Reserving Compute Engine zonal resources

Create reservations for Virtual Machine (VM) instances in a specific zone, using custom or predefined machine types, with or without additional GPUs or local SSDs, to ensure resources are available for your workloads when you need them. After you create a reservation, you begin paying for the reserved resources immediately, and they remain available for your project to use indefinitely, until the reservation is deleted.

Use reservations to ensure that your project has resources for future increases in demand, including planned or unplanned spikes, migrating a large number of VMs, backup and disaster recovery, or for planned growth and buffer.

When you no longer need a reservation, delete the reservation to stop incurring charges for it. Because reservations consume resources just like normal VMs, each reservation is charged based on existing on-demand rates, which include sustained use discounts, and are eligible for committed use discounts.

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.

Benefits

Reservations provide the following benefits:

• Machines are always available, even if you don't use them immediately.

• No time constraints. Create a reservation at any time and delete it at any time to stop paying for it.

  Note: If the reservation is attached to a commitment, then the reservation cannot be deleted until the commitment contract has expired.

• Discounts. Because we bill for reservations in the same way and at the same rate as the resources that they are reserving, reserved resources qualify for committed- and sustained-use discounts.

Limitations and restrictions

Reservations have the following limitations and restrictions:

• Reservations apply only to Compute Engine, Dataproc, and Google Kubernetes Engine VM usage.
• Reservations don't apply to f1-micro or g1-small machine types, preemptible VMs, sole-tenant nodes, or other services not listed above, like Cloud SQL and Dataflow.
• You can reserve up to 1,000 VM instances per reservation.
• You must have sufficient quota in your project for the resources you are reserving. If the reservation is successful, quota for that resource is charged accordingly.
• Resources are allocated when the reservation is created. If there aren't enough resources in the zone at the time of the request, the reservation fails with an insufficient capacity error.
• When combined with a committed use discount:
  • For committed use discounted pricing for GPUs and local SSDs, a reservation must be created when purchasing the commitment.
  • If the reservation is attached to a commitment, the reservation cannot be deleted.
  • You can only buy a 1-year commitment on K80 GPUs.
• Reservations that aren't attached to a committed use discount can be held for any duration, but the same minimum charge of 1 minute applies as for regular instances.

How reservations work

Create reservations to reserve the VM instances you need. After you create a reservation, the reservation ensures that those resources are always available for you to use. During the creation process, you can choose how a reservation is to be used. For example, you can choose for a reservation be automatically applied to any new or existing instances that match the reservation's properties (the default behavior), or specify that only a specific VM instance can consume a reservation.

In all cases, a VM instance can only use a reservation if its properties exactly match the properties of the reservation. Specifically, for an instance to consume a reservation, the instance must match the reservation's properties for:

• Zone
• Machine type (cores and memory)
• Minimum CPU platform
• GPU type
• GPU count
• Local SSD interface
• Local SSD size

If a VM instance does match a reservation's properties, the default behavior is for all existing and new VM instances to consume the reservation automatically unless otherwise specified. For example, by default, if you create a reservation for 10 custom-8-10240 instances, and you already have 5 matching custom-8-10240 instances, then those 5 instances consume 5 of the reservations. If you create an additional 3 matching instances, then another 3 reservations are consumed.

You can override this default behaviour by specifying certain options when creating reservations and instances.

• To create a reservation that you don't want to be consumed automatically, use the specificReservationRequired option. These reservations can only be used if you explicitly specify the reservation during instance creation.
• To specify that you want an instance to consume a reservation, use the reservationAffinity option when creating the instance or in the instance template. You can specify that an instance automatically consumes from any matching reservation, doesn't automatically consume from matching reservations, or only consumes from a specific reservation.

If you stop or delete an instance that is using a reservation, the instance no longer counts against the reservation and the reserved resources are available again.

If you delete a reservation but don't delete the instances that are using the reserved resources, the instances persist and you pay for them as usual.

Creating a reservation

Create a reservation for VM instances using the console, the gcloud command-line tool, or the API. By default, any VM instances that match a reservation's properties automatically consume that reservation. To create a reservation that isn't consumed automatically, use the specificReservationRequired option.

Note: After a reservation is successfully created, you are charged for the resources until the reservation is deleted. If you delete a reservation but don't delete the instances that are using the reserved resources, the instance continues to incur charges. See the billing section for additional details.

You don't need to create a commitment in order to reserve resources. However, if you want committed use discount pricing on GPUs or local SSDs, you must create a reservation that includes those resources at the same time when purchasing your commitment. For more information, see Combining reservations with committed use discounts.

Permissions required for this task

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

• compute.reservations.create on the project

Console

1. In the Google Cloud Console, go to the Committed use discounts page.

   Go to Committed use discounts

2. Click Create reservation to create a standalone reservation that has no parent commitment.

   Note: For information about how to purchase a commitment with an attached reservation, which is required if you want committed use discount pricing on GPUs and local SSDs, see Combining reservations with committed use discounts.

3. Name your reservation.

4. If you only want to use this reservation's resources when creating matching instances that specifically target this reservation by name, under Use with VM instance, choose Select specific reservation.

5. Choose the Region and Zone where you want to reserve resources.

6. Specify the Number of VM instances that you want to reserve.

7. Specify the resources that you want to reserve for each instance:

   • If you have an instance template, click Use instance template and select an instance template from the list.
   • Otherwise, click Specify machine type.
     1. For predefined machine types, select what you need from the list.
     2. For custom machine types, including a minimum CPU platform, or to add GPUs, click Customize and make your selections.
     3. Optionally, specify the number of Local SSD disks that you want to add to each instance and specify the Interface type to use.

