Deployment for Elixir and Erlang

edeliver is based on deliver and enables you to build and deploy Elixir and Erlang applications and perform hot-code upgrades.

The erlang releases are built on a remote host that is similar to the production machines - or in a docker container. After being built, the release can then be deployed to one or more production machines.

Once built, the release contains the full erts (erlang runtime system), all dependencies (erlang or elixir applications), the Elixir runtime, native port drivers, and your erlang/elixir application(s) in a standalone embedded node.

• Issues
• #deployment on Slack
• Community Wiki — feel free to contribute!

• Quick Start
• Installation
• Configuration
• Build Commands
• Deploy Commands
• Admin Commands
• Help
• Examples
• License

Assuming an Elixir project, you already have a build server and a staging server, and you've created a database on your staging server already (there is no ecto.create, we skip straight to migrations).

Add edeliver and your build tool (distillery) to your project dependencies in mix.exs:

def application, do: [
  applications: [
  	 ...
    # Add edeliver to the END of the list
    :edeliver
  ]
]

defp deps do
  [
    ...
    {:edeliver, ">= 1.9.2"},
    {:distillery, "~> 2.1", warn_missing: false},
  ]
end

If this is a Phoenix project, upload your prod.secret.exs to your build server somewhere. Let's say it's at /home/builder/prod.secret.exs.

In your project, create the file .deliver/config

# .deliver/config

APP="myapp"

BUILD_HOST="my-build-server.myapp.com"
BUILD_USER="builder"
BUILD_AT="/tmp/edeliver/myapp/builds"

STAGING_HOSTS="stage.myapp.com"
STAGING_USER="web"
DELIVER_TO="/home/web"

# For *Phoenix* projects, symlink prod.secret.exs to our tmp source
pre_erlang_get_and_update_deps() {
  local _prod_secret_path="/home/builder/prod.secret.exs"
  if [ "$TARGET_MIX_ENV" = "prod" ]; then
    __sync_remote "
      ln -sfn '$_prod_secret_path' '$BUILD_AT/config/prod.secret.exs'
    "
  fi
}

Add the release directory to your gitignore

echo ".deliver/releases/" >> .gitignore

Commit everything, compile the new dependencies:

git add -A && git commit -m "Setting up edeliver"
mix do deps.get, compile

Now you can release with edeliver!

mix edeliver update
mix edeliver start
mix edeliver migrate

Because it is based on deliver, it uses only shell scripts and has no further dependencies except the Erlang/Elixir build system.

It can be used with any one of these build systems:

• mix in conjunction with distillery for elixir/erlang releases (recommended)
• mix in conjunction with relx for elixir/erlang releases
• rebar3 for pure erlang releases or in conjunction with rebar_mix plugin to build also Elixir sources and dependencies
• rebar for legacy pure erlang releases

Edeliver tries to autodetect which system to use:

• If a ./mix.exs and a rel/config.exs file exists, mix is used fetch the dependencies, compile the sources and distillery is used to generate the releases / upgrades.
• If a ./relx.config file exists in addition to a ./mix.exs file, mix is used fetch the dependencies, compile the sources and relx is used to generate the releases / upgrades.
• If a ./rebar.config file exists but no ./relx.config, rebar3 is used to fetch the dependencies, compile the sources and to build the release
• Otherwise rebar is used to fetch the dependencies, compile the sources and generate the releases / upgrades. It is recommended to migrate to rebar3 in that case.

This can be overridden by the config variables BUILD_CMD=rebar3|rebar|mix, RELEASE_CMD=rebar3|rebar|mix|relx and USING_DISTILLERY=true|false in .deliver/config.

Edeliver uses ssh and scp to build and deploy the releases. It is recommended that you use ssh and scp with key+passphrase only. You can use ssh-add if you don't want to enter your passphrase every time.

