Let's say I have pulled the official mysql:5.6.21 image.
I have deployed this image by creating several docker containers.
These containers have been running for some time until MySQL 5.6.22 is released. The official image of mysql:5.6 gets updated with the new release, but my containers still run 5.6.21.
How do I propagate the changes in the image (i.e. upgrade MySQL distro) to all my existing containers? What is the proper Docker way of doing this?
After evaluating the answers and studying the topic I'd like to summarize.
The Docker way to upgrade containers seems to be the following:
Application containers should not store application data. This way you can replace app container with its newer version at any time by executing something like this:
docker pull mysql
docker stop my-mysql-container
docker rm my-mysql-container
docker run --name=my-mysql-container --restart=always \
-e MYSQL_ROOT_PASSWORD=mypwd -v /my/data/dir:/var/lib/mysql -d mysql
You can store data either on host (in directory mounted as volume) or in special data-only container(s). Read more about it
Upgrading applications (eg. with yum/apt-get upgrade) within containers is considered to be an anti-pattern. Application containers are supposed to be immutable, which shall guarantee reproducible behavior. Some official application images (mysql:5.6 in particular) are not even designed to self-update (apt-get upgrade won't work).
I'd like to thank everybody who gave their answers, so we could see all different approaches.
I don't like mounting volumes as a link to a host directory, so I came up with a pattern for upgrading docker containers with entirely docker managed containers. Creating a new docker container with --volumes-from <container> will give the new container with the updated images shared ownership of docker managed volumes.
docker pull mysql
docker create --volumes-from my_mysql_container [...] --name my_mysql_container_tmp mysql
By not immediately removing the original my_mysql_container yet, you have the ability to revert back to the known working container if the upgraded container doesn't have the right data, or fails a sanity test.
At this point, I'll usually run whatever backup scripts I have for the container to give myself a safety net in case something goes wrong
docker stop my_mysql_container
docker start my_mysql_container_tmp
Now you have the opportunity to make sure the data you expect to be in the new container is there and run a sanity check.
docker rm my_mysql_container
docker rename my_mysql_container_tmp my_mysql_container
The docker volumes will stick around so long as any container is using them, so you can delete the original container safely. Once the original container is removed, the new container can assume the namesake of the original to make everything as pretty as it was to begin.
There are two major advantages to using this pattern for upgrading docker containers. Firstly, it eliminates the need to mount volumes to host directories by allowing volumes to be directly transferred to an upgraded containers. Secondly, you are never in a position where there isn't a working docker container; so if the upgrade fails, you can easily revert to how it was working before by spinning up the original docker container again.
Just for providing a more general (not mysql specific) answer...
Synchronize with service image registry (https://docs.docker.com/compose/compose-file/#image):
docker-compose pull
Recreate container if docker-compose file or image have changed:
docker-compose up -d
Container image management is one of the reason for using docker-compose (see https://docs.docker.com/compose/reference/up/)
If there are existing containers for a service, and the service’s configuration or image was changed after the container’s creation, docker-compose up picks up the changes by stopping and recreating the containers (preserving mounted volumes). To prevent Compose from picking up changes, use the --no-recreate flag.
Data management aspect being also covered by docker-compose through mounted external "volumes" (See https://docs.docker.com/compose/compose-file/#volumes) or data container.
This leaves potential backward compatibility and data migration issues untouched, but these are "applicative" issues, not Docker specific, which have to be checked against release notes and tests...
I would like to add that if you want to do this process automatically (download, stop and restart a new container with the same settings as described by @Yaroslav) you can use WatchTower. A program that auto updates your containers when they are changed https://github.com/v2tec/watchtower
Consider for this answers:
app_schema app_dbroot123 This is considered a bad practice, because if you lose the container, you will lose the data. Although it is a bad practice, here is a possible way to do it:
1) Do a database dump as SQL:
docker exec app_db sh -c 'exec mysqldump app_schema -uroot -proot123' > database_dump.sql
2) Update the image:
docker pull mysql:5.6
3) Update the container:
docker rm -f app_db
docker run --name app_db --restart unless-stopped \
-e MYSQL_ROOT_PASSWORD=root123 \
-d mysql:5.6
4) Restore the database dump:
docker exec app_db sh -c 'exec mysql -uroot -proot123' < database_dump.sql
Using an external volume is a better way of managing data, and it makes easier to update MySQL. Loosing the container will not lose any data. You can use docker-compose to facilitate managing multi-container Docker applications in a single host:
1) Create the docker-compose.yml file in order to manage your applications:
version: '2'
services:
app_db:
image: mysql:5.6
restart: unless-stopped
volumes_from: app_db_data
app_db_data:
volumes: /my/data/dir:/var/lib/mysql
2) Update MySQL (from the same folder as the docker-compose.yml file):
docker-compose pull
docker-compose up -d
Note: the last command above will update the MySQL image, recreate and start the container with the new image.
Here's what it looks like using docker-compose when building a custom Dockerfile.
docker build -t imagename:version . This will store your new version locally.docker-compose downdocker-compose.yml file to reflect the new image name you set at step 1.docker-compose up -d. It will look locally for the image and use your upgraded one.-EDIT-
My steps above are more verbose than they need to be. I've optimized my workflow by including the build: . parameter to my docker-compose file. The steps looks this now:
docker-compose builddocker-compose up -dI didn't realize at the time, but docker-compose is smart enough to simply update my container to the new image with the one command, instead of having to bring it down first.
If you do not want to use Docker Compose, I can recommend portainer. It has a recreate function that lets you recreate a container while pulling the latest image.
Taking from http://blog.stefanxo.com/2014/08/update-all-docker-images-at-once/
You can update all your existing images using the following command pipeline:
docker images | awk '/^REPOSITORY|\<none\>/ {next} {print $1}' | xargs -n 1 docker pull
Make sure you are using volumes for all the persistent data (configuration, logs, or application data) which you store on the containers related to the state of the processes inside that container. Update your Dockerfile and rebuild the image with the changes you wanted, and restart the containers with your volumes mounted at their appropriate place.
This is something I've also been struggling with for my own images. I have a server environment from which I create a Docker image. When I update the server, I'd like all users who are running containers based on my Docker image to be able to upgrade to the latest server.
Ideally, I'd prefer to generate a new version of the Docker image and have all containers based on a previous version of that image automagically update to the new image "in place." But this mechanism doesn't seem to exist.
So the next best design I've been able to come up with so far is to provide a way to have the container update itself--similar to how a desktop application checks for updates and then upgrades itself. In my case, this will probably mean crafting a script that involves Git pulls from a well-known tag.
The image/container doesn't actually change, but the "internals" of that container change. You could imagine doing the same with apt-get, yum, or whatever is appropriate for you environment. Along with this, I'd update the myserver:latest image in the registry so any new containers would be based on the latest image.
I'd be interested in hearing whether there is any prior art that addresses this scenario.
Update
This is mainly to query the container not to update as building images is the way to be done
I had the same issue so I created docker-run, a very simple command-line tool that runs inside a docker container to update packages in other running containers.
It uses docker-py to communicate with running docker containers and update packages or run any arbitrary single command
Examples:
docker run --rm -v /var/run/docker.sock:/tmp/docker.sock itech/docker-run exec
by default this will run date command in all running containers and return results but you can issue any command e.g. docker-run exec "uname -a"
To update packages (currently only using apt-get):
docker run --rm -v /var/run/docker.sock:/tmp/docker.sock itech/docker-run update
You can create and alias and use it as a regular command line e.g.
alias docker-run='docker run --rm -v /var/run/docker.sock:/tmp/docker.sock itech/docker-run'