8. Click Create to create the reservation.

gcloud

Use the gcloud compute reservations create command to create a new reservation.

gcloud compute reservations create RESERVATION_NAME \
    --machine-type=MACHINE_TYPE \
    --min-cpu-platform MINIMUM_CPU_PLATFORM \
    --vm-count=NUMBER_OF_VMS \
    --accelerator=count=NUMBER_OF_ACCELERATORS,type=ACCELERATOR_TYPE \
    --local-ssd=size=SIZE_IN_GB,interface=INTERFACE \
    --require-specific-reservation \
    --zone=ZONE

Replace the following:

• RESERVATION_NAME: the name of the reservation to create.
• MACHINE_TYPE: the machine type. For example, n1-standard-1. For custom machine types, use the format custom-CPUS-MEMORY, replacing the following:
  • CPUS: the number of vCPUs.
  • MEMORY: the total memory for this instance. Memory must be a multiple of 256 MB and must be supplied in MB; for example, 5 GB of memory is 5120 MB: custom-4-5120. For a full list of restrictions, read the Specifications for custom machine types.
• MINIMUM_CPU_PLATFORM: the minimum CPU to use for each instance.
• NUMBER_OF_VMS: the quantity of VM instances to reserve.
• NUMBER_OF_ACCELERATORS: the number of GPUs to add, per instance.
• ACCELERATOR_TYPE: the type of accelerator.
• SIZE_IN_GB: the amount, in GB, of the local SSD to attach to each instance. The amount must be a multiple of 375.
• INTERFACE: the type of interface you want the local SSD for each instance to use. Valid options are: scsi and nvme.
• ZONE: the zone in which to reserve resources.

Optionally, add the --require-specific-reservation flag to indicate that only VM instances that explicitly target this reservation can use it. For more information about configuration options for consuming reserved resources, see How reservations work.

For example, to make a reservation in us-central1-a that can only be used when this reservation is specifically targeted, use a command similar to the following. This example reserves 10 custom machines, each with 8 Intel Haswell (or more recent) vCPUs, 10 GB of memory, 2 V100 GPUs, and 375 GB local SSD:

gcloud compute reservations create my-reservation \
    --machine-type=custom-8-10240 \
    --min-cpu-platform "Intel Haswell" \
    --vm-count=10 \
    --accelerator=count=2,type=nvidia-tesla-v100 \
    --local-ssd=size=375,interface=scsi \
    --require-specific-reservation \
    --zone us-central1-c

To see a list of all available flags, see the gcloud compute reservations create reference.

API

In the API, construct a POST request to the reservations.insert method. In the request body, include the following parameters:

POST https://compute.googleapis.com/compute/v1/projects/PROJECT_ID/zones/ZONE/reservations

{
  "name":"RESERVATION_NAME",
  "specificReservation":{
    "count":"NUMBER_OF_VMS",
    "instanceProperties":{
      "machineType":"MACHINE_TYPE",
      "minCpuPlatform": "MINIMUM_CPU_PLATFORM",
      "guestAccelerators":[
        {
          "acceleratorCount":"NUMBER_OF_ACCELERATORS",
          "acceleratorType":"ACCELERATOR_TYPE"
        }
      ],
      "localSsds":[
        {
          "diskSizeGb":"SIZE_IN_GB",
          "interface":"INTERFACE"
        }
      ]
    }
  }
}

Replace the following:

• PROJECT_ID: the project ID for the request.
• ZONE: the zone in which to reserve resources.
• RESERVATION_NAME: the name of the reservation to create.
• NUMBER_OF_VMS: the quantity of VM instances to reserve.
• MACHINE_TYPE: a predefined or custom machine type. For example, n1-standard-1. For custom machine types, use the format custom-CPUS-MEMORY, replacing the following:
  • CPUS: the number of vCPUs.
  • MEMORY: the total memory for this instance. Memory must be a multiple of 256 MB and must be supplied in MB; for example, 5 GB of memory is 5120 MB: custom-4-5120. For a full list of restrictions, read the Specifications for custom machine types.
• MINIMUM_CPU_PLATFORM: the minimum CPU to use for each instance.
• NUMBER_OF_ACCELERATORS: the number of GPUs to add, per instance.
• ACCELERATOR_TYPE: the type of accelerator.
• SIZE_IN_GB: the amount, in GB, of the local SSD to attach to each instance. The amount must be a multiple of 375.
• INTERFACE: the type of interface that you want the local SSD for each instance to use. Valid options are: scsi and nvme.

Optionally, include the specificReservationRequired Boolean field to specify that only VM instances that specifically target this reservation by name can use this reservation. For more information about configuration options for using the instances in your reservations, see How reservations work.

For example, to make a reservation in us-central1-a that can only be used when this reservation is specifically targeted, use a command similar to the following. This example reserves 10 custom machines, each with 8, Intel Haswell (or more recent) vCPUs, 10 GB of memory, 2 V100 GPUs, and a 375 GB local SSD:

POST https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central1-a/reservations

{
  "name": "reservation-1",
  "specificReservation":
  {
    "count": "10",
    "instanceProperties":
    {
      "machineType": "custom-8-10240",
      "minCpuPlatform": "Intel Haswell",
      "guestAccelerators":
      [
        {
          "acceleratorCount": 2,
          "acceleratorType": "nvidia-tesla-v100"
        }
      ],
      "localSsds":
      [
        {
          "diskSizeGb": "375",
          "interface": "SCSI"
        }
      ]
    }
  },
  "specificReservationRequired": true
}

Note: If you have instances running, but your reservation did not consume them, or if you have a reservation but cannot create matching instances, see Ensuring utilization of reservations.

Consuming reserved instances

When you create an instance, set the instance's reservation affinity flag to choose if the instance can use any available matching reservation (default), use a specific reservation, or use no reservation at all. For an instance to consume from a specific reservation, that reservation must be created with the corresponding specific reservation required flag.

