How do I achieve with tune.run
today?
- on_experiment_begin_callback(ray_config) which runs some prep script only on the head node using the ray_config
- on_experiment_end_callback(ray_config) which runs some clean up script only on the head node using the ray_config
Can you say more about what you are trying to achieve here?
I’m trying to create and teardown a custom cluster of machines in the respective cbs
Is this the same cluster that Tune is running on? If it is, I don’t think these callbacks are intended for that.
Yes, they are in the same cluster. I’m thinking of writing a wrapper on top of tune.run
like so,
def my_tune_run(cluster_config, **kwargs):
cid = my_start_cluster(cluster_config)
tune.run(**kwargs)
my_stop_cluster(cid)
Got this idea while brainstorming with your here