Is it possible to start a shell session in a runni

问题:

I was naively expecting this command to run a bash shell in a running container :

docker run "id of running container" /bin/bash

it looks like it's not possible, I get the error :

2013/07/27 20:00:24 Internal server error: 404 trying to fetch remote history for 27d757283842

So, if I want to run bash shell in a running container (ex. for diagnosis purposes)

do I have to run an SSH server in it and loggin via ssh ?

回答1:

EDIT: Now you can use docker exec -it "id of running container" bash (doc)

Previously, the answer to this question was:

If you really must and you are in a debug environment, you can do this: sudo lxc-attach -n <ID> Note that the id needs to be the full one (docker ps -notrunc).

However, I strongly recommend against this.

notice: -notrunc is deprecated, it will be replaced by --no-trunc soon.

回答2:

With docker 1.3, there is a new command docker exec. This allows you to enter a running docker:

docker exec -it "id of running container" bash

回答3:

Just do

docker attach container_name

As mentioned in the comments, to detach from the container without stopping it, type Ctrlpthen Ctrlq.

回答4:

Since things are achanging, at the moment the recommended way of accessing a running container is using nsenter.

You can find more information on this github repository. But in general you can use nsenter like this:

PID=$(docker inspect --format {{.State.Pid}} <container_name_or_ID>)
nsenter --target $PID --mount --uts --ipc --net --pid

or you can use the wrapper docker-enter:

docker-enter <container_name_or_ID>

A nice explanation on the topic can be found on Jérôme Petazzoni's blog entry: Why you don't need to run sshd in your docker containers

回答5:

First thing you cannot run

docker run "existing container" command

Because this command is expecting an image and not a container and it would anyway result in a new container being spawned (so not the one you wanted to look at)

I agree with the fact that with docker we should push ourselves to think in a different way (so you should find ways so that you don't need to log onto the container), but I still find it useful and this is how I work around it.

I run my commands through supervisor in DEAMON mode.

Then I execute what I call docker_loop.sh The content is pretty much this:

#!/bin/bash
/usr/bin/supervisord
/usr/bin/supervisorctl
while ( true )
    do
    echo "Detach with Ctrl-p Ctrl-q. Dropping to shell"
    sleep 1
    /bin/bash
done

What it does is that it allows you to "attach" to the container and be presented with the supervisorctl interface to stop/start/restart and check logs. If that should not suffice, you can Ctrl+D and you will drop into a shell that will allow you to have a peek around as if it was a normal system.

PLEASE DO ALSO TAKE INTO ACCOUNT that this system is not as secure as having the container without a shell, so take all the necessary steps to secure your container.

回答6:

Keep an eye on this pull request: https://github.com/docker/docker/pull/7409

Which implements the forthcoming docker exec <container_id> <command> utility. When this is available it should be possible to e.g. start and stop the ssh service inside a running container.

There is also nsinit to do this: "nsinit provides a handy way to access a shell inside a running container's namespace", but it looks difficult to get running. https://gist.github.com/ubergarm/ed42ebbea293350c30a6

回答7:

You can use

docker exec -it <container_name> bash

回答8:

There is actually a way to have a shell in the container.

Assume your /root/run.sh launches the process, process manager (supervisor), or whatever.

Create /root/runme.sh with some gnu-screen tricks:

# Spawn a screen with two tabs
screen -AdmS 'main' /root/run.sh
screen -S 'main' -X screen bash -l
screen -r 'main'

Now, you have your daemons in tab 0, and an interactive shell in tab 1. docker attach at any time to see what's happening inside the container.

Another advice is to create a "development bundle" image on top of the production image with all the necessary tools, including this screen trick.

回答9:

here is my solution

part of DOckerfile:

...
RUN mkdir -p /opt
ADD initd.sh /opt/
RUN chmod +x /opt/initd.sh
ENTRYPOINT ["/opt/initd.sh"]

part of "initd.sh"

#!/bin/bash
...
/etc/init.d/gearman-job-server start
/etc/init.d/supervisor start
#very important!!!
/bin/bash

after image is built you have two options using exec and attach:

1. with exec (which i use), run:

docker run --name $CONTAINER_NAME -dt $IMAGE_NAME

then

docker exec -it $CONTAINER_NAME /bin/bash

and use

CTRL+D to detach

2. with attach, run:

docker run --name $CONTAINER_NAME -dit $IMAGE_NAME

then

docker attach $CONTAINER_NAME

and use

CTRL+P and CTRL+Q to detach

difference between options is in parameter -i

回答10:

There are two ways.

With attach

$ sudo docker attach 665b4a1e17b6 #by ID

With exec

$ sudo docker exec - -t 665b4a1e17b6 #by ID

回答11:

It's useful assign name when running container. You don't need refer container_id.

docker run --name container_name yourimage docker exec -it container_name bash

回答12:

If the goal is to check on the application's logs, this post shows starting up tomcat and tailing the log as part of CMD. The tomcat log is available on the host using 'docker logs containerid'.

http://blog.trifork.com/2013/08/15/using-docker-to-efficiently-create-multiple-tomcat-instances/

回答13:

Maybe you were mislead like myself into thinking in terms of VMs when developing containers. My advice: Try not to.

Containers are just like any other process. Indeed you might want to "attach" to them for debugging purposes (think of /proc//env or strace -p ) but that's a very special case.

Normally you just "run" the process, so if you want to modify the configuration or read the logs, just create a new container and make sure you write the logs outside of it by sharing directories, writing to stdout (so docker logs works) or something like that.

For debugging purposes you might want to start a shell, then your code, then press CTRL-p + CTRL-q to leave the shell intact. This way you can reattach using:

docker attach <container_id>

If you want to debug the container because it's doing something you haven't expect it to do, try to debug it: https://serverfault.com/questions/596994/how-can-i-debug-a-docker-container-initialization

回答14:

No. This is not possible. Use something like supervisord to get an ssh server if that's needed. Although, I definitely question the need.