Creating Your Docker Container and Command Line Interface (with docopt)

Tengfei Yin <tengfei.yin@sevenbridges.com>

2017-09-28

1 Introduction

In progress

In this tutorial, we will go through ways to

There are many fun ways to do it, here I am more focused for R developers.

1.1 Existing Docker repos

Before you create any, make sure you don’t re-invent the wheel and use the best base image for your container as your tool chain may save your lots of time later on.

1.1.1 Rocker Project

Official R Docker images is called “Rocker”" project and is on GitHub, please visit the page to find more details and Dockerfile.

Image Description
rocker/r-base base package to build from
rocker/r-devel base + R-devel from SVN
rocker/rstudio base + RStudio Server
rocker/hadleyverse rstudio + Hadley’s packages, LaTeX
rocker/ropensci hadleyverse + rOpenSci packages
rocker/r-devel-san base, SVN’s R-devel and SAN

1.1.2 Bioconductor Images

Bioconductor have a nice page about the official Docker images, please read for more details.

Image (release branch) Image (development branch)
bioconductor/release_base bioconductor/devel_base
bioconductor/release_core bioconductor/devel_core
bioconductor/release_flow bioconductor/devel_flow
bioconductor/release_microarray bioconductor/devel_microarray
bioconductor/release_proteomics bioconductor/devel_proteomics
bioconductor/release_sequencing bioconductor/devel_sequencing
bioconductor/release_metabolomics bioconductor/devel_metabolomics

To understand the image quickly here is the short instruction for the image name:

  • release images are based on rocker/rstudio
  • devel images are based on rocker/rstudio-daily

  • base: Contains R, RStudio, and BiocInstaller + system dependencies.
  • core: base + a selection of core.
  • flow: core + all packages tagged with the FlowCytometry biocView.
  • microarray: core + all packages tagged with the Microarray biocView.
  • proteomics: core + all packages tagged with the Proteomics biocView.
  • sequencing: core + all packages tagged with the Sequencing biocView.
  • metabolomics: core + all packages tagged with the Metabolomics biocView.

1.1.3 Docker Hub

Docker Hub also provide public/private repos, you can search existing tools without building yourself, it’s very likely some popular tool already have Docker container well maintained there.

1.1.4 Seven Bridges Docker Registry

Tutorial coming soon.

Example Seven Bridges registry:

  • SevenBridges : images.sbgenomics.com/<repository>[:<tag>]
  • Cancer Genomics Cloud: cgc-images.sbgenomics.com/<repository>[:<tag>]

2 Tutorial: Random Number Generator

Our goal here is to making a CWL app to generate uniform random numbers, yes, the core function is runif(), it’s a native function in R.

dput(runif)
## function (n, min = 0, max = 1) 
## .Call(C_runif, n, min, max)
set.seed(1001)
runif(10)
##  [1] 0.985688783 0.412628483 0.429539246 0.419172236 0.426506559
##  [6] 0.887797565 0.006096034 0.081215761 0.288657362 0.765342145

2.1 Using docopt Package

In R, we also have a nice implementation in a package called docopt, developed by Edwin de Jonge. Check out its tutorial on GitHub.

So let’s quickly create a command line interface for our R scripts with a dummy example. Let’s turn the uniform distribution function runif into a command line tool.

when you check out the help page for runif, here is the key information you want to markdown.

Usage

runif(n, min = 0, max = 1)

Arguments

n
number of observations. If length(n) > 1, the length is taken to be the number required.

min, max
lower and upper limits of the distribution. Must be finite.

I will add one more parameter to set seed, here is the R script file called runif.R.

At the beginning of the command line script, I use the docopt standard to write my tool help.

