7 sept. 2015

Meteor Devops on OSX with Docker set for Ubuntu 15.04

Introduction

While using Meteor in development is an easy task and deploying it on Meteor's infrastructure is a no brainer, things may start to get messy if you need to deploy it, secure it and scale it on your cloud. Especially if your customer imposes you a specific constraint on cloud sovereignty. The best way to achieve easy deployment is using the excellent Meteor Up tool. But if it fails or if you need to go a bit further in your infrastructure deployment, I recommend that you start using Docker to get familiar with this handy DevOps tool.
I hope that this tutorial will lead you on the appropriate tracks.

Versions applied in this tutorial

As you may need to update this tutorial for your own DevOps use cases, here is the complete list of versions used in this tutorial:
  • OSX 10.10.5 as the development platform
  • Ubuntu 15.04 as Docker host system
  • Debian Jessie 7 with latest updates as Docker container system
  • Docker 1.8.1
  • Docker Registry 2
  • Docker Machine 0.4.1
  • Docker Compose 1.4.0
  • VirtualBox 5.0.2
  • Meteor 1.1.0.3
  • NGinx 1.9.4-1
  • NodeJS 0.10.40
  • Mongo 3.0.6 - WiredTiger
Software architecture
Why Debian Jessie instead of Debian Wheezie? Simple, a gain of 30MB of footprint. Note that we could have set this tutorial on other even smaller Linux distributions for our Docker Images, like Alpine Linux. But as time of this writing, these smaller distributions do not offer the package required for installing Meteor (namely, MongoDB and node-fibers).

Installing the tooling

If you have never done it before install Homebrew and its plugin Caskroom.
ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
brew install caskroom/cask/brew-cask
Then install VirtualBox and Vagrant:
brew cask install virtualbox vagrant
Now install Docker and its tools:
brew install docker docker-machine docker-compose
For easing the access to VM and servers, we are using an SSH key installer:
brew install ssh-copy-id
For parsing and querying JSON produced by Docker, we are using ./jq:
brew install jq

Some file structure

For differentiating the Meteor project from the DevOps project, we store our files like so:
.
├── app
└── docker
The app folder contains the root of Meteor sources and the docker folder contains the root of DevOps sources.

Create your virtual machines as Docker Machine

Create a Vagrantfile that matches your production environment. Here, we are using an Ubuntu 15.04 with Docker pre-installed.
hosts = {
  "dev" => "192.168.1.50",
  "pre" => "192.168.1.51"
}

Vagrant.configure(2) do |config|
  config.vm.box = "ubuntu/vivid64"
  config.ssh.insert_key = false
  hosts.each do |name, ip|
    config.vm.define name do |vm|
      vm.vm.hostname = "%s.example.org" % name
      #vm.vm.network "private_network", ip: ip
      vm.vm.network "public_network", bridge: "en0: Wi-Fi (AirPort)", ip: ip
      vm.vm.provider "virtualbox" do |v|
        v.name = name
      end
      vm.vm.provision "shell", path: "provisioning.sh"
    end
  end
end
I've provided 2 network configurations here. The first one is a private network leading to 2 virtual machines that are not accessible to your local network ( only your local OSX). The second bridges your local OSX network driver so that your VMs gain public access within your LAN. Note that for both of these network configurations, I've used static IPs.
Before creating our virtual machine, we need to setup a provisioning.sh:
#!/bin/bash
# Overriding bad Systemd default in Docker startup script
sudo mkdir -p /etc/systemd/system/docker.service.d
echo -e '[Service]\n# workaround to include default options\nEnvironmentFile=-/etc/default/docker\nExecStart=\nExecStart=/usr/bin/docker -d -H fd:// $DOCKER_OPTS' | sudo tee /etc/systemd/system/docker.service.d/ubuntu.conf
# Set Docker daemon with the following properties:
# * Daemon listen to external request and is exposed on port 2376, the default Docker port.
# * Docker uses the AUFS driver for file storage.
# * Daemon uses Docker's provided certification chain.
# * Dameon has a generic label.
# * Daemon is able to resolve DNS query using Google's DNS.
echo 'DOCKER_OPTS="-H tcp://0.0.0.0:2376 -H unix:///var/run/docker.sock --storage-driver aufs --tlsverify --tlscacert /etc/docker/ca.pem --tlscert /etc/docker/server.pem --tlskey /etc/docker/server-key.pem --label provider=generic --dns 8.8.8.8 --dns 8.8.4.4"'  | sudo tee /etc/default/docker
sudo systemctl daemon-reload
sudo systemctl restart docker
# Enable Docker on server reboot
sudo systemctl enable docker
# Remove and clean unused packages
sudo apt-get autoremove -y
sudo apt-get autoclean -y
Now, we are starting our virtual hosts and declare it as a Docker Machine:
vagrant up --no-provision
Throughout this terminal sessions, we need some environment variables. We store them in a local_env.sh file that we fill step by step and source each time we open a new terminal session:
export HOST_IP_DEV='192.168.1.50'
export HOST_IP_PRE='192.168.1.51'
# Use preferably your FQDN (example.org)
export HOST_IP_PROD='X.X.X.X'
If you are using Fish like me, use the following content:
export HOST_IP_DEV='192.168.1.50'
export HOST_IP_PRE='192.168.1.51'
# Use preferably your FQDN (example.org)
export HOST_IP_PROD='X.X.X.X'
This should provide an easy access to all parts of the following network architecture: Network architecture
Open 3 terminal sessions. In the first session, launch the following commands:
docker-machine -D create -d generic \
  --generic-ip-address $HOST_IP_DEV \
  --generic-ssh-user vagrant \
  --generic-ssh-key ~/.vagrant.d/insecure_private_key \
  dev
In the second session, launch the following commands:
docker-machine -D create -d generic \
  --generic-ip-address $HOST_IP_PRE \
  --generic-ssh-user vagrant \
  --generic-ssh-key ~/.vagrant.d/insecure_private_key \
  pre
Now, in the last session, wait for the 2 previous sessions to be blocked on the following repeated message Daemon not responding yet: dial tcp XX.XX.XX.XX:2376: connection refused and issue the following command:
vagrant provision
What's going on here? Actually, the current state of Docker for Ubuntu 15.04 doesn't support DOCKER_OPTS. This is due to the transition in Ubuntu from upstart to Systemd. Plus, when we are creating our Docker Machine in our local OSX, Docker Machine re-install Docker on the host. Thus, we end up with a screwed installation on the host unable to speak to the outside world (leading to the message Daemon not responding yet: dial tcp XX.XX.XX.XX:2376: connection refused). Basically, the vagrant provisioning script patches both vagrant virtual servers. You can reuse the content of this script on your production server when you create the associated Docker Machine. For this, you can use the following command:
ssh root@$HOST_IP_PROD "bash -s" < ./provisioning.sh
In this last section, we will finish our configuration of our development and pre-production hosts by installing Docker Machine and securing their open ports with simple firewall rules. The script that we are using is named postProvisioning.sh.
#!/bin/bash
# Install Docker Machine
curl -L https://github.com/docker/machine/releases/download/v0.4.0/docker-machine_linux-amd64 | sudo tee /usr/local/bin/docker-machine > /dev/null
sudo chmod u+x /usr/local/bin/docker-machine

