tl;dr: If you are deploying docker containers to production, be sure to run
docker system prune -f before pulling new images so you don't run out of disk
When I'm building a new project, I generally learn towards using docker compose during development. When coupled with docker machine I have a quick and easy way to deploy my containers to the cloud provider of my choice. Overall, it works really well, but there's one important thing to consider when using docker for production: running out of disk space.
Images, containers, networks, and volumes will continue to grow on a production
VM which will inevitably lead to the hard drive running out of memory. This
seems obvious, but it wasn't obvious to me when I first started using
I recently deployed a new app listifi to google cloud compute. Here's a brief overview of my deployment process:
- Develop using
- Build features locally
- Use a production tuned
docker-composeyml file to build images locally
- Push the images to Google's Container Registry
- Then I run
eval $(docker-machine <name> env)to tunnel into my production VM's docker
- Then I run
docker-compose -f production.yml pull --ignore-pull-failuresto download the new images
- Then I run
docker-compose -f production.yml up --no-deps -dto restart my containers with the new images
This process works great. I don't need to setup CI for a new project but it still provides me with the flexibility to deploy to my own VM and inspect its health with docker.
Things are working great, I'm iterating on feature development and deploying in between major changes. Only, the last deploy I tried to perform failed. The reason: hard drive is out of space. Hmm, my VM has 16GB of diskspace, why is it out of memory?
When I run
docker system df the problem becomes clear: I have unused images
soaking up all of my hard drive space. I have scoured docker deployment
strategies and never came across documentation that references this issue. There
are plenty of StackOverflow issues referencing the problem with a solution, but
I never really made the connection to my production VM until I hit this problem.
Before I pull my new images in production, I run another command:
1docker system prune -f
You can read the documentation about this command here. Once I added that to my deployment step, things are working much more smoothly.
I completely forgot that as I deployed new changes to production, there was lingering old docker images laying dormant on my production VM, increasing in size of time.
It never clicked for me until I ran into the problem, hopefully this short blog article will help others avoid this problem in the future as well.