compute-mig

GCE Managed Instance Group module

This module allows creating a managed instance group supporting one or more application versions via instance templates. Optionally, a health check and an autoscaler can be created, and the managed instance group can be configured to be stateful.

This module can be coupled with the compute-vm module which can manage instance templates, and the net-lb-int module to assign the MIG to a backend wired to an Internal Load Balancer. The first use case is shown in the examples below.

Stateful disks can be created directly, as shown in the last example below.

• Examples
  • Simple Example
  • Multiple Versions
  • Health Check and Autohealing Policies
  • Autoscaling
  • Update Policy
  • Stateful MIGs - MIG Config
  • Stateful MIGs - Instance Config
• Variables
• Outputs
• Fixtures

Examples

Simple Example

This example shows how to manage a simple MIG that leverages the compute-vm module to manage the underlying instance template. The following sub-examples will only show how to enable specific features of this module, and won't replicate the combined setup.

module "cos-nginx" {
  source = "./fabric/modules/cloud-config-container/nginx"
}

module "nginx-template" {
  source     = "./fabric/modules/compute-vm"
  project_id = var.project_id
  name       = "nginx-template"
  zone       = "${var.region}-b"
  tags       = ["http-server", "ssh"]
  network_interfaces = [{
    network    = var.vpc.self_link
    subnetwork = var.subnet.self_link
    nat        = false
    addresses  = null
  }]
  boot_disk = {
    initialize_params = {
      image = "projects/cos-cloud/global/images/family/cos-stable"
    }
  }
  create_template = true
  metadata = {
    user-data = module.cos-nginx.cloud_config
  }
}

module "nginx-mig" {
  source            = "./fabric/modules/compute-mig"
  project_id        = var.project_id
  location          = "${var.region}-b"
  name              = "mig-test"
  target_size       = 2
  instance_template = module.nginx-template.template.self_link
}
# tftest modules=2 resources=2 inventory=simple.yaml e2e

Multiple Versions

If multiple versions are desired, use more compute-vm instances for the additional templates used in each version (not shown here), and reference them like this:

module "cos-nginx" {
  source = "./fabric/modules/cloud-config-container/nginx"
}

module "nginx-template" {
  source     = "./fabric/modules/compute-vm"
  project_id = var.project_id
  name       = "nginx-template"
  zone       = "${var.region}-b"
  tags       = ["http-server", "ssh"]
  network_interfaces = [{
    network    = var.vpc.self_link
    subnetwork = var.subnet.self_link
    nat        = false
    addresses  = null
  }]
  boot_disk = {
    initialize_params = {
      image = "projects/cos-cloud/global/images/family/cos-stable"
    }
  }
  create_template = true
  metadata = {
    user-data = module.cos-nginx.cloud_config
  }
}

module "nginx-mig" {
  source            = "./fabric/modules/compute-mig"
  project_id        = var.project_id
  location          = "${var.region}-b"
  name              = "mig-test"
  target_size       = 3
  instance_template = module.nginx-template.template.self_link
  versions = {
    canary = {
      instance_template = module.nginx-template.template.self_link
      target_size = {
        fixed = 1
      }
    }
  }
}
# tftest modules=2 resources=2 inventory=multiple.yaml e2e

Health Check and Autohealing Policies

Autohealing policies can use an externally defined health check, or have this module auto-create one:

module "cos-nginx" {
  source = "./fabric/modules/cloud-config-container/nginx"
}

module "nginx-template" {
  source     = "./fabric/modules/compute-vm"
  project_id = var.project_id
  name       = "nginx-template"
  zone       = "${var.region}-b"
  tags       = ["http-server", "ssh"]
  network_interfaces = [{
    network    = var.vpc.self_link,
    subnetwork = var.subnet.self_link,
    nat        = false,
    addresses  = null
  }]
  boot_disk = {
    initialize_params = {
      image = "projects/cos-cloud/global/images/family/cos-stable"
    }
  }
  create_template = true
  metadata = {
    user-data = module.cos-nginx.cloud_config
  }
}

module "nginx-mig" {
  source            = "./fabric/modules/compute-mig"
  project_id        = var.project_id
  location          = "${var.region}-b"
  name              = "mig-test"
  target_size       = 3
  instance_template = module.nginx-template.template.self_link
  auto_healing_policies = {
    initial_delay_sec = 30
  }
  health_check_config = {
    enable_logging = true
    http = {
      port = 80
    }
  }
}
# tftest modules=2 resources=3 inventory=health-check.yaml e2e