# Install Firewall
sudo apt-get install -y ufw
# Allow SSH
sudo ufw allow ssh
# Allow HTTP and WS
sudo ufw allow 80/tcp
# Allow HTTPS and WSS
sudo ufw allow 443/tcp
# Allow Docker daemon port and forwarding policy
sudo ufw allow 2376/tcp
sudo sed -i -e "s/^DEFAULT_FORWARD_POLICY=\"DROP\"/DEFAULT_FORWARD_POLICY=\"ACCEPT\"/" /etc/default/ufw
# Enable and reload
yes | sudo ufw enable
sudo ufw reload
We execute this script on both VM using simple SSH commands like so:
ssh -i ~/.vagrant.d/insecure_private_key vagrant@$HOST_IP_DEV "bash -s" < ./postProvisioning.sh
ssh -i ~/.vagrant.d/insecure_private_key vagrant@$HOST_IP_PRE "bash -s" < ./postProvisioning.sh
Now you can access your VM either via Docker, Vagrant and plain SSH. To finish our VM configuration, we are going to allow full root access to the VM without requiring to use password. For that, you need a public and a private SSH keys on your local machine. If you haven't done it before simply use the following command:
ssh-keygen -t rsa
Now, using Vagrant, copy the content of your ~/.ssh/id_rsa.pub in each of the VM's /root/.ssh/authorized_key.

Reference your production host as a Docker Machine

In this example, we are using a VPS from OVH with a pre-installed Ubuntu 15.04 with Docker. These VPS starts at 2.99€ (around $3.5) per month and comes with interesting features such as Anti-DDos, real time monitoring, ...
Preinstalled VPS comes with an OpenSSH access. Therefore, we will be using the generic-ssh driver for our Docker Machine just like we did for the Vagrant VM for development and pre-production. And like before, we are using 2 terminal sessions to overcome the Docker installation issue on Ubuntu 15.04.
In the first terminal session, we setup a root SSH access without password like so:
ssh-copy-id root@$HOST_IP_PROD
# Now, you should check if your key is properly copied
ssh root@$HOST_IP_PROD "cat /root/.ssh/authorized_keys"
cat ~/.ssh/id_rsa.pub
# These 2 last commands should return the exact same key
I've been tricked by some ssh-user-agent issue there. Docker wasn't reporting any issue even in debug mode and was just exiting with a default error code. So, be careful that your public key is exactly the same on your local machine, your VM and your production host.
Next and still on the same terminal session, we declare our production host :
docker-machine -D create -d generic \
  --generic-ip-address $HOST_IP_PROD \
  --generic-ssh-user root \
  prod
And on the second terminal session, when the message Daemon not responding yet: dial tcp X.X.X.X:2376: connection refused appears on the first session, we launch:
ssh root@$HOST_IP_PROD "bash -s" < ./provisioning.sh
The last remaining step consists into solidifying our security by enabling a firewall on the host and removing the old packages:
ssh root@$HOST_IP_PROD "bash -s" < ./postProvisioning.sh

Creating your own registry

Basically, what we want to achieve is micro-services oriented to stick to a multi-tiers architecture: Docker architecture
This architecture could then be spread over a Docker Swarm of multiple servers or kept on a single one. But playing with multiple containers in development is quickly a pain. We can leverage the power of Docker Compose and a local registry to fasten our development of Docker images.
In your first terminal session, activate your development Docker Machine:
eval "$(docker-machine env dev)"
# In Fish
eval (docker-machine env dev)
Create a Docker Compose file registry.yml:
registry:
  restart: always
  image: registry:2
  ports:
    - 5000:5000
  environment:
    REGISTRY_STORAGE_FILESYSTEM_ROOTDIRECTORY: /var/lib/registry
  volumes:
    - /var/lib/registry:/var/lib/registry
Now, we will use the development Docker Machine as our local registry:
ssh root@$HOST_IP_DEV "mkdir /var/lib/registry"
docker-compose -f registry.yml up -d
For making it visible to our preproduction VM, we need to update our default firewall rules:
ssh root@$HOST_IP_DEV ufw allow 5000
Now we are editing our /etc/default/docker configuration file for adding this insecure registry in both our development and preproduction VM with this new flag:
# On 192.168.1.50 & 192.168.1.51, in /etc/default/docker, we add in DOCKER_OPTS:
--insecure-registry 192.168.1.50:5000
We need to restart our Docker daemon and restart the Docker registry on the development VM:
ssh root@$HOST_IP_DEV systemctl restart docker
ssh root@$HOST_IP_PRE systemctl restart docker
eval "$(docker-machine env dev)"
Our final step in the registry management is to login your preproduction VM and your production server to Docker Hub using your Docker credential.
eval "$(docker-machine env pre)"
docker login
eval "$(docker-machine env prod)"
docker login
Note that our registry isn't published outside our LAN. This makes it unusable for our production host. This development chain uses Docker Hub for publishing your images. Exposing this private registry to the outside world would require some additional configurations to tighten its security and server with a publicly exposed IP. While you could solely rely on Docker Hub for publishing your images, pushing and pulling to the outside world of your LAN are lengthy operations though lighten since Docker 1.6 and Docker Registry 2.