To consume the reservation, the instances you create must match that reservation's instance properties. See How reservations work.

Consuming instances from any matching reservation

In this model, existing and new instances automatically count against the reservation if those instance's properties match the reserved instance's properties. This model is useful when you create and delete lots of VMs, and you want the reservations to be used whenever possible.

When creating your reservations, exclude the --require-specific-reservation flag, so that matching instances can consume from these reservations automatically.

Permissions required for this task

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

• compute.reservations.create on the project

• compute.instances.create on the project
• compute.instances.updateShieldedVmConfig if you plan to create a Shielded VM instance and you want to be able to change any of the Shielded VM settings.
• compute.networks.use on the project if using a legacy network
• compute.subnetworks.use either on the whole project or on the chosen subnet (VPC networks)
• compute.networks.useExternalIp on the project if you need to assign an external IP address (either ephemeral or static) to the instance using a legacy network
• compute.subnetworks.useExternalIp either on the whole project or on the chosen subnet if you need to assign an external IP address (either ephemeral or static) to the instance using a VPC network
• compute.addresses.use on the project if specifying a static address in the project
• compute.instances.setMetadata if setting metadata
• compute.instances.setTags on the instance if setting tags
• compute.instances.setLabels on the instance if setting labels
• compute.images.useReadOnly on the image if creating a new root persistent disk
• compute.disks.create on the project if creating a new root persistent disk with this instance
• compute.disks.useReadOnly on the disk if attaching an existing persistent disk in read-only mode
• compute.disks.use on the disk if attaching an existing disk in read-write mode
• compute.disks.setLabels on the disk if setting labels
• compute.snapshots.create on the project to create a new snapshot if creating an instance from a snapshot
• compute.snapshots.useReadOnly on the snapshot if creating an instance from a snapshot

Console

1. Create an open reservation.

   1. In the Cloud Console, go to the Committed use discounts page.

      Go to Committed use discounts

   2. To create a standalone reservation that has no parent commitment, click Create reservation.

      Note: For information about how to purchase a commitment with an attached reservation, which is required if you want committed use discount pricing on GPUs and local SSDs, see Combining reservations with committed use discounts.

   3. Name your reservation, for example: reservation-01.

   4. Under Use with VM instance, select Use reservation automatically, so that any matching instances can automatically use this reservation.

   5. Choose the Region and Zone where you want to reserve resources.

   6. Specify the Number of VM instances that you want to reserve.

   7. Specify the resources that you want each instance to have:

      • Number of vCPUs
      • Minimum CPU platform
      • Amount of memory
      • GPUs
      • Local SSD (as required)

   8. Click the Create button to create the reservation.

2. Create a VM instance that targets any open reservation and that matches the instance properties in reservation-01, including the zone, machine type (vCPU count and memory amount), minimum CPU platform, GPU amount and type, and local SSD interface and size.

   1. In the Cloud Console, go to the VM instances page.

      Go to VM instances

   2. Select your project and click Continue.

   3. Click Create instance.

   4. Specify a Name for your instance.

   5. Specify a Machine type that matches the properties of reservation-01. For example, if you specified a minimum CPU platform or GPUs in the reservation, then, to consume the reservation, you must customize this instance to match the reservation.

   6. If your reservation includes local SSDs, under Disks, click Add new disk to add local SSDs that match the reservation's local SSD interface and amount.

   7. Under Management, under Reservations, select Automatically use created reservation.

   8. Click Create.

gcloud

1. Create an open reservation called reservation-01.

   gcloud compute reservations create reservation-01 \
       --machine-type=n1-standard-32 \
       --min-cpu-platform "Intel Skylake" \
       --vm-count=2 \
       --zone=us-central1-a
   

2. Create a VM instance that targets any open reservation and that matches the instance properties in reservation-01, including the zone, machine type (vCPU count and memory amount), minimum CPU platform, GPU amount and type, and local SSD interface and size.

   gcloud compute instances create instance-1 \
       --machine-type=n1-standard-32 \
       --min-cpu-platform "Intel Skylake" \
       --zone=us-central1-a \
       --reservation-affinity=any
   

API

1. Create an open reservation named reservation-01.

   POST https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central1-a/reservations
   
   {
     "name":"reservation-01",
     "specificReservation":{
       "count":"2",
       "instanceProperties":{
         "machineType":"n1-standard-32",
         "minCpuPlatform": "Intel Haswell",
       }
     },
     "specificReservationRequired": false
   }
   

2. Create a VM instance that targets any open reservation and that matches the instance properties in reservation-01, including the zone, machine type (vCPU count and memory amount), minimum CPU platform, GPU amount and type, and local SSD interface and size.

   POST https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central1-a/instances
   
   {
     "name": "instance-1",
     "machineType": "zones/us-central1-a/machineTypes/n1-standard-32",
     "minCpuPlatform": "Intel Haswell",
     "reservationAffinity":
     {
       "consumeReservationType": "ANY_RESERVATION"
     },
     ...
   }
   

Consuming instances from a specific reservation

In this model, only new instances that target a specific reservation by name consume that reservation. This model is useful, for example, when you want to hold a certain amount of capacity as backup for special events.

When creating the reservation, specify that the reservation should not be used automatically. Then create VMs that specifically target that reservation and that match its properties.

If you are using an instance template to create VMs in a regional managed instance group, create identical reservations with the same name in each applicable zone. Then, target the reservations by name in the group's instance template.

For example, create a specific reservation named reservation-02, and then create a matching instance that targets that reservation and that matches the reservation's instance properties.

Console

1. Create a reservation with resources that can only be used by instances that specifically target this reservation by name.

   1. In the Cloud Console, go to the Committed use discounts page.

      Go to Committed use discounts

   2. Click Create reservation to create a standalone reservation that has no parent commitment.

      Note: For information about how to purchase a commitment with an attached reservation, which is required if you want committed use discount pricing on GPUs and local SSDs, see Combining reservations with committed use discounts.

