This submission uses a Jupyter Notebook to explore the data in csv files included with the repo.
All the csv files at setup/data/*.csv are imported into a notebook named Visualization under the notebooks folder.
Due to the large file size of the slug generated for the Jupyter Notebook, deployment to Predix's free plan fails but it works on Heroku.
This document will describe how to run the Jupyter Notebook locally as well as on Heroku.
Exploration of the csv data revealed that there are 176,534 datapoints in the timeseries data and only 63,456 datapoints have non-null values but there are only 36,278 unique timestamps.
To summarize, there are:
- 11 unique nodes (nodes 'N1'-'N10' and a 'Real-time Seismic Amplitude Measurement' or 'RSAM' node);
- 61 unique sensors (each of node 'N1'-'N10' had 6 sensors plus the 'RSAM' node sensor i.e. 6 * 10 + 1)
- 7 unique data types
You must clone the repo and follow the steps below to be able to successfully run the app under your Predix account.
This is just an overview of the steps I took on Windows 8.1 to get the repo working under my Predix account. I had to make a few minor changes to the generated manifest.yml file as can be seen in my submission. Those changes were: setting the instances to 2 and explicitly setting the memory to 256MB for deployment to Predix to succeed i.e:
applications:
- name: my-predix-app01
memory: 256MB
instances: 2
- Download the
cfCLI binary for your platform then login with your Predix credentials.
cf login -a https://api.system.aws-usw02-pr.ice.predix.io
- Next create a space called
volcanothen set it as default.
cf create-space volcano
cf target -s volcano
- Follow the remaining commands (tested on Python 2.7.11) to deploy the app under your Predix account.
# Clone and install the repo
git clone https://github.com/PredixDev/predix-volcano-app
cd predix-volcano-app/
pip install -r requirements.txt
# Publishes the data to Predix and generates your `manifest.yml`
cd setup/
python create_services.py
python ingest_data.py
# Generate the HTML assets for the Dashboard
cd ../app/dashboard
npm install
node .\node_modules\bower\lib\bin\bower.js install # bower install
node .\node_modules\gulp\bin\gulp.js # gulp
# Edit the manifest.yml to set the memory and instance as above.
vim manifest.yml
# Publish the assets to Predix
cd ../..
cf push
- Visit the printed URL to view the demo (in my case this was https://my-predix-app01.run.aws-usw02-pr.ice.predix.io/).
Before you begin, please overlay all the changes from my submission in your local clone of the repo.
Most of the files listed in the change log are newly added files.
-
Change Log:
- environment.yml
- jupyter-manifest.yml
- jupyter-Procfile
- Jupyter_README.md
- manifest.yml (changes explained above)
- notebooks\
- start_jupyter
- app.json
- .jupyter\
- .gitignore (existing file)
-
Please check out the repo from GitHub https://github.com/PredixDev/predix-volcano-app
-
Next follow the
Predix Preparationinstructions above to generate a validmanifest.ymlfor Predix. -
Ensure you have modified the
manifest.ymlto fix the memory issue as described above. -
Next overlay the newly added files and related changes in my submission over your working copy.
-
Next rename
jupyter-ProcfiletoProcfile.
- To deploy the Jupyter Notebook locally, please follow the instructions below.
cd D:\Apps\Python_Environments
python -m pip install --upgrade pip
pip install virtualenv
virtualenv py2env
.\py2env\Scripts\activate
# Python 3 instead use:
# python -m venv py2env
# .\py2env\Scripts\activate
cd predix-volcano-app/
pip install jupyter pandas matplotlib numpy
# Activate the env
cd D:\Apps\Python_Environments
.\py2env\Scripts\activate
# Change to the submission folder
cd predix-volcano-app/
jupyter notebook
- Next visit the URL in the output to see the Notebook (in my case this was http://localhost:8888/).
An outbound connection to GitHub during buildpack compilation times out causing deployment to fail.
I suspect that the timeout error is due to the strict resource limits imposed by the free Predix plan.
It turns out that Predix does not yet support Docker so pushing a docker image of the Python buildpack / dependencies as a workaround will not work either.
# First attempt using https://github.com/pl31/heroku-buildpack-conda.git
...
notebook-4.4.1 100% |###############################| Time: 0:00:00 56.39 MB/s
Collecting git+git://github.com/ipython-contrib/jupyter_contrib_nbextensions.git
Cloning git://github.com/ipython-contrib/jupyter_contrib_nbextensions.git to /tmp/pip-2brus181-build
fatal: unable to connect to github.com:
github.com[0: 192.30.255.113]: errno=Connection timed out
github.com[1: 192.30.255.112]: errno=Connection timed out
Command "git clone -q git://github.com/ipython-contrib/jupyter_contrib_nbextensions.git /tmp/pip-2brus181-build" failed with error code 128 in None
CondaValueError: pip returned an error.
widgetsnbexten 100% |###############################| Time: 0:00:00 26.67 MB/s
ipywidgets-7.0 100% |###############################| Time: 0:00:00 33.09 MB/s
Staging failed: Buildpack compilation step failed
FAILED
Error restarting application: BuildpackCompileFailed
# Second attempt using python_minconda_buildpack (turns out it is disabled on Predix)
Starting app jupyter-predix-app01 in org xxxxxxxx@gmail.com / space volcano as xxxxxxxx@gmail.com...
-----> Downloaded app package (14M)
Staging failed: Buildpack compilation step failed
FAILED
Error restarting application: BuildpackCompileFailed
# cf buildpacks
Getting buildpacks...
buildpack position enabled locked filename
custom_node_buildpack 1 true false
java_buildpack 2 true false java-buildpack-v3.6.zip
ruby_buildpack 3 true false ruby_buildpack-cached-v1.6.34.zip
nodejs_buildpack 4 true false nodejs_buildpack-cached-v1.5.29.zip
python_buildpack 5 true false python_buildpack-cached-v1.5.15.zip
predix_openresty_buildpack 6 true false staticfile_buildpack-cached-v1.2.1.zip
php_buildpack 7 true false php_buildpack-cached-v4.3.27.zip
staticfile_buildpack 8 true false staticfile_buildpack-cached-v1.3.17.zip
go_buildpack 9 true false go_buildpack-cached-v1.7.18.zip
java_buildpack_large_heap 10 true false java_buildpack_large_heap.zip
matlab_buildpack_r2011b 11 true false matlab-buildpack-r2011b.zip
matlab_buildpack_r2012a 12 true false matlab-buildpack-r2012a.zip
matlab_buildpack_r2015a 13 true false matlab-buildpack-r2015a.zip
binary_buildpack 14 true false binary_buildpack-cached-v1.0.9.zip
java-buildpack 15 true false java-buildpack-v3.13.zip
java-offline-buildpack 16 true false java-buildpack-offline-v3.13.zip
python_minconda_buildpack 17 false false python_minconda_buildpack.zip
dotnet_core_buildpack 18 true false dotnet-core_buildpack-cached-v1.0.11.zip
Fortunately, the Jupyter Notebook deployed successfully on Heroku.
- If you have access to a paid plan on Predix (i.e. higher than the 512MB limit), these are the commands to deploy the Jupyter Notebook
cd predix-volcano-app/
chmod +x ./start_jupyter # please do $(dos2unix start_jupyter) if on Windows
cf login -a https://api.system.aws-usw02-pr.ice.predix.io
cf push -f jupyter-manifest.yml
cf set-env jupyter-predix-app01 JUPYTER_NOTEBOOK_PASSWORD password
cf restage jupyter-predix-app01
cf logs jupyter-predix-app01 --recent
- Visit https://jupyter-predix-app01.run.aws-usw02-pr.ice.predix.io/ then login with 'password'.
- The commands below assume you have an existing Heroku app named
xreso please update accordingly.
cd predix-volcano-app/
chmod +x ./start_jupyter # please do $(dos2unix start_jupyter) if on Windows
heroku git:remote -a xre
heroku buildpacks:set https://github.com/pl31/heroku-buildpack-conda.git -a xre
heroku config:set JUPYTER_NOTEBOOK_PASSWORD=password -a xre
git push -f heroku master
- Visit https://xre.herokuapp.com/ then login with 'password'.
- Note the cells in the Notebook are not interactive, Python dependencies cause the slug to exceed the 500MB cap for free Heroku plans.