It may be required to install and configure git on your build host. You may also have to clone the repository initially at the BUILD_AT path, although edeliver will try to take care of this for you. Erlang and Elixir must be installed and available on the BUILD_HOST. The default shell for the build user should be bash or zsh on your build host (usually already the default on most systems).

The build host must be similar to the production/staging hosts. For example, if you want to deploy to a production system based on Linux, the release must also be built on a Linux system.

The Erlang runtime (OTP) and the Elixir runtime are packaged with the release—you do not have to install Erlang or Elixir separately on your production/staging servers.

If using mix, add edeliver and your build tool (distillery as hex package to your mix.exs config:

defp deps do
  [
    {:edeliver, ">= 1.9.2"},
    {:distillery, "~> 2.1", warn_missing: false},
  ]
end

Run mix do deps.get, deps.compile. Edeliver is then available as a mix task: mix edeliver.

Some edeliver commands used for server administration require that edeliver be running on the server itself so that it can respond. These commands include version, migrate, show migrations, etc. To enable this, add edeliver as application to be started in mix.exs. It should be added last at the end of the list:

def application, do: [
  applications: [
    # ...
    :edeliver,
  ],
]

When using rebar3, edeliver can be added as rebar3 dependency. Just add it to your rebar.config (and ensure that a ./rebar3 binary/link is in your project directory):

{deps, [
  % ...
  {edeliver, {git, "git:https://github.com/edeliver/edeliver.git", {tag, "1.9.2"}}}
]}.

And link the edeliver binary to the root of your project directory:

wget https://s3.amazonaws.com/rebar3/rebar3 && chmod +x rebar3
./rebar3 get-deps
ln -s ./_build/default/lib/edeliver/bin/edeliver ./edeliver 

Then use the linked binary ./edeliver to build and deploy releases. The default rebar3 profile can be overridden by setting the REBAR_PROFILE environment variable in the edeliver config e.g. to prod.

When using rebar, edeliver can be added as rebar dependency. Just add it to your rebar.config (and ensure that a ./rebar binary/link is in your project directory):

{deps, [
  % ...
  {edeliver, "1.9.2",
    {git, "git:https://github.com/boldpoker/edeliver.git", {branch, master}}}
]}.

And link the edeliver binary to the root of your project directory:

./rebar get-deps # when using rebar, or ...
ln -s ./deps/edeliver/bin/edeliver .

Then use the linked binary ./edeliver instead of the mix edeliver tasks from the examples.

Create a .deliver directory in your project folder and add the config file:

#!/usr/bin/env bash

APP="your-erlang-app" # name of your release

BUILD_HOST="build-system.acme.org" # host where to build the release, or "docker"
BUILD_USER="build" # local user at build host
BUILD_AT="/tmp/erlang/my-app/builds" # build directory on build host

STAGING_HOSTS="test1.acme.org test2.acme.org" # staging / test hosts separated by space
STAGING_USER="test" # local user at staging hosts
TEST_AT="/test/my-erlang-app" # deploy directory on staging hosts. default is DELIVER_TO

PRODUCTION_HOSTS="deploy1.acme.org deploy2.acme.org" # deploy / production hosts separated by space
PRODUCTION_USER="production" # local user at deploy hosts
DELIVER_TO="/opt/my-erlang-app" # deploy directory on production hosts

To use different configurations on different hosts, you can configure edeliver to link the vm.args and/or the sys.config files in the release package by setting the LINK_VM_ARGS=/path/to/vm.args and/or LINK_SYS_CONFIG=/path/to/sys.config variables in the edeliver config if you use mix and distillery to build the releases.

Another strategy is to use runtime environment variable evaluation (available for distillery and relx). For more information on this technique, see Plataformatec - Deploying Elixir with edeliver

This strategy relies on exporting an environment variable in your deployment environment to signal that environment variable replacement should be performed, as well as exporting all of the environment variables your configuration relies on.

For relx, export RELX_REPLACE_OS_VARS=true. For distillery, export REPLACE_OS_VARS=true.

