File server demo in a single Haskell file

Sneak peek: Run docker run --rm -p 8080:8080 snoyberg/file-server-demo and open http://localhost:8080.

We've all been there. We need to write some non-trivial piece of functionality, and end up doing it in bash or perl because that's what we have on the server we'll be deploying to. Or because it's the language we can most easily rely on being present at a consistent version on our coworkers' machines. We'd rather use a different language and leverage more advanced, non-standard libraries, but we can't do that reliably.

One option is to create static executables or to ship around Docker images. This is great for many use cases, and we are going to have a follow-up blog post about using Docker and Alpine Linux to make such static executables. But there are at least two downsides to this approach:

Said another way: there are good reasons why people like to use scripting languages. This blog post is going to demonstrate doing some non-trivial work with Haskell, and do so with a fully reproducible and trivially installed toolchain, supported on multiple operating systems.

Why Haskell?

Haskell is a functional programming language with high performance, great safety features, and a large ecosystem of open source libraries to choose from. Haskell programs are high level enough to be readable and modifiable by non-experts, making it ideal for these kinds of shared scripts. If you're new to Haskell, learn more on haskell-lang.org.

The task

We're going to put together a simple file server with upload capability. We're going to assume a non-hostile environment (like a corporate LAN with no external network access), and therefore not put in security precautions like upload size limits. We're going to use the relatively low-level Web Application Interface instead of a web framework. While it makes the code a bit longer, there's no magic involved. Common frameworks in Haskell include Yesod and Servant. We're going to host this all with the blazingly fast Warp web server.

Get Stack

Stack is a cross-platform program for developing Haskell projects. While it has many features, in our case the most important bit is that it can:

Check out the Get Started page on haskell-lang.org to get Stack on your system.

The code

You can see the full source code on Github. Let's step through the important parts here.

Script interpreter

We start off our file with something that is distinctly not Haskell code:

#!/usr/bin/env stack
{- stack
    --resolver lts-6.11
    --install-ghc
    runghc
    --package shakespeare
    --package wai-app-static
    --package wai-extra
    --package warp
 -}

With this header, we've made our file executable from the shell. If you chmod +x the source file, you can run ./FileServer.hs. The first line is a standard shebang. After that, we have a comment that provides Stack with the relevant command line options. These options tell it to:

For more information on Stack's script interpreter support, see the Stack user guide.

Command line argument parsing

Very often with these kinds of tools, we need to handle command line arguments. Haskell has some great libraries for doing this in an elegant way. For example, see the optparse-applicative library tutorial. However, if you want to go simple, you can also just use the getArgs function to get a list of arguments. We're going to add support for a sanity argument, which will allow us to sanity-check that running our application works:

main :: IO ()
main = do
    args <- getArgs
    case args of
        ["sanity"] -> putStrLn "Sanity check passed, ready to roll!"
        [] -> do
            putStrLn "Launching application"
            -- Run our application (defined below) on port 8080
            run 8080 app
        _ -> error $ "Unknown arguments: " ++ show args

Routing

We're going to support three different routes in our application:

Thanks to pattern matching in Haskell, getting this to work is very straightforward:

app :: Application
app req send =
    -- Route the request based on the path requested
    case pathInfo req of
        -- "/": send the HTML homepage contents
        [] -> send $ responseBuilder
                status200
                [("Content-Type", "text/html; charset=utf-8")]
                (runIdentity $ execHtmlT homepage)

        -- "/browse/...": use the file server to allow directory
        -- listings and downloading files
        ("browse":rest) ->
            -- We create a modified request that strips off the
            -- "browse" component of the path, so that the file server
            -- does not need to look inside a /browse/ directory
            let req' = req { pathInfo = rest }
             in fileServer req' send

        -- "/upload": handle a file upload
        ["upload"] -> upload req send

        -- anything else: 404
        _ -> send $ responseLBS
            status404
            [("Content-Type", "text/plain; charset=utf-8")]
            "Not found"

The most complicated bit above is the path modification for the /browse tree, which is something a web framework would handle for us automatically. Remember: we're doing this low level to avoid extra concepts, real world code is typically even easier than this!

Homepage content

An area that Haskell really excels at is Domain Specific Languages (DSLs). We're going to use the Hamlet for HTML templating. There are many other options in the Haskell world favoring other syntax, such as Lucid library (which provides a Haskell-based DSL), plus implementations of language-agnostic templates, like mustache.

