#!/usr/bin/env expect
############################################################################
# Purpose:  Test of Slurm functionality
#           Test GPU resource limits with various allocation options
#
# Requires: AccountingStorageEnforce=limits
#           AccountingStorageTRES=gres/gpu
#           SelectType=select/cons_tres
#           Administrator permissions
############################################################################
# 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
source ./globals_accounting

set acct           "test$test_id\_acct"
set cluster        [get_cluster_name]
set exit_code      0
set file_in        "test$test_id.input"
set file_out1      "test$test_id.output1"
set file_out2      "test$test_id.output2"
set one_task_pc    0
set user           [get_my_user_name]

proc setup { gpu_limit } {
	global acct cluster exit_code user

	set acct_req(cluster) $cluster
	set acct_req(parent)  "root"
	set acct_req(maxtres) "gres/gpu=$gpu_limit"

	set user_req(cluster) $cluster
	set user_req(account) $acct

	if { [add_acct $acct [array get acct_req]] } {
		log_error "Child account was not added"
		incr exit_code
		return 1
	}

	if { [add_user $user [array get user_req]] } {
		log_error "User was not added to child account"
		incr exit_code
		return 1
	}

	return 0
}

proc cleanup { } {
	global acct

	remove_acct "" $acct
}

set store_tres [string tolower [get_acct_store_tres]]
set store_mps [string first "gres/gpu:" $store_tres]
if {$store_mps != -1} {
	skip "This test requires homogeneous GPU accounting (NO Type)"
}
set store_gpu [string first "gres/gpu" $store_tres]
if {$store_gpu == -1} {
	skip "This test requires accounting for GPUs"
} elseif { [test_using_slurmdbd] == 0 } {
	skip "This test can't be run without AccountStorageType=slurmdbd"
} elseif { [test_enforce_limits] == 0 } {
	skip "This test can't be run without AccountingStorageEnforce=limits"
} elseif {[test_front_end]} {
	skip "This test is incompatible with front-end systems"
} elseif {[string compare [check_accounting_admin_level] "Administrator"]} {
	skip "This test can't be run without being an Accounting administrator"
}

if {[test_cons_tres]} {
	log_debug "Valid configuration, using select/cons_tres"
} else {
	skip "This test is only compatible with select/cons_tres"
}
if {[test_select_type_params "CR_ONE_TASK_PER_CORE"]} {
	set one_task_pc 1
}

set nb_nodes [get_node_cnt_in_part]
log_debug "Default partition node count is $nb_nodes"
if {$nb_nodes > 1} {
	set nb_nodes 2
}
set gpu_cnt [get_highest_gres_count $nb_nodes "gpu"]
if {$gpu_cnt < 0} {
	fail "Error getting GPU count"
}
if {$gpu_cnt < 2} {
	skip "This test requires 2 or more GPUs in the default partition"
}

set node_name [get_nodes_by_request "--gres=gpu:1 -n1 -t1"]
if { [llength $node_name] != 1 } {
	skip "This test need to be able to submit jobs with at least --gres=gpu:1"
}
set cpus_per_node     [get_node_param $node_name "CPUTot"]
set sockets_per_node  [get_node_param $node_name "Sockets"]
set cores_per_socket  [get_node_param $node_name "CoresPerSocket"]
set cpus_per_socket   [expr $cpus_per_node / $sockets_per_node]

log_debug "GPUs per node is $gpu_cnt"
log_debug "Sockets per node is $sockets_per_node"
log_debug "CPUs per socket is $cpus_per_socket"
log_debug "CPUs per node is $cpus_per_node"

if {$cpus_per_node < 3} {
	skip "This test requires 3 or more CPUs per node in the default partition"
}

# Remove any vestigial test account
cleanup

# Add parent account (off root)
set gpu_limit [expr $gpu_cnt * $nb_nodes]
if {$gpu_limit > 8} {
	set gpu_limit 8
} else {
	incr gpu_limit -1
}
if {[setup $gpu_limit]} {
	fail "Setup failed (gpu_limit = $gpu_limit)"
}


make_bash_script $file_in "
	$scontrol -dd show job \${SLURM_JOBID} | grep gpu
	exit 0"

#
# Test --gpus option by job (first job over limit, second job under limit)
#
log_info "TEST 1: --gpus option by job (first job over limit, second job under limit)"

set timeout $max_job_delay
exec $bin_rm -f $file_out1  $file_out2
set gpu_fail_cnt [expr $gpu_limit + 1]
set job_id1 0
spawn $sbatch --account=$acct --gres=craynetwork:0 --gpus=$gpu_fail_cnt -t1 -o $file_out1 -J "test$test_id" ./$file_in
expect {
	-re "Submitted batch job ($number)" {
		set job_id1 $expect_out(1,string)
		exp_continue
	}
	timeout {
		log_error "sbatch not responding"
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$job_id1 == 0} {
	log_error "Job not submitted"
	set exit_code 1
}

set job_id2 0
spawn $sbatch --account=$acct --gres=craynetwork:0 --gpus=$gpu_limit -t1 -o $file_out2 -J "test$test_id" ./$file_in
expect {
	-re "Submitted batch job ($number)" {
		set job_id2 $expect_out(1,string)
		exp_continue
	}
	timeout {
		log_error "sbatch not responding"
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$job_id2 == 0} {
	log_error "Job not submitted"
	set exit_code 1
}

if {[wait_for_job $job_id2 "DONE"] != 0} {
	log_error "Job $job_id2 did not complete"
	cancel_job $job_id2
	set exit_code 1
}

set match 0
spawn $scontrol show job $job_id1
expect {
	-re "JobState=PENDING" {
		incr match
		exp_continue
	}
	-re "Reason=.*AssocMaxGRESPerJob" {
		incr match
		exp_continue
	}
	timeout {
		log_error "scontrol not responding"
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$match != 2} {
	log_error "Job $job_id1 state is bad"
	set exit_code 1
}
cancel_job $job_id1
fail_on_error "Test failed due to previous errors (\$exit_code = $exit_code)"