For example in ~/.profile

export REPLACE_OS_VARS=true
export MY_CUSTOM_DATABASE_PORT=5433

For build commands the following configuration variables must be set:

• APP: the name of your release which should be built
• BUILD_HOST: the host where to build the release, or "docker" to build in a docker container
• BUILD_USER: the local user at build host
• BUILD_AT: the directory on build host where to build the release. must exist.

The built release is then copied to your local directory .deliver/releases and can then be delivered to your production servers by using one of the deploy commands.

If compiling and generating the release build was successful, the release is copied from the remote build host to the release store. The default release store is the local .deliver directory but you can configure any destination with the RELEASE_STORE= environment variables, also remote ssh destinations (in your server network) like [email protected]:/releases/, amazon s3 locations like s3:https://AWS_ACCESS_KEY_ID@AWS_SECRET_ACCESS_KEY:bucket or as a docker image like docker:https://edeliver/echo-server. The release is copied from the remote build host using the RELEASE_DIR= environment variable. If this is not set, the default directory is found by finding the subdirectory that contains the generated RELEASES file and has the $APP name in the path. e.g. if $APP=myApp and the RELEASES file is found at rel/myApp/myApp/releases/RELEASE the rel/myApp/myApp is copied to the release store.

To build releases and upgrades faster, you might adjust the GIT_CLEAN_PATHS variable in your config file e.g. to something like ="_build rel priv/generated" which defaults to .. That value means, that everything from the last build is reset (beam files, release files, deps, generated assets etc.) before the next build is started to ensure that no conflicts with old (e.g. removed or renamed) files might arise. You can also use the command line option --skip-git-clean to skip this step completely and in addition with the --skip-mix-clean option for full incremental builds.

mix edeliver build release [--revision=<git-revision>|--tag=<git-tag>] [--branch=<git-branch>]

Builds an initial release that can be deployed to the production hosts. If you want to build a different tag or revision, use the --revision= or the --tag argument. If you want to build a different branch or the tag / revision is in a different branch, use the --branch= argument.

If BUILD_HOST is set to "docker", edeliver builds the release in a docker container instead of building on a build host. It uses the docker image set as DOCKER_BUILD_IMAGE which defaults to elixir:1.13.3 with erlang 24, but can be overridden in your .deliver/config. When building in a docker container, the git repository to build is pushed to the local dir .docker-build which is then mounted into the container and edelivers build commands are executed as docker exec commands in the container.

If RELEASE_STORE is a (private) docker image in a docker registry like docker:https://edeliver/echo-server the built release will be embedded into a docker image based on DOCKER_RELEASE_BASE_IMAGE (which defaults to edeliver/release-base:1.0) and pushed with that image name from the RELEASE_STORE to your registry (if --push is used). It creates (and optionally pushes) three image tags: release version + latest, release version + git sha and release version + branch. The release can then be started on a host authenticated at the same docker registry like this:

docker start -ti edeliver/echo-server:1.0-latest -p 8080:8080 echo-server/bin/echo-server console

mix edeliver build upgrade --from=<git-tag-or-revision>|--with=<release-version-from-store>
                          [--to=<git-tag-or-revision>] [--branch=<git-branch>]

Builds an upgrade package that can be deployed to production hosts with running nodes without restarting them. To build an upgrade package you need the release or upgrade package (when using distillery) of the running release. If it is available (in the release store), you can build the upgrade to the new version by passing the old version to the --with=<old-version> option. If not, you can build the old release and the live upgrade from it in a single step by using the --from=<git-tag-or-revision> option. If you don't want to build an upgrade to the current head of the given branch (master is the default), you can use the --to=<git-tag-or-revision> option. If the upgrade package is built, you might want to modify the generated upgrade instructions (relup) as described in the next section or (more advanced) automatically patch the relup file by implementing your own Edeliver.Relup.Modificationbehaviour to automate this step.

