Using multiple policies

When you create an autoscaler with multiple policies, the autoscaler scales based on the policy that provides the largest number of virtual machine (VM) instances in the managed instance group (MIG). This ensures that there are always enough VMs to handle your application workloads and lets you scale applications with multiple possible bottlenecks.

To learn more about the different types of autoscaling policies, see policies.

Before you begin

If you want to use the command-line examples in this guide:
1. Install or update to the latest version of the gcloud command-line tool.
2. Set a default region and zone.
If you want to use the API examples in this guide, set up API access.
Read about autoscaling policies.

How autoscaler handles multiple policies

The autoscaler handles multiple policies by calculating the recommended number of VMs for each policy and then picking the policy that results in the largest number of VMs in the MIG.

An autoscaler can handle one policy per metric type except in the case of Cloud Monitoring metrics and scaling schedules (preview). You can choose up to 5 policies for Cloud Monitoring metrics and create up to 128 schedules per MIG. For example, you can create an autoscaler that uses up to 1 CPU utilization policy, up to 1 load balancing policy, up to 5 custom metric policies, and up to 128 scaling schedules.

For example, you can define an autoscaler with all of the following policy definitions:

cpuUtilization with target of 0.8
loadBalancingUtilization with target of 0.6
customMetricUtilization for metric1 with target of 1000
customMetricUtilization for metric2 with target of 2000
scalingSchedules
- A minimum of 6 VMs every Saturday and Sunday at 12:00 AM for 24 hours
- A minimum of 15 VMs every Monday through Friday at 9:00 AM for 8 hours

For those policies, suppose the autoscaler measures the following average utilization values for a MIG that contains 10 VMs at the time of measurement:

0.5 for CPU utilization
0.4 load balancing utilization
1100 for metric1
2700 for metric2

Using those values, the autoscaler calculates the recommended number of VMs based on each policy:

7 VMs based on CPU utilization
7 VMs based on load balancing utilization
11 VMs based on metric1
14 VMs based on metric2
6 VMs every Saturday and Sunday for the entire day based on the first scaling schedule
15 VMs every Monday through Friday from 9:00 AM to 5:00 PM based on the second scaling schedule

The autoscaler picks the policy that results in the largest number of VMs in the MIG and sets the MIG's recommended size to that value. In this case, the autoscaler resizes the MIG to 15 VMs if the second schedule is currently active or to 14 VMs otherwise.

Create an autoscaler with multiple policies

Creating an autoscaler with multiple policies is slightly different depending on whether you are using the Cloud Console, gcloud tool, or the Compute Engine API.

Permissions required for this task

To perform this task, you must have the following permissions:

To create a new autoscaler for an existing MIG:
- compute.autoscalers.create on the project
- compute.instanceGroupManagers.use on the MIG
To edit an existing autoscaler:
- compute.autoscalers.update on the project

Console

In the Cloud Console, go to the Instance groups page.

Go to Instance groups
Click on the name of a MIG from the list. Then click Edit group.
On the edit page under Autoscaling mode, turn on autoscaling by selecting Autoscale from the drop-down menu.
For each metric-based autoscaling policy that you want to add, under Autoscaling policy, click Add new metric.
- To create an autoscaling policy based on average CPU utilization, under Metric type select CPU Utilization from the drop-down menu. Then, enter a Target CPU utilization and click Done. For more information, see Scaling based on CPU utilization.
- To create an autoscaling policy based on HTTP load balancing serving capacity, under Metric type select HTTP load balancing utilization from the drop-down menu. Then, enter a Target HTTP load balancing utilization and click Done. For more information, see Scaling based on load balancing serving capacity.
- To create an autoscaling policy based on Cloud Monitoring metrics, under Metric type select Stackdriver Monitoring from the drop-down menu. Then, complete the remaining fields and click Done. For more information, see Scaling based on Cloud Monitoring metrics.
For each schedule-based autoscaling policy that you want to add, under Autoscaling policy, click Add new scaling schedule.

Note: If you are unable to click Add new scaling schedule, you might need to save this autoscaler first. Click Save at the bottom the page and then click Edit group. Then, click Add new scaling schedule.

In the new Create a schedule pane, complete the required fields and click Save. Then, in the Scaling schedules pane, click Done. For more information, see Scaling based on schedules.
Under Cool down period, enter the number of seconds that your application needs to initialize—this lets your VMs finish initializing before the autoscaler begins collecting usage information from them. For more information, see cool down periods.
Under Minimum number of instances, enter the minimum number of VMs that you want for this MIG. The minimum size of this MIG is always maintained regardless of the recommended size generated by its autoscaling policies.
Under Maximum number of instances enter the maximum number of VMs that you want for this MIG. The maximum size of this MIG is never exceeded regardless of the recommended size generated by its autoscaling policies.
Optional: to enable scale-in controls, select the Enable Scale In Controls checkbox. For more information, see scale-in controls
Click Save.

gcloud

To create an autoscaler with multiple policies, pass in multiple policy specifications using their respective command-line flags. If you want to create an autoscaler that includes scaling schedules, use the gcloud beta compute instance-groups managed set-autoscaling command. Otherwise, use the gcloud compute instance-groups managed set-autoscaling command.