#
# Test --gpus-per-node option by job (first job over limit, second job under limit)
#
log_info "TEST 2: --gpus-per-node option by job (first job over limit, second job under limit)"

set timeout $max_job_delay
exec $bin_rm -f $file_out1  $file_out2
set gpu_good_cnt [expr $gpu_limit / $nb_nodes]
if {$nb_nodes == 1} {
	set gpu_fail_cnt [expr $gpu_limit + 1]
} else {
	set gpu_fail_cnt [expr $gpu_good_cnt + 1]
}
set job_id1 0
spawn $sbatch --account=$acct --gres=craynetwork:0 --gpus-per-node=$gpu_fail_cnt -N$nb_nodes -t1 -o $file_out1 -J "test$test_id" ./$file_in
expect {
	-re "Submitted batch job ($number)" {
		set job_id1 $expect_out(1,string)
		exp_continue
	}
	timeout {
		log_error "sbatch not responding"
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$job_id1 == 0} {
	log_error "Job not submitted"
	set exit_code 1
}

set job_id2 0
spawn $sbatch --account=$acct --gres=craynetwork:0 --gpus-per-node=$gpu_good_cnt -N$nb_nodes -t1 -o $file_out2 -J "test$test_id" ./$file_in
expect {
	-re "Submitted batch job ($number)" {
		set job_id2 $expect_out(1,string)
		exp_continue
	}
	timeout {
		log_error "sbatch not responding"
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$job_id2 == 0} {
	log_error "Job not submitted"
	set exit_code 1
}

if {[wait_for_job $job_id2 "DONE"] != 0} {
	log_error "Job $job_id2 did not complete"
	cancel_job $job_id2
	set exit_code 1
}

set match 0
spawn $scontrol show job $job_id1
expect {
	-re "JobState=PENDING" {
		incr match
		exp_continue
	}
	-re "Reason=.*AssocMaxGRESPerJob" {
		incr match
		exp_continue
	}
	timeout {
		log_error "scontrol not responding"
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$match != 2} {
	log_error "Job $job_id1 state is bad"
	set exit_code 1
}
cancel_job $job_id1
fail_on_error "Test failed due to previous errors (\$exit_code = $exit_code)"

#
# Test --gpus-per-task option by job (first job over limit, second job under limit)
#
log_info "TEST 3: --gpus-per-task option by job (first job over limit, second job under limit)"

set total_cores [expr $cores_per_socket * $sockets_per_node]
if {$one_task_pc && $cpus_per_node > $total_cores} {
	set ntasks_per_core [expr $cpus_per_node / $total_cores]
	set extra_opt "--ntasks-per-core=$ntasks_per_core"
} else {
	set extra_opt "-t1"
}

set timeout $max_job_delay
exec $bin_rm -f $file_out1  $file_out2
set gpu_good_cnt $gpu_limit
set gpu_fail_cnt [expr $gpu_limit + 1]
set job_id1 0
spawn $sbatch --account=$acct --gres=craynetwork:0 --gpus-per-task=1 -n$gpu_fail_cnt $extra_opt -t1 -o $file_out1 -J "test$test_id" ./$file_in
expect {
	-re "Submitted batch job ($number)" {
		set job_id1 $expect_out(1,string)
		exp_continue
	}
	timeout {
		log_error "sbatch not responding"
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$job_id1 == 0} {
	log_error "Job not submitted"
	set exit_code 1
}

set job_id2 0
spawn $sbatch --account=$acct --gres=craynetwork:0 --gpus-per-task=1 -n$gpu_good_cnt $extra_opt -t1 -o $file_out2 -J "test$test_id" ./$file_in
expect {
	-re "Submitted batch job ($number)" {
		set job_id2 $expect_out(1,string)
		exp_continue
	}
	timeout {
		log_error "sbatch not responding"
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$job_id2 == 0} {
	log_error "Job not submitted"
	set exit_code 1
}

if {[wait_for_job $job_id2 "DONE"] != 0} {
	log_error "Job $job_id2 did not complete"
	cancel_job $job_id2
	set exit_code 1
}

set match 0
spawn $scontrol show job $job_id1
expect {
	-re "JobState=PENDING" {
		incr match
		exp_continue
	}
	-re "Reason=.*AssocMaxGRESPerJob" {
		incr match
		exp_continue
	}
	timeout {
		log_error "scontrol not responding"
		set exit_code 1
	}
	eof {
		wait
	}
}
if {$match != 2} {
	log_error "Job $job_id1 state is bad"
	set exit_code 1
}
cancel_job $job_id1
fail_on_error "Test failed due to previous errors (\$exit_code = $exit_code)"

if {$exit_code == 0} {
	exec $bin_rm -f $file_in $file_out1 $file_out2
}