Building Mongo

Our mongo/Dockerfile is based on Mongo's official one. It adds to the picture the configuration of small ReplicaSet for making OPLOG available:
# Based on: https://github.com/docker-library/mongo/blob/d5aca073ca71a7023e0d4193bd14642c6950d454/3.0/Dockerfile
FROM debian:wheezy
MAINTAINER Pierre-Eric Marchandet 

# Update system
ENV DEBIAN_FRONTEND noninteractive
RUN apt-get update && \
    apt-get upgrade -y -qq --no-install-recommends && \
    apt-get install -y -qq --no-install-recommends apt-utils && \
    apt-get install -y -qq --no-install-recommends \
      ca-certificates curl psmisc apt-utils && \
    apt-get autoremove -y -qq && \
    apt-get autoclean -y -qq && \
    rm -rf /var/lib/apt/lists/*

# Grab gosu for easy step-down from root
RUN gpg --keyserver ha.pool.sks-keyservers.net --recv-keys B42F6819007F00F88E364FD4036A9C25BF357DD4 && \
    curl -sS -o /usr/local/bin/gosu -L "https://github.com/tianon/gosu/releases/download/1.2/gosu-$(dpkg --print-architecture)" && \
    curl -sS -o /usr/local/bin/gosu.asc -L "https://github.com/tianon/gosu/releases/download/1.2/gosu-$(dpkg --print-architecture).asc" && \
    gpg --verify /usr/local/bin/gosu.asc && \
    rm /usr/local/bin/gosu.asc && \
    chmod +x /usr/local/bin/gosu

# Install MongoDB
ENV MONGO_MAJOR 3.0
ENV MONGO_VERSION 3.0.6
RUN groupadd -r mongodb && \
    useradd -r -g mongodb mongodb && \
    apt-key adv --keyserver ha.pool.sks-keyservers.net --recv-keys 492EAFE8CD016A07919F1D2B9ECBEC467F0CEB10 && \
    echo "deb http://repo.mongodb.org/apt/debian wheezy/mongodb-org/$MONGO_MAJOR main" > /etc/apt/sources.list.d/mongodb-org.list && \
    apt-get update && \
    apt-get install -y -qq --no-install-recommends \
      mongodb-org=$MONGO_VERSION \
      mongodb-org-server=$MONGO_VERSION \
      mongodb-org-shell=$MONGO_VERSION \
      mongodb-org-mongos=$MONGO_VERSION \
      mongodb-org-tools=$MONGO_VERSION && \
    rm -rf /var/lib/apt/lists/* && \
    rm -rf /var/lib/mongodb && \
    mv /etc/mongod.conf /etc/mongod.conf.orig && \
    apt-get autoremove -y -qq && \
    apt-get autoclean -y -qq && \
    rm -rf /var/lib/apt/lists/* && \
    # Prepare environment for Mongo daemon: Use a Docker Volume container
    mkdir -p /db && chown -R mongodb:mongodb /db

# Launch Mongo
COPY mongod.conf /etc/mongod.conf
CMD ["gosu", "mongodb", "mongod", "-f", "/etc/mongod.conf"]
We need a configuration file for this Docker image to be built mongo/mongod.conf:
storage:
  dbPath: "/db"
  engine: "wiredTiger"
  wiredTiger:
    engineConfig:
      cacheSizeGB: 1
    collectionConfig:
      blockCompressor: snappy
replication:
  oplogSizeMB: 128
  replSetName: "rs0"
net:
  port: 27017
  wireObjectCheck : false
  unixDomainSocket:
    enabled : true
We could build this image and run it, but I prefer using a Docker Compose file. These file eases the process of build, run and deploys of your Docker images acting as a project file when multiple Docker images are required to work together for an application. Here's the minimal docker/docker-compose.yml that we will enrich in the next steps of this tutorial:
# Persistence layer: Mongo
db:
  build: mongo
  volumes:
    - /var/db:/db
  expose:
    - "27017"
Before building or launching this Docker image, we need to prepare the volume on each host that receives and persists Mongo's data:
ssh root@$HOST_IP_DEV "rm -rf /var/db; mkdir /var/db; chmod go+w /var/db"
ssh root@$HOST_IP_PRE "rm -rf /var/db; mkdir /var/db; chmod go+w /var/db"
ssh root@$HOST_IP_PROD "rm -rf /var/db; mkdir /var/db; chmod go+w /var/db"
For building our Mongo Docker image:
docker-compose build db
# Or even faster, for building and running
docker-compose up -d db
And once it's running, initialize a single instance ReplicaSet for making Oplog tailing available:
docker-compose run db mongo db:27017/admin --quiet --eval "rs.initiate(); rs.conf();"
Some useful commands while developing a container:
# Access to a container in interactive mode
docker run -ti -P docker_db

# Delete all stopped containers
docker rm $(docker ps -a -q)
# Delete all images that are not being used in a running container
docker rmi $(docker images -q)
# Delete all images that failed to build (untagged images)
docker rmi $(docker images -f "dangling=true" -q)

