REST Resource: autoscalers

Resource: Autoscaler

Represents an Autoscaler resource.

Google Compute Engine has two Autoscaler resources:

Use autoscalers to automatically add or delete instances from a managed instance group according to your defined autoscaling policy. For more information, read Autoscaling Groups of Instances.

For zonal managed instance groups resource, use the autoscaler resource.

For regional managed instance groups, use the regionAutoscalers resource.

JSON representation
  "id": string,
  "creationTimestamp": string,
  "name": string,
  "description": string,
  "target": string,
  "autoscalingPolicy": {
    "minNumReplicas": integer,
    "maxNumReplicas": integer,
    "scaleInControl": {
      "maxScaledInReplicas": {
        "fixed": integer,
        "percent": integer,
        "calculated": integer
      "timeWindowSec": integer
    "coolDownPeriodSec": integer,
    "cpuUtilization": {
      "utilizationTarget": number,
      "predictiveMethod": enum
    "customMetricUtilizations": [
        "metric": string,
        "filter": string,
        "utilizationTargetType": enum,

        "utilizationTarget": number,
        "singleInstanceAssignment": number
    "loadBalancingUtilization": {
      "utilizationTarget": number
    "mode": enum,
    "scalingSchedules": {
      string: {
        "minRequiredReplicas": integer,
        "schedule": string,
        "timeZone": string,
        "durationSec": integer,
        "disabled": boolean,
        "description": string
  "zone": string,
  "region": string,
  "selfLink": string,
  "status": enum,
  "statusDetails": [
      "message": string,
      "type": enum
  "recommendedSize": integer,
  "scalingScheduleStatus": {
    string: {
      "nextStartTime": string,
      "lastStartTime": string,
      "state": enum
  "kind": string

string (uint64 format)

[Output Only] The unique identifier for the resource. This identifier is defined by the server.



[Output Only] Creation timestamp in RFC3339 text format.



Name of the resource. Provided by the client when the resource is created. The name must be 1-63 characters long, and comply with RFC1035. Specifically, the name must be 1-63 characters long and match the regular expression [a-z]([-a-z0-9]*[a-z0-9])? which means the first character must be a lowercase letter, and all following characters must be a dash, lowercase letter, or digit, except the last character, which cannot be a dash.



An optional description of this resource. Provide this property when you create the resource.



URL of the managed instance group that this autoscaler will scale. This field is required when creating an autoscaler.



The configuration parameters for the autoscaling algorithm. You can define one or more signals for an autoscaler: cpuUtilization, customMetricUtilizations, and loadBalancingUtilization.

If none of these are specified, the default will be to autoscale based on cpuUtilization to 0.6 or 60%.



The minimum number of replicas that the autoscaler can scale in to. This cannot be less than 0. If not provided, autoscaler chooses a default value depending on maximum number of instances allowed.



The maximum number of instances that the autoscaler can scale out to. This is required when creating or updating an autoscaler. The maximum number of replicas must not be lower than minimal number of replicas.





Maximum allowed number (or %) of VMs that can be deducted from the peak recommendation during the window autoscaler looks at when computing recommendations. Possibly all these VMs can be deleted at once so user service needs to be prepared to lose that many VMs in one step.



Specifies a fixed number of VM instances. This must be a positive integer.



Specifies a percentage of instances between 0 to 100%, inclusive. For example, specify 80 for 80%.



[Output Only] Absolute value of VM instances calculated based on the specific mode.

  • If the value is fixed, then the calculated value is equal to the fixed value.
  • If the value is a percent, then the calculated value is percent/100 * targetSize. For example, the calculated value of a 80% of a managed instance group with 150 instances would be (80/100 * 150) = 120 VM instances. If there is a remainder, the number is rounded.


How far back autoscaling looks when computing recommendations to include directives regarding slower scale in, as described above.



The number of seconds that the autoscaler waits before it starts collecting information from a new instance. This prevents the autoscaler from collecting information when the instance is initializing, during which the collected usage would not be reliable. The default time autoscaler waits is 60 seconds.

Virtual machine initialization times might vary because of numerous factors. We recommend that you test how long an instance may take to initialize. To do this, create an instance and time the startup process.



Defines the CPU utilization policy that allows the autoscaler to scale based on the average CPU utilization of a managed instance group.



The target CPU utilization that the autoscaler maintains. Must be a float value in the range (0, 1]. If not specified, the default is 0.6.

If the CPU level is below the target utilization, the autoscaler scales in the number of instances until it reaches the minimum number of instances you specified or until the average CPU of your instances reaches the target utilization.

If the average CPU is above the target utilization, the autoscaler scales out until it reaches the maximum number of instances you specified or until the average utilization reaches the target utilization.