   3. Name your reservation, for example: reservation-02.

   4. Under Use with VM instance, check Select specific reservation, so that only matching instances that specifically target this reservation by name can use this reservation.

   5. Choose the Region and Zone where you want to reserve resources.

   6. Specify the Number of VM instances that you want to reserve.

   7. Specify the resources that you want each instance to have:

      • Number of vCPUs
      • Minimum CPU platform
      • Amount of memory
      • GPUs
      • Local SSD

   8. Click Create.

2. Create a VM instance that targets that specific reservation by name.

   Ensure that the instance's properties match the instance properties of that specific reservation, including the zone, machine type (vCPU count and memory amount), minimum CPU platform, GPU amount and type, and local SSD interface and size.

   1. In the Cloud Console, go to the VM instances page.

      Go to VM instances

   2. Select your project and click Continue.

   3. Click the Create instance button.

   4. Specify a Name for your instance.

   5. Specify a Machine type that matches the properties of reservation-01. For example, if you specified a minimum CPU platform in the reservation, then, to consume the reservation, you must customize this instance to match the reservation.

   6. If your reservation includes local SSDs, under Disks, click Add new disk to add local SSDs that match the reservation's local SSD interface and amount.

   7. Under Management, under Reservations, select an available reservation with matching properties, for example: reservation-02.

   8. Click Create to create the instance.

gcloud

1. Create a reservation named reservation-02 with the --require-specific-reservation flag. These reserved resources can be used only by instances that specifically target this reservation by name.

   gcloud compute reservations create reservation-02 \
       --machine-type=n1-standard-32 \
       --min-cpu-platform "Intel Skylake" \
       --vm-count=10 \
       --zone=us-central1-a \
       --require-specific-reservation
   

2. Create a VM instance that targets reservation-02 by name by using the --reservation-affinity and --reservation flags.

   Ensure that the instance's properties match the reservation's instance properties, including the zone, machine type (vCPU count and memory amount), minimum CPU platform, GPU amount and type, and local SSD interface and size.

   gcloud compute instances create instance-2 \
       --machine-type=n1-standard-32 \
       --min-cpu-platform "Intel Skylake" \
       --zone=us-central1-a \
       --reservation-affinity=specific \
       --reservation=reservation-02
   

API

1. Create a reservation named reservation-02 with the specificReservationRequired field set to true.

   POST https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central1-a/reservations
   
   {
     "name":"reservation-02",
     "specificReservation":{
       "count":"10",
       "instanceProperties":{
         "machineType":"n1-standard-32",
         "minCpuPlatform": "Intel Haswell",
       }
     },
     "specificReservationRequired": true
   }

2. Create a VM instance that targets reservation-02 by name by using the reservationAffinity field.

   Ensure that the instance's properties match the reservation's instance properties, including the zone, machine type (vCPU count and memory amount), minimum CPU platform, GPU amount and type, and local SSD interface and size.

   POST https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central1-a/instances
   
   {
     "name": "instance-2",
     "machineType": "zones/us-central1-a/machineTypes/n1-standard-32",
     "minCpuPlatform": "Intel Haswell",
     "reservationAffinity":
     {
       "consumeReservationType": "SPECIFIC_RESERVATION",
       "key": "compute.googleapis.com/reservation-name",
       "values":
       ["reservation-02"
       ]
     },
     ...
   }

Creating instances without consuming reservations

To explicitly avoid consuming resources from any reservation, set the reservation affinity flag to "none" when creating the VM. Or, to implicitly avoid using up any of your reservations, create an instance that doesn't match the instance properties of any of your reservations.

Console

1. Create an instance that explicitly doesn't consume from a reservation.

   1. In the Cloud Console, go to the VM instances page.

      Go to VM instances

   2. Click Create instance.

   3. Create an instance as usual.

   4. Under Management, under Reservations, select Don't use to ensure that this instance does not consume from any existing reservations.

   5. Click Create.

gcloud

Create an instance that explicitly does not consume from a reservation.

gcloud compute instances create instance-3 --reservation-affinity=none

Or create an instance that does not match the instance properties in any of your reservations. To review your reservation properties, see listing and describing reservations.

API

Create an instance that explicitly does not consume from a reservation.

POST https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central1-a/instances

{
  "machineType": "zones/us-central1-a/machineTypes/n1-standard-16",
  "name": "instance-3",
  "reservationAffinity":
  {
    "consumeReservationType": "NO_RESERVATION"
  },
  ...
}

Or create an instance that does not match the instance properties in any of your reservations. To review your reservation properties, see listing and describing reservations.

Ensuring utilization of reservations

If you aren't able to consume your reserved resources, review the reservation's properties and the instance's properties. To test that the instance's properties match the reservation's properties, create a specific reservation, and then create an instance that targets that specific reservation.

Verify reservation properties

The reservation's instance properties must match the properties in your instance creation command. To verify:

1. Check the reservation's properties specifically for:

   • zone
   • specificReservationRequired
     • If set to true, your instance creation command must target this reservation by name.
   • machineType
   • guestAccelerators.acceleratorType
   • guestAccelerators.acceleratorCount
   • minCpuPlatform
     • If you have a VM instance with a minCpuPlatform, and you want to create a reservation that uses that instance, create the reservation with matching properties, including a matching minCpuPlatform.
     • If you have a reservation with a minCpuPlatform, and you want to create an instance that uses that reservation, create the instance with matching properties, including a matching minCpuPlatform.
   • localSsds.diskSizeGb
   • localSsds.interface

2. Create an instance that matches those properties.

3. Describe the instance to verify that its properties match the reservation's. For example, with the gcloud tool, use the instances describe command.

