#!/usr/bin/env expect ############################################################################ # Purpose: Test of Slurm functionality # Test accounting for GPU resources with various allocation 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 . # 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 "39.19" set exit_code 0 set file_in1 "test$test_id.input1" set file_in2 "test$test_id.input2" set file_out "test$test_id.output" # # Validate the job, batch step and step 0 of a job have the proper GPU counts # No step to test if step_gpus == -1 # # NOTE: AllocGRES and ReqGRES values for all steps (including batch step) # are reported based upon the job specification # proc test_acct { job_id job_gpus step_gpus have_gpu_types } { global alpha_numeric_under number global bin_cat bin_rm bin_grep exit_code file_out sacct if {$job_id == 0} { return } sleep 2 send_user "\nJob $job_id Expected job GPUs:$job_gpus Step GPUs:$step_gpus\n" exec $bin_rm -f $file_out exec >$file_out $sacct --job=$job_id --parsable2 --start=today -o JobID,AllocGRES,ReqGRES,AllocTRES spawn $bin_cat $file_out expect { eof { wait } } log_user 0 if {$step_gpus != -1} { set match 0 spawn $bin_grep $job_id.0 $file_out expect { -re "gpu:($number)" { incr match if {$expect_out(1,string) != $job_gpus} { send_user "\nFAILURE: bad step GPU count reported by sacct ($expect_out(1,string) != $step_gpus)\n" set exit_code 1 } exp_continue } -re "gres/gpu=($number)" { incr match if {$expect_out(1,string) != $step_gpus} { send_user "\nFAILURE: bad step GPU count reported by sacct ($expect_out(1,string) != $step_gpus)\n" set exit_code 1 } exp_continue } -re "gres/gpu:($alpha_numeric_under)=($number)" { if {$have_gpu_types != 0} { incr match if {$expect_out(2,string) != $step_gpus} { send_user "\nFAILURE: bad step GPU count reported by sacct ($expect_out(2,string) != $step_gpus)\n" set exit_code 1 } } exp_continue } eof { wait } } if {$match != 3} { send_user "\nFAILURE: Missing step GPU count report from sacct ($match != 3)\n" } } set match 0 spawn $bin_grep $job_id.batch $file_out expect { -re "gpu:($number)" { incr match if {$expect_out(1,string) != $job_gpus} { send_user "\nFAILURE: bad batch GPU count reported by sacct ($expect_out(1,string) != $job_gpus)\n" set exit_code 1 } exp_continue } eof { wait } } if {$match != 2} { send_user "\nFAILURE: missing batch GPU count report from sacct ($match != 2)\n" } set match 0 spawn $bin_grep $job_id| $file_out expect { -re "gpu:($number)" { incr match if {$expect_out(1,string) != $job_gpus} { send_user "\nFAILURE: Bad job GPU count reported by sacct ($expect_out(1,string) != $job_gpus)\n" set exit_code 1 } exp_continue } -re "gres/gpu=($number)" { incr match if {$expect_out(1,string) != $job_gpus} { send_user "\nFAILURE: Bad job GPU count reported by sacct ($expect_out(1,string) != $job_gpus)\n" set exit_code 1 } exp_continue } -re "gres/gpu:($alpha_numeric_under)=($number)" { if {$have_gpu_types != 0} { incr match if {$expect_out(2,string) != $job_gpus} { send_user "\nFAILURE: Bad job GPU count reported by sacct ($expect_out(2,string) != $job_gpus)\n" set exit_code 1 } } exp_continue } eof { wait } } if {$match != 3} { send_user "\nFAILURE: Missing job GPU count report from sacct ($match != 3)\n" } exec $bin_rm -f $file_out log_user 1 } # # Validate the job, batch step and step 0 of a job have the proper GPU counts # No step to test if step_gpus == -1 # # NOTE: AllocGRES and ReqGRES values for all steps (including batch step) # are reported based upon the job specification # proc test_out_file { file_out target } { global alpha_numeric_under number bin_cat exit_code if {[wait_for_file $file_out] != 0} { send_user "\nFAILURE: no output file\n" exit 1 } set match 0 spawn $bin_cat $file_out expect { -re "TRES=.