def show(id):
'''
show [id] show task [id] details.
'''
(id, ), err_msg = check_input(((id, int), ))
if err_msg:
return 1, err_msg
task = task_manage.get_task(id)
if task is None:
return 1, f'[error]: can not found id {id} in task queue.'
table = pt.PrettyTable(['name', 'value'])
table.border = False
table.align = 'l'
table.add_row(['task id:', task.id])
table.add_row(['task pid:', task.pid])
table.add_row(['priority:', task.priority])
table.add_row(['use gpu:', task.gpu])
table.add_row(['run times:', task.run_times])
table.add_row(['status:', task.status])
table.add_row(['out file:', task.out_path])
table.add_row(['pwd:', task.pwd])
table.add_row(['cmds:', " ".join(task.cmds)])
return 0, str(table)
def ls(*, all=False):
'''
ls ls GPU task queue status
Options:
--all default ls only show <80 commands,
use `all` to show all commands.
--sort show by different sort type.
can use: ['id', 'priority', 'show', 'run']
'''
(all, ), err_msg = check_input(((all, bool), ))
if err_msg: return 1, err_msg
tasks = task_manage.tasks
table = pt.PrettyTable(
['[ID]', 'num', 'status', 'run_times', 'pwd', 'cmds'])
table.border = False
for i, task in enumerate(tasks):
status = str(task.status
) + f'(GPU:{task.gpu})' if task.gpu is not None else str(
task.status)
if not all:
table.add_row([
task.id, i, status, task.run_times, task.pwd + '#',
" ".join(task.cmds)[:80]
])
else:
table.add_row([
task.id, i, status, task.run_times, task.pwd + '#',
" ".join(task.cmds)
])
return 0, str(table)
def rm(id):
'''
rm [id] remove task [id] from manage, if task is running, kill it.
'''
(id, ), err_msg = check_input(((id, int), ))
if err_msg:
return 1, err_msg
if task_manage.rm_task(id):
return 0, f'[info]: del task {id}'
else:
return 1, f'[error]: can not found {id} in task queue.'
def kill(id):
'''
kill [id] kill task [id]
'''
(id, ), err_msg = check_input(((id, int), ))
if err_msg:
return 1, err_msg
task = task_manage.get_task(id)
if task is None:
return 1, f'[error]: can not found id {id} in task queue.'
return task.kill()
def mv(id, index=0, *args, **kwargs):
'''
mv [id] [index(default=0)] move [id] to [index]
'''
(id, index), err_msg = check_input(((id, int), (index, int)), args, kwargs)
if err_msg:
return 1, err_msg
if index > len(task_manage):
return 2, f'[error]: index {index} is bigger than task queue length({len(task_manage)})'
if task_manage.mv_task(id, index):
return 0, f'[info]: move {id} to the first'
else:
return 1, f'[error]: can not found task {id}'
def start(id=None):
'''
start [iddefalut=None] Force start task(s).
Information:
gpulimit start running `check_and_start`, and auto start new task.
gpulimit start 1 Force start task [id].
'''
if id is None:
return task_manage.scheduling.user_start_scheduling(task_manage)
(id, ), err_msg = check_input(((id, int), ), )
if err_msg:
return 1, err_msg
return task_manage.scheduling.user_start_scheduling(task_manage, id)
def debug(id):
'''
debug [id] if task [id] is `CMD_ERROR`, use this show error traceback.
Example:
debug 1 show task 1 error traceback.
'''
(id, ), err_msg = check_input(((id, int), ), )
if err_msg:
return 1, err_msg
task = task_manage.get_task(id)
if task is None:
return 1, f'[error]: can not found task[{id}]'
return 0, str(task.debug_msg)
def get_output_filename(id):
'''
log [id] show [id] output.
Example:
gpulimit log 1 show task(id=1) output.
gpulimit log main show manage background log info.
'''
if id == 'main':
return 0, task_manage.log_file
(id, ), err_msg = check_input(((id, int), ), )
if err_msg:
return 1, err_msg
task = task_manage.get_task(id)
if task is None:
return 1, 'Error'
return 0, os.path.abspath(task.out_path)