reprex-connect-20200414's Introduction

Reprex Connect 20200414

Setup

Install the connectapi package

remotes::install_github("rstudio/connectapi")

Note that the user/password for Connect are admin/admin0
Create the reprex environment

docker-compose up -d

# add python packages to the site library
docker exec reprex-connect-20200414_rstudio-connect_1 /opt/python/3.6.5/bin/pip install flask-restx sklearn

# deploy the example
R -f deploy.R

# tear down the server
docker-compose down

# build it again
docker-compose up -d

Problems

Notice that the content is now broken, because the underlying site library has changed.

This is undesirable for a number of reasons:

Reproducibility of content is reduced
As an admin, I may be lax about managing the site library
As an admin, I did not opt into this behavior, and Connect usually isolates content / manages deployments for me / etc.
As a publisher, this error is very confusing and I have no way to remedy

2020/04/14 13:03:15.973098700 Error while loading Python API: Could not import
module 'predict'. Connect expected to find a file in the application directory
named 'predict.py'. Please check the entrypoint specified in the manifest.json
file or on the rsconnect-python command line. Reason: No module named
'sklearn'.

Even figuring out "what" the issue is requires shell access on the server
As a publisher, I have no real recourse to remedy the issue other than try to intuit how I need to change / redeploy my content
The issue could pop up at any time that the python system library changes, between nodes, etc.
Connect's contract of "content bundle reproducible over time" is much reduced without a way to purge these cached environments

Recommend Projects