# In Fish
# Delete all stopped containers
docker rm (docker ps -a -q)
# Delete all images that are not being used in a running container
docker rmi (docker images -q)
# Delete all images that failed to build (dangling images)
docker rmi (docker images -f "dangling=true" -q)

Building Meteor

Meteor is fairly easy to build. It's a simple NodeJS app. We start by creating our docker/meteor/Dockerfile:
# Based on: https://github.com/joyent/docker-node/blob/master/0.10/wheezy/Dockerfile
FROM debian:wheezy
MAINTAINER Pierre-Eric Marchandet 

# Update system
ENV DEBIAN_FRONTEND noninteractive
RUN apt-get update && \
    apt-get upgrade -qq -y --no-install-recommends && \
    apt-get install -qq -y --no-install-recommends \
      # CURL
      ca-certificates curl wget \
      # SCM
      bzr git mercurial openssh-client subversion \
      # Build
      build-essential && \
    apt-get autoremove -qq -y && \
    apt-get autoclean -qq -y && \
    rm -rf /var/lib/apt/lists/*

# Install NodeJS
ENV NODE_VERSION 0.10.40
ENV NPM_VERSION 2.13.3
RUN curl -sSLO "https://nodejs.org/dist/v$NODE_VERSION/node-v$NODE_VERSION-linux-x64.tar.gz" && \
    tar -xzf "node-v$NODE_VERSION-linux-x64.tar.gz" -C /usr/local --strip-components=1 && \
    rm "node-v$NODE_VERSION-linux-x64.tar.gz" && \
    npm install -g npm@"$NPM_VERSION" && \
    npm cache clear

# Add PM2 for process management and PhantomJS
RUN npm install -g pm2 phantomjs

# Import sources
COPY bundle /app

# Install Meteor's dependencies
WORKDIR /app
RUN (cd programs/server && npm install)

# Launch application
COPY startMeteor.sh /app/startMeteor.sh
CMD ["./startMeteor.sh"]
Before building this Docker image, we need to prepare the volume on each host that receives its settings.json used for storing your secrets in Meteor:
ssh root@$HOST_IP_DEV "mkdir /etc/meteor"
ssh root@$HOST_IP_PRE "mkdir /etc/meteor"
ssh root@$HOST_IP_PROD "mkdir /etc/meteor"
Now copy your settings.json files on each hosts using a regular SCP. Mine are slightly different depending on the target where I deploy my Meteor apps.
# Just an exammple, adapt it to suit your needs
scp ../app/development.json root@$HOST_IP_DEV:/etc/meteor/settings.json
scp ../app/development.json root@$HOST_IP_DEV:/etc/meteor/settings.json
scp ../app/production.json root@$HOST_IP_DEV:/etc/meteor/settings.json
Note that we do not include our secrets, nor in our code repository by using a .gitgignore file, nor in our Docker Images.
We import our Meteor sources using a shared script docker/buildMeteor.sh for the Meteor container and the NGinx container:
#!/bin/bash
rm -rf meteor/bundle nginx/bundle
cd ../app
meteor build --architecture os.linux.x86_64 --directory ../docker/meteor
cd -
cp -R meteor/bundle nginx
In order to avoid importing too much files in our Docker image, we create a docker/meteor/.dockerignore file which removes the parts dedicated to the clients wich will be serverd by NGinx:
bundle/README
bundle/packages/*/.build*
bundle/packages/*/.styl
bundle/*/*.md*
bundle/programs/web.browser/app
Our last required file is a script docker/meteor/startMeteor.sh for starting Meteor with the private settings that we add as a specific volume:
#!/bin/bash
METEOR_SETTINGS=$(cat /etc/meteor/settings.json) pm2 start -s --no-daemon --no-vizion main.js
Note that we launch Meteor with PM2. As we will see it, it's not a mandatory step as we are using Docker's restart policy in our Docker images. However this process management utility could be used to get some metrics on NodeJS's status.
For building and launching, we are extending our /docker/docker-compose.yml file:
# Application server: NodeJS (Meteor)
server:
  build: meteor
  environment:
    MONGO_URL: "mongodb://db:27017"
    MONGO_OPLOG_URL: "mongodb://db:27017/local"
    PORT: 3000
    ROOT_URL: "https://192.168.1.50"
  volumes:
    - /etc/meteor:/etc/meteor
  expose:
    - "3000"
For building and launching our Meteor Docker image:
docker-compose up -d db server

Building NGinx

It's up to our front container to be created. Let's start with our docker/nginx/Dockerfile:
# Based on: https://github.com/nginxinc/docker-nginx/blob/master/Dockerfile
FROM debian:wheezy
MAINTAINER Pierre-Eric Marchandet 

# Add NGinx official repository
RUN apt-key adv --keyserver pgp.mit.edu --recv-keys 573BFD6B3D8FBC641079A6ABABF5BD827BD9BF62
RUN echo "deb http://nginx.org/packages/mainline/debian/ wheezy nginx" >> /etc/apt/sources.list
ENV NGINX_VERSION 1.9.4-1~wheezy

# Update system
ENV DEBIAN_FRONTEND noninteractive
RUN groupadd -r www && \
    useradd -r -g www www && \
    apt-get update && \
    apt-get upgrade -qq -y --no-install-recommends && \
    apt-get install -qq -y --no-install-recommends \
      ca-certificates nginx=${NGINX_VERSION} && \
    apt-get autoremove -qq -y && \
    apt-get autoclean -qq -y && \
    rm -rf /var/lib/apt/lists/*

# Forward request and error logs to Docker log collector
RUN ln -sf /dev/stdout /var/log/nginx/access.log
RUN ln -sf /dev/stderr /var/log/nginx/error.log

# Configuration files
COPY nginx.conf /etc/nginx/nginx.conf
COPY conf host-specific /etc/nginx/conf/

# Mount points for volumes
RUN mkdir -p /etc/certs /var/cache/nginx /var/tmp

# Source
# Raw source files exposed as HTTP and HTTPS
COPY raw /www/
# Project files exposed as HTTPS
COPY  bundle/programs/web.browser/*.js \
      bundle/programs/web.browser/*.css \
      bundle/programs/web.browser/packages \
      bundle/programs/web.browser/app \
      /www/

