gcloud alpha compute instances set-scheduling

NAME
gcloud alpha compute instances set-scheduling - set scheduling options for Compute Engine virtual machines
SYNOPSIS
gcloud alpha compute instances set-scheduling INSTANCE_NAME [--clear-min-node-cpu] [--[no-]graceful-shutdown] [--graceful-shutdown-max-duration=GRACEFUL_SHUTDOWN_MAX_DURATION] [--host-error-timeout-seconds=HOST_ERROR_TIMEOUT_SECONDS] [--local-ssd-recovery-timeout=LOCAL_SSD_RECOVERY_TIMEOUT] [--min-node-cpu=MIN_NODE_CPU] [--[no-]preemptible] [--provisioning-model=PROVISIONING_MODEL] [--[no-]restart-on-failure] [--zone=ZONE] [--clear-discard-local-ssds-at-termination-timestamp     | --discard-local-ssds-at-termination-timestamp=DISCARD_LOCAL_SSDS_AT_TERMINATION_TIMESTAMP] [--clear-instance-termination-action     | --instance-termination-action=INSTANCE_TERMINATION_ACTION] [--clear-max-run-duration     | --max-run-duration=MAX_RUN_DURATION] [--clear-node-affinities     | --node=NODE     | --node-affinity-file=PATH_TO_FILE     | --node-group=NODE_GROUP] [--clear-termination-time     | --termination-time=TERMINATION_TIME] [--maintenance-policy=MAINTENANCE_POLICY     | --on-host-maintenance=MAINTENANCE_POLICY] [GCLOUD_WIDE_FLAG]
DESCRIPTION
(ALPHA) $gcloud alpha compute instances set-scheduling is used to update scheduling options for VM instances. You can only call this method on a VM instance that is stopped (a VM instance in a TERMINATED state).
EXAMPLES
To set instance to be terminated during maintenance, run:
gcloud alpha compute instances set-scheduling example-instance --maintenance-policy=TERMINATE --zone=us-central1-b
POSITIONAL ARGUMENTS
INSTANCE_NAME
Name of the instance to operate on. For details on valid instance names, refer to the criteria documented under the field 'name' at: https://cloud.google.com/compute/docs/reference/rest/v1/instances
FLAGS
--clear-min-node-cpu
Removes the min-node-cpu field from the instance. If specified, the instance min-node-cpu will be cleared. The instance will not be overcommitted and utilize the full CPU count assigned.
--[no-]graceful-shutdown
If set to true, enables graceful shutdown for the instance. Use --graceful-shutdown to enable and --no-graceful-shutdown to disable.
--graceful-shutdown-max-duration=GRACEFUL_SHUTDOWN_MAX_DURATION
Specifies time needed to gracefully shutdown the instance. After that time, the instance goes to STOPPING even if graceful shutdown is not completed.e.g. 300s, 1h. See $ gcloud topic datetimes for information on duration formats.
--host-error-timeout-seconds=HOST_ERROR_TIMEOUT_SECONDS
The timeout in seconds for host error detection. The value must be set with 30 second increments, with a range of 90 to 330 seconds. If unset, the default behavior of the host error recovery is used.
--local-ssd-recovery-timeout=LOCAL_SSD_RECOVERY_TIMEOUT
Specifies the maximum amount of time a Local Ssd Vm should wait while recovery of the Local Ssd state is attempted. Its value should be in between 0 and 168 hours with hour granularity and the default value being 1 hour.
--min-node-cpu=MIN_NODE_CPU
Minimum number of virtual CPUs this instance will consume when running on a sole-tenant node.
--[no-]preemptible
If provided, instances will be preemptible and time-limited. Instances might be preempted to free up resources for standard VM instances, and will only be able to run for a limited amount of time. Preemptible instances can not be restarted and will not migrate. Use --preemptible to enable and --no-preemptible to disable.
--provisioning-model=PROVISIONING_MODEL
Specifies provisioning model, which determines price, obtainability, and runtime for the VM instance. PROVISIONING_MODEL must be one of:
SPOT
Spot VMs are spare capacity; Spot VMs are discounted to have much lower prices than standard VMs but have no guaranteed runtime. Spot VMs are the new version of preemptible VM instances, except Spot VMs do not have a 24-hour maximum runtime.
STANDARD
Default. Standard provisioning model for VM instances, which has user-controlled runtime but no Spot discounts.
--[no-]restart-on-failure
The instances will be restarted if they are terminated by Compute Engine. This does not affect terminations performed by the user. This option is mutually exclusive with --preemptible. Use --restart-on-failure to enable and --no-restart-on-failure to disable.
--zone=ZONE
Zone of the instance to operate on. If not specified, you might be prompted to select a zone (interactive mode only). gcloud attempts to identify the appropriate zone by searching for resources in your currently active project. If the zone cannot be determined, gcloud prompts you for a selection with all available Google Cloud Platform zones.

