Deploying to self-hosted Shiny Server

[GregSutcliffe]

Hi there - new to Golem but enjoying the workflow :)

I'm looking for guidance on deploying a Golem-based Shiny app to a self-hosted Shiny server. As far as I can see in 03_deploy.R the functions are all aimed at RStudio hosting or Docker, so how would I go about deploying this to my Linux VM?

For contrast, in my non-Golem apps, I would just rsync the project over to /srv/shiny-server/apps/ and make sure the dependencies are available. What's the right approach for Golem? Do I need to write a minimal app.R which is similar to the one generated in golem::add_shinyserver_file and get Shiny to launch that (given that I will have installed my Golem package too)?

Thanks!

[ColinFay]

Hey @GregSutcliffe,

Thanks for giving {golem} a spin!

Deploying to Shiny Server should be pretty straightfoward if you did this with a standard application before. Right now there are two ways to do it: copy the full golem folder, or use the built package.

With the first solution, the idea is to have the whole {golem} folder, with the app.R generated by golem::add_shinyserver_file(), at the right spot on the Shiny Server.

The app.R generated by this function behaves as a drop-in replacement for your previous app.R.

So to sum up :

• Create your golem but don't build the tar.gz
• generate the app.R with golem::add_shinyserver_file()
• copy the root folder of your app to the right folder on the server scp ~/mygolem /distant/server/folder (or with rsync or anything)
• (don't forget to touch restart.txt)

And this should do the trick.

Second option:

• create the golem && build the tar.gz
• upload the tar.gz and install it on the server
• create an app.R that simply has thenameofyourapp::run_app()

Let me know if this works

[GregSutcliffe]

Thanks for the fast reply!

So if I understand (1) correctly, that means app.R is in a subdirectory, from Shiny Server's perspective, right? Will it find it there (in app/) or do I need to move it to the project root?

I'm just getting my head around Shiny modules - as soon as I have a basic app running, I'll try the deploy and report back!

EDIT: Update - my mistake, app.R is in the root dir already. I misread where it had ended up. Everything makes a lot more sense now :)

[GregSutcliffe]

It's probably worth mentioning that my non-Golem apps use renv to keep dependencies under control - if I use renv within Golem, does that affect either option?

[GregSutcliffe]

So I have a vary basic app (one static plot in a module, one table in app_server/ui.R), and it works in RStudio, via run_dev.R and via app.R.

When I use method (1) to deploy to my server (rsync the project, renv::restore the deps), and then go to the URL in my browser, I'm getting errors (the UI starts but then disconnects after drawing the h2 header, which suggests a code error) - but Shiny doesn't appear to be logging the errors. Even if I set golem.app.prod = FALSE in app.R, I don't get a logfile in /var/log/shiny-server/apps.

I know this is likely something weird on my end, but do I need to do anything special for logging from a Golem app?

[GregSutcliffe]

Just had a try with method (2) - installed the built package tgz on my server and I'm seeing the same behaviour. The package works locally, but on the server it disconnects, and I cannot get it to log why. Any pointers welcome :)

[GregSutcliffe]

OK, I have the app working now - there was, of course, a typo in some of the startup code. There are logs coming from the reactive parts of the app (i.e things like "stat_bin() using bins = 30. Pick better value with binwidth." from geom_histogram() etc.

Having fixed that, I've re-tested both methods, and both work fine. I'll probably go with option (2) as I can push the package to a Git repo and use renv and remotes::install_gitlab to get the package.

Thanks!

[cparsania]

I am facing problem to what mentioned above. I am using workflow 2) as generated by the function golem::add_dockerfile() given in the 03_deploy.R. I am using rocker/r-apt:bionic instead of rocker/tidyverse and also added necessary libraries to the Dockerfile. THe final line on the Dockerfile is CMD R -e "options('shiny.port'=8080,shiny.host='127.0.0.1');myApp::run_app()" The image build successfully. When I start container using docker run myApp:tagName it generates logs as expected but stops at http://127.0.0.1:8080. When I hit url in the browser it says can't connect to the server. Am I missing something here ?? I haven't tried workflow 1) as I find workflow 2) more handy.

[VincentGuyader]

Hi

Can you please show us :

Your dockerfike
Le complete docker run instruction you use

And the docker logs <conaiterid> you have after launching your docker.

[cparsania]

Hi, it was very silly mistake. After taking dig into container-host port expose I figured it out. I did not mentioned option -p while running the container. So final docker file i used is

FROM rocker/r-apt:bionic

RUN apt-get update && \
  apt-get install libcurl4-openssl-dev libv8-3.14-dev  libssl-dev libxml2-dev libxslt-dev libudunits2-dev libproj-dev libgdal-dev libfontconfig1-dev libcairo2-dev -y

COPY ./requirement_bin.txt .
RUN cat requirement_bin.txt | xargs apt-get install -y -qq

COPY ./myApp.tar.gz /app.tar.gz
RUN R -e 'devtools::install_local("/app.tar.gz" , dependencies=TRUE,  build = FALSE ,repos=BiocManager::repositories() )'
EXPOSE 80
CMD R -e "options('shiny.port'=80,shiny.host='0.0.0.0');myApp::run_app()"

and command to launch the container is docker run -p 3232:80 myApp:tag

This solved the problem.

Thanks a lot.

========= EDIT=======

It is worth mentioning that, the above setup does not attach packages which are under the Depends tag of package DESCRIPTION file. In my case I explicitly load shinyBS package using library call. I changed final line of Dockerfile to

CMD R -e "options('shiny.port'=80,shiny.host='0.0.0.0');library(shinyBS); myApp::run_app()"

[ColinFay]

Thanks for your inputs :)

This seems solved, closing now, feel free to reopen if needed!