# Ensure proper rights on static assets
RUN chown -R www:www /www /var/cache /var/tmp

# Launch NGinx
COPY startNginx.sh /startNginx.sh
RUN chmod u+x /startNginx.sh
CMD ["/startNginx.sh"]
Like the Meteor container, we are using the same import script. This time, we remove the server part of our container using the same technic in the docker/nginx/.dockerignore:
bundle/README
bundle/packages/*/.build*
bundle/packages/*/.styl
bundle/*/*.md*
bundle/programs/server
For building it, we enhance pour docker/docker-compose.yml file:
# Front layer, static file, SSL, proxy cache: NGinx
front:
  build: nginx
  links:
    - server
  environment:
    # Can be: dev, pre, prod
    HOST_TARGET: "dev"
  volumes:
    - /etc/certs:/etc/certs
    - /var/cache:/var/cache
    - /var/tmp:/var/tmp
  ports:
    - "80:80"
    - "443:443"
Our NGinx requires certificates set on the hosts in /etc/certs. For the production host, you require SSL certificates from a certificate authority know by the browser vendors. For the development and the preproduction hosts, we can use self signed certificate that we create on our hosts:
ssh root@$HOST_IP_DEV "mkdir -p /etc/certs; openssl req -nodes -new -x509 -keyout /etc/certs/server.key -out /etc/certs/server.crt -subj '/C=FR/ST=Paris/L=Paris/CN=$HOST_IP_DEV'"
ssh root@$HOST_IP_PRE "mkdir -p /etc/certs; openssl req -nodes -new -x509 -keyout /etc/certs/server.key -out /etc/certs/server.crt -subj '/C=FR/ST=Paris/L=Paris/CN=$HOST_IP_PRE'"
We need 2 additional volumes exposed on each host, one for NGinx's cache and another one for NGinx temporary files:
ssh root@$HOST_IP_DEV "mkdir /var/cache; chmod go+w /var/cache; mkdir /var/tmp; chmod go+w /var/tmp"
ssh root@$HOST_IP_PRE "mkdir /var/cache; chmod go+w /var/cache; mkdir /var/tmp; chmod go+w /var/tmp"
ssh root@$HOST_IP_PROD "mkdir -p /etc/certs; mkdir /var/cache; chmod go+w /var/cache; mkdir /var/tmp; chmod go+w /var/tmp"
In our Docker Container, we have already imported the static part of our Meteor app that will be exposed through HTTPS. Our NGinx server will also act as a static file server in HTTP.Simply put your static assets in the docker/nginx/raw folder for that.
While serving HTTP file for our Meteor application has no interest, it could be usefull to expose some static assets without protection (this is sometime required by SSL certificate provider).
We now need the configuration files our front. This configuration is mostly forked and customized from HTML5's boilerplate for NGinx servers. I will not explained all of them, simply the interesting parts that Meteor and our multi hosts configuration require. Our entry points is docker/nginx/nginx.conf.
# Run as a less privileged user for security reasons.
user www www;
# How many worker threads to run;
# The maximum number of connections for Nginx is calculated by:
# max_clients = worker_processes * worker_connections
worker_processes 1;
# Maximum open file descriptors per process;
# should be > worker_connections.
worker_rlimit_nofile 8192;
events {
  # When you need > 8000 * cpu_cores connections, you start optimizing your OS,
  # and this is probably the point at which you hire people who are smarter than
  # you, as this is *a lot* of requests.
  worker_connections 8000;
}
# Default error log file
# (this is only used when you don't override error_log on a server{} level)
error_log  /var/log/nginx/error.log warn;
pid        /var/run/nginx.pid;
# Main configuration
http {
  # Hide nginx version information.
  server_tokens off;
  # Proxy cache definition
  proxy_cache_path /var/cache/nginx levels=1:2 keys_zone=one:8m max_size=3000m inactive=600m;
  proxy_temp_path /var/tmp;
  # Define the MIME types for files.
  include conf/mimetypes.conf;
  default_type application/octet-stream;
  # Update charset_types due to updated mime.types
  charset_types text/xml text/plain text/vnd.wap.wml application/x-javascript application/rss+xml text/css application/javascript application/json;
  # Format to use in log files
  log_format  main  '$remote_addr - $remote_user [$time_local] "$request" '
                    '$status $body_bytes_sent "$http_referer" '
                    '"$http_user_agent" "$http_x_forwarded_for"';
  # Default log file
  # (this is only used when you don't override access_log on a server{} level)
  access_log /var/log/nginx/access.log main;
  # How long to allow each connection to stay idle; longer values are better
  # for each individual client, particularly for SSL, but means that worker
  # connections are tied up longer. (Default: 65)
  keepalive_timeout 20;
  # Speed up file transfers by using sendfile() to copy directly
  # between descriptors rather than using read()/write().
  sendfile        on;
  # Tell Nginx not to send out partial frames; this increases throughput
  # since TCP frames are filled up before being sent out. (adds TCP_CORK)
  tcp_nopush      on;
  # GZip Compression
  include conf/gzip.conf;
  # Error pages redirections
  error_page 404 /404.html;
  error_page 500 502 503 504  /50x.html;
  # HTTP server
  server {
    # Server name
    include conf/servername.conf;
    # Protocol HTTP
    listen [::]:80 ipv6only=on;
    listen 80;
    # Static files with fallback to HTTPS redirect
    include conf/staticfile-with-fallback.conf;
    # Redirect non-SSL to SSL
    location @fallback {
      rewrite  ^ https://$server_name$request_uri? permanent;
    }
  }
  # Upstream server for the web application server and load balancing
  include conf/upstream-server-and-load-balancing.conf;
  # Upgrade proxy web-socket connections
  include conf/websocket-upgrade.conf;
  # HTTPS server
  server {
    # Server name
    include conf/servername.conf;
    # Protocols HTTPS, SSL, SPDY
    listen [::]:443 ipv6only=on ssl spdy;
    listen 443 ssl spdy;
    # SSL configuration
    include conf/ssl.conf;
    # SPDY configuration
