Using Habitat to Create Rancher Catalog Templates


Today, Chef announced the release of Habitat, a new approach to automate applications. Habitat shifts the focus of application management and configuration from the infrastructure to the application itself. In a nutshell, it allows users to create packages that encapsulate the application logic itself, runtime dependencies, and configuration. These packages can then auto-update accordingly to policies set by your organization. In this article, I will show you how to leverage the runtime configuration and service member discovery capabilities of Habitat to build a Rancher Catalog template. For illustration purposes, we will look at the Habitat plan and the Rancher Catalog item for RabbitMQ.

Habitat plans

To create a Habitat package, you’ll need to start with a Habitat plan. A Habitat plan is a collection of shell scripts that define what the package contains, how it is built, and how it can be configured. The Habitat plan for RabbitMQ discussed in this post is available here, in GitHub. Let’s walk through each of the files in the RabbitMQ plan.

default.toml (available here)

The default.toml file specifies which configuration options Habitat should expose for this package, as well as their default values. In our case, the first five options are rabbitmq-server configuration options, and are used to populate the “config/rabbitmq.conf” template. For example, the following section sets up the configuration variables that are referred in the {{cfg.default_user}} and {{cfg_default_pass}} expressions in the config/rabbitmq-server.conf template:

# Default user name
default_user = "guest"

# Default user password
default_pass = "guest"

The last two configuration variables:

# Erlang cookie
# Used for authenticating nodes in a cluster
# Must be an alphanumeric string of any length
erlang_cookie = "SUPERSECRETSTRING"

# Enable the management plugin (HTTP API on port 15672)
enable_management="false"

are not used in the configuration template but in the init hook shell script to create the Erlang cookie required for clustering and to enable the RabbitMQ management plugin.

plan.sh (available here)

The top most part of the plan.sh file specifies basic information about the plan. To learn what all of those mean, check out the Habitat documentation for the plan syntax. For now, let’s just look at the following settings:

pkg_source=https://www.rabbitmq.com/releases/rabbitmq-server/v${pkg_version}/rabbitmq-server-generic-unix-${pkg_version}.tar.xz
pkg_deps=(core/coreutils core/erlang/18.3)
pkg_expose=(4369 25672 5672 15672)

As you can see, pkg_source points to the release archive for the 3.6.2 version of RabbitMQ. When building the package, Habitat will automatically download and extract this archive. In pkg_deps, we specified the runtime dependencies for our package. Since RabbitMQ is an Erlang application, we reference the Erlang package available from the “core” depot (which is a public repository containing a number of key dependency packages). pkg_expose lists all the ports on which RabbitMQ exposes services. When Habitat builds a Docker image from this package, each of the specified ports will be set as an EXPOSE attribute of the image. Next, let’s take a look at the do_install() block:

do_install()
 cat > ${pkg_prefix}/etc/rabbitmq/rabbitmq-env.conf << EOF
CONFIG_FILE=$pkg_svc_config_path/rabbitmq
LOGS=-
SASL_LOGS=-
EOF
 chmod 644 ${pkg_prefix}/etc/rabbitmq/rabbitmq-env.conf
}

Here, we first copy the extracted RabbitMQ source files from the temporary build path into the actual package directory. Then we create a rabbitmq-env.conf file and populate it with the required options to make the rabbitmq-server use the dynamically generated rabbitmq-server.config, and write its log to stdout instead of to syslog.

init (available here)

The init file is executed by the Habitat supervisor before the application is started. For our purposes, it’s the perfect place to write the Erlang cookie so it’s picked up when the rabbitmq-server is started. As you can see, the value we write to the erlang.cookie file is once again a handlebar expression referring to a variable in the default.toml file:

erlang_cookie_file={{pkg.svc_path}}/var/lib/rabbitmq/.erlang.cookie
echo -n "{{cfg.erlang_cookie}}" > $erlang_cookie_file
chmod 600 $erlang_cookie_file

Finally, the init script also takes care of enabling the RabbitMQ management plugin.

run (available here)

The script in the “run” file takes care of starting the rabbitmq-server. Note that this sets the HOME environment variable to the directory in which we wrote the Erlang cookie earlier.

health_check (available here)

