Accommodating more channels in sagemaker PIPE mode

I am training a classification model using AWS Sagemaker -TensorFlow. My training dataset is huge and distributed in 4 folders in the same AWS s3 bucket.

I defined input channels like this
inputs = {
'train1' : folder1,
'train2' : folder2,
'train3': folder3,
'train4':folder4,

'valid':folder
}

I am passing these channels 'ids' into my main train code and then reading the data using PIPE mode like this,
all_data = []
if mode = train:
for id in ids:
data = PipeModeDataset(channel=id, format = 'TFRecord')
data = parsing data here
all_data.append(data)

Now I am using _all_data_ as my whole data and doing augmentation in it and then passing it to the training script.
I got an error while doing this, (error related to data). sometimes training hangs.

What I want to know is the correct way of using multiple channels for single training using PIPE mode

Thanks

Start with the training script's PipeModeDataset usage and the AWS SageMaker input-channel setup shown in the issue. Reproduce the four-channel PIPE-mode run, inspect the reported data error and hangs, and establish the supported multi-channel behavior as the done condition.

Written by the indexing model from the issue text.