4. Describe your reservation to check that its inUseCount has incremented.

   • If the inUseCount has not changed, the properties are mismatched.
   • If the inUseCount has incremented, your reservation is being used.

Target a specific reservation

If you attempt to create an instance that targets a specific reservation, the instance cannot be created unless it matches the reservation's properties. Use this method to verify that you have the correct instance properties.

1. Create a reservation with the specificReservationRequired option.
2. Create an instance that explicitly matches the reservation's instance properties and that targets the reservation by name through the reservation affinity property.
   • If instance properties are mismatched, you see an error.
   • If instance properties match, your reservation is being used.

Listing and describing reservations

List and view details about your reservation using the console, the gcloud tool, or the API.

Permissions required for this task

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

• compute.reservations.list on the project to list reservations
• compute.reservations.get on the project or on the reservation, to describe a reservation

Console

1. In the Cloud Console, go to the Committed use discounts page.

   Go to Committed use discounts

2. Click Reservations.

The list of reservations includes details about each reservation. You can use the Machines used (total) column to determine the utilization of each reservation.

gcloud

List your reservations with the gcloud compute reservations list command:

gcloud compute reservations list [--filter="zone:('ZONE')"]

NAME             IN_USE_COUNT   COUNT   ZONE
reservation-04   25             50      us-central1-a
reservation-05   0              100     us-central1-b

Describe your reservations with the gcloud compute reservations describe command:

gcloud compute reservations describe RESERVATION_NAME --zone=ZONE

kind: compute#reservation
name: reservation-05
selfLink: https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-east1-d/reservations/reservation-05
specificReservation:
  count: '50'
  inUseCount: '25'
  instanceProperties:
    guestAccelerators:
    - acceleratorCount: 1
      acceleratorType: nvidia-tesla-k80
    localSsds:
    - diskSizeGb: '375'
      interface: SCSI
    machineType: n1-standard-1
    minCpuPlatform: Any CPU Platform
specificReservationRequired: false
status: READY
zone: https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-east1-d    

You can use the count and inUseCount values to determine the utilization of each reservation. In this example, 50 instances are reserved, and 25 of them are currently used.

To create an instance that uses this reservation, remember to match the reservation's instance properties. For example:

gcloud compute instances create my-instance \
    --accelerator=type=nvidia-tesla-k80,count=1 \
    --maintenance-policy terminate \
    --local-ssd=interface=SCSI \
    --machine-type n1-standard-1 \
    --zone us-east1-d 

API

In the API, list your reservations by making a GET request to the reservations.list method.

GET https://compute.googleapis.com/compute/v1/projects/PROJECT_ID/zones/ZONE/reservations

Describe a reservation by calling the reservations.get method.

GET https://compute.googleapis.com/compute/v1/projects/PROJECT_ID/zones/ZONE/reservations/RESERVATION_NAME

{
  "id": "2533514314332214789",
  "creationTimestamp": "2019-09-27T08:21:14.707-07:00",
  "selfLink": "https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-east1-d/reservations/reservation-05",
  "zone": "https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-east1-d",
  "name": "reservation-05",
  "specificReservationRequired": false,
  "status": "READY",
  "kind": "compute#reservation",
  "specificReservation": {
    "instanceProperties": {
      "machineType": "n1-standard-1",
      "guestAccelerators": [
        {
          "acceleratorType": "nvidia-tesla-k80",
          "acceleratorCount": 1
        }
      ],
      "minCpuPlatform": "Any CPU Platform",
      "localSsds": [
        {
          "diskSizeGb": "375",
          "interface": "SCSI"
        }
      ]
    },
    "count": "50",
    "inUseCount": "25"
  }
}

You can use the count and inUseCount values to determine the utilization of each reservation. In this example 50 instances are reserved, and 25 of them are currently used.

Modifying reservations

You can resize or delete a reservation if it is not attached to a commitment.

Resizing a reservation

You can resize the number of VMs in a reservation that are not tied to a commitment by using the gcloud tool or the API.

Permissions required for this task

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

• compute.reservations.resize on the reservation

gcloud

Resize your reservation by using the gcloud compute reservations update command. For example:

Create a reservation for two VM instances:

gcloud compute reservations create reservation-01 \
    --machine-type=n1-standard-32 \
    --zone=us-central1-a \
    --vm-count=2

Resize the reservation from two to ten VM instances:

gcloud compute reservations update reservation-01 \
    --zone=us-central1-a \
    --vm-count=10

API

In the API, construct a POST request to the reservations.resize method and include the new specificSkuCount in the request body. The following request body updates the reservation's VM count to 10.

POST https://compute.googleapis.com/compute/beta/projects/PROJECT_ID/zones/ZONE/reservations/RESERVATION_NAME/resize

{"specificSkuCount": "10"}

The update request succeeds if, at the time of the request, there are enough resources in the target zone and sufficient quota in the target region.

Deleting a reservation

You can delete reservations that aren't tied to a commitment by using the console, the gcloud tool, or the API.

Permissions required for this task

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

• compute.reservations.delete on the reservation

Console

1. In the Cloud Console, go to the Committed use discounts page.

   Go to Committed use discounts

2. Click Reservations to see a list of your reservations.

3. Select the checkbox next to each reservation that you want to delete.

   Note: If the reservation has a parent commitment, it cannot be deleted.

4. Click Delete reservation.

gcloud

You can release reservations by using the delete command:

gcloud compute reservations delete RESERVATION_NAME --zone ZONE

API

In the API, construct a DELETE request to the reservation.delete method.

DELETE https://compute.googleapis.com/compute/v1/projects/PROJECT_ID/zones/ZONE/reservations/RESERVATION_NAME

After the delete command completes successfully, you are no longer charged for the reservation, and the resources are no longer reserved for you. Deleting a reservation has no effect on any running instances that were tied to that reservation; you are still charged for those instances.