Autoscaling

The module can create and manage an autoscaler associated with the MIG. When using autoscaling do not set the target_size variable or set it to null. Here we show a CPU utilization autoscaler, the other available modes are load balancing utilization and custom metric, like the underlying autoscaler resource.

module "cos-nginx" {
  source = "./fabric/modules/cloud-config-container/nginx"
}

module "nginx-template" {
  source     = "./fabric/modules/compute-vm"
  project_id = var.project_id
  name       = "nginx-template"
  zone       = "${var.region}-b"
  tags       = ["http-server", "ssh"]
  network_interfaces = [{
    network    = var.vpc.self_link
    subnetwork = var.subnet.self_link
    nat        = false
    addresses  = null
  }]
  boot_disk = {
    initialize_params = {
      image = "projects/cos-cloud/global/images/family/cos-stable"
    }
  }
  create_template = true
  metadata = {
    user-data = module.cos-nginx.cloud_config
  }
}

module "nginx-mig" {
  source            = "./fabric/modules/compute-mig"
  project_id        = var.project_id
  location          = "${var.region}-b"
  name              = "mig-test"
  target_size       = 3
  instance_template = module.nginx-template.template.self_link
  autoscaler_config = {
    max_replicas    = 3
    min_replicas    = 1
    cooldown_period = 30
    scaling_signals = {
      cpu_utilization = {
        target = 0.65
      }
    }
  }
}
# tftest modules=2 resources=3 inventory=autoscaling.yaml e2e

Update Policy

module "cos-nginx" {
  source = "./fabric/modules/cloud-config-container/nginx"
}

module "nginx-template" {
  source     = "./fabric/modules/compute-vm"
  project_id = var.project_id
  name       = "nginx-template"
  zone       = "${var.region}-b"
  tags       = ["http-server", "ssh"]
  network_interfaces = [{
    network    = var.vpc.self_link
    subnetwork = var.subnet.self_link
    nat        = false
    addresses  = null
  }]
  boot_disk = {
    initialize_params = {
      image = "projects/cos-cloud/global/images/family/cos-stable"
    }
  }
  create_template = true
  metadata = {
    user-data = module.cos-nginx.cloud_config
  }
}

module "nginx-mig" {
  source            = "./fabric/modules/compute-mig"
  project_id        = var.project_id
  location          = "${var.region}-b"
  name              = "mig-test"
  target_size       = 3
  instance_template = module.nginx-template.template.self_link
  update_policy = {
    minimal_action = "REPLACE"
    type           = "PROACTIVE"
    min_ready_sec  = 30
    max_surge = {
      fixed = 1
    }
  }
}
# tftest modules=2 resources=2 inventory=policy.yaml e2e

Stateful MIGs - MIG Config

Stateful MIGs have some limitations documented here. Enforcement of these requirements is the responsibility of users of this module.

You can configure a disk defined in the instance template to be stateful for all instances in the MIG by configuring in the MIG's stateful policy, using the stateful_disk_mig variable. Alternatively, you can also configure stateful persistent disks individually per instance of the MIG by setting the stateful_disk_instance variable. A discussion on these scenarios can be found in the docs.

An example using only the configuration at the MIG level can be seen below.

Note that when referencing the stateful disk, you use device_name and not disk_name. Specifying an existing disk in the template (and stateful config) only allows a single instance to be managed by the MIG, typically coupled with an autohealing policy (shown in the examples above).