'usage: runif.R [--n=<int> --min=<float> --max=<float> --seed=<float>]

options:
 --n=<int>        number of observations. If length(n) > 1, the length is taken to be the number required [default: 1].
 --min=<float>   lower limits of the distribution. Must be finite [default: 0].
 --max=<float>   upper limits of the distribution. Must be finite [default: 1].
 --seed=<float>  seed for set.seed() function [default: 1]' -> doc

library("docopt")

Let’s first do some testing in your R session before you make it a full functional command line tool.

docopt(doc) #with no argumetns provided
## List of 8
##  $ --n   : chr "1"
##  $ --min : chr "0"
##  $ --max : chr "1"
##  $ --seed: chr "1"
##  $ n     : chr "1"
##  $ min   : chr "0"
##  $ max   : chr "1"
##  $ seed  : chr "1"
## NULL
docopt(doc, "--n 10 --min=3 --max=5")
## List of 8
##  $ --n   : chr "10"
##  $ --min : chr "3"
##  $ --max : chr "5"
##  $ --seed: chr "1"
##  $ n     : chr "10"
##  $ min   : chr "3"
##  $ max   : chr "5"
##  $ seed  : chr "1"
## NULL

Looks like it works, now let’s add main function script for this command line tool.

opts <- docopt(doc)
set.seed(opts$seed)
runif(n = as.integer(opts$n),
      min = as.numeric(opts$min),
      max = as.numeric(opts$max))
## [1] 0.2655087

Add Shebang at the top of the file, this is a complete example for runif.R command line will be like this

#!/usr/bin/Rscript
'usage: runif.R [--n=<int> --min=<float> --max=<float> --seed=<float>]

options:
 --n=<int>        number of observations. If length(n) > 1, the length is taken to be the number required [default: 1].
 --min=<float>   lower limits of the distribution. Must be finite [default: 0].
 --max=<float>   upper limits of the distribution. Must be finite [default: 1].
 --seed=<float>  seed for set.seed() function [default: 1]' -> doc

library("docopt")
opts <- docopt(doc)
set.seed(opts$seed)
runif(n = as.integer(opts$n),
      min = as.numeric(opts$min),
      max = as.numeric(opts$max))

Let’s test this command line.

$ runif.R --help
Loading required package: methods
usage: runif.R [--n=<int> --min=<float> --max=<float> --seed=<float>]

options:
 --n=<int>        number of observations. If length(n) > 1, the length is taken to be the number required [default: 1].
 --min=<float>   lower limits of the distribution. Must be finite [default: 0].
 --max=<float>   upper limits of the distribution. Must be finite [default: 1].
 --seed=<float>  seed for set.seed() function [default: 1]
$ runif.R
Loading required package: methods
[1] 0.2655087
$ runif.R
Loading required package: methods
[1] 0.2655087
$ runif.R --seed=123 --n 10 --min=1 --max=100
Loading required package: methods
 [1] 29.470174 79.042208 41.488715 88.418723 94.106261  5.510093 53.282443
 [8] 89.349485 55.592066 46.204859

For full example you can check my GitHub example.

2.2 Quick Command Line Interface with commandArgs (Position and Named Args)

For advanced users, please read another tutorial “Creating Your Docker Container and Command Line Interface (with docopt)”, “docopt” is more formal way to construct your command line interface, but there is a quick way to make command line interface here using just commandArgs.

Suppose I already have a R script like this using position mapping the arguments

  1. numbers
  2. min
  3. max
fl <- system.file("docker/sevenbridges/src", "runif2spin.R", package = "sevenbridges")
cat(readLines(fl), sep = '\n')
#' ---
#' title: "Uniform random number generator example"
#' output:
#'     html_document:
#'     toc: true
#' number_sections: true
#' ---

#' # summary report
#'
#' This is a random number generator

#+
args = commandArgs(TRUE)

r = runif(n   = as.integer(args[1]),
          min = as.numeric(args[2]),
          max = as.numeric(args[3]))
head(r)
summary(r)
hist(r)

Ignore the comment part, I will introduce spin/stich later. My base command will be something like