*,gres/gpu=($number)" { set match $expect_out(1,string) exp_continue } -re "TRES=.*,gres/gpu:($alpha_numeric_under)=($number)" { set match $expect_out(2,string) exp_continue } eof { wait } } if {$match != $target} { send_user "\nFAILURE: Failed to account for proper GPU count ($match != $target)\n" set exit_code 1 } } print_header $test_id if { [test_account_storage] == 0 } { send_user "\nWARNING: This test can't be run without a usable AccountStorageType\n" exit 0 } set store_tres [string tolower [get_acct_store_tres]] set store_gpu [string first "gres/gpu" $store_tres] if {$store_gpu == -1} { send_user "\nWARNING: This test requires accounting for GPUs\n" exit $exit_code } elseif {[test_front_end]} { send_user "\nWARNING: This test is incompatible with front-end systems\n" exit $exit_code } 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 } set def_part_name [default_partition] set nb_nodes [get_node_cnt_in_part $def_part_name] send_user "\nDefault partition node count is $nb_nodes\n" if {$nb_nodes > 2} { 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 < 2} { send_user "\nWARNING: This test requires 2 or more GPUs in the default partition\n" exit 0 } get_node_config set cpus_per_node [expr $sockets_per_node * $cpus_per_socket] set sockets_with_gpus [get_gpu_socket_count $gpu_cnt $sockets_per_node] send_user "GPUs per node 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 "CPUs per socket is $cpus_per_socket\n" send_user "CPUs per node is $cpus_per_node\n" if {$cpus_per_node < 3} { send_user "\nWARNING: This test requires 3 or more CPUs per node in the default partition\n" exit 0 } # # Test --gpus-per-node option by job # make_bash_script $file_in1 " $scontrol -dd show job \${SLURM_JOBID} | grep gpu exit 0" send_user "\n\nTEST 1: --gpus-per-node option by job\n" set target [expr $nb_nodes * 2] exec $bin_rm -f $file_out set job_id 0 set timeout $max_job_delay spawn $sbatch --gres=craynetwork:0 --gpus-per-node=2 -N$nb_nodes -t1 -o $file_out -J "test$test_id" ./$file_in1 expect { -re "Submitted batch job ($number)" { set job_id $expect_out(1,string) exp_continue } timeout { send_user "\nFAILURE: sbatch not responding\n" set exit_code 1 } eof { wait } } if {$job_id == 0} { send_user "\nFAILURE: job not submitted\n" exit 1 } if {[wait_for_job $job_id "DONE"] != 0} { send_user "\nFAILURE: job $job_id did not complete\n" cancel_job $job_id exit 1 } if {[wait_for_file $file_out] != 0} { send_user "\nFAILURE: no output file\n" exit 1 } set have_gpu_types 0 set match 0 spawn $bin_cat $file_out expect { -re "TRES=.*,gres/gpu=($number)" { set match $expect_out(1,string) exp_continue } -re "TRES=.*,gres/gpu:($alpha_numeric_under)=($number)" { if {$match == 0} { set have_gpu_types 1 set match $expect_out(2,string) } exp_continue } eof { wait } } set target [expr $nb_nodes * 2] if {$match != $target} { send_user "\nFAILURE: failed to account for proper GPU count ($match != $target)\n" set exit_code 1 } test_acct $job_id $target -1 $have_gpu_types # # Test --gpus option by job # send_user "\n\nTEST 2: --gpus option by job\n" exec $bin_rm -f $file_out if {$nb_nodes >= 2 || $gpu_cnt >= 2} { set target 3 } else { set target 2 } set job_id 0 spawn $sbatch --gres=craynetwork:0 --gpus=$target -N$nb_nodes -t1 -o $file_out -J "test$test_id" ./$file_in1 expect { -re "Submitted batch job ($number)" { set job_id $expect_out(1,string) exp_continue } timeout { send_user "\nFAILURE: sbatch not responding\n" set exit_code 1 } eof { wait } } if {$job_id == 0} { send_user "\nFAILURE: job not submitted\n" exit 1 } if {[wait_for_job $job_id "DONE"] != 0} { send_user "\nFAILURE: job $job_id did not complete\n" cancel_job $job_id exit 1 } test_out_file $file_out $target test_acct $job_id $target -1 $have_gpu_types # # Test --gpus-per-task option by job # send_user "\n\nTEST 3: --gpus-per-task option by job\n" exec $bin_rm -f $file_out if {$cpus_per_node >= 2 && $nb_nodes >= 2} { set nb_tasks 3 } elseif {$cpus_per_node >= 2 || $nb_nodes >= 2} { set nb_tasks 2 } else { set nb_tasks 1 } set job_id 0 spawn $sbatch --gres=craynetwork:0 --gpus-per-task=1 -N$nb_nodes -n$nb_tasks -t1 -o $file_out -J "test$test_id" ./$file_in1 expect { -re "Submitted batch job ($number)" { set job_id $expect_out(1,string) exp_continue } timeout { send_user "\nFAILURE: sbatch not responding\n" set exit_code 1 } eof { wait } } if {$job_id == 0} { send_user "\nFAILURE: job not submitted\n" exit 1 } if {[wait_for_job $job_id "DONE"] != 0} { send_user "\nFAILURE: job $job_id did not complete\n" cancel_job $job_id exit 1 } test_out_file $file_out $nb_tasks test_acct $job_id $nb_tasks -1 $have_gpu_types # # Test --gpus-per-socket option by job # send_user "\n\nTEST 4: --gpus-per-socket option by job\n" exec $bin_rm -f $file_out if {$sockets_with_gpus >= 2} { set nb_sockets 2 set cpus_per_task $cpus_per_socket } else { set nb_sockets 1 set cpus_per_task 1 } set job_id 0 spawn $sbatch --gres=craynetwork:0 --gpus-per-socket=1 -N$nb_nodes --ntasks=$nb_nodes --sockets-per-node=$nb_sockets --cpus-per-task=$cpus_per_task -t1 -o $file_out -J "test$test_id" ./$file_in1 expect { -re "Submitted batch job ($number)" { set job_id $expect_out(1,string) exp_continue } timeout { send_user "\nFAILURE: sbatch not responding\n" set exit_code 1 } eof { wait } } if {$job_id == 0} { send_user "\nFAILURE: job not submitted\n" exit 1 } if {[wait_for_job $job_id "DONE"] != 0} { send_user "\nFAILURE: job $job_id did not complete\n" cancel_job $job_id exit 1 } set target [expr $nb_nodes * $nb_sockets] test_out_file $file_out $target test_acct $job_id $target -1 $have_gpu_types # # Test --gpus-per-node option by step # make_bash_script $file_in1 " $srun ./$file_in2 exit 0" make_bash_script $file_in2 " if \[ \$SLURM_PROCID -eq 0 \]; then $scontrol show step \${SLURM_JOBID}.\${SLURM_STEPID} fi exit 0" send_user "\n\nTEST 5: --gpus-per-node option by step\n" set target [expr $nb_nodes * 2] exec $bin_rm -f $file_out set job_id 0 set timeout $max_job_delay spawn $sbatch --gres=craynetwork:0 --gpus-per-node=2 -N$nb_nodes -t1 -o $file_out -J "test$test_id" ./$file_in1 expect { -re "Submitted batch job ($number)" { set job_id $expect_out(1,string) exp_continue } timeout { send_user "\nFAILURE: sbatch not responding\n" set exit_code 1 } eof { wait } } if {$job_id == 0} { send_user "\nFAILURE: job not submitted\n" exit 1 } if {[wait_for_job $job_id "DONE"] != 0} { send_user "\nFAILURE: job $job_id did not complete\n" cancel_job $job_id exit 1 } set target [expr $nb_nodes * 2] test_out_file $file_out $target test_acct $job_id $target $target $have_gpu_types # # Test --gpus option by step # send_user "\n\nTEST 6: --gpus option by step\n" exec $bin_rm -f $file_out if {$nb_nodes >= 2 || $gpu_cnt >= 2} { set target 3 } else { set target 2 } set job_id 0 spawn $sbatch --gres=craynetwork:0 --gpus=$target -N$nb_nodes -t1 -o $file_out -J "test$test_id" ./$file_in1 expect { -re "Submitted batch job ($number)" { set job_id $expect_out(1,string) exp_continue } timeout { send_user "\nFAILURE: sbatch not responding\n" set exit_code 1 } eof { wait } } if {$job_id == 0} { send_user "\nFAILURE: job not submitted\n" exit 1 } if {[wait_for_job $job_id "DONE"] != 0} { send_user "\nFAILURE: job $job_id did not complete\n" cancel_job $job_id exit 1 } test_out_file $file_out $target test_acct $job_id $target $target $have_gpu_types # # Test --gpus-per-task option by step # send_user "\n\nTEST 7: --gpus-per-task option by step\n" exec $bin_rm -f $file_out if {$cpus_per_node >= 2 && $nb_nodes >= 2} { set nb_tasks 3 } elseif {$cpus_per_node >= 2 || $nb_nodes >= 2} { set nb_tasks 2 } else { set nb_tasks 1 } set job_id 0 spawn $sbatch --gres=craynetwork:0 --gpus-per-task=1 -N$nb_nodes -n$nb_tasks -t1 -o $file_out -J "test$test_id" ./$file_in1 expect { -re "Submitted batch job ($number)" { set job_id $expect_out(1,string) exp_continue } timeout { send_user "\nFAILURE: sbatch not responding\n" set exit_code 1 } eof { wait } } if {$job_id == 0} { send_user "\nFAILURE: job not submitted\n" exit 1 } if {[wait_for_job $job_id "DONE"] != 0} { send_user "\nFAILURE: job $job_id did not complete\n" cancel_job $job_id exit 1 } test_out_file $file_out $nb_tasks test_acct $job_id $nb_tasks $nb_tasks $have_gpu_types # # Test --gpus-per-socket option by step # send_user "\n\nTEST 8: --gpus-per-socket option by step\n" exec $bin_rm -f $file_out if {$sockets_with_gpus >= 2} { set nb_sockets 2 set cpus_per_task $cpus_per_socket } else { set nb_sockets 1 set cpus_per_task 1 } set job_id 0 spawn $sbatch --gres=craynetwork:0 --gpus-per-socket=1 -N$nb_nodes --ntasks=$nb_nodes --sockets-per-node=$nb_sockets --cpus-per-task=$cpus_per_task -t1 -o $file_out -J "test$test_id" ./$file_in1 expect { -re "Submitted batch job ($number)" { set job_id $expect_out(1,string) exp_continue } timeout { send_user "\nFAILURE: sbatch not responding\n" set exit_code 1 } eof { wait } } if {$job_id == 0} { send_user "\nFAILURE: job not submitted\n" exit 1 } if {[wait_for_job $job_id "DONE"] != 0} { send_user "\nFAILURE: job $job_id did not complete\n" cancel_job $job_id exit 1 } set target [expr $nb_nodes * $nb_sockets] test_out_file $file_out $target test_acct $job_id $target $target $have_gpu_types # # Test --gpus-per-task option by step # send_user "\n\nTEST 9: --gpus-per-task option by step\n" exec $bin_rm -f $file_out if {$cpus_per_node >= 2 && $nb_nodes >= 2} { set job_tasks 3 set step_tasks 2 } elseif {$cpus_per_node >= 2 || $nb_nodes >= 2} { set job_tasks 2 set step_tasks 2 } else { set job_tasks 1 set step_tasks 1 } if {$nb_nodes >= 2} { set step_nodes 2 } else { set step_nodes 1 } make_bash_script $file_in1 " $srun -N$step_nodes -n$step_tasks ./$file_in2 exit 0" set job_id 0 spawn $sbatch --gres=craynetwork:0 --gpus-per-task=1 -N$nb_nodes -n$job_tasks -t1 -o $file_out -J "test$test_id" ./$file_in1 expect { -re "Submitted batch job ($number)" { set job_id $expect_out(1,string) exp_continue } timeout { send_user "\nFAILURE: sbatch not responding\n" set exit_code 1 } eof { wait } } if {$job_id == 0} { send_user "\nFAILURE: job not submitted\n" exit 1 } if {[wait_for_job $job_id "DONE"] != 0} { send_user "\nFAILURE: job $job_id did not complete\n" cancel_job $job_id exit 1 } test_out_file $file_out $step_tasks test_acct $job_id $job_tasks $step_tasks $have_gpu_types if {$exit_code == 0} { exec $bin_rm -f $file_in1 $file_in2 $file_out send_user "\nSUCCESS\n" } exit $exit_code