mix edeliver edit relup [--version=<upgrade-version>]

From the auto-generated appup instructions of all included and updated applications, a relup file is generated during the build upgrade command and included in the upgrade package. It contains the upgrade instructions for the new release version. If there are dependencies between modules or applications, it might be necessary to modify this file, e.g. changing the order of the applications or modules that are reloaded. If there are repeating steps to adjust the relup for your application, you can automate this step by implementing your own Edeliver.Relup.Modification behavior.

edeliver provides a way to automatically increment the current version for the current build and/or to append metadata to the version (such as the current git sha). Having unique versions for each release is important especially if you build hot code upgrades. It also helps to determine exactly which version is running when using mix edeliver version. For more information check the --auto-version= option described e.g in mix edeliver help upgrade or in the wiki.

To build upgrades, there must be only one release in the release directory (rel) of your project as configured in your rebar.config. E.g. if you want to build two different releases project-dir/rel/release_a and project-dir/rel/release_b you need two rebar.config files that refer only to either one of that release directories in the sub_dirs section. You can then pass the config file to use by setting the environment REBAR_CONFIG= at the command line. The reason for that is, that when the upgrade is build with rebar, rebar tries to find the old version in both release directories.

mix edeliver deploy release|upgrade [[to] staging|production] [--version=<release-version>] [Options]

Deploy commands deliver the builds that were created with a build command to your staging or production hosts. They can also perform a live code upgrade. Built releases or upgrades are available in your local directory .deliver/releases. To deploy releases the following configuration variables must be set:

• APP: the name of your release which should be built

• PRODUCTION_HOSTS: the production hosts to deploy to, separated by space

• PRODUCTION_USER: the local users at the production hosts

• DELIVER_TO: the directory at the production hosts to deploy the release at

• STAGING_HOSTS: the staging hosts to test the releases at, separated by space

• STAGING_USER: the local users at the staging hosts

• TEST_AT: the directory at the staging hosts. if not set, the DELIVER_TO is used as directory

Deploying to staging can be used to test your releases and upgrades before deploying them to the production hosts. Staging is the default target if you don't pass the [to] production argument.

If the RELEASE_STORE is a docker image, the deploy command pulls and starts the image with the given tag as version. See section Deploy Docker Releases below for details.

mix edeliver deploy

Deploys an initial release at the production hosts. This requires that the build release command was executed before.

If there are several releases in the release store, you will be asked which release to deploy or you can pass the version by the --version= argument variable. If the nodes on the remote deploy hosts are up, the running old release is not affected—the new release will be available only after starting or restarting the nodes on the deploy hosts.

mix edeliver upgrade

Deploys an upgrade at the production hosts and upgrades the running nodes to the new version. This requires that the build upgrade command was executed before, and that there is already an initial release deployed to the production hosts, and that the node is running.

Release archives in your release store that were created by the build release command cannot be used to deploy an upgrade.

This command requires that your release start script was generate by a recent rebar version that supports the upgrade command in addition to the start|stop|ping|attach commands. Releases generated with mix and distillery always contain the upgrade command.

If using rebar, make sure that the install_upgrade.escript file which was generated by rebar is included in your release. So ensure, that the following line is in your reltool.config:

{overlay, [ ...
       {copy, "files/install_upgrade.escript", "bin/install_upgrade.escript"}
]}.

When embedding releases into docker containers, the deploy command pulls the docker image from the registry defined as RELEASE_STORE and extracts the boot script from /$APP/bin/start_container (can be configured in CONTAINER_START_SCRIPT) to $DELIVER_TO/bin while replacing the string {{edeliver-version}} with the version which is deployed. The script should use that value to always start that tag of the image.

The start script should handle the same commands as the extended start script from relx/rebar, at least the start, stop and version commands.