Indicates whether predictive autoscaling based on CPU metric is enabled. Valid values are:

  • NONE (default). No predictive method is used. The autoscaler scales the group to meet current demand based on real-time metrics.
  • OPTIMIZE_AVAILABILITY. Predictive autoscaling improves availability by monitoring daily and weekly load patterns and scaling out ahead of anticipated demand.


Configuration parameters of autoscaling based on a custom metric.



The identifier (type) of the Stackdriver Monitoring metric. The metric cannot have negative values.

The metric must have a value type of INT64 or DOUBLE.



A filter string, compatible with a Stackdriver Monitoring filter string for TimeSeries.list API call. This filter is used to select a specific TimeSeries for the purpose of autoscaling and to determine whether the metric is exporting per-instance or per-group data.

For the filter to be valid for autoscaling purposes, the following rules apply:

  • You can only use the AND operator for joining selectors.
  • You can only use direct equality comparison operator (=) without any functions for each selector.
  • You can specify the metric in both the filter string and in the metric field. However, if specified in both places, the metric must be identical.
  • The monitored resource type determines what kind of values are expected for the metric. If it is a gce_instance, the autoscaler expects the metric to include a separate TimeSeries for each instance in a group. In such a case, you cannot filter on resource labels.

    If the resource type is any other value, the autoscaler expects this metric to contain values that apply to the entire autoscaled instance group and resource label filtering can be performed to point autoscaler at the correct TimeSeries to scale upon. This is called a per-group metric for the purpose of autoscaling.

    If not specified, the type defaults to gce_instance.

Try to provide a filter that is selective enough to pick just one TimeSeries for the autoscaled group or for each of the instances (if you are using gce_instance resource type). If multiple TimeSeries are returned upon the query execution, the autoscaler will sum their respective values to obtain its scaling value.



The target value of the metric that autoscaler maintains. This must be a positive value. A utilization metric scales number of virtual machines handling requests to increase or decrease proportionally to the metric.

For example, a good metric to use as a utilizationTarget is The autoscaler works to keep this value constant for each of the instances.



If scaling is based on a per-group metric value that represents the total amount of work to be done or resource usage, set this value to an amount assigned for a single instance of the scaled group. Autoscaler keeps the number of instances proportional to the value of this metric. The metric itself does not change value due to group resizing.

A good metric to use with the target is for example or a custom metric exporting the total number of requests coming to your instances.

A bad example would be a metric exporting an average or median latency, since this value can't include a chunk assignable to a single instance, it could be better used with utilizationTarget instead.



Defines how target utilization value is expressed for a Stackdriver Monitoring metric. Either GAUGE, DELTA_PER_SECOND, or DELTA_PER_MINUTE.



Configuration parameters of autoscaling based on load balancer.



Fraction of backend capacity utilization (set in HTTP(S) load balancing configuration) that the autoscaler maintains. Must be a positive float value. If not defined, the default is 0.8.



Defines operating mode for this policy.


map (key: string, value: object)

Scaling schedules defined for an autoscaler. Multiple schedules can be set on an autoscaler, and they can overlap. During overlapping periods the greatest minRequiredReplicas of all scaling schedules is applied. Up to 128 scaling schedules are allowed.

An object containing a list of "key": value pairs. Example: { "name": "wrench", "mass": "1.3kg", "count": "3" }.



The minimum number of VM instances that the autoscaler will recommend in time intervals starting according to schedule. This field is required.



The start timestamps of time intervals when this scaling schedule is to provide a scaling signal. This field uses the extended cron format (with an optional year field). The expression can describe a single timestamp if the optional year is set, in which case the scaling schedule runs once. The schedule is interpreted with respect to timeZone. This field is required. Note: These timestamps only describe when autoscaler starts providing the scaling signal. The VMs need additional time to become serving.



The time zone to use when interpreting the schedule. The value of this field must be a time zone name from the tz database: This field is assigned a default value of “UTC” if left empty.



The duration of time intervals, in seconds, for which this scaling schedule is to run. The minimum allowed value is 300. This field is required.



A boolean value that specifies whether a scaling schedule can influence autoscaler recommendations. If set to true, then a scaling schedule has no effect. This field is optional, and its value is false by default.



A description of a scaling schedule.



[Output Only] URL of the zone where the instance group resides (for autoscalers living in zonal scope).



[Output Only] URL of the region where the instance group resides (for autoscalers living in regional scope).



[Output Only] The status of the autoscaler configuration. Current set of possible values:

  • PENDING: Autoscaler backend hasn't read new/updated configuration.
  • DELETING: Configuration is being deleted.
  • ACTIVE: Configuration is acknowledged to be effective. Some warnings might be present in the statusDetails field.
  • ERROR: Configuration has errors. Actionable for users. Details are present in the statusDetails field.
New values might be added in the future.



[Output Only] Human-readable details about the current state of the autoscaler. Read the documentation for Commonly returned status messages for examples of status messages you might encounter.