Persistent volumes with Docker - Data-only container pattern16 Dec 2013
Docker has an option to allow specific folders in a container to be mapped to the normal filesystem on the host. This allows us to have data in the container without making the data part of the Docker image, and without being bound to AUFS.
There are a couple of issues with using volumes in certain scenarios:
- Volumes are container specific, when you create a new container, even with the same image, you do not have access to the data.
- Since image layers are built using containers, data saved to folders which have been specified as a volume does not make it to the next layer, or your final container
If you want data to persist between updated containers you have to manually map data to the host outside the normal, container-specific mount points.
For example, when creating a container:
docker run -d -v /var/container_data/mysql:/var/lib/mysql me/awesome_mysql_image
This, however, is generally not a good idea as you are tying the container to that host and you lose one of the things that makes Docker great: portability.
In addition to this, you've now created a container storage location that is not under Docker's control.
But... data must be able to be persisted, especially in the use case above, so new versions of an image can be used to replace the current container... so we take the risk and do it anyway.
Introducing: Data-only containers
Volumes are still great! We can still use them and use them as intended by our Docker overlords!
Instead of manually setting these mount points on the docker host, let's take the concept of SRP (Single Responsibility Principle) a bit farther.
We can create a container which is solely for storing data for another container:
Create the data-only container:
# docker run -d -v /var/lib/mysql -name data-mysql --enterypoint /bin/echo mysql data-only container for mysql
Great, so now we have a container which has a volume for /var/lib/mysql... now what.
Docker allows us to pull in volumes from another container to use in our own... using the above mysql example this would look like:
docker run -d -volumes-from data-mysql -e MYSQL_ROOT_PASS="muchsecurity" mysql
Here, all data being saved by mysql will be stored in the volume specified by
data-mysql container likely won't ever need to be updated, and if it
does we can easily handle moving the data around as needed, we essentially
work-around the issues listed above and we still have good portability.
We can now create as many mysql instances as we can handle and use volumes from
data-mysql style containers as we want as well (provided unique naming
or use of container ID's). This can much more easily be scripted than mounting
folders ourselves since we are letting docker do the heavy lifting.
One thing that's really cool is that these data-only containers don't even nee to be running, it just needs to exist.
This pattern definitely does not fit all use cases, but it may fit yours! Try it out!