Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/begriffs/heroku-buildpack-ghc
Deploy Haskell apps to Heroku
https://github.com/begriffs/heroku-buildpack-ghc
Last synced: about 2 months ago
JSON representation
Deploy Haskell apps to Heroku
- Host: GitHub
- URL: https://github.com/begriffs/heroku-buildpack-ghc
- Owner: begriffs
- Created: 2013-09-28T21:01:45.000Z (almost 11 years ago)
- Default Branch: master
- Last Pushed: 2019-05-27T04:43:55.000Z (over 5 years ago)
- Last Synced: 2024-06-22T09:42:20.562Z (3 months ago)
- Language: Shell
- Homepage:
- Size: 147 KB
- Stars: 272
- Watchers: 11
- Forks: 59
- Open Issues: 9
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- awesome-cabal - Heroku buildpack GHC
README
Deploy Haskell apps to Heroku
This buildpack supports frameworks like Yesod, Snap, and Happstack with
the latest stable GHC binaries. Putting Haskell web applications online
should be easy, and now it is. Try it for yourself.
**Note**, this buildpack does a basic build from scratch. For a
faster build, try
[mietek/haskell-on-heroku](https://github.com/mietek/haskell-on-heroku)
which detects your project dependencies and fetches a custom pre-built
sandbox.
### Example: deploying a Snap app
Here's how to go from zero to "hello world" on Heroku. You'll need to
install the [Haskell Platform](http://www.haskell.org/platform/) and
the [Heroku Toolbelt](https://toolbelt.herokuapp.com/) on your local
machine, then do this:
```sh
# Generate a barebones snap app called snapdemo
mkdir snapdemo && cd $_
cabal sandbox init
cabal install snap
cabal exec snap init barebones
# Tell Heroku how to start the server
echo 'web: cabal run -- -p $PORT' > Procfile
# Create a git repo and deploy!
git init .
echo "dist\n.cabal-sandbox\ncabal.sandbox.config" > .gitignore
git add *
git commit -m 'Initial commit'
heroku create --stack=cedar-14 --buildpack https://github.com/begriffs/heroku-buildpack-ghc.git
git push heroku master
```
**The first deploy is slowest** as the environment downloads and
bootstraps. Subsequent deploys use cached binaries and cached cabal
packages to go faster.
### Beating the Fifteen-Minute Build Limit
The first time you try to deploy a big framework like Yesod the
compilation can take so long that Heroku cuts it off. If this happens
fear not, you can build your app with an Anvil server.
```sh
# Enable Anvil builds
heroku plugins:install https://github.com/ddollar/heroku-anvil
# Move big build artifacts out of the way or else the upload
# to Anvil will be very slow
mkdir -p /tmp/deploy-stash ; mv .cabal-sandbox /tmp/deploy-stash ; mv dist /tmp/deploy-stash
# Build your slug and cache without any time limits
heroku build -r -b https://github.com/begriffs/heroku-buildpack-ghc.git
# Use Anvil-generated cache next time we do a regular git push to Heroku
heroku config:set EXTERNAL_CACHE=$(cat .anvil/cache)
# Bring your sandbox etc back
mv /tmp/deploy-stash/.cabal-sandbox . ; mv /tmp/deploy-stash/dist .
```
After the first deploy using Anvil you can go back to the regular
deploy process. This is because the cabal sandbox etc are cached
by Anvil and will be retrieved, making future builds incremental
and fast.
Remember to `heroku config:unset EXTERNAL_CACHE` after your first
successful regular (post-Anvil) git push.
### Locking Package Versions
Cabal sometimes gets confused on Heroku and tries installing outdated
packages. If you have your app working locally you can constrain the
remote package versions to match your local environment. Just do this:
```sh
cabal freeze
git add cabal.config
# commit and push to fix remote build
```
### Configuring the Build
You can change build settings through Heroku environment variables.
```sh
# set the variable of your choice
heroku config:set VARIABLE=value
```
Here are the options
namedescriptiondefault
CLEAR_CACHE
Force everything to reinstall from scratch by setting to 1.
0
EXTERNAL_CACHE
Url of replacement buildpack cache tarball. Useful for Anvil.