Rscript runif2spin.R 10 30 50

I just describe my tool in this way

library("sevenbridges")
library("readr")
fd <- fileDef(name = "runif.R",
              content = read_file(fl))

rbx <- Tool(id = "runif",
            label = "runif",
            hints = requirements(docker(pull = "rocker/r-base"),
                                 cpu(1), mem(2000)),
            requirements = requirements(fd),
            baseCommand = "Rscript runif.R",
            stdout = "output.txt",
            inputs = list(input(id = "number",
                                type = "integer",
                                position = 1),
                          input(id = "min",
                                type = "float",
                                position = 2),
                          input(id = "max",
                                type = "float",
                                position = 3)),
            outputs = output(id = "random", glob = "output.txt"))

Now copy-paste the JSON into your project app and run it in the cloud to test it

How about named arguments? I will still recommend use “docopt” package, but for simple way.

fl <- system.file("docker/sevenbridges/src", "runif_args.R", package = "sevenbridges")
cat(readLines(fl), sep = '\n')
#' ---
#' title: "Uniform random number generator example"
#' output:
#'     html_document:
#'     toc: true
#' number_sections: true
#' ---

#' # summary report
#'
#' This is a random number generator

#+
args <- commandArgs(TRUE)

## quick hack to split named arguments
splitArgs <- function(x) {
    res <- do.call(rbind, lapply(x, function(i){
        res <- strsplit(i, "=")[[1]]
        nm <- gsub("-+", "",res[1])
        c(nm, res[2])
    }))
    .r <- res[,2]
    names(.r) <- res[,1]
    .r
}
args <- splitArgs(args)

#+
r <- runif(n   = as.integer(args["n"]),
           min = as.numeric(args["min"]),
           max = as.numeric(args["max"]))
summary(r)
hist(r)
write.csv(r, file = "out.csv")
Rscript runif_args.R --n=10 --min=30 --max=50

I just describe my tool in this way, note, I use separate=FALSE and add = to my prefix as a hack.

library("readr")
fd <- fileDef(name = "runif.R",
              content = read_file(fl))

rbx <- Tool(id = "runif",
            label = "runif",
            hints = requirements(docker(pull = "rocker/r-base"),
                                 cpu(1), mem(2000)),
            requirements = requirements(fd),
            baseCommand = "Rscript runif.R",
            stdout = "output.txt",
            inputs = list(input(id = "number",
                                type = "integer",
                                separate = FALSE,
                                prefix = "--n="),
                          input(id = "min",
                                type = "float",
                                separate = FALSE,
                                prefix = "--min="),
                          input(id = "max",
                                type = "float",
                                 separate = FALSE,
                                prefix = "--max=")),
            outputs = output(id = "random", glob = "output.txt"))

2.3 Quick Report: Spin and Stich

Alternative, you can use spin/stich from knitr to generate report directly from an R script with special format. For example, let’s use the above example

fl <- system.file("docker/sevenbridges/src", "runif_args.R", package = "sevenbridges")
cat(readLines(fl), sep = '\n')
#' ---
#' title: "Uniform random number generator example"
#' output:
#'     html_document:
#'     toc: true
#' number_sections: true
#' ---

#' # summary report
#'
#' This is a random number generator

#+
args <- commandArgs(TRUE)

## quick hack to split named arguments
splitArgs <- function(x) {
    res <- do.call(rbind, lapply(x, function(i){
        res <- strsplit(i, "=")[[1]]
        nm <- gsub("-+", "",res[1])
        c(nm, res[2])
    }))
    .r <- res[,2]
    names(.r) <- res[,1]
    .r
}
args <- splitArgs(args)

#+
r <- runif(n   = as.integer(args["n"]),
           min = as.numeric(args["min"]),
           max = as.numeric(args["max"]))
summary(r)
hist(r)
write.csv(r, file = "out.csv")

You command is something like this

Rscript -e "rmarkdown::render(knitr::spin('runif_args.R', FALSE))" --args --n=100 --min=30 --max=50

And so I describe my tool like this with Docker image rocker/hadleyverse this contains knitr and rmarkdown package.