To avoid prompting when this flag is omitted, the user can set the compute/zone property:

gcloud config set compute/zone ZONE

A list of zones can be fetched by running:

gcloud compute zones list

To unset the property, run:

gcloud config unset compute/zone

Alternatively, the zone can be stored in the environment variable CLOUDSDK_COMPUTE_ZONE.

Discard Local SSDs At Termination Timestamp

At most one of these can be specified:

--clear-discard-local-ssds-at-termination-timestamp
Removes the discard-local-ssds-at-termination-timestamp field from the scheduling options.
--discard-local-ssds-at-termination-timestamp=DISCARD_LOCAL_SSDS_AT_TERMINATION_TIMESTAMP
Required to be set to true and only allowed for VMs that have one or more local SSDs, use --instance-termination-action=STOP, and use either --max-run-duration or --termination-time.

This flag indicates the value that you want Compute Engine to use for the --discard-local-ssd flag in the automatic gcloud compute instances stop command. This flag only supports the true value, which discards local SSD data when automatically stopping this VM during its terminationTimestamp.

For more information about the --discard-local-ssd flag, see https://cloud.google.com/compute/docs/disks/local-ssd#stop_instance.

Instance Termination Action

At most one of these can be specified:

--clear-instance-termination-action
Disables the termination action for this VM if allowed OR sets termination action to the default value. Depending on a VM's availability settings, a termination action is either required or not allowed. This flag is required when you are updating a VM such that it's previously specified termination action is no longer allowed. If you use this flag when a VM requires a termination action, it's termination action is just set to the default value (stop).
--instance-termination-action=INSTANCE_TERMINATION_ACTION
Specifies the termination action that will be taken upon VM preemption (--provisioning-model=SPOT) or automatic instance termination (--max-run-duration or --termination-time).

INSTANCE_TERMINATION_ACTION must be one of:

DELETE
Permanently delete the VM.
STOP
Default only for Spot VMs. Stop the VM without preserving memory. The VM can be restarted later.
Max Run Duration

At most one of these can be specified:

--clear-max-run-duration
Removes the max-run-duration field from the scheduling options.
--max-run-duration=MAX_RUN_DURATION
Limits how long this VM instance can run, specified as a duration relative to the last time when the VM began running. Format the duration, MAX_RUN_DURATION, as the number of days, hours, minutes, and seconds followed by d, h, m, and s respectively. For example, specify 30m for a duration of 30 minutes or specify 1d2h3m4s for a duration of 1 day, 2 hours, 3 minutes, and 4 seconds. Alternatively, to specify a timestamp, use --termination-time instead.

If neither --max-run-duration nor --termination-time is specified (default), the VM instance runs until prompted by a user action or system event. If either is specified, the VM instance is scheduled to be automatically terminated at the VM's termination timestamp (terminationTimestamp) using the action specified by --instance-termination-action.

