Table of Contents
The problem #
Sunday afternoon I spend playing with docker backups and exporting them locally. My problem was that I had remote machine with running containers in Docker. Containers had docker volumes which weren’t mapped directly to local filesystem, simply:
Because it’s not mapped on local filesystem this gives freedom to operate with data inside these volumes.
Now we need those data inside container
/src/models/data and since we are lazy we don’t want to log to machine copy them to machine filesystem and then copy them to local machine.
Create docker context #
First we can create local docker context:
docker context create machine01 --docker "host=tcp://192.168.28.100:2375"
This can help us to connect docker containers and volumes while staying on local machine. Bear in mind to use TCP connection with port 2375 you need to set docker API first. How setup this take a look here.
How to copy files to local machine #
With the help of context we can list volumes like (as well as access to containers*):
docker volume ls
Now we can access to containers, volumes on the remote host via docker api. All we need now is to smuggle data from volume to our local machine. Here is the script I created to solve this problem:
Is does that lists volumes. Using tool fzf (a great tool I strongly recommend to use) have the ability to scroll these listed volume. After you choose volume it’s mounted inside
busybox container. Script is going to compress whole directory and at last download it via STDOUT to your local directory.
Which compression algorithm is best #
When I worked on the last part of the script, I had two problems. First I need to do it effectively since we are moving whole directory. We can pack whole directory into single file using
tar. The second problem is that we need to choose compression algorithm.
Using the gzip format file size was
420MB instead of
And the time was:
time bash backup-docker-volume.sh
tar: removing leading '/' from member names
Executed in 57.13 secs fish external
usr time 0.74 secs 0.11 millis 0.74 secs
sys time 4.62 secs 1.39 millis 4.62 secs
zstd compression, the file size was
470MB instead of
And the time was:
tar: Removing leading `/' from member names
Executed in 61.08 secs fish external
usr time 0.43 secs 0.13 millis 0.43 secs
sys time 2.64 secs 1.88 millis 2.64 secs
Final try was the xz compress algorithm. Although I was unable to use it on either of those images (busybox and custom image). There might have the biggest compression from those three, but at the time of writing I could not find any minimal docker image containing this compression algorithm.
Final thoughs #
I was disappointed with
zstd. First in my test it was slower than
gzip and bigger. Meaning there is not upper side to choose this compress algorithm. Plus
gzip can be used with the
busybox image. No need any special unstrusty image (though you could build one your self and pushit to registry). Very light simple image for this use case.