    include conf/spdy.conf;
    # Static files with fallback to proxy server
    include conf/staticfile-with-fallback.conf;
    # Proxy pass to server node with websocket upgrade
    location @fallback {
      include conf/proxy-pass-and-cache.conf;
    }
  }
}
Depending on which host launches NGinx, we need a method to set a proper sever name. For this, we create 3 files:
  • docker/nginx/host-specific/servername-dev.conf: sh # Server name server_name 192.168.1.50;
  • docker/nginx/host-specific/servername-pre.conf: sh # Server name server_name 192.168.1.51;
  • docker/nginx/host-specific/servername-prod.conf: sh # Server name (the real FQDN of your production server) server_name example.org;
For accessing the static files exposed over HTTP, we use simply declare the root of the front and we use a @fallback function in case no file has been found. This is declared in the docker/nginx/staticfile-with-fallback.conf:
# Serve static file and use a fallback otherwise
location / {
  charset utf-8;
  root /www;
  # Basic rules
  include conf/basic.conf;
  # Try static files and redirect otherwise
  try_files $uri @fallback;
  # Expiration rules
  include conf/expires.conf;
}
In our HTTP part of our main configuration, you can see that the trafic is redirected to HTTPS via URL rewriting technic. Our SSL configuration docker/nginx/conf/ssl.conf uses the exposed Docker Volume /etc/certs:
# SSL configuration
ssl on;
# SSL key paths
ssl_certificate /etc/certs/server.crt;
ssl_certificate_key /etc/certs/server.key;
# Trusted cert must be made up of your intermediate certificate followed by root certificate
# ssl_trusted_certificate /path/to/ca.crt;
# Optimize SSL by caching session parameters for 10 minutes. This cuts down on the number of expensive SSL handshakes.
# The handshake is the most CPU-intensive operation, and by default it is re-negotiated on every new/parallel connection.
# By enabling a cache (of type "shared between all Nginx workers"), we tell the client to re-use the already negotiated state.
# Further optimization can be achieved by raising keepalive_timeout, but that shouldn't be done unless you serve primarily HTTPS.
ssl_session_cache shared:SSL:10m; # a 1mb cache can hold about 4000 sessions, so we can hold 40000 sessions
ssl_session_timeout 1m;
# Use a higher keepalive timeout to reduce the need for repeated handshakes
keepalive_timeout 300; # up from 75 secs default
# Protect against the BEAST and POODLE attacks by not using SSLv3 at all. If you need to support older browsers (IE6) you may need to add
# SSLv3 to the list of protocols below.
ssl_protocols TLSv1 TLSv1.1 TLSv1.2;
# Ciphers set to best allow protection from Beast, while providing forwarding secrecy, as defined by Mozilla (Intermediate Set)
# - https://wiki.mozilla.org/Security/Server_Side_TLS#Nginx
ssl_ciphers ECDH+AESGCM:ECDH+AES256:ECDH+AES128:DES-CBC3-SHA:!ADH:!AECDH:!MD5;
ssl_prefer_server_ciphers on;
ssl_session_cache shared:SSL:10m;
# OCSP stapling...
ssl_stapling on;
ssl_stapling_verify on;
# DNS resolution on Google's DNS and DynDNS
resolver 8.8.8.8 8.8.4.4 216.146.35.35 216.146.36.36 valid=60s;
resolver_timeout 2s;
# HSTS (HTTP Strict Transport Security)
# This header tells browsers to cache the certificate for a year and to connect exclusively via HTTPS.
add_header Strict-Transport-Security "max-age=31536000;";
# This version tells browsers to treat all subdomains the same as this site and to load exclusively over HTTPS
#add_header Strict-Transport-Security "max-age=31536000; includeSubdomains;";
add_header X-Frame-Options DENY;
We have also added SPDY to our HTTPS configuration in the docker/nginx/conf/spdy.conf:
# SPDY configuration
add_header Alternate-Protocol  443:npn-spdy/3;
# Adjust connection keepalive for SPDY clients:
spdy_keepalive_timeout 300; # up from 180 secs default
# enable SPDY header compression
spdy_headers_comp 9;
HTTP/2 support is on its way. When integrated to NGinx, this configuration will be upgraded for taking advantage of it.
No that SSL and SPDY are set, we can serve the static file exposed via HTTPS with the same configuration as before for HTTP. But this time, the fallback mecanism redirect the trafic to our Meteor application (our server container). If no static file is found, the trafic is send to our Meteor application using a proxy with cache:
proxy_http_version 1.1;
proxy_pass http://server;
proxy_headers_hash_max_size 1024;
proxy_headers_hash_bucket_size 128;
proxy_redirect off;
# Upgrade proxy web-socket connections
proxy_set_header Upgrade $http_upgrade; # allow websockets
proxy_set_header Connection $connection_upgrade;
proxy_set_header X-Forward-Proto http;
proxy_set_header Host $http_host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forward-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forward-Proto http;
proxy_set_header X-Nginx-Proxy true;
proxy_cache one;
proxy_cache_key prj$request_uri$scheme;
proxy_cache_bypass $http_upgrade;
# Expiration rules
if ($uri != '/') {
  expires 30d;
}
Our proxy cache needs to upgrade the HTTPS connections to WSS. This is achieved in our docker/nginx/conf/upstream-server-and-load-balancing.conf:
# Upstream server for the web application server
upstream server {
  # server is included in each dynamic /etc/hosts by Docker
  server server:3000;
  # Load balancing could be done here, if required.
}
For directing our NGinx on the appropriate configuration, we use an simple environment variables HOST_TARGET that can be dev, pre or prod and a script docker/nginx/startNginx.sh for using this variable:
#!/bin/bash
if [ ! -f /etc/nginx/conf/servername.conf ]
then
  ln -s /etc/nginx/conf/servername-$HOST_TARGET.conf /etc/nginx/conf/servername.conf
fi
nginx -g "daemon off;"
Like before for the other containers, we build it and launch it with:
docker-compose up -d
You should now have a full development host.

