Ansible cannot import docker-py even though it is

I checked this post and followed the fix in both answers and neither worked. I'm opening a new post partly because of that and partly because I'm getting a slightly different error even though the problem might be the same.

Ansible host:

$ ansible --version
ansible 2.1.0.0
  config file = /etc/ansible/ansible.cfg
  configured module search path = Default w/o overrides

Destination client myserver:

$ pip list | egrep 'six|docker|websocket_client'
docker-py (1.2.3)
six (1.10.0)

test.yml:

---
- hosts: myserver
  remote_user: root
  tasks:
  - name: stop any running docker registries
    docker_container:
      name: registry
      state: stopped
...

Ansible server (ansible-playbook aliased to ap):

$ ap -vvvv test.yml

The output:

(probably extraneous output, snipped):

fatal: [myserver]: FAILED! => {
    "changed": false,
    "failed": true,
    "invocation": {
        "module_args": {
            "api_version": null,
            "blkio_weight": null,
            "cacert_path": null,
            "capabilities": null,
            "cert_path": null,
            "command": null,
            "cpu_period": null,
            "cpu_quota": null,
            "cpu_shares": null,
            "cpuset_cpus": null,
            "cpuset_mems": null,
            "debug": false,
            "detach": true,
            "devices": null,
            "dns_opts": null,
            "dns_search_domains": null,
            "dns_servers": null,
            "docker_host": null,
            "entrypoint": null,
            "env": null,
            "etc_hosts": null,
            "exposed_ports": null,
            "filter_logger": false,
            "force_kill": false,
            "groups": null,
            "hostname": null,
            "image": null,
            "interactive": false,
            "ipc_mode": null,
            "keep_volumes": true,
            "kernel_memory": null,
            "key_path": null,
            "kill_signal": null,
            "labels": null,
            "links": null,
            "log_driver":
            "json-file",
            "log_options": null,
            "mac_address": null,
            "memory": "0",
            "memory_reservation": null,
            "memory_swap": null,
            "memory_swappiness": null,
            "name": "registry",
            "network_mode": null,
            "networks": null,
            "oom_killer": null,
            "paused": false,
            "pid_mode": null,
            "privileged": false,
            "published_ports": null,
            "pull": false,
            "read_only": false,
            "recreate": false,
            "restart": false,
            "restart_policy": null,
            "restart_retries": 0,
            "security_opts": null,
            "shm_size": null,
            "ssl_version": null,
            "state": "stopped",
            "stop_signal": null,
            "stop_timeout": null,
            "timeout": null,
            "tls": null,
            "tls_hostname": null,
            "tls_verify": null,
            "trust_image_content": false,
            "tty": false,
            "ulimits": null,
            "user": null,
            "uts": null,
            "volume_driver": null,
            "volumes": null,
            "volumes_from": null
        },
        "module_name": "docker_container"
    },
    "msg":

(the pertinent error):

"Failed to import docker-py - cannot import name NotFound. Try pip install docker-py"}

I get the same error when I downgrade the docker-py module to 1.1.0 as per the first answer in the referenced post. I also tried to chmod the directories and it made no difference:

(/usr/lib/python2.7/site-packages) myserver$ ls -lad docker*
drwxr-xr-x. 6 root root 4096 Jul  4 10:57 docker/
drwxr-xr-x. 2 root root 4096 Jul  4 10:57 docker_py-1.2.3-py2.7.egg-info/

from chmod -R go+rx docker*.

Has anyone seen this before? I have tried using the pip ansible module to install the modules and then after removing them manually, reinstalled them manually as in the referenced post. I'm also using 2.1.0.0. as you can see, which was supposed to fix this issue.

回答1:

This is because new versions of python modules docker and docker-py that ansible uses are incompatible. I had to revert back and explicitly specify the following versions of PIP packages:

docker: 2.0.0
docker-py: 1.10.6

Sample playbook task for these:

- name: install certain python modules for docker
  pip:
    name: "{{ item.name }}"
    version: "{{ item.version }}"
    state: present
  with_items:
  - { name: docker, version: 2.0.0 }
  - { name: docker-py, version: 1.10.6 }

All my play books work fine since then.

回答2:

For me specifying the path to docker-py worked.

- hosts: <host>
  environment:
    PYTHONPATH: "/home/path/.local/lib/python2.7/site-packages"

Basically Ansible was looking in the wrong directory.

Full credit to Clay Graham for his great article on the issue:

https://medium.com/dronzebot/ansible-and-docker-py-path-issues-and-resolving-them-e3834d5bb79a

回答3:

The requirements for the docker-py module on ansible's documentation for the docker* modules is not really up-to-date, but generally here are some ansible - docker-py pairs that should work:

Ansible 2.1.0.0 requires docker-py version 1.7.0 at least, which in turn requires at least docker 1.9 installed on the client.
If you need to use an older version of docker, then you can use docker 1.7 with ansible 2.0.1.0 and docker-py 1.4.0.
If you need an even older version of docker, like 1.6 then you are stuck with ansible 1.9