args = parser.parse_args() os.environ['TRAINS_PROC_MASTER_ID'] = f"1:{args.trains_id}" os.environ['OMPI_COMM_WORLD_NODE_RANK'] = str(randint(0, sys.maxsize)) trains_task = trains.Task.init(project_name=f'{ALG_NAME}-inference', task_name='inference')
failes on init
how can I pass to task_init the task_id? should it also be in some env?
Ok, so it is master node and not consumer producer pattern..
Notice that you have to Have the task already started by the Master process
ValueError: Task object can only be updated if created or in_progress
so if i plot image with matplot lib..it would not upload? i need use the logger.
yes you are correct, OS environment:TRAINS_PROC_MASTER_ID=1:task_id_here
It has to be alive so all the "child nodes" could report to it....
FranticCormorant35 As far as I understand what you have going is a multi-node setup, that you manage yourself. Something like Horovod Torch distributed or any MPI setup. Since Trains support all of the above standard multi-node. The easiest way is to do the following:
On the master Node set OS environment:OMPI_COMM_WORLD_NODE_RANK=0
Then on any client node:OMPI_COMM_WORLD_NODE_RANK=unique_client_node_number
In all processes you can Call Task.init - with all the automagic kicking in. The Master node will be the only one registering the execution section of the experiment (i.e. git arg parser etc.) while all the rest will be logged as usual (console output, tensorboard matplotlib etc.)
How does that sound?
should OMPI_COMM_WORLD_NODE_RANK be number or can be some guid?
so if i plot image with matplot lib..it would not upload? i need use the logger.
Correct, if you have no "main" task , no automagic 😞
so how can i make it run with the "auto magic"
Automagic logs a single instance... unless those are subprocesses, in which case, the main task takes care of "copying" itself to the subprocess.
Again what is the use case for multiple machines?
Should work out of the box, as long as the task was started. You can forcefully start the task with:task.mark_started()
so how can i make it run with the "auto magic"
hhhhmm..then i can not get master task params
os.environ['TRAINS_PROC_MASTER_ID'] = args.trains_id os.environ['OMPI_COMM_WORLD_NODE_RANK'] = str(randint(0, sys.maxsize)) trains_task = trains.Task.init(project_name=f'{ALG_NAME}-inference', task_name='inference') print(type(trains_task))
<class 'trains.task.Task.init.<locals>._TaskStub'>
so if i load task and not init it..so it is not the main one?
os.environ['TRAINS_PROC_MASTER_ID'] = '1:da0606f2e6fb40f692f5c885f807902a' os.environ['OMPI_COMM_WORLD_NODE_RANK'] = '1' task = Task.init(project_name="examples", task_name="Manual reporting") print(type(task))
Should be: <class 'trains.task.Task'>
os.environ['TRAINS_PROC_MASTER_ID'] = args.trains_id
it should be '1:'+args.trains_id
os.environ['TRAINS_PROC_MASTER_ID'] = '1:{}'.format(args.trains_id)
Also str(randint(1, sys.maxsize))
Correct, and that also means the code the runs is not auto-magically logged.
Logger.current_logger()
Will return the logger for the "main" Task.
The "Main" task is the task of this process, a singleton for the process.
All other instances create Task object. you can have multiple Task objects and log different things to them, but you can only have a single "main" Task (the one created with Task.init).
All the auto-magic stuff is logged automatically to the "main" task.
Make sense ?