GHC_VER
GHC version to download or build
7.10.3
CABAL_VER
Version of cabal-install
1.22.8.0
PREBUILT
Base url for the prebuilt binary cache
https://s3.amazonaws.com/heroku-ghc
PRE_SCRIPT
Path to an executable, relative to checkout root. If present, this will be chmodded (+x) and ran before anything else.
POST_SCRIPT
Path to an executable, relative to checkout root. If present, this will be chmodded (+x) and ran after anything else.
### Interacting with a running app
```sh
heroku run bash # shell access
heroku run cabal repl # Haskell repl with all app modules loaded
```
### Benefits of this buildpack
* **Latest binaries: GHC 7.10.3, cabal-install 1.22.8.0**
* Uses cabal >=1.20 features to run the app and repl
* Exposes Haskell platform binaries to your app and scripts
* Uses prebuilt binaries for speed but...
* ...can fall back to building the standard GHC distribution
### Contributing
There are a number of ways to improve this buildpack. Please see the
Github issues for ideas.
In order to contribute to the build script it will help to understand
how Heroku's deployment process works, and how that affects GHC. Heroku
provides three areas for storing files during build: a cache directory,
a working directory, and a build directory.
The cache, called `$CACHE_DIR` in the script, persists between
deployments. We use it to avoid building binaries more than once. The
working directory, called `$WORKING_HOME`, does not persist between
builds, or even after the build script is done. Seems we should avoid
this area, right? Well GHC has some idiosyncracies that make this area
quite useful as you will see. Finally the build destination directory,
`$BUILD_DIR`, holds the git repo the user pushes and gets copied into
what will be `/app` in the deployed application. During build it lives
in a weird nonce filename.
We want to use GHC binaries at build time to compile the app, and we
would also like those binaries to be available in the app environment
after deployment (so people can use `runhaskell` or `cabal repl`). Seems
like we should install directly to `$BUILD_DIR`. There's one problem:
GHC breaks if you move it to a new path after installation because many
of its binaries are just scripts with hard-coded full paths in them to
other GHC files. And as you remember, Heroku is going to move things in
`$BUILD_DIR` to `/app`. So the trick will be to install to `/app` in the
working directory, use GHC there, then copy that installation to the
build directory which will be renamed in the deployed application and
not notice it has been moved.
GHC is also sensitive to having `libgmp` named just right. We don't
have privileges to adjust `/usr/lib` in the deployed app so we create a
symbolic link in a place we are permitted and set linker variables in
the shell so that everything can build.
### External dependencies
Some packages have external dependencies (i.e. non-Haskell dependencies which cannot be satisfied by `cabal`). If you come across such a package, check in `contribs` to see if someone has already created patches for the dependencies. If they have, you should be able to
```sh
patch -p0 < contribs/DEP.patch
```
to patch the buildpack and proceed. If not, you'll need to modify `bin/compile` yourself. Contributing these changes back as patches is appreciated.
### Building new binaries for Heroku
As new versions of GHC and Cabal are released we should build them for
Heroku and put them on S3 to speed up future deploys for everyone. Luckily
the buildpack can do the building too.
Adjust the `GHC_VER` and `CABAL_VER` environment vars and then
deploy. It will build the new binaries from the standard GHC
distribution. Then copy the results to S3 like this:
```sh
heroku run bash
# now SSH'd into the server
cd /app/vendor
url "https://s3.amazonaws.com/aws-cli/awscli-bundle.zip" -o "awscli-bundle.zip"
unzip awscli-bundle.zip
awscli-bundle/install
~/.local/lib/aws/bin/aws configure
# ^^^ answer the configuration questions
tar zcf heroku-ghc-[VERSION].tar.gz ghc-[VERSION]/
tar zcf heroku-cabal-install-[VERSION].tar.gz cabal-install-[VERSION]/
~/.local/lib/aws/bin/aws s3 cp heroku-ghc-[VERSION].tar.gz s3://[BUCKET]
~/.local/lib/aws/bin/aws s3 cp heroku-cabal-install-[VERSION].tar.gz s3://[BUCKET]
```
### Thanks
Thanks to Brian McKenna and others for their work on
[heroku-buildpack-haskell](https://github.com/puffnfresh/heroku-buildpack-haskell)
which inspired and informed this buildpack. For a history of that project's
contributions and ideas see [this article]
(http://blog.begriffs.com/2013/08/haskell-on-heroku-omg-lets-get-this.html).