Application logging

When launching, stopping, refreshing our services, Docker produces a log for each container that you can easily access in your CLI:
docker-compose logs
# Or only for the db
docker-compose logs db
# Or only for the server
docker-compose logs server
# Or only for the server and the front...
docker-compose logs server front
# ...
As you can see it, it can start to be a bit verbose. Still, you can inspect any Docker container log with a tail like this:
$ docker ps
CONTAINER ID        IMAGE               COMMAND                  CREATED             STATUS              PORTS                                      NAMES
82a7489e41a0        docker_front        "/startNginx.sh"         4 hours ago         Up 4 hours          0.0.0.0:80->80/tcp, 0.0.0.0:443->443/tcp   docker_front_1
4b0656669213        docker_server       "./startMeteor.sh"       27 hours ago        Up 4 hours          3000/tcp                                   docker_server_1
fe6a7238328a        docker_db           "mongod -f /etc/mongo"   45 hours ago        Up 4 hours          27017/tcp                                  docker_db_1
1a878c646094        registry:2          "/bin/registry /etc/d"   46 hours ago        Up 46 hours         0.0.0.0:5000->5000/tcp                     docker_registry_1

$ docker logs --tail 4 -f docker_db_1
2015-09-03T12:20:49.298+0000 I NETWORK  [initandlisten] connection accepted from 172.17.0.64:49051 #22 (18 connections now open)
2015-09-03T12:20:49.314+0000 I NETWORK  [initandlisten] connection accepted from 172.17.0.64:49052 #23 (19 connections now open)
2015-09-03T12:20:49.315+0000 I NETWORK  [initandlisten] connection accepted from 172.17.0.64:49053 #24 (20 connections now open)
2015-09-03T16:36:13.666+0000 I QUERY    [conn10] g...
Docker logs are not regular /var/log entries. They are specific to each ones of your container. There's an important risk to fill up your disk pretty fast depending on you log usages. Fortunately, since Docker 1.8, specific log driver can be added to our running container. We are using logrotate here but you could setup a specific server for an ELK stack or any other of your favorite log solution. For configuring our logrotate on each host, add a new configuration for Docker:
ssh root@$HOST_IP_DEV "echo -e '/var/lib/docker/containers/*/*.log {  \n  rotate 7\n  daily\n  compress\n  size=1M\n  missingok\n  delaycompress\n  copytruncate\n}' > /etc/logrotate.d/docker"
ssh root@$HOST_IP_PRE "echo -e '/var/lib/docker/containers/*/*.log {  \n  rotate 7\n  daily\n  compress\n  size=1M\n  missingok\n  delaycompress\n  copytruncate\n}' > /etc/logrotate.d/docker"
ssh root@$HOST_IP_PROD "echo -e '/var/lib/docker/containers/*/*.log {  \n  rotate 7\n  daily\n  compress\n  size=1M\n  missingok\n  delaycompress\n  copytruncate\n}' > /etc/logrotate.d/docker"
Now, we are updating our docker/docker-compose.yml and set our Docker containers to use the json-file log driver so that it doesn't stay buried in /var/lib/docker/containers/[CONTAINER ID]/[CONTAINER_ID]-json.log:
# Persistence layer: Mongo
db:
  build: mongo
  log_driver: "json-file"
  volumes:
    - /var/db:/db
  expose:
    - "27017"
# Application server: NodeJS (Meteor)
server:
  build: meteor
  log_driver: "json-file"
  environment:
    MONGO_URL: "mongodb://db:27017"
    MONGO_OPLOG_URL: "mongodb://db:27017/local"
    PORT: 3000
    ROOT_URL: "https://192.168.1.50"
  volumes:
    - /etc/meteor:/etc/meteor
  expose:
    - "3000"
# Front layer, static file, SSL, proxy cache: NGinx
front:
  build: nginx
  log_driver: "json-file"
  links:
    - server
  environment:
    # Can be: dev, pre, prod
    HOST_TARGET: "dev"
  volumes:
    - /etc/certs:/etc/certs
    - /var/cache:/var/cache
    - /var/tmp:/var/tmp
  ports:
    - "80:80"
    - "443:443"
  log_driver: "json-file"
For taking this new logging configuration, just issue the following commands:
# This stops the current running containers
docker-compose stop
# This rebuild all images
docker-compose build
# This starts all containers
docker-compose up -d

Push to your local registry

When your are satisfied with the development of your container, you can save your Docker images into your local registry for deploying them to preproduction.
For Mongo:
docker tag -f docker_db $HOST_IP_DEV:5000/mongo:v1.0.0
docker push $HOST_IP_DEV:5000/mongo:v1.0.0
docker tag -f docker_db $HOST_IP_DEV:5000/mongo:latest
docker push $HOST_IP_DEV:5000/mongo:latest
For Meteor:
docker tag -f docker_server $HOST_IP_DEV:5000/meteor:v1.0.0
docker push $HOST_IP_DEV:5000/meteor:v1.0.0
docker tag -f docker_server $HOST_IP_DEV:5000/meteor:latest
docker push $HOST_IP_DEV:5000/meteor:latest
For NGinx:
docker tag -f docker_front $HOST_IP_DEV:5000/nginx:v1.0.0
docker push $HOST_IP_DEV:5000/nginx:v1.0.0
docker tag -f docker_front $HOST_IP_DEV:5000/nginx:latest
docker push $HOST_IP_DEV:5000/nginx:latest

