pl distributed training on cluster needs configuration

When calling pl.Trainer with accelerator='ddp', the training on cluster will not stop properly