For example, the following command creates an autoscaler that uses CPU utilization, load balancing serving capacity, two custom metrics, and one scaling schedule. The --max-num-replicas and --min-num-replicas flags restrict how big or small the MIG can get, regardless of each policy's recommended size.

gcloud beta compute instance-groups managed set-autoscaling MIG_NAME \
   --target-cpu-utilization=0.8 \
   --target-load-balancing-utilization=0.6 \
   --custom-metric-utilization metric=PATH_TO_CUSTOM_METRIC_1,utilization-target=1000.0,utilization-target-type=GAUGE \
   --custom-metric-utilization metric=PATH_TO_CUSTOM_METRIC_2,utilization-target=2000.0,utilization-target-type=DELTA_PER_SECOND \
   --set-schedule=workday-capacity \
   --schedule-cron="30 8 * * Mon-Fri" \
   --schedule-duration-sec=30600 \
   --schedule-min-required-replicas=10 \
   --schedule-description="Have at least 10 VMs every Monday through Friday from 8:30 AM to 5 PM UTC"
   --min-num-replicas=1
   --max-num-replicas=50

Replace the following:

MIG_NAME: the name of a MIG.
PATH_TO_CUSTOM_METRIC_1 and PATH_TO_CUSTOM_METRIC_2: Optional: the paths to custom metrics with a format similar to custom.cloudmonitoring.googleapis.com/path/to/metric1 and custom.cloudmonitoring.googleapis.com/path/to/metric2. For more information, see Custom metrics.

With the exception of the custom metric flags, you can only pass in one flag maximum for each metric type per gcloud command. For custom metrics, you can provide multiple custom metrics in a single command by repeating the --custom-metric-utilization flag.

For more information about how to configure the flags for each type of autoscaling policy, see the following pages:

API

To create an autoscaler, use the autoscalers.insert method for a zonal MIG or the regionAutoscalers.insert method for a regional MIG. If you want to create an autoscaler that includes scaling schedules, you must use the beta autoscalers.insert method for a zonal MIG or the beta regionAutoscalers.insert method for a regional MIG.

For example, the following request creates an autoscaler that uses CPU utilization, load balancing serving capacity, two custom metrics, and two scaling schedules. The maxNumReplicas and minNumReplicas fields restrict how big or small the MIG can get, regardless of each policy's recommended size.

POST https://compute.googleapis.com/compute/beta/projects/PROJECT/zones/ZONE/autoscalers

{
  "autoscalingPolicy" : {
    "cpuUtilization":{
      "utilizationTarget": 0.8
    },
    "loadBalancingUtilization":{
      "utilizationTarget": 0.6
    },
    "customMetricUtilizations":[
      {
        "metric": "PATH_TO_CUSTOM_METRIC_1",
        "utilizationTarget": 1000,
        "utilizationTargetType":"GAUGE"
      },
      {
        "metric": "PATH_TO_CUSTOM_METRIC_2",
        "utilizationTarget": 2000,
        "utilizationTargetType": "DELTA_PER_SECOND"
      }
    ],
    "scalingSchedules": {
      "workday-capacity": {
        "minRequiredReplicas": 10,
        "schedule": "30 8 * * Mon-Fri",
        "durationSec": 30600,
        "description": "Have at least 10 VMs every Monday through Friday from 8:30 AM to 5 PM UTC"
      },
      "january-30-2030-schedule": {
        "minRequiredReplicas": 30,
        "schedule": "0 0 30 1 * 2030",
        "timeZone": "America/New_York",
        "durationSec": 86400,
        "description": "Schedule a minimum of 30 VMs all day for January 30, 2030"
      }
    },
    "maxNumReplicas": 50,
    "minNumReplicas": 1
  },
  "target": "https://www.googleapis.com/compute/v1/projects/myproject/zones/us-central1-a/instanceGroupManagers/MIG_NAME",
  "name": "MIG_NAME"
}

Replace the following:

PROJECT: your project id.
ZONE: the zone where your MIG is located.
MIG_NAME: the name of a MIG.
PATH_TO_CUSTOM_METRIC_1 and PATH_TO_CUSTOM_METRIC_2: Optional: the paths to custom metrics with a format similar to custom.cloudmonitoring.googleapis.com/path/to/metric1 and custom.cloudmonitoring.googleapis.com/path/to/metric2. For more information, see Custom metrics.

As shown in this example, you can provide multiple custom metrics and scaling schedules in the same request.

For more information about how to configure the fields for each type of autoscaling policy, see the following pages: