Dockerfile.RBuildEnv issues with dependencies about incidence-mapper HOT 8 CLOSED

But if I install the dependencies from the terminal in RStudio, all is good:

terminal:

sudo apt install -y libproj-dev libv8-dev libjq-dev libprotobuf-dev protobuf-compiler

console:

> install.packages('geojsonio')
...
...
installing vignettes
** testing if installed package can be loaded
* DONE (geojsonio)

from incidence-mapper.

The most up-to-date build environment is in the connect-to-real-data branch: https://github.com/seattleflu/incidence-mapper/blob/connect-to-real-data/Dockerfile.RBuildEnv

from incidence-mapper.

I believe I've fixed some of these package issues on my PR #35. Someone should check that it also works for them.

from incidence-mapper.

@devclinton I rebuilt the image on our machine from trs/simplify-db-connection. Seems mostly good (and better than before). One issue is this line doesn't work

from incidence-mapper.

@famulare Docker RUN commands happen inside the container, so you need to first COPY the install_local_packages.R file into the container somewhere and then run it using its path inside the container.

from incidence-mapper.

(I didn't notice that failing for me, but it likely did!)

from incidence-mapper.

@devclinton I'll return this to your list, as I'm a little confused about copying in something that is at the same level as the dockerfile. Thanks!

from incidence-mapper.

@tsibley For now I removed the running of install_local_packages.R on the building on the container.

This is more because of how the future workflow will actually execute. At the moment, the container is being used as a development environment but later the workflow will be more along the lines you provided where we need to build each package. For now, let's make that a script the user can execute from within the R environment after they start the environment.

Later, I envision a workflow like the following

1. Build the package building/training container. This will be used to create the packages in the form of tar.gz file for now. I am working to get a local CRAN server up or alternatively maybe later we could publish to the general CRAN servers.

2. From There we build 2 new containers

   1. Training container with incidenceMapR will be installed. Most likely dbViewer as well. That container will then be used to train all the various iterations of the the models. Having a separate training container will be slightly lighter but is optional. The main thing I want to be able to do is begin to highly parallelize the training step and having it as a separate step makes that slightly easier. so as long as I can have a container that at runtime has the incidenceMapR available, I should be good.

   2. The second container will then just install the modelServ package. This will act as the execution worker for the API server. Really, this layer should rarely change since it is just loading a model and then running it. I suspect the training containers will change the most. Also, we could re-use the build container later in development process as the training container as well.

from incidence-mapper.