In previous posts we started talking about using Cinder as a Block Storage provider for consumers other than Nova. In particular our first example has been how to use Cinder by itself for Docker. You can catch the first part of that here Cinder providing block storage for more than just Nova
So there we walked through how to install the Cinder Docker Driver and configure things in a simple stand-alone setup. In this post we’re going to follow up on a few things and consume Cinder resources from a Swarm Cluster.
If you’ve never built a Swarm cluster, it’s super easy using Docker 1.12. Even better, if you have an OpenStack Cloud internally, you can check out my post How to Swarm on OpenStack, or watch the Screen Cast and try this out for yourself in about 10 minutes or so.
A little Background
We’re going to start from the poing where we already have a Swarm Cluster set up and ready to use. For simplicity I’m going to use the Swarm deployed on my OpenStack Cloud. That means I’m using Nova Instances as my Swarm Nodes, BUT Nova doesn’t know anything about Cinder in this case.
The whole idea here is how to use Cinder just as Cinder by itself. This all works the same whether you’re Swarm is on
- Bare Metal
Whoa, hold up!!
Did I say use Cinder to provide storage for my Containers running in vSphere???
YES I DID!! and YES YOU CAN!
The whole point of this post is that we only really need Cinder.
Installing/Configuring the cinder-docker-driver on each Docker Node
The following is much the same as we did in part 1 of our post, we’ll go through the same steps, but we’re going to do everything through docker-machine. You could certainly do this with your own bash, python or ansible or whatever, but docker-machine and the command line in my case works really well.
Firs we’ll use curl to fetch and run the install script for the Plugin
$ for each in $(docker-machine ls -q); do; docker-machine ssh $each "curl -sSl https://raw.githubusercontent.com/j-griffith/cinder-docker-driver/master/install.sh | sudo sh -"; done
Remember, we’re manually creating conf files because of the HostUUID issue. Since we’re using OpenStack Instances
here, might as well use their UUID’s for this. I’ve created the files I need,
they’re all the same with the exception of the
$ for each in 1 2 3 4; do; docker-machine scp cdd.config.swarm-$each.json swarm-$each:~/config.json; done
Now, just move the files into the default cinder dockerdriver directory using sudo…
$ for each in $(docker-machine ls -q); do; docker-machine ssh $each "sudo cp ~/config.json /var/lib/cinder/dockerdriver/config.json"; done
Start the plugin daemon on each Node
$ for each in $(docker-machine ls -q); do; docker-machine ssh $each "sudo service cinder-docker-driver start"; done
Verify the services all started up ok…
$ for each in $(docker-machine ls -q); do; docker-machine ssh $each "sudo service cinder-docker-driver status"; done ● cinder-docker-driver.service - "Cinder Docker Plugin daemon" Loaded: loaded (/etc/systemd/system/cinder-docker-driver.service; enabled; vendor preset: enabled) Active: active (running) since Tue 2016-09-13 23:12:52 UTC; 8s ago Main PID: 26875 (cinder-docker-d) Tasks: 5 Memory: 1.2M CPU: 11ms CGroup: /system.slice/cinder-docker-driver.service └─26875 /usr/bin/cinder-docker-driver & Sep 13 23:12:52 swarm-1 systemd: Started "Cinder Docker Plugin daemon". Sep 13 23:12:52 swarm-1 cinder-docker-driver: time="2016-09-13T23:12:52Z" level=info msg="Using config file: /var/lib/cinder/dockerdriver/config.json" Sep 13 23:12:52 swarm-1 cinder-docker-driver: time="2016-09-13T23:12:52Z" level=info msg="Set InitiatorIFace to: default" Sep 13 23:12:52 swarm-1 cinder-docker-driver: time="2016-09-13T23:12:52Z" level=info msg="Set node InitiatorIP to: " Sep 13 23:12:52 swarm-1 cinder-docker-driver: time="2016-09-13T23:12:52Z" level=info msg="Set DefaultVolSz to: 1 GiB" Sep 13 23:12:52 swarm-1 cinder-docker-driver: time="2016-09-13T23:12:52Z" level=info msg="Set Endpoint to: http://172.16.140.243:5000/v2.0" Sep 13 23:12:52 swarm-1 cinder-docker-driver: time="2016-09-13T23:12:52Z" level=info msg="Set Username to: jdg" Sep 13 23:12:52 swarm-1 cinder-docker-driver: time="2016-09-13T23:12:52Z" level=info msg="Set TenantID to: 3dce5dd10b414ac1b942aba8ce8558e7" ......
Easy enough… we’ve now succesfully added/configured the Cinder Plugin to each of our Swarm nodes. We’re ready to deploy Swarm services and consume our Cinder storage.
Our first Swarm Service that Consumes Cinder
The first thing we’ll do now is set our env back to our Swarm Manager Node, and create a Swarm network to use. I just use the basic overlay driver, this network is really important. We’ll specify this in our services and it will do things like enable the containers to communicate across nodes, and also do things like allow us to access a running container from any one of the Swarm Nodes IP addresses (regardless of where the Container is actually running).
$ eval $(docker-machine env swarm-1) $ docker network create --driver overlay swarm-net 93huy770ajjyvk87hc3kmdabr $ docker network list NETWORK ID NAME DRIVER SCOPE 4bc75c8415f5 bridge bridge local a4b3019e592a docker_gwbridge bridge local 4fce4f1ddbe0 host host local 0by3ei97kci5 ingress overlay swarm afafb56eb377 none null local 93huy770ajjy swarm-net overlay swarm
So you’ll notice we did a ```docker network list`` there, we want to make sure we have a network we can use with the SCOPE==swarm.
At this point we’re going to build an “app”, albeit a very simple app. Our app is going to consist of: 1. A redis service (that persists data to a Cinder Volume) 2. A web service that takes input via a simple web-page and stores the info to our Redis service
The app is a modified version of the Moby-Counter app you’ve probably seen. One thing about Docker is that “counting” apps seem to be THE DEFACTO example, so why break with tradition here.
Start our Redis Service
docker service create --name redis --network swarm-net --mount type=volume,src=redis-1,dst=/data,volume-driver=cinder -p 6379:6379 redis
So that’s pretty simple, the syntax for service as it relates to Volumes is a little different, but it makes sense once you kinda walk through it.
We want to “mount” something to the container when it starts up
That thing we want to mount is a “volume”
src=redis-1 Name of the src volume to create, or just use one if
it already exists.
dst=/data That’s where we’re going to mount the src volume in our
volume-driver=cinder Tells the Docker Volume API which Volume
Plugin/Service to use.
That’s all ther is to it. Swarm will figure out where to place the Container (since we’re not giving it any constraints anywhere one of the Swarm nodes is fair game), And it will communicate with the Cinder Plugin Daemon ON the Node that it places the Container.
The Plugin takes care of all the details:
- Check if Volume already exists
- Create it if it doesn’t
- Perform an iSCSI attach to the Swarm Node
- Format the volume (only if it’s a new Volume that it created)
- Mount the volume on the Swarm Node
- Pass the Mount point into the Container.
Now… if that Container “fails” for some reason, it will be restarted on another node. In that case the above steps are repeated on the new Node. So, you may have lost a node, or your container died, but your storage and “important” data are now efficiently portable across your Swarm Nodes!!!
Start up our web server
$ docker service create --name web --network swarm-net -p 80:80 \ jgriffith/jgriffith-webbase
That’s all there is to it, now you should be able to do
docker service ps <service-id> to get details about where
your service is running and it’s status.