Note: The terminationTimestamp is removed whenever the VM is stopped or suspended and redefined whenever the VM is rerun. For --max-run-duration specifically, the terminationTimestamp is the sum of MAX_RUN_DURATION and the time when the VM last entered the RUNNING state, which changes whenever the VM is rerun.

Sole Tenancy.

At most one of these can be specified:

--clear-node-affinities
Removes the node affinities field from the instance. If specified, the instance node settings will be cleared. The instance will not be scheduled onto a sole-tenant node.
--node=NODE
The name of the node to schedule this instance on.
--node-affinity-file=PATH_TO_FILE
The JSON/YAML file containing the configuration of desired nodes onto which this instance could be scheduled. These rules filter the nodes according to their node affinity labels. A node's affinity labels come from the node template of the group the node is in.

The file should contain a list of a JSON/YAML objects. For an example, see https://cloud.google.com/compute/docs/nodes/provisioning-sole-tenant-vms#configure_node_affinity_labels. The following list describes the fields:

key
Corresponds to the node affinity label keys of the Node resource.
operator
Specifies the node selection type. Must be one of: IN: Requires Compute Engine to seek for matched nodes. NOT_IN: Requires Compute Engine to avoid certain nodes.
values
Optional. A list of values which correspond to the node affinity label values of the Node resource.

Use a full or relative path to a local file containing the value of node_affinity_file.

--node-group=NODE_GROUP
The name of the node group to schedule this instance on.
Termination Time

At most one of these can be specified:

--clear-termination-time
Removes the termination-time field from the scheduling options.
--termination-time=TERMINATION_TIME
Limits how long this VM instance can run, specified as a time. Format the time, TERMINATION_TIME, as a RFC 3339 timestamp. For more information, see https://tools.ietf.org/html/rfc3339. Alternatively, to specify a duration, use --max-run-duration instead.

If neither --termination-time nor --max-run-duration is specified (default), the VM instance runs until prompted by a user action or system event. If either is specified, the VM instance is scheduled to be automatically terminated at the VM's termination timestamp (terminationTimestamp) using the action specified by --instance-termination-action.

Note: The terminationTimestamp is removed whenever the VM is stopped or suspended and redefined whenever the VM is rerun. For --termination-time specifically, the terminationTimestamp remains the same whenever the VM is rerun, but any requests to rerun the VM fail if the specified timestamp is in the past.

Maintenance Behavior.

At most one of these can be specified:

--maintenance-policy=MAINTENANCE_POLICY
(DEPRECATED) Specifies the behavior of the VMs when their host machines undergo maintenance. The default is MIGRATE. For more information, see https://cloud.google.com/compute/docs/instances/host-maintenance-options.

The --maintenance-policy flag is now deprecated. Please use --on-host-maintenance instead. MAINTENANCE_POLICY must be one of:

MIGRATE
The instances should be migrated to a new host. This will temporarily impact the performance of instances during a migration event.
TERMINATE
The instances should be terminated.
--on-host-maintenance=MAINTENANCE_POLICY
Specifies the behavior of the VMs when their host machines undergo maintenance. The default is MIGRATE. For more information, see https://cloud.google.com/compute/docs/instances/host-maintenance-options. MAINTENANCE_POLICY must be one of:
MIGRATE
The instances should be migrated to a new host. This will temporarily impact the performance of instances during a migration event.
TERMINATE
The instances should be terminated.
GCLOUD WIDE FLAGS
These flags are available to all commands: --access-token-file, --account, --billing-project, --configuration, --flags-file, --flatten, --format, --help, --impersonate-service-account, --log-http, --project, --quiet, --trace-token, --user-output-enabled, --verbosity.

Run $ gcloud help for details.

NOTES
This command is currently in alpha and might change without notice. If this command fails with API permission errors despite specifying the correct project, you might be trying to access an API with an invitation-only early access allowlist. These variants are also available:
gcloud compute instances set-scheduling
gcloud beta compute instances set-scheduling