Skip to content

Latest commit

 

History

History
211 lines (184 loc) · 6.77 KB

migration.md

File metadata and controls

211 lines (184 loc) · 6.77 KB
Raw

Magic Castle Migration Guide

Magic Castle 10.x to 11.x

Here are some helpful tips for migrating from using Magic Castle 10.x to 11.x.

Note that live clusters cannot be simply migrated from 10 to 11. This guide is only meant to help you migrating existing Terraform code, not cloud resources.

main.tf: instances

The first major change is the structure for the instances variable. In MC 10, the instances were defined like this:

instances = {
    mgmt  = { type = "p4-6gb", count = 1 }
    login = { type = "p2-3gb", count = 1 }
    node  = [
        { type = "p2-3gb", count = 1 }
        { type = "g1-10gb-60", count = 1, prefix = "gpu" }
    ]
}

In MC 11, the same architecture is defined like this:

instances = {
    mgmt     = { type = "p4-6gb", tags = ["puppet", "mgmt", "nfs"] }
    login    = { type = "p2-3gb", tags = ["login", "public", "proxy"] }
    node     = { type = "p2-3gb", tags = ["node"], count = 1 }
    gpu-node = { type = "g1-8gb-c4-22gb", tags = ["node"], count = 1 }
}

List of changes:

  1. Each instance' purpose in the cluster is now defined by a list of tags instead of using its hostname. The role of each tag is documented here. The hostname of the instances is therefore no longer limited to mgmt, login or node.
  2. To define heterogeneous compute nodes, prefix is no longer used. Instead, a new entry with a distinct hostname prefix is defined.
  3. The count parameter is now optional and has a default value of 1. Instance hostnames still include the index number, even when count is not defined.

Note that it is possible to move the tags, and create more instances with these tags. For example, to move the Puppet main server on an instance separate of mgmt1, one could do:

instances = {
    puppet   = { type = "p4-6gb", tags = ["puppet"]}
    mgmt     = { type = "p4-6gb", tags = ["mgmt", "nfs"] }
    login    = { type = "p2-3gb", tags = ["login", "public", "proxy"] }
    node     = { type = "p2-3gb", tags = ["node"], count = 1 }
}

main.tf: storage -> volumes

The second major change is storage has been replaced by the volumes. In MC 10, it was possible to define only three volumes attached to mgmt1 with the following code:

storage = {
    type         = "nfs"
    home_size    = 100
    project_size = 100
    scratch_size = 100
}

In MC 11, the same volumes attached to mgmt1 are defined like this

volumes = {
    nfs = {
        home    = { size = 100 }
        project = { size = 100 }
        scratch = { size = 100, type = "volumes-ssd" }
    }
}

List of changes:

  1. type is removed. Instead, a set of volumes gets to share the same arbitrary tag, in the case above nfs.
  2. For each instance with a tag matching of the volume tag, Magic Castle will create the set of matching volume and attach them to the instance.
  3. It is now possible to define the volume type. If left unspecified, the cloud provider default for volume type is used.
  4. Removing an instance will delete the corresponding volumes. It is no longer possible to maintain the home, project and scratch volumes while setting the mgmt count at 0.

main.tf: Other changes

  • root_disk_size variable has been removed. You can define the root disk size of each instance type independently by defining disk_size in the instance attributes' map.
  • Azure managed_disk_type has been removed. The disk type can be defined for instances using the disk_type attribute in the instance attributes' map and the disk type for attached volume can be defined with type in the volumes map. This attribute works for every cloud provider, not only Azure.

Puppet: terraform_data.yaml format

Instead of fetching a template of a YAML file, Magic Castle now writes its own YAML file that is than integrated in the Puppet data hierarchy.

In MC 10, terraform_data.yaml looked like this when populated on mgmt1

---
profile::base::sudoer_username: "centos"

profile::consul::acl_api_token: "123-abc-def-456"

profile::freeipa::base::admin_passwd: "aaaaBBBBcccDDee"
profile::freeipa::base::domain_name: "phoenix.calculquebec.cloud"
profile::freeipa::mokey::passwd: "aaaaBBBBcccDDee"

profile::accounts::guests::passwd: "sneaky.sensitive.girafe"
profile::accounts::guests::nb_accounts: 10

profile::slurm::base::cluster_name: "phoenix"
profile::slurm::base::munge_key: "aaaBbbCc1231aF"
profile::slurm::accounting::password: "aaaaBBBBcccDDee"

profile::freeipa::client::server_ip: "10.0.0.0.2"
profile::consul::client::server_ip: "10.0.0.2"
profile::nfs::client::server_ip: "10.0.0.2"

profile::nfs::server::home_devices: "/dev/disk/by-id/virtio-*"
profile::nfs::server::project_devices: "/dev/disk/by-id/virtio-*"
profile::nfs::server::scratch_devices: "/dev/disk/by-id/virtio-*"

profile::reverse_proxy::domain_name: "phoenix.calculquebec.cloud"

In MC 11, the terraform_data.yaml matching the preceding instances and volumes configuration could look like this:

terraform:
  instances:
    mgmt1:
      local_ip: "10.0.0.1"
      public_ip: ""
      tags: ["puppet", "nfs", "mgmt"]
      hostkeys:
        rsa: "ssh-rsa ..."
    login1:
      local_ip: "10.0.0.2"
      public_ip: "132.219.29.10"
      tags: ["login", "public", "proxy"]
      hostkeys:
        rsa: "ssh-rsa ..."
    node1:
      local_ip: "10.0.0.3"
      public_ip: ""
      tags: ["node"]
      hostkeys:
        rsa: "ssh-rsa ..."
    gpu-node1:
      local_ip: "10.0.0.4"
      public_ip: ""
      tags: ["node"]
      hostkeys:
        rsa: "ssh-rsa ..."
  volumes:
    nfs:
      home:
        - "/dev/disk/by-id/virtio-*"
      project:
        - "/dev/disk/by-id/virtio-*"
      scratch:
        - "/dev/disk/by-id/virtio-*"
  tag_ip:
    mgmt:
      - 10.0.0.1
    nfs:
      - 10.0.0.1
    puppet:
      - 10.0.0.1
    login:
      - 10.0.0.2
    node:
      - 10.0.0.3
    gpu-node:
      - 10.0.0.4
  data:
    cluster_name: "phoenix"
    consul_token: "123-abc-def-456"
    domain_name: "phoenix.calculquebec.cloud"
    freeipa_passwd: "aaaaBBBBcccDDee"
    guest_passwd: "sneaky.sensitive.girafe"
    munge_key: "aaaBbbCc1231aF"
    nb_users: 10
    sudoer_username: "centos"

The values from terraform_data.yaml are input in Puppet classes hieradata using aliases in common.yaml. For example, to number of guest accounts that corresponds to profile::accounts::guests::nb_accounts is set to the nb_users value like this in common.yaml:

profile::accounts::guests::nb_accounts: "%{alias('terraform.data.nb_users')}"

Important note: Because terraform_data.yaml now includes information about the resources, it will be overwritten every time something is modified in main.tf. It is therefore no longer recommended to edit it manually. All changes to this file should be perform by Terraform. The documentation has been updated accordingly.