Docker 将一个容器的目录挂载到另一个容器

Question

I have an container Container-A which has directory /opt/test/data which contains files which get created during building image. Now I want to share /opt/test/data(container-a directory) to Container-B as it requires to use those files during its lifecycle. I have tried below option

docker run -it -d -v shareddata:/opt/test/data --name container-a test

and then run container-B as

docker run -it --volumes-from container-a --name container-b test-prod:latest

The data got shared but it creates an named volume shareddata on the host which contains a duplicate data of /opt/test/data directory of container-a.

I have tried another option of creating a name volume data and mounting it in container-a. Then I created a symlink of /opt/test/data on mounted volume as below

ln -s /data/data /opt/test/data

Now mounted the /data volume to container-b but the symlink didn't work on container-b as it shows it as broken links.

Is there a way I can share container-a directory to container-b without duplicating the data as we want to keep data which is static and quite big approx 2.5G to one image and then an instance keep running which is shared across multiple container of test-prod image?

Answer 1

If you are willing and can replace docker with podman , it should be able to do it:

1. podman create says its --mount option supports type=image , so something like

   podman run --mount type=image,source=test,destination=/opt/test/data test-prod:latest

   should work.

2. It also has podman mount that allows you to access the container filesystem from the host, which you can then bind into another container.

If you can't go to podman, the low level ctr command that controls the containerd (that now is underneath docker and also most kubernetes installs) has a ctr image mount command that should work similar to the podman mount one. Unfortunately I couldn't find proper documentation. Also the command may not be installed on your system.

Answer 2

Another option is to access it through the /proc filesystem when sharing a pid namespace.

$ docker run -d --name container-a test sh -c 'echo $$; while :; do sleep 1; done'
$ docker run -it --pid container:container-a --name container-b -e "A_PID=$(docker logs container-a)" test-prod:latest

and then the directory should be available at /proc/$A_PID/root/opt/test/data .

Besides sharing the PID namespace the containers have to be using the same UID for this to work.

Explanation:

• The filesystem view a specific process sees is available under /proc/ pid /root . It can only be accessed by processes with the same UID (the same rules as for ptrace should apply).
• You can make two containers share the PID namespace with the --pid container: container flag, which makes the /proc/ pid accessible from the other container.
• The PID is not necessarily fixed, so the container providing the data must print the PID of a process it is running and then enter an endless loop.

Answer 3

The usual way to do this is to create those files while starting the container, instead while building. You can check first if the files are created, and if not, create them (or move them from a /tmp or wathever)

This is the way official image for mysql do that, for example.

Then, you can mount a volume in that directory, and share it between containers.

Answer 4

Try this, you could share data directly from one container to another without using the host machines directory,

$ docker run -v /opt/test/data --name container-a test
$ docker run --volumes-from container-a test-prod:latest

Hope, this solves your problem. As the data size is larger and also you are sharing this between multiple containers, probably go for better storage orchestration solution like rexray which you could share easily between n number of containers.

Answer 5

If files already exist in the directory of the container you're mounting as a volume, the contents of the directory will be copied to the volume. See Populate a volume using a container .

There's no way to mount a container's directory as a volume without copying. An alternative is to not store the data in a container and instead create a volume on the host and populate it with the data. This way, the data will only exist on once.