[docker] How to edit Docker container files from the host?

Now that I found a way to expose host files to the container (-v option) I would like to do kind of the opposite:

How can I edit files from a running container with a host editor?

sshfs could probably do the job but since a running container is already some kind of host directory I wonder if there is a portable (between aufs, btrfs and device mapper) way to do that?

This question is related to docker

The answer is


The best way is:

  $ docker cp CONTAINER:FILEPATH LOCALFILEPATH
  $ vi LOCALFILEPATH
  $ docker cp LOCALFILEPATH CONTAINER:FILEPATH

Limitations with $ docker exec: it can only attach to a running container.

Limitations with $ docker run: it will create a new container.


The following worked for me

docker run -it IMAGE_NAME /bin/bash

eg. my image was called ipython/notebook

docker run -it ipython/notebook /bin/bash


Here's the script I use:

#!/bin/bash
IFS=$'\n\t'
set -euox pipefail


CNAME="$1"
FILE_PATH="$2"

TMPFILE="$(mktemp)"
docker exec "$CNAME" cat "$FILE_PATH" > "$TMPFILE"
$EDITOR "$TMPFILE"
cat "$TMPFILE" | docker exec -i "$CNAME" sh -c 'cat > '"$FILE_PATH"
rm "$TMPFILE"

and the gist for when I fix it but forget to update this answer: https://gist.github.com/dmohs/b50ea4302b62ebfc4f308a20d3de4213


docker run -it -name YOUR_NAME IMAGE_ID /bin/bash

$>vi path_to_file

If you think your volume is a "network drive", it will be easier. To edit the file located in this drive, you just need to turn on another machine and connect to this network drive, then edit the file like normal.

How to do that purely with docker (without FTP/SSH ...)?

  1. Run a container that has an editor (VI, Emacs). Search Docker hub for "alpine vim"

Example:

docker run -d --name shared_vim_editor \
 -v <your_volume>:/home/developer/workspace \
jare/vim-bundle:latest
  1. Run the interactive command:

docker exec -it -u root shared_vim_editor /bin/bash

Hope this helps.


We can use another way to edit files inside working containers (this won't work if container is stoped).

Logic is to:
-)copy file from container to host
-)edit file on host using its host editor
-)copy file back to container

We can do all this steps manualy, but i have written simple bash script to make this easy by one call.

/bin/dmcedit:

#!/bin/sh
set -e

CONTAINER=$1
FILEPATH=$2
BASE=$(basename $FILEPATH)
DIR=$(dirname $FILEPATH)
TMPDIR=/tmp/m_docker_$(date +%s)/

mkdir $TMPDIR
cd $TMPDIR
docker cp $CONTAINER:$FILEPATH ./$DIR
mcedit ./$FILEPATH
docker cp ./$FILEPATH $CONTAINER:$FILEPATH
rm -rf $TMPDIR

echo 'END'
exit 1;

Usage example:

dmcedit CONTAINERNAME /path/to/file/in/container

The script is very easy, but it's working fine for me.

Any suggestions are appreciated.


I use sftp plugin from my IDE.

  1. Install ssh server for your container and allow root access.
  2. Run your docker container with -p localport:22
  3. Install from your IDE a sftp plugin

Example using sublime sftp plugin: https://www.youtube.com/watch?v=HMfjt_YMru0


The way I am doing is using Emacs with docker package installed. I would recommend Spacemacs version of Emacs. I would follow the following steps:

1) Install Emacs (Instruction) and install Spacemacs (Instruction)

2) Add docker in your .spacemacs file

3) Start Emacs

4) Find file (SPC+f+f) and type /docker:<container-id>:/<path of dir/file in the container>

5) Now your emacs will use the container environment to edit the files


There are two ways to mount files into your container. It looks like you want a bind mount.

Bind Mounts

This mounts local files directly into the container's filesystem. The containerside path and the hostside path both point to the same file. Edits made from either side will show up on both sides.

  • mount the file:
? echo foo > ./foo
? docker run --mount type=bind,source=$(pwd)/foo,target=/foo -it debian:latest
# cat /foo
foo # local file shows up in container
  • in a separate shell, edit the file:
? echo 'bar' > ./foo # make a hostside change
  • back in the container:
# cat /foo
bar # the hostside change shows up
# echo baz > /foo # make a containerside change
# exit
 
? cat foo
baz # the containerside change shows up

Volume Mounts

  • mount the volume
? docker run --mount type=volume,source=foovolume,target=/foo  -it debian:latest
root@containerB# echo 'this is in a volume' > /foo/data
  • the local filesystem is unchanged
  • docker sees a new volume:
? docker volume ls
DRIVER    VOLUME NAME
local     foovolume
  • create a new container with the same volume
? docker run --mount type=volume,source=foovolume,target=/foo  -it debian:latest
root@containerC:/# cat /foo/data
this is in a volume # data is still available

syntax: -v vs --mount

These do the same thing. -v is more concise, --mount is more explicit.

bind mounts

-v /hostside/path:/containerside/path
--mount type=bind,source=/hostside/path,target=/containerside/path

volume mounts

-v /containerside/path
-v volumename:/containerside/path
--mount type=volume,source=volumename,target=/containerside/path

(If a volume name is not specified, a random one is chosen.)

The documentaion tries to convince you to use one thing in favor of another instead of just telling you how it works, which is confusing.