It could start the container with e.g. like this

VERSION="{{edeliver-version}}"
docker run --rm --detatch \
           --workdir "/${APP}" \
           --publish 127.0.0.1:8080:8080 \
           --env ERL_DIST_PORT="${ERL_DIST_PORT:-9999}" \
           --env INET_DIST_USE_INTERFACE='{0,0,0,0}' \
           --mount type=bind,source=$HOME/.erlang.cookie,target=/root/.erlang.cookie
       my-registry/my-image:$VERSION /$APP/bin/$APP console

edeliver has a set of commands to check up on your running nodes:

mix edeliver ping production # shows which nodes are up and running
mix edeliver version production # shows the release version running on the nodes
mix edeliver show migrations on production # shows pending database migrations
mix edeliver migrate production # run database migrations
mix edeliver restart production # or start or stop

If something goes wrong, retry with the --verbose option. If you want to see everything, try the --debug option.

For detailed information about the edeliver commands and their options, try mix edeliver help <command>.

For advanced usage have a look also at the wiki.

Definitely join the #deployment channel in the Elixir Slack community as well.

your-app/                              <- project root dir
  + rebar                              <- rebar binary
  + mix                                <- optional mix binary when compiling with mix
  + relx                               <- optional relx binary if rebar is not used
  + edeliver                           <- edeliver binary linking to deps/deliver/bin/deliver
  + rebar.config                       <- should have "rel/your-app" in the sub_dirs section
  + mix.exs                            <- if present, mix is used for dependencies and compile
  + relx.config                        <- if present, relx is used for releases
  + .deliver                           <- default release store
  |  + releases/*.tar.gz               <- the built releases / upgrade packages
  |  + appup/OldVsn-NewVsn/*.apppup    <- generated appup files
  |  + config                          <- deliver configuration
  + src/                               <- erlang source files
  |  + *.erl
  |  + your-app.app.src
  + lib/                               <- elixir source files
  |  + *.ex
  + priv/
  + deps/
  |  + edeliver/
  + rel/
     + your-app/
         + files/
         |   + your-app                <- binary to start|stop|upgrade your app
         |   + nodetool                <- helper for your-app binary
         |   + install-upgrade.escript <- helper for the upgrade task of your-app binary
         |   + sys.config              <- app configuration for the release build
         |   + vm.args                 <- erlang vm args for the node
         + reltool.config              <- should have the install_upgrade.escript in overlay section

Build a release and deploy it to your production hosts:

mix edeliver build release --branch=feature
mix edeliver deploy release to production
mix edeliver start production

Or execute the above steps with a single command:

mix edeliver update production --branch=feature --start-deploy

Build a live upgrade from v1.0 to v2.0 for a release and deploy it to production:

# build upgrade from tag v1.0 to v2.0

mix edeliver build upgrade --from=v1.0 --to=v2.0
mix edeliver deploy upgrade to production

# or if you have the old release in your release store,
# you can build the upgrade with that old release instead of the old git revision/tag

mix edeliver build upgrade --with=v1.0 --to=v2.0
mix edeliver deploy upgrade to production

# run ecto migrations manually:
mix edeliver migrate production
# or automatically during upgrade when upgrade is built with --run-migrations

The deployed upgrade will be available immediately, without restarting your application. If the generated upgrade instructions (relup) for the hot code upgrade are not sufficient, you can modify these files before installing the upgrade by using the edeliver edit relup command.

To execute that steps by a single command and upgrade e.g. all production nodes automatically from their running version to the current version using hot code upgrade without restarting, you can use the upgrade command:

mix edeliver upgrade production

This performs the following steps automatically:

• Detect current version on all running nodes
• Validate that all nodes run the same version
• Build new upgrade from that version to the current version
• Auto-patch the relup file
• Deploy (hot code) upgrade while nodes are running
• Validate that all nodes run the upgraded version
• Deploy the release to not running nodes

(The MIT license)

Copyright (c) Gerhard Lazu

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.