Here's what our HTML page looks like in Hamlet:

homepage :: Html ()
homepage = [shamlet|
$doctype 5
<html>
    <head>
        <title>File server
    <body>
        <h1>File server
        <p>
            <a href=/browse/>Browse available files

        <form method=POST action=/upload enctype=multipart/form-data>
            <p>Upload a new file
            <input type=file name=file>
            <input type=submit>
|]

Note that Hamlet - like Haskell itself - uses significant whitespace and indentation to denote nesting.

The rest

We're not going to cover the rest of the code in the Haskell file. If you're interested in the details, please read the comments there, and feel free to ask questions about any ambiguous bits (hopefully the inline comments give enough clarity on what's going on).

Running

Download the FileServer.hs file contents (or copy-paste, or clone the repo), make sure the file is executable (chmod +x FileServer.hs), and then run:

$ ./FileServer.hs

If you're on Windows, you can instead run:

> stack FileServer.hs

That's correct: the same source file will work on POSIX systems and Windows as well. The only requirement is Stack and GHC support. Again, to get Stack on your system, please see the Get Started page.

The first time you run this program, it will take a while to complete. This is because Stack will need to download and install GHC and necessary libraries to a user-local directory. Once complete, the results are kept on your system, so subsequent runs will be almost instantaneous.

Once running, you can view the app on localhost:8080.

Dockerizing

Generally, I wouldn't recommend Dockerizing a source file like this; it makes more sense to Dockerize a compiled executable. We'll cover how to do that another time (though sneak preview: Stack has built in support for generating Docker images). For now, let's actually Dockerize the source file itself, complete with Stack and the GHC toolchain.

You can check out the Dockerfile on Github. That file may be slightly different from what I cover here.

FROM ubuntu:16.04
MAINTAINER Michael Snoyman

Nothing too interesting...

ADD https://github.com/Yelp/dumb-init/releases/download/v1.1.3/dumb-init_1.1.3_amd64 /usr/local/bin/dumb-init
RUN chmod +x /usr/local/bin/dumb-init

While interesting, this isn't Haskell-specific. We're just using an init process to get proper handling for signals. For more information, see dumb-init's announcement blog post.

ADD https://get.haskellstack.org/get-stack.sh /usr/local/bin/
RUN sh /usr/local/bin/get-stack.sh

Stack has a shell script available to automatically install it on POSIX systems. We just download that script and then run it. This is all it takes to have a Haskell-ready system set up: we're now ready to run script interpreter based files like our FileServer.hs!

COPY FileServer.hs /usr/local/bin/file-server
RUN chmod +x /usr/local/bin/file-server

We're copying over the source file we wrote and then ensuring it is executable. Interestingly, we can rename it to not include a .hs file extension. There is plenty of debate in the world around whether scripts should or should not include an extension indicating their source language; Haskell is allowing that debate to perpetuate :).

RUN useradd -m www && mkdir -p /workdir && chown www /workdir
USER www

While not strictly necessary, we'd rather not run our executable as the root user, for security purposes. Let's create a new user, create a working directory to store files in, and run all subsequent commands as the new user.

RUN /usr/local/bin/file-server sanity

As I mentioned above, that initial run of the server takes a long time. We'd like to do the heavy lifting of downloading and installing during the Docker image build rather than at runtime. To make this happen, we run our program once with the sanity command line argument, so that it immediately exits after successfully starting up.

CMD ["/usr/local/bin/dumb-init", "/usr/local/bin/file-server"]
WORKDIR /workdir
EXPOSE 8080

Finally, we use CMD, WORKDIR, and EXPOSE to make it easier to run. This Docker image is available on Docker Hub, so if you'd like to try it out without doing a full build on your local machine:

docker run --rm -p 8080:8080 snoyberg/file-server-demo

You should be able to play with the application on http://localhost:8080.

What's next

As you can see, getting started with Haskell as a scripting language is easy. You may be interested in checking out the turtle library, which is a Shell scripting DSL written in Haskell.

If you're ready to get deeper into Haskell, I'd recommend:

FP Complete both supports the open source Haskell ecosystem, as well as provides commercial support for those seeking it. If you're interested in learning more about how FP Complete can help you and your team be more successful in your development and devops work, you can learn about what services we offer or contact us for a free consultation.

Subscribe to our blog via email
Email subscriptions come from our Atom feed and are handled by Blogtrottr. You will only receive notifications of blog posts, and can unsubscribe any time.

Do you like this blog post and need help with Next Generation Software Engineering, Platform Engineering or Blockchain & Smart Contracts? Contact us.