Deployment in pre-production

For deploying to production, we are going to refactor our docker/docker-compose.yml a bit to avoid repetition on Docker Compose file depending on the host that you're tageting.
We create a docker/common.yml file which centralized value used for all hosts:
# Persistence layer: Mongo
db:
  build: mongo
  log_driver: "json-file"
  volumes:
    - /var/db:/db
  expose:
    - "27017"
# Application server: NodeJS (Meteor)
server:
  build: meteor
  log_driver: "json-file"
  environment:
    MONGO_URL: "mongodb://db:27017"
    MONGO_OPLOG_URL: "mongodb://db:27017/local"
    PORT: 3000
  volumes:
    - /etc/meteor:/etc/meteor
  expose:
    - "3000"
# Front layer, static file, SSL, proxy cache: NGinx
front:
  log_driver: "json-file"
  build: nginx
  volumes:
    - /etc/certs:/etc/certs
    - /var/cache:/var/cache
    - /var/tmp:/var/tmp
  ports:
    - "80:80"
    - "443:443"
Now, we can refactor our docker/docker-compose.yml to only set the remaining Docker command required for development:
# Persistence layer: Mongo
db:
  extends:
    file: common.yml
    service: db
# Application server: NodeJS (Meteor)
server:
  extends:
    file: common.yml
    service: server
  links:
    - db
  environment:
    ROOT_URL: "https://192.168.1.50"
# Front layer, static file, SSL, proxy cache: NGinx
front:
  extends:
    file: common.yml
    service: front
  links:
    - server
  environment:
    # Can be: dev, pre, prod
    HOST_TARGET: "dev"
Now for easing the deployment on the pre-production hosts, we are using our common configuration in a docker/deploy-pre.yml file that ease the pull and launch of your services:
# Persistence layer: Mongo
db:
  image: 192.168.1.50:5000/mongo:v1.0.0
  extends:
    file: common.yml
    service: db
  restart: always
# Application server: NodeJS (Meteor)
server:
  image: 192.168.1.50:5000/meteor:v1.0.0
  extends:
    file: common.yml
    service: server
  links:
    - db
  environment:
    ROOT_URL: "https://192.168.1.51"
  restart: always
# Front layer, static file, SSL, proxy cache: NGinx
front:
  image: 192.168.1.50:5000/nginx:v1.0.0
  extends:
    file: common.yml
    service: front
  links:
    - server
  environment:
    # Can be: dev, pre, prod
    HOST_TARGET: "pre"
  restart: always
Connect Docker Machine to your pre-production host, start your services and ensure that your ReplicationSet creation is applied:
eval "$(docker-machine env pre)"
docker-compose -f deploy-pre.yml up -d
docker-compose -f deploy-pre.yml run --rm db mongo db:27017/admin --quiet --eval "rs.initiate(); rs.conf();"
Once you are satisfied with you containers, it's time to make them available to your production server.

Push to Docker Hub

Now we go back on our development host for publishing these container on the public Docker Hub:
eval "$(docker-machine env dev)"
And we publish our containers for Mongo:
docker tag -f docker_db YOUR_DOCKER_HUB_LOGIN/mongo:v1.0.0
docker push YOUR_DOCKER_HUB_LOGIN/mongo:v1.0.0
docker tag -f docker_db YOUR_DOCKER_HUB_LOGIN/mongo:latest
docker push YOUR_DOCKER_HUB_LOGIN/mongo:latest
For Meteor:
docker tag -f docker_server YOUR_DOCKER_HUB_LOGIN/meteor:v1.0.0
docker push YOUR_DOCKER_HUB_LOGIN/meteor:v1.0.0
docker tag -f docker_server YOUR_DOCKER_HUB_LOGIN/meteor:latest
docker push YOUR_DOCKER_HUB_LOGIN/meteor:latest
For NGinx:
docker tag -f docker_front YOUR_DOCKER_HUB_LOGIN/nginx:v1.0.0
docker push YOUR_DOCKER_HUB_LOGIN/nginx:v1.0.0
docker tag -f docker_front YOUR_DOCKER_HUB_LOGIN/nginx:latest
docker push YOUR_DOCKER_HUB_LOGIN/nginx:latest

Deployment in production

Like the deployment in pre-production, we are leveraging the capabilities of Docker Compose for easing the pulling and running of Docker containers. For this, we create a docker/deploy-prod.yml file:
# Persistence layer: Mongo
db:
  image: YOUR_DOCKER_HUB_LOGIN/mongo:v1.0.0
  extends:
    file: common.yml
    service: db
  restart: always
# Application server: NodeJS (Meteor)
server:
  image: YOUR_DOCKER_HUB_LOGIN/meteor:v1.0.0
  extends:
    file: common.yml
    service: server
  links:
    - db
  environment:
    ROOT_URL: "https://YOUR_SITE_FQDN"
  restart: always
# Front layer, static file, SSL, proxy cache: NGinx
front:
  image: YOUR_DOCKER_HUB_LOGIN/nginx:v1.0.0
  extends:
    file: common.yml
    service: front
  links:
    - server
  environment:
    # Can be: dev, pre, prod
    HOST_TARGET: "prod"
  restart: always
Before running everything in production, we must pull our images. Behind the scene so that our users doesn't notice the changes, then we will stop our current running containers, launch our new ones and finish by a ReplicaSet configuration.
eval "$(docker-machine env prod)"
docker-compose -f deploy-prod.yml pull
docker stop "$(docker ps -a -q)"
docker-compose -f deploy-prod.yml up -d
docker-compose -f deploy-prod.yml run --rm db mongo db:27017/admin --quiet --eval "rs.initiate(); rs.conf();"

Links

Sources for this tutorial:
Informations used for this tutorial:
Going further: