Thank you for your post. We noticed you have not filled out the following field in the issue template. Could you update them if they are relevant in your case, or leave them as N/A? Thanks.
Have I written custom code
OS Platform and Distribution
TensorFlow installed from
TensorFlow version
Bazel version
CUDA/cuDNN version
GPU model and memory
Exact command to reproduce

I have a similar issue -
I'm trying to load a trained model for inference. In the code snippet below, features:0 is the name of one of the tensors returned by dataset iterator. Is it possible to directly feed the tensor without having to initialize the iterator ?

Code

with tf.Session(graph=tf.Graph()) as session:
    graph_meta = tf.train.latest_checkpoint(model_dir) + '.meta'
    saver = tf.train.import_meta_graph(os.path.join(model_dir, graph_meta))
    saver.restore(session, tf.train.latest_checkpoint(model_dir))
    feed_dict = {
        'features:0': x_test # shape 102x13
    }
    predictions = session.run('logits:0', feed_dict)
    print(predictions.shape)

Error

FailedPreconditionError: GetNext() failed because the iterator has not been initialized. Ensure that you have run the initializer operation for this iterator before getting the next element.
	 [[Node: IteratorGetNext = IteratorGetNext[output_shapes=[[?,13], [?], [?]], output_types=[DT_DOUBLE, DT_DOUBLE, DT_DOUBLE], _device="/job:localhost/replica:0/task:0/device:CPU:0"](Iterator)]]

@taehyunkim1527 Can you share a complete reproducible example of the problem? I was unable to reproduce the problem with the code fragment in your example, although it was possible to reproduce it by adding arguments to tf.train.export_meta_graph(). I am in the process of fixing the latter problems, but it would be great to confirm that the fix works for your issue too.

@suryasumukh I think you're running into a different problem. You certainly can feed values over the tensors returned from Iterator.get_next(), but you need to ensure that you feed all of the tensors returned by the iterator that might be used in the run() call. If you continue to have problems with this, please post a question on Stack Overflow.

@mrry When i use that series of code of 'exporting and importing' in tensorflow/benchmarks's ImageNet tasks, I experienced this error message.
I tried to implement a simple code to make the same error message today, but it failed.
I think it is not a problem of tf.data.Dataset() and let me debug again.
Thank you for your kindness.

Thanks for confirming that the problem doesn't arise with that exact code! I have a change in the pipeline that will make this path work in more cases (e.g. when using clear_devices=True or a scope prefix), so I'll reopen this issue until it lands.

@mrry
Hi, I'm who opened this question formerly.
And i found what was the problem.
if we give arg 'clear_devices' as True for the function 'export_scoped_meta_graph()' of python\framework\meta_graph.py
It starts to make empty graph_def and copy previous node_defs newly.
In this process, this code does not consider about graph_def.library.
But the information about 'map_function' resides in graph_def.library

@suryasumukh In order to retrain the pre-trained model, the initializer of data iterator can be declared as a tf.operation with a name while training for the first time.

    data_iter = dataset.make_initializable_iterator()
    data_iter_init = data_iter.make_initializer(dataset, name='Data_itr_init')
    next_batch = data_iter.get_next()

Then, it can be sess.run with the name and fed with training data.
sess.run('Data_itr_init', feed_dict={"Model_in:0": train_X, "Model_out:0": train_Y})

@suryasumukh In order to retrain the pre-trained model, the initializer of data iterator can be declared as a tf.operation with a name while training for the first time.

    data_iter = dataset.make_initializable_iterator()
    data_iter_init = data_iter.make_initializer(dataset, name='Data_itr_init')
    next_batch = data_iter.get_next()

Then, it can be sess.run with the name and fed with training data.
sess.run('Data_itr_init', feed_dict={"Model_in:0": train_X, "Model_out:0": train_Y})

Not the point