The supervisor calls the “health_check” hook to check whether the application has fully started. Here we leverage the “rabbitmqctl” tool’s “node_health_check” command to check whether the rabbitmq-server is fully started. We use the “health_check” hook for the RabbitMQ package because it allows us to block the followers in our census from starting their rabbitmq nodes until RabbitMQ node on the leader has fully started up and is able to accept clustering requests.

rabbitmq.conf (available here)

Finally, the config/rabbitmq.config file includes the Erlang syntax and handlebars expressions which the supervisor uses to dynamically create a valid configuration file for the rabbitmq-server. Let’s look at the most interesting section here:

{{~#if svc.me.follower}}
  {mnesia_table_loading_timeout, 30000},
{{~#with svc.leader}}
  {cluster_nodes, {['rabbit@{{hostname}}'], disc}}
{{~/with}}
{{else}}
  {mnesia_table_loading_timeout, 30000}
{{~/if}}

As you can see, we are using if/else expressions to selectively enable automatic clustering for followers using the hostname of the leader (svc.leader).

Building a Docker image from a Habitat plan

Since our goal is to deploy this Habitat package as a service in Rancher, we first need to build a Docker image using the following steps:

cd ~/plans/rabbitmq
hab studio -k rancher enter
[default:/src:0]# build
hab pkg export docker rancher/rabbitmq

This will result in a Docker image tagged as rancher/rabbitmq:latest.

Rancher Catalog Template

Next, let’s see how we can use this image in a Rancher Catalog template to deploy a RabbitMQ cluster. The template is available on GitHub here. Let’s have a closer look at the docker-compose.ymp for this template, starting with the rabbitmq-master service:

rabbitmq-master:
  image: janeczku/rabbitmq:dev
  command:
  - "--topology"
  - "initializer"
  - "--listen-peer"
  - "0.0.0.0"
  ports:
  - "15672:15672"
  - "${AMQP_PORT}:${AMQP_PORT}"
  expose:
    - "9634/udp"
  environment:
   HAB_RABBITMQ: 'enable_management="${ENABLE_MANAGEMENT}" default_user="${USER_NAME}" default_pass="${USER_PASS}" (...)
  labels:
    io.rancher.container.hostname_override: container_name

Note the command passed to the Habitat container. The arguments are configuration flags for the supervisor, which runs as PID1 in Habitat service containers. We configure the supervisor with a topology of type “initializer” (which is a leader-follower topology where the followers are blocked until the leader’s application is fully initialized). This ensures that the RabbitMQ node running on the supervisor elected as leader is fully started, so the other nodes don’t fail to join the cluster with it. Finally, notice how we assign the values entered in the Catalog UI to the variables specified in the plan’s default.toml file in a single environment variable named HAB_APPLICATIONNAME. Next, let’s check out the “rabbitmq-slave” service:

rabbitmq-slave:
  image: janeczku/rabbitmq:dev
  command:
    - "--topology"
    - "initializer"
    - "--listen-peer"
    - "0.0.0.0"
    - "--peer"
    - "rabbitmq-master"
  expose:
    - "9634/udp"
  environment:
   HAB_RABBITMQ: 'enable_management="${ENABLE_MANAGEMENT}" default_user="${USER_NAME}" default_pass="${USER_PASS}" (...)
  labels:
    io.rancher.container.hostname_override: container_name
    io.rancher.scheduler.affinity:container_label_soft_ne: io.rancher.stack_service.name=$${stack_name}/$${service_name
  links:
    - rabbitmq-master

As you’ll notice, we are passing the additional flag --peer rabbitmq-master in the command instruction to the container. This makes both rabbitmq-slave services and the rabbitmq-master service members of the same census, and eventually elect a leader to become the first RabbitMQ node started. When you launch this Catalog stack, a single instance of the rabbitmq-master service is going to be started first, followed by the configured number of rabbitmq-slave instances. The census of supervisors will elect a leader, which becomes the first to run a RabbitMQ node. Once this first node has become fully operational, the other service instances (followers) are going to join this node and form a RabbitMQ cluster. And with that, we’ve launched RabbitMQ with Habitat and the Rancher Catalog! If you have additional questions, please reach out to me (jan@rancher.com), or on Twitter @Rancher_Labs.

快速开启您的Rancher之旅