus_per_node $gpus_per_node
}
set srun_pid [spawn $srun --gres=craynetwork --cpus-per-gpu=1 --gpus-per-node=$use_gpus_per_node --nodes=$nb_nodes -t1 -J "test$test_id" -l ./$file_in]
expect {
	-re "CUDA_VISIBLE_DEVICES:($number_commas)" {
		incr match_gpu [cuda_count $expect_out(1,string)]
		exp_continue
	}
	-re "craynetwork.CNT:($number)" {
		if {$expect_out(1,string) == 1} {
			incr match_craynetwork
		} else {
			set match_craynetwork -9999
		}
		exp_continue
	}
	timeout {
		send_user "\nFAILURE: srun not responding\n"
		slow_kill $srun_pid
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$match_craynetwork < 1} {
	send_user "\nFAILURE: srun --gres=craynetwork failure ($match_craynetwork < 1)\n"
	set exit_code 1
}
set expected_gpus [expr $use_gpus_per_node * $nb_nodes]
if {$match_gpu != $expected_gpus} {
	send_user "\nFAILURE: srun --gpus-per-node failure ($match_gpu != $expected_gpus)\n"
	set exit_code 1
}

#
# Test --gpus-per-socket options using a subset of GPUs actually available on the node
#
send_user "\n\nTEST: --gpus-per-socket option\n"
set match_craynetwork 0
set match_gpu 0
if {$sockets_with_gpus > 1} {
	set cpus_per_task $cpus_per_socket
} else {
	set cpus_per_task 1
}
# Every node requires at least 1 GPU
if {$use_gpus_per_job < $nb_nodes} {
	set nb_nodes $use_gpus_per_job
}
set srun_pid [spawn $srun --gres=craynetwork --gpus-per-socket=1 --sockets-per-node=$sockets_with_gpus --nodes=$nb_nodes --ntasks=$nb_nodes --cpus-per-task=$cpus_per_task -t1 -J "test$test_id" -l ./$file_in]
expect {
	-re "CUDA_VISIBLE_DEVICES:($number_commas)" {
		incr match_gpu [cuda_count $expect_out(1,string)]
		exp_continue
	}
	-re "craynetwork.CNT:($number)" {
		if {$expect_out(1,string) == 1} {
			incr match_craynetwork
		} else {
			set match_craynetwork -9999
		}
		exp_continue
	}
	timeout {
		send_user "\nFAILURE: srun not responding\n"
		slow_kill $srun_pid
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$match_craynetwork < 1} {
	send_user "\nFAILURE: srun --gres=craynetwork failure ($match_craynetwork < 1)\n"
	set exit_code 1
}
set expected_gpus [expr $sockets_with_gpus * $nb_nodes]
if {$match_gpu != $expected_gpus} {
	send_user "\nFAILURE: srun --gpus-per-socket failure ($match_gpu != $expected_gpus)\n"
	set exit_code 1
}

#
# Test --gpus-per-task options using a subset of GPUs actually available on the node
#
send_user "\n\nTEST: --gpus-per-task option\n"
set match_craynetwork 0
set match_gpu 0
if {$gpu_cnt > 1} {
	set use_gpus_per_node [expr $gpu_cnt - 1]
} else {
	set use_gpus_per_node $gpu_cnt
}
set srun_pid [spawn $srun --gres=craynetwork --cpus-per-gpu=1 --gpus-per-task=1 -N1 --ntasks=$use_gpus_per_node -t1 -J "test$test_id" -l ./$file_in]
expect {
	-re "CUDA_VISIBLE_DEVICES:($number_commas)" {
		incr match_gpu [cuda_count $expect_out(1,string)]
		exp_continue
	}
	-re "craynetwork.CNT:($number)" {
		if {$expect_out(1,string) == 1} {
			incr match_craynetwork
		} else {
			set match_craynetwork -9999
		}
		exp_continue
	}
	timeout {
		send_user "\nFAILURE: srun not responding\n"
		slow_kill $srun_pid
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$match_craynetwork < 1} {
	send_user "\nFAILURE: srun --gres=craynetwork failure ($match_craynetwork < 1)\n"
	set exit_code 1
}
set expected_gpus [expr $use_gpus_per_node * $use_gpus_per_node]
if {$match_gpu != $expected_gpus} {
	send_user "\nFAILURE: srun --gpus-per-task failure ($match_gpu != $expected_gpus)\n"
	set exit_code 1
}

if {$exit_code == 0} {
	exec $bin_rm -f $file_in
	send_user "\nSUCCESS\n"
}
exit $exit_code