cloud-ca / docker-machine-driver-cloudca Goto Github PK

Currently, we cannot provision hosts from public IPs on Rancher. The default flow is to have public IPs and because some of the port forwarding rules are missing, Rancher fails to provision the host correctly. You would have to check the Private IP flag and change the settings for the provision host which is not straight-forward.

One solution is to use static NAT instead of port-forwarding so that any port that is open, gets exposed correctly.

When CoreOS is deployed using docker-machine-driver-cloudca, after an automatic upgrade of CoreOS, docker daemon is not restarted by default, default behavior of CoreOS.

Maybe we should push the command via cloud-init

echo "docker ps" >> /etc/rc.local

as describe in rancher/rancher#5069

The only UUID in the list of parameters is --cloudca-network-id ($CLOUDCA_NETWORK_ID).

It could be more user friendly to split this into 2 options that have names. This would allow us to provide default values for 4 different options, and would help customers getting started in a trial environment, as all of the names are standardized (i.e. compute-on / test_area / default-vpc / Frontend).

I noticed that the driver doesn't populate the IPAddress field https://github.com/docker/machine/blob/15fd4c70403bab784d91031af02d9e169ce66412/libmachine/drivers/base.go#L17 in the base driver config, instead it populates its own variable https://github.com/cloud-ca/docker-machine-driver-cloudca/blob/master/cloudca.go#L85 PublicIp

The user-data pushed by the driver as cloud-init config is not good.

user-data sent is

#!/bin/sh
mkfs -t ext4 /dev/xvdb
mkdir -p /var/lib/docker
mount -t ext4 /dev/xvdb /var/lib/docker

The data volume is not remounted automatically after a reboot, and on CoreOS, the volume is formated at each reboot, which is dangerous because coreos will reboot automatically after a channel upgrade.

Good cloud-init configuration for coreos, ubuntu and centos:
Tested on ubuntu16.04 and coreos:

#cloud-config
fs_setup:
   - label:  data
     filesystem: 'ext4'
     device: '/dev/xvdb'
mounts:
 - [ xvdb, /var/lib/docker, "ext4", "defaults", "0", "0" ]
coreos:
  units:
    - name: format-datavolume.service
      command: start
      content: |
        [Unit]
        Description=Formats the data volume
        After=dev-xvdb.device
        Requires=dev-xvdb.device
        ConditionPathExists=!/var/lib/docker.btrfs
        [Service]
        Type=oneshot
        RemainAfterExit=yes
        ExecStart=/usr/sbin/mkfs.btrfs /dev/xvdb
        ExecStart=/usr/bin/mkdir /var/lib/docker.btrfs
    - name: var-lib-docker.mount
      command: start
      content: |
        [Unit]
        Description=Mount data volume to /var/lib/docker
        Requires=format-datavolume.service
        After=format-datavolume.service
        [Mount]
        What=/dev/xvdb
        Where=/var/lib/docker
        Type=btrfs
    - name: docker.service
      drop-ins:
        - name: 10-wait-docker.conf
          content: |
            [Unit]
            After=var-lib-docker.mount
            Requires=var-lib-docker.mount

To avoid confusion with the cloud.ca terminology, we should rename the cloudca-network-id option to cloudca-tier-id.