library("readr")
fd <- fileDef(name = "runif.R",
              content = read_file(fl))

rbx <- Tool(id = "runif",
            label = "runif",
            hints = requirements(docker(pull = "rocker/hadleyverse"),
                                 cpu(1), mem(2000)),
            requirements = requirements(fd),
            baseCommand = "Rscript -e \"rmarkdown::render(knitr::spin('runif.R', FALSE))\" --args",
            stdout = "output.txt",
            inputs = list(input(id = "number",
                                type = "integer",
                                 separate = FALSE,
                                prefix = "--n="),
                          input(id = "min",
                                type = "float",
                                 separate = FALSE,
                                prefix = "--min="),
                          input(id = "max",
                                type = "float",
                                separate = FALSE,
                                prefix = "--max=")),
            outputs = list(output(id = "stdout", type = "file", glob = "output.txt"),
                           output(id = "random", type = "file", glob = "*.csv"),
                           output(id = "report", type = "file", glob = "*.html")))

You will get a report in the end.

2.4 Executable Report with R Markdown (Advanced)

We cannot really make a R Markdown file executable in it by simply put

#!/bin/bash/Rscript

In your markdown document.

Of course, we can figure out a way to do it in liftr or knitr. But R Markdown allow you to pass parameters to your R Markdown template, please read this tutorial Parameterized Reports. This doesn’t solve my problem that I want to directly describe command line interface in the markdown template. However, here is alternative method:

Create an command line interface to pass params from docopt into rmarkdown::render() function. In this way, we can pass as many as possible parameters from command line interface into our R Markdown template.

So here we go, here is updated methods and it’s also what I use for another tutorial about RNA-seq workflow.

fl <- system.file("docker/sevenbridges/src/", "runif.R", package = "sevenbridges")

Here is the current content of command line interface

#!/usr/local/bin/Rscript
'usage: runif.R [--n=<int> --min=<float> --max=<float> --seed=<float>]

options:
--n=<int>        number of observations. If length(n) > 1, the length is taken to be the number required [default: 1].
--min=<float>   lower limits of the distribution. Must be finite [default: 0].
--max=<float>   upper limits of the distribution. Must be finite [default: 1].
--seed=<float>  seed for set.seed() function [default: 1]'  -> doc

library("docopt")
opts <- docopt(doc)

## create param list
lst <- list(n = as.integer(opts$n),
            min = as.numeric(opts$min),
            max = as.numeric(opts$max),
            seed = as.numeric(opts$seed))

## execute your Rmarkdown with these parameters
rmarkdown::render("/report/report.Rmd", rmarkdown::html_document(toc = TRUE),
                  output_dir = ".", params = lst)

And here is the report template

---
title: "Uniform random number generator example"
output:
  rmarkdown::html_document:
    toc: true
    number_sections: true
    css: style.css
params:
  seed: 1
  n: 1
  min: 0
  max: 1
---

## Summary
```{r}
set.seed(params$seed)
r = runif(n   = as.integer(params$n),
          min = as.numeric(params$min),
          max = as.numeric(params$max))
summary(r)
hist(r)
```

3 Setup Docker Hub Automated Build

To make things more reproducible and explicit and automatic, you can do a autohook to automatically build your container/image on Docker Hub. Here is what I do

  1. I created some project called ‘docker’ on my GitHub and it has all container that crated from a Dockerfile, for example, tengfei/docker/runif, please go here to check it out
  2. This folder root has a Dockerfile and subfolders for extra materials I added at build time, like script or report template.
  3. Log into your Docker Hub account, following this tutorial to make “automated build” from your GitHub account. Make sure you input the right location for your Dockerfile, by customizing it.
  4. Then you will have auto-build every time you push a new update in GitHub.
  5. Start using your Docker image like tengfei/runif.
  6. Feel free to push it onto your Seven Bridges platform registry as well.

4 More Examples

There are more examples under inst/docker folder, you can check out how to describe command line and build Docker, how to make report template. You may read the online GitHub code. Or you could read another tutorial about how we wrap RNA-seq workflow from Bioconductor.