Combining reservations with Committed use discounts

A committed use discount provides a 1- or 3-year discounted price agreement, but it does not reserve capacity in a specific zone. A reservation ensures that capacity is held in a specific zone even if the reserved VMs are not running. By combining a reservation with a commitment, you can get discounted, reserved resources.

By default, when you create a reservation, any applicable committed use discounts for cores and memory are automatically applied to your VM instances. However, to get committed use discount rates for GPUs and local SSDs, you must create a reservation for those resources when purchasing the commitment.

After you purchase a commitment, it isn't possible to cancel it. For more information, see Committed use discounts.

Purchasing a commitment for GPUs or local SSDs

To purchase a commitment for GPUs or local SSDs:

• You must purchase a general-purpose N1 commitment. GPUs are only supported with N1 machine types.
• You must create a reservation that includes either GPUs or local SSDs at the same time when you are purchasing your commitment.
• You do not need to commit to vCPUs or memory.

You must purchase commitments for specific GPU types. For example, you might purchase either Tesla P100s or Tesla V100s, but you cannot purchase commitments for Tesla P100 GPUs and expect to apply them to other GPU types.

The amount of GPUs and local SSDs that you reserve must be equal to the amount that you commit to. For example, if you want to reserve 4 V100 GPUs, then you must also commit to 4 V100 GPUs. However, the amount of vCPU and memory that you reserve can be more or less than what you commit to.

If you only want to commit to GPUs or local SSDs, you can specify 0 for vCPU and memory commitments. But the reservation that you attach to your commitment must contain the same GPUs and local SSDs as the commitment, as well as the machine types (with vCPUs and memory) that you want to reserve.

Note: You do not need to specify a commitment type because the commitment automatically defaults to general purpose. You can only purchase GPUs and local SSDs for general-purpose commitments.

Purchase a commitment with an attached reservation using the Cloud Console, the gcloud tool, or the API.

Console

1. In the Google Cloud Console, go to the Committed use discounts page.

   Go to Committed use discounts

2. To purchase a new commitment, click Purchase commitment.

3. Name your commitment, and choose the region where you want it to apply.

4. For Commitment type, select General Purpose.

5. Choose a commitment Duration of 1 or 3 years.

6. Enter the quantity of Cores and Memory that you want to purchase. If you only want to commit to and reserve GPUs or local SSDs without committing to vCPUs and memory, specify 0 for the vCPU and memory commitment quantities.

7. Click Add GPUs and select the GPU type and Number of GPUs that you want to commit to.

8. Click Add local SSD and specify the number of disks that you want to commit to.

9. Click Add new reservation to create one or more reservations for instances that will use the GPUs and local SSDs.

   1. Name your reservation.
   2. Under Use with VM instance:
      • If you want to use this reservation's resources only when creating matching instances that specifically target this reservation by name, select Select specific reservation.
      • If you want matching instances to automatically use this reservation, select Use reservation automatically.
   3. Choose the Zone where you want to reserve resources.
   4. Specify the Number of VM instances that you want to reserve.
   5. Specify the resources that you want to reserve for each instance:
      • If you have an instance template, click Use instance template and select an instance template from the list.
      • Otherwise, click Specify machine type.
        1. Click Customize to use the slider to select the Cores and Memory for your machine type.
        2. Specify the CPU platform.
        3. To add GPUs, enter the number of GPUs and the GPU type.
        4. Optionally, specify the number of Local SSDs that you want to add to each instance and specify the Interface type to use.
   6. Click the Done button to create the reservation.
   Note: The amount of GPUs and local SSDs that you reserve must be equal to the amount that you commit to.

10. Click the Purchase button to purchase the commitment.

11. Read the terms of purchase and click Purchase when you are ready to purchase the commitment.

gcloud

Use the gcloud compute commmitments create command to purchase a commitment, and include flags to create an attached reservation.

For example, the following commitment includes 4 GPUs and a new reservation for those 4 GPUs to be used across 2 n1-standard-32 instances in us-central1-a.

gcloud compute commitments create commitment-01 \
    --region=us-central1 \
    --resources=vcpu=96,memory=624GB \
    --resources-accelerator=type=nvidia-tesla-v100,count=4 \
    --plan 12-month \
    --reservation=reservation-01 \
    --reservation-zone=us-central1-a \
    --machine-type=n1-standard-16 \
    --accelerator=type=nvidia-tesla-v100,count=2 \
    --vm-count=2

If you want to commit to and reserve GPUs or local SSDs without committing to vCPUs and memory, specify 0 for the vCPU and memory commitment quantities.

gcloud compute commitments create commitment-01 \
    --region=us-west2 \
    --resources=vcpu=0,memory=0 \
    --resources-accelerator=type=nvidia-tesla-p4,count=1 \
    --plan 12-month \
    --reservation=reservation-01 \
    --reservation-zone=us-west2-b \
    --machine-type=n1-standard-1 \
    --accelerator=type=nvidia-tesla-p4,count=1 \
    --vm-count=1

To create multiple reservations when purchasing a commitment, use a YAML file that contains the reservation properties. For example:

gcloud compute commitments create commitment-01 \
    --region=us-central1 \
    --resources=vcpu=96,memory=624,local-ssd=750 \
    --resources-accelerator=type=nvidia-tesla-v100,count=1 \
    --plan 12-month \
    --reservations-from-file=YAML_FILE

For example, the following YAML file contains 2 reservations. The first reservation, res-01, contains 1 n1-standard-1 instance with 1 GPU, and it is a targeted reservation, which means that you must specifically target that reservation by name to use its reserved instances. The second reservation, res-02, contains 1 n1-standard-1 VM instance with 2 types of attached local SSDs.