module "cos-nginx" {
  source = "./fabric/modules/cloud-config-container/nginx"
}

module "nginx-template" {
  source        = "./fabric/modules/compute-vm"
  project_id    = var.project_id
  name          = "nginx-template"
  zone          = "${var.region}-b"
  tags          = ["http-server", "ssh"]
  instance_type = "e2-small"
  network_interfaces = [{
    network    = var.vpc.self_link
    subnetwork = var.subnet.self_link
  }]
  boot_disk = {
    initialize_params = {
      image = "projects/cos-cloud/global/images/family/cos-stable"
    }
  }
  attached_disks = [{
    source_type = "attach"
    name        = "data-1"
    size        = 10
    source      = google_compute_disk.test-disk.name
  }]
  create_template = true
  metadata = {
    user-data = module.cos-nginx.cloud_config
  }
}

module "nginx-mig" {
  source            = "./fabric/modules/compute-mig"
  project_id        = var.project_id
  location          = "${var.region}-b"
  name              = "mig-test-2"
  target_size       = 1
  instance_template = module.nginx-template.template.self_link
  stateful_disks = {
    data-1 = false
  }
}
# tftest modules=2 resources=3 fixtures=fixtures/attached-disks.tf inventory=mig-config.yaml e2e

Stateful MIGs - Instance Config

Here is an example defining the stateful config at the instance level. As in the example above, specifying an existing disk in the template (and stateful config) only allows a single instance to be managed by the MIG, typically coupled with an autohealing policy (shown in the examples above).

module "cos-nginx" {
  source = "./fabric/modules/cloud-config-container/nginx"
}

module "nginx-template" {
  source        = "./fabric/modules/compute-vm"
  project_id    = var.project_id
  name          = "nginx-template"
  zone          = "${var.region}-b"
  tags          = ["http-server", "ssh"]
  instance_type = "e2-small"
  network_interfaces = [{
    network    = var.vpc.self_link
    subnetwork = var.subnet.self_link
  }]
  boot_disk = {
    initialize_params = {
      image = "projects/cos-cloud/global/images/family/cos-stable"
    }
  }
  attached_disks = [{
    source_type = "attach"
    name        = "data-1"
    size        = 10
    source      = google_compute_disk.test-disk.name
  }]
  create_template = true
  metadata = {
    user-data = module.cos-nginx.cloud_config
  }
}

module "nginx-mig" {
  source            = "./fabric/modules/compute-mig"
  project_id        = var.project_id
  location          = "${var.region}-b"
  name              = "mig-test"
  instance_template = module.nginx-template.template.self_link
  stateful_config = {
    instance-1 = {
      minimal_action                 = "NONE",
      most_disruptive_allowed_action = "REPLACE"
      preserved_state = {
        disks = {
          data-1 = {
            source = google_compute_disk.test-disk.id
          }
        }
        metadata = {
          foo = "bar"
        }
      }
    }
  }
}
# tftest modules=2 resources=4 fixtures=fixtures/attached-disks.tf  inventory=stateful.yaml e2e

Variables

name	description	type	required	default
instance_template	Instance template for the default version.	string	✓
location	Compute zone or region.	string	✓
name	Managed group name.	string	✓
project_id	Project id.	string	✓
all_instances_config	Metadata and labels set to all instances in the group.	object({…})		null
auto_healing_policies	Auto-healing policies for this group.	object({…})		null
autoscaler_config	Optional autoscaler configuration.	object({…})		null
default_version_name	Name used for the default version.	string		"default"
description	Optional description used for all resources managed by this module.	string		"Terraform managed."
distribution_policy	DIstribution policy for regional MIG.	object({…})		null
health_check_config	Optional auto-created health check configuration, use the output self-link to set it in the auto healing policy. Refer to examples for usage.	object({…})		null
named_ports	Named ports.	map(number)		null
stateful_config	Stateful configuration for individual instances.	map(object({…}))		{}
stateful_disks	Stateful disk configuration applied at the MIG level to all instances, in device name => on permanent instance delete rule as boolean.	map(bool)		{}
target_pools	Optional list of URLs for target pools to which new instances in the group are added.	list(string)		[]
target_size	Group target size, leave null when using an autoscaler.	number		null
update_policy	Update policy. Minimal action and type are required.	object({…})		null
versions	Additional application versions, target_size is optional.	map(object({…}))		{}
wait_for_instances	Wait for all instances to be created/updated before returning.	object({…})		null

Outputs

name	description	sensitive
autoscaler	Auto-created autoscaler resource.
group_manager	Instance group resource.
health_check	Auto-created health-check resource.
id	Fully qualified group manager id.

Fixtures

• attached-disks.tf