Note that I have not touched this prototype since October 2014, it might need some adjustments to run in a recent Docker environment. https://docs.docker.com/compose/ now replaces Fig, for example.
Experimental Sling/Oak cluster running on Docker. My main goal so far is to create elastic Sling clusters in a simple way, as a playground for making Sling more operations-friendly.
See also my http://www.slideshare.net/bdelacretaz/modern-operations-with-apache-sling-2014-adaptto-version slides for additional information.
This is using http://www.fig.sh/ to build and start the Docker images. The only real dependency on fig is the fig.yml
definition file, which can easily be translated to Docker commands if you don't want to use fig.
A Vagrant box is provided which should keep setup to the minimum, so you can just get https://www.vagrantup.com/ (I'm currently using 1.6.2) and you should be good, see instructions below.
If you already have a Docker environment with Fig you can start the cluster as follows:
export HOST_IP=(public IP of your Docker host)
cd docker
fig build
fig up
In both cases you can then use fig scale sling=N to start N Sling instances which announce themselves to the HAProxy
front-end via etcd. To check that HAProxy is actually using several Sling instances as its back-end you can use
export HOST_IP=(public IP of your Docker host)
export PORT=80
for i in 1 2 3 4 5 6 7 8 9 0
do
curl -s -u admin:admin http://$HOST_IP:$PORT/system/console/status-slingsettings.json | grep Name
done | sort -u
This should display something like (here with 3 Sling instances up):
"Sling Name = Instance 88461e68-91ef-495f-9d38-0746c48bf9e8",
"Sling Name = Instance d9a10ec2-2d6d-4a22-97ad-42c33c6242da",
"Sling Name = Instance fcba04db-1a00-4c0d-99bb-55d7790ea9c6",
Which shows that you are hitting 3 different Sling instances, each with their own instance ID.
If using the Vagrant box HOST_IP is the host that's running Vagrant, and PORT is 9080 as exposed in the Vagrantfile.
The following ports (81, 82 or 9081, 9082) expose the HAProxy and Graphite status pages, respectively. Graphite is not getting stats from the Sling instances as I write this, for some reason.
Note that this might Download The Web (TM) on first startup, using many fine ways of doing that: downloading a Vagrant image, building Java modules with Maven starting on an empty local repository, and getting the required Docker images which are not optimized so far.
Should you accept doing that, install Vagrant 1.6.2 or later and run this from the folder that contains the Vagrantfile:
vagrant up
vagrant ssh
# You're in a dark and scary Vagrant box now. Inception starts.
sudo bash
cd /dsc/docker
export HOST_IP=$(ifconfig eth0 | grep "inet addr" | cut -d: -f2- | cut -d' ' -f1)
fig build
fig up
You can then open another vagrant ssh session to use fig ps or other fig commands to see details
of what's happening, or use fig scale sling=N to start more Sling instances.
Brutally stopping the mongo Docker box might prevent it from restarting, if that happens see http://docs.mongodb.org/manual/tutorial/recover-data-following-unexpected-shutdown/ for how to fix it.
The Mongo data is stored on the Vagrant box, the following works for me (but it's kinda brute force):
Use find / -name mongod.lock to find that lock file.
Use > full_path_of/mongod.lock to make it zero length.
Restart the Mongo box with fig up mongo (requires restarting the Sling boxes), or restart the whole cluster with
fig up if you stopped everything.