‐ reservation: res-01
  reservation_zone: us-central1-a
  require_specific_reservation: true
  vm_count: 1
  machine_type: n1-standard-1
  accelerator:
  ‐ count: 1
    type: nvidia-tesla-v100
‐ reservation: res-02
  reservation_zone: us-central1-a
  vm_count: 1
  machine_type: n1-standard-1
  local_ssd:
  ‐ interface: scsi
    size: 375
  ‐ interface: nvme
    size: 375

API

Use the regionCommitments.insert method and include the reservations field to define the reservation's properties. For example, the following commitment includes 4 GPUs and a reservation for those 4 GPUs to be used across 2 instances in us-central1-a.

POST https://compute.googleapis.com/compute/v1/projects/PROJECT_ID/regions/REGION/commitments

{
  "name": "commitment-01",
  "plan": "TWELVE_MONTH",
  "resources":
  [
    {
      "amount": "96",
      "type": "VCPU"
    },
    {
      "amount": "638976",
      "type": "MEMORY"
    },
    {
      "acceleratorType": "nvidia-tesla-v100",
      "amount": "4",
      "type": "ACCELERATOR"
    }
  ],
  "reservations":
  [
    {
      "name": "reservation-01",
      "specificReservation":
      {
        "count": "2",
        "instanceProperties":
        {
          "guestAccelerators":
          [
            {
              "acceleratorCount": 2,
              "acceleratorType": "nvidia-tesla-v100"
            }
          ],
          "machineType": "n1-standard-8"
        }
      },
      "specificReservationRequired": false,
      "zone": "us-central1-a"
    }
  ]
}

If you only want to commit to and reserve GPUs or local SSDs without committing to vCPUs and memory, specify 0 for the vCPU and memory commitment quantities. For example:

POST https://compute.googleapis.com/compute/v1/projects/PROJECT_ID/regions/REGION/commitments

{
  "name": "commitment-01",
  "plan": "TWELVE_MONTH",
  "resources":
  [
    {
      "amount": "0",
      "type": "VCPU"
    },
    {
      "amount": "0",
      "type": "MEMORY"
    },
    {
      "acceleratorType": "nvidia-tesla-v100",
      "amount": "4",
      "type": "ACCELERATOR"
    }
  ],
  "reservations":
  [
    {
      "name": "reservation-01",
      "specificReservation":
      {
        "count": "2",
        "instanceProperties":
        {
          "guestAccelerators":
          [
            {
              "acceleratorCount": 2,
              "acceleratorType": "nvidia-tesla-v100"
            }
          ],
          "machineType": "n1-standard-8"
        }
      },
      "specificReservationRequired": false,
      "zone": "us-central1-a"
    }
  ]
}

The commitment is successfully created only if there are enough resources in the target zone, and sufficient quota, at the time of the request.

When you create a commitment with an attached reservation, you cannot delete the reservation for the duration of the commitment. When your commitment expires, Compute Engine automatically deletes any attached reservations.

If you need to transfer GPUs or local SSDs across committed reservations, see Modifying reservations that are attached to commitments.

Modifying reservations that are attached to commitments

If a commitment has one or more reservations with GPUs or local SSDs, you can use the gcloud tool or the API to transfer GPUs or local SSDs across those reservations. For example, you might move GPUs from one reservation to a second, new reservation.

Limitations:

• You can only move resources between two reservations, where one is the source reservation and the other is the destination. The source reservation must already exist; the destination can be an existing or new reservation.
• The source and destination reservations must be in the same zone.
• You cannot modify the resource properties in existing reservations; you can only transfer existing resources from one reservation to another.
• The total amount of GPUs and local SSDs must remain constant.
• You can only change 100 VMs per request. If you want to update more, call the API multiple times or reach out to Google Cloud Support.

• You cannot transfer vCPUs or memory or modify other properties of attached reservations.

gcloud

Use the gcloud beta compute commitments update-reservations command.

Create a commitment with one reservation:

• res-1 has 4 n1-standard-1 instances, each with 1 P100 GPU (4 vCPUs and 4 P100 GPUs). The total reserved resources includes 4 P100 GPUs.

gcloud beta compute commitments create my-commitment-with-reservations \
    --region=asia-east1 \
    --resources=vcpu=10,memory=32GB \
    --resources-accelerator=type=nvidia-tesla-p100,count=4 \
    --plan 12-month \
    --reservations-from-file=one-reservation.yaml

The one-reservation.yaml file contains:

- reservation: res-1
  reservation_zone: asia-east1-a
  require_specific_reservation: true
  vm_count: 4
  machine_type: n1-standard-1
  accelerator:
  - count: 1
    type: nvidia-tesla-p100

Update the commitment to transfer some resources from res-1 to a new, second reservation, res-2. In this example:

• Modify res-1 to have 2 n1-standard-1 instances, each with 2 P100 GPUs (2 vCPU and 2 P100 GPUs).
• Add res-2 to have 2 n1-standard-2 instances each with 1 P100 GPU (4 vCPUs and 2 P100 GPUs).

The total number of reserved GPUs within the commitment remains constant at 4.

gcloud compute beta commitments update-reservations my-commitment-with-reservations \
    --region=us-central1 \
    --reservations-from-file=two-reservations.yaml

The two-reservations.yaml file contains:

- reservation: res-1
  reservation_zone: asia-east1-a
  require_specific_reservation: true
  vm_count: 2
  machine_type: n1-standard-1
  accelerator:
  - count: 1
    type: nvidia-tesla-p100
- reservation: res-2
  reservation_zone: asia-east1-a
  require_specific_reservation: true
  vm_count: 2
  machine_type: n1-standard-2
  accelerator:
  - count: 1
    type: nvidia-tesla-p100

API

Use the regionCommitments.updateReservations method.

For example, create a commitment with the following reservation:

• res-1 has 4 n1-standard-1 instances each with 1 nvidia-tesla-p100 GPU (4 vCPUs and 4 P100 GPUs).

