Update packages and tests in the straight dope nightly (apache#12744)

* [apache#12345] Enabling two tests in the Straight Dope Nightly.

Two straight dope notebook tests were disabled due to a timeout so they
were disabled. I've updated one of the notebooks (rnn-gluon) to use the
gpu instead of the cpu so it takes ~ 5 minutes on a p3.2xl, and verified
the other notebook takes a minute and was a false alarm (visual-qa). The
PR in the Straight Dope is:
zackchase/mxnet-the-straight-dope#540

* Add dependency for IPython update.

* Detect errors in notebook execution failure.

* Clean up of naming in retry code.

ci/docker/install/ubuntu_nightly_tests.sh

@@ -32,5 +32,5 @@ apt-get -y install time
 apt-get install -y subversion maven -y #>/dev/null
 
 # Packages needed for the Straight Dope Nightly tests.
-pip2 install pandas scikit-image
-pip3 install pandas scikit-image
+pip2 install pandas scikit-image prompt_toolkit
+pip3 install pandas scikit-image prompt_toolkit

tests/nightly/straight_dope/test_notebooks_single_gpu.py

@@ -35,13 +35,11 @@
     'chapter02_supervised-learning/environment',
     'chapter03_deep-neural-networks/kaggle-gluon-kfold',
     'chapter04_convolutional-neural-networks/deep-cnns-alexnet',  # > 10 mins.
-    'chapter05_recurrent-neural-networks/rnns-gluon', # > 10 mins.
     'chapter06_optimization/gd-sgd-scratch',  # Overflow warning is intended.
     'chapter06_optimization/gd-sgd-gluon',  # Overflow warning is intended.
     'chapter07_distributed-learning/multiple-gpus-scratch',
     'chapter07_distributed-learning/multiple-gpus-gluon',
     'chapter07_distributed-learning/training-with-multiple-machines',
-    'chapter08_computer-vision/visual-question-answer', # > 10 mins.
     'chapter11_recommender-systems/intro-recommender-systems',  # Early draft, non-working.
     'chapter12_time-series/intro-forecasting-gluon',
     'chapter12_time-series/intro-forecasting-2-gluon',
@@ -178,6 +176,9 @@ def test_lstm_scratch(self):
     def test_gru_scratch(self):
         assert _test_notebook('chapter05_recurrent-neural-networks/gru-scratch')
 
+    def test_rnn_gluon(self):
+        assert _test_notebook('chapter05_recurrent-neural-networks/rnns-gluon')
+
     # Chapter 6
 
     def test_optimization_intro(self):
@@ -227,6 +228,10 @@ def test_object_detection(self):
     def test_fine_tuning(self):
         assert _test_notebook('chapter08_computer-vision/fine-tuning')
 
+    def test_visual_qa(self):
+        assert _test_notebook('chapter08_computer-vision/visual-question-answer')
+
+
     # Chapter 9
 
     def test_tree_lstm(self):

tests/utils/notebook_test/__init__.py

@@ -32,7 +32,7 @@
 
 IPYTHON_VERSION = 4  # Pin to ipython version 4.
 TIME_OUT = 10*60  # Maximum 10 mins/test. Reaching timeout causes test failure.
-RETRIES = 8
+ATTEMPTS = 8
 KERNEL_ERROR_MSG = 'Kernel died before replying to kernel_info'
 
 
@@ -80,23 +80,28 @@ def run_notebook(notebook, notebook_dir, kernel=None, no_cache=False, temp_dir='
         else:
             eprocessor = ExecutePreprocessor(timeout=TIME_OUT)
 
+        success = False
         # There is a low (< 1%) chance that starting a notebook executor will fail due to the kernel
         # taking to long to start, or a port collision, etc.
-        for i in range(RETRIES):
+        for i in range(ATTEMPTS):
             try:
                 nb, _ = eprocessor.preprocess(notebook, {'metadata': {'path': working_dir}})
+                success = True
             except RuntimeError as rte:
                 # We check if the exception has to do with the Jupyter kernel failing to start. If
-                # not, we rethrow to prevent the notebook from erring RETRIES times. It is not ideal
-                # to inspect the exception message, but necessary for retry logic, as Jupyter client
-                # throws the generic RuntimeError that can be confused with other Runtime errors.
+                # not, we rethrow to prevent the notebook from erring ATTEMPTS times. It is not
+                # ideal to inspect the exception message, but necessary for retry logic, as Jupyter
+                # client throws the generic RuntimeError that can be confused with other Runtime
+                # errors.
                 if str(rte) != KERNEL_ERROR_MSG:
                     raise rte
 
-                logging.info("Error starting preprocessor: {}. Attempt {}/{}".format(str(rte), i+1, RETRIES))
+                logging.info("Error starting preprocessor: {}. Attempt {}/{}".format(str(rte), i+1, ATTEMPTS))
                 time.sleep(1)
                 continue
             break
+        if not success:
+            errors.append("Error: Notebook failed to run after {} attempts.".format(ATTEMPTS))
     except Exception as err:
         err_msg = str(err)
         errors.append(err_msg)