tf.data.Dataset.from_generator() empties after use even with repeat()

[mr-ubik]

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): Yes
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Arch Linux
• TensorFlow installed from (source or binary): PyPI binary
• TensorFlow version (use command below): 2.0.0-dev20190526
• Python version: 3.7

Describe the current behavior

A tf.data.Dataset created using from_generator() is consumed and not recreated/repeated even when explicitly invoking the repeat() method.

Describe the expected behavior

If repeat() is being invoked, repeat the dataset.

Code to reproduce the issue

#25254 (comment)

Other info / logs

The issue is present both when manually iterating over it in a custom training scenario and when using tf.keras.Model.fit().

[mr-ubik]

Actually, it was my fault 🤦‍♂️ I had a generator emptying and not being recreated way way upstream in my pipeline. Sorry ❤️

[zhenghh04]

I encountered the same issue.
I had a python generator: gen
tf.data.Dataset.from_generator(lambda: gen, ....)
It does not support repeat. But if I define a lambda generator outside of tf.data.from_generator(gen_lambda, ...) as follows

gen_lambda = lambda: gen
tf.data.Dataset.from_generator(gen_lambda,  ....) solved the issue. 

It supports repeat method.