In this example, the total reserved resources includes 4 P100 GPUs.

POST https://compute.googleapis.com/compute/beta/projects/my-project/regions/asia-east1/commitments/my-commitment-with-reservation/updateReservations

{
  "name": "my-commitment-with-reservation",
  "plan": "TWELVE_MONTH",
  "reservations":
  [
    {
      "name": "res-1",
      "specificReservation":
      {
        "count": "4",
        "instanceProperties":
        {
          "guestAccelerators":
          [
            {
              "acceleratorCount": 1,
              "acceleratorType": "nvidia-tesla-p100"
            }
          ],
          "machineType": "n1-standard-1"
        }
      },
      "specificReservationRequired": true,
      "zone": "asia-east1-a"
    }
  ],
  "resources":
  [
    {
      "amount": "10",
      "type": "VCPU"
    },
    {
      "amount": "32768",
      "type": "MEMORY"
    },
    {
      "acceleratorType": "nvidia-tesla-p100",
      "amount": "4",
      "type": "ACCELERATOR"
    }
  ]
}

Update the commitment to transfer some resources from res-1 to a new, second reservation, res-2. In this example:

• Modify res-1 to have 2 n1-standard-1 instances, each with 2 P100 GPUs (2 vCPU and 2 P100 GPUs).
• Add res-2 to have 2 n1-standard-2 instances each with 1 P100 GPU (4 vCPUs and 2 P100 GPUs).

The total number of reserved GPUs within the commitment remains constant at 4.

POST https://compute.googleapis.com/compute/beta/projects/my-project/regions/asia-east1/commitments/my-commitment-with-reservation/updateReservations

{
  "reservations":
  [
    {
      "name": "res-1",
      "specificReservation":
      {
        "count": "2",
        "instanceProperties":
        {
          "guestAccelerators":
          [
            {
              "acceleratorCount": 1,
              "acceleratorType": "nvidia-tesla-p100"
            }
          ],
          "machineType": "n1-standard-1"
        }
      },
      "specificReservationRequired": true,
      "zone": "asia-east1-a"
    },
    {
      "name": "res-2",
      "specificReservation":
      {
        "count": "2",
        "instanceProperties":
        {
          "guestAccelerators":
          [
            {
              "acceleratorCount": 1,
              "acceleratorType": "nvidia-tesla-p100"
            }
          ],
          "machineType": "n1-standard-2"
        }
      },
      "specificReservationRequired": true,
      "zone": "asia-east1-a"
    }
  ]
}

Viewing reservation usage reports

Export detailed reports of your Compute Engine usage to a Cloud Storage bucket using the usage export feature. See Viewing usage reports for instructions.

The usage report shows:

• Reserved resources that are in use. These entries appear as normal vCPU, memory, GPU, and local SSD resources.
• Reserved resources that are not in use. These entries have normal SKU names and reservation resource URIs.
• Total reserved resources. These entries have reservation SKU names and reservation resource URIs. There are no costs associated with these entries. Use these entries to calculate how much of your reservations you are using.

Measurement	MeasurementId format	Resource URI format
Reserved resources that are in use	com.google.cloud/services/compute-engine/SKU_NAME	https://compute.googleapis.com/compute/v1/projects/PROJECT_ID/zones/ZONE/RESOURCE_TYPE/RESOURCE_NAME. For example, https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central1-a/instances/my-instance
Reserved resources that are not in use	com.google.cloud/services/compute-engine/SKU_NAME	https://compute.googleapis.com/compute/v1/projects/PROJECT_ID/zones/ZONE/reservations/RESERVATION_NAME. For example, https://compute.googleapis.com/compute/v1/projects/my-project/zones/europe-west1-a/reservations/my-reservation
Total reserved resources	com.google.cloud/services/compute-engine/ReservationSKU_NAME	https://compute.googleapis.com/compute/v1/projects/PROJECT_ID/zones/ZONE/reservations/RESERVATION_NAME. For example, https://compute.googleapis.com/compute/v1/projects/my-project/zones/europe-west1-a/reservations/my-reservation

For example, in the following snippet from a usage report for a reservation named my-reservation:

• Row 1 shows reserved RAM that is currently in use. The row's ResourceId shows that this RAM is used by an instance named my-instance.
• Row 2 shows reserved RAM that is not in use. The row's ResourceId shows that this reserved RAM is held by my-reservation; it is not yet used by any instance.
• Row 3 shows the reservation's total reserved RAM.

Report Date,MeasurementId,Quantity,Unit,Resource URI,ResourceId,Location
2019-06-06,com.google.cloud/services/compute-engine/VmimageN1StandardRam,166970074857472,byte-seconds,https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central2-a/instances/my-instance,1775485842510981624,us-central2-a
2019-06-06,com.google.cloud/services/compute-engine/VmimageN1StandardRam,166970074857472,byte-seconds,https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central2-a/reservations/my-reservation,7.58809E+17,us-central2-a
2019-06-06,com.google.cloud/services/compute-engine/ReservationN1StandardRam,333940149714944,byte-seconds,https://compute.googleapis.com/compute/v1/projects/my-project/zones/us-central2-a/reservations/my-reservation,7.58809E+17,us-central2-a
...

Reservation billing

Reservations are billed at the same rate as the resources that they are reserving. Sustained use discounts, committed use discounts and custom pricing apply as they would for regular VMs.

Take the following example:

• You have a 3-vCPU commitment in us-central1
• You are running 5 vCPUs in us-central1-a
• You have a 10-vCPU reservation in us-central1-a

You will be billed as follows:

Covered by	# of vCPUs
Committed use discount price	3
On-demand price (2 vCPU used reservations + 5 vCPU unused reservations)	7

What's next

• Learn more about committed use discounts