A simple R package to generate a Dockerfile and corresponding Docker image from an analysis directory. tugboat uses the renv package to automatically detect all the packages necessary to replicate your analysis and will generate a Dockerfile that contains an exact copy of your entire directory with all the packages installed.
tugboat transforms an unstructured analysis folder into a renv.lock file and constructs a Docker image that includes all your essential R packages based on this lockfile.
tugboat may be of use, for example, when preparing a replication package for research. With tugboat, you can take a directory on your local computer and quickly generate a Dockerfile and Docker image that contains all the code and the necessary software to reproduce your findings.
Installation
Install tugboat from CRAN:
install.packages("tugboat")Or install the development version from GitHub:
# install.packages("pak")
pak::pkg_install("dmolitor/tugboat")Usage
tugboat only has two exported functions; one to create a Dockerfile from your analysis directory, and one to build the corresponding Docker image.
Create the Dockerfile
The primary function from tugboat is create(). This function converts your analysis directory into a Dockerfile that includes all your code and essential R packages.
This function scans all files in the current analysis directory, attempts to detect all R packages, and installs these packages in the resulting Docker image. It also copies the entire contents of the analysis directory into the Docker image. For example, if your analysis directory is named incredible_analysis, the corresponding location of your code and data files in the generated Docker image will be /incredible_analysis.
For the most common use-cases, there are a couple of arguments in this function that are particularly important:
-
project: This argument tells tugboat which directory is the one to generate the Dockerfile from. You can set this value yourself, or you can just use the default value. By default, tugboat uses thehere::herefunction to determine what directory is the analysis directory. To get a detailed understanding of exactly how this works take a look at the here package. In general, this “just works”! -
as: This argument tells tugboat where to save the Dockerfile. In general you don’t need to set this and tugboat will just save the Dockerfile in theprojectdirectory from above. -
exclude: A vector of files or sub-directories in your analysis directory that should NOT be included in the Docker image. This is particularly important when you have, for example, a sub-directory with large data files that would make the resulting Docker image extremely large if included. You can tell tugboat to exclude this sub-directory and then simply mount it to a Docker container as needed.
Below I’ll outline a couple examples.
library(tugboat)
# The simplest scenario where your analysis directory is your current
# active project, you are fine with the default base "r-base:latest"
# Docker image, and you want to include all files/directories:
create()
# Suppose your analysis directory is actually a sub-directory of your
# main project directory:
create(project = here::here("sub-directory"))
# Suppose that you specifically need a Docker base image that has RStudio
# installed so that you can interact with your analysis within a Docker
# container. To do this, we will pass additional arguments directly to the
# `dockerfiler::dock_from_renv function.
create(FROM = "rocker/rstudio")
# Finally, suppose that we want to include all files except a couple
# particularly data-heavy sub-directories:
create(exclude = c("data/big_directory_1", "data/big_directory_2"))Build the Docker image
Once the Dockerfile has been created, we can build the Docker image with the build() function. By default this will infer the Dockerfile directory using here::here. This function assumes a little knowledge about Docker; if you aren’t sure where to start, this is a great starting point.
The following example will do the simplest thing and will build the image locally.
build(image_name = "awesome_analysis")Suppose that, like above, your analysis directory is a sub-directory of your main project directory:
Push to DockerHub
If, instead of just building the Docker image locally, you want to build the image and then push to DockerHub, you can make a couple small additions to the code above:
build(
image_name = "awesome_analysis",
push = TRUE,
dh_username = Sys.getenv("DH_USERNAME"),
dh_password = Sys.getenv("DH_PASSWORD")
)Note: If you choose to push, you also need to provide your DockerHub username and password. Typically you don’t want to pass these in directly and should instead use environment variables (or a similar method) instead.
Why tugboat? 🚢
There are a few available packages with similar goals, so why tugboat? tugboat is minimal and builds directly on top of renv and pak. Each of these packages is actively maintained and provides specific utilities that the tugboat utilizes for maximum convenience. tugboat aims to leverage packages that are likely to remain actively maintained and handle dependency management as seamlessly as possible.
containerit is a robust package that is directly comparable to tugboat. However, it implements its own method for discovering R package dependencies instead of using renv. It also relies on sysreqsdb for system dependency discovery, which has been archived in favor of r-system-requirements, which pak is built on. It also isn’t super actively maintained and isn’t on CRAN.
holepunch is related but focuses on making GitHub repositories Binder-compatible. It currently relies on MRAN, which is now obsolete, and does not use pak for system dependency management. It is also not actively maintained and is not on CRAN.
automagic focuses on automatically detecting and installing R package dependencies but uses its own method rather than relying on renv. automagic also has no utilities for creating/building Docker images.