A Compute Engine instance transitions through different states from the time you create it until you delete it. This document explains the compute instance lifecycle to help you troubleshoot issues, manage resources, plan migrations, and understand costs.
To check the current state of your compute instances, see instead the following:
Compute instance states
When you create a compute instance, it starts in the PROVISIONING state, moves
to STAGING, and then enters the RUNNING state. Flex-start virtual machine
(VM) instances with a wait time (requestValidForDuration) and TPU instances
start instead in the PENDING state. From the RUNNING state, you can move a
compute instance to the SUSPENDED or TERMINATED states, or you can delete
it.
The following diagram shows the different states that Compute Engine can set a compute instance to:
The states shown in the preceding diagram are described in the following sections:
PENDING state
After you create a compute instance, it enters the PENDING state if it's one
of the following. Otherwise, the compute instance directly transitions to the
PROVISIONING state.
A Flex-start VM with a wait time (
requestValidForDuration) that is 90 seconds or longer.A standalone TPU instance.
A Flex-start VM or TPU instance that is part of a managed instance group (MIG), unless the MIG has creation retries disabled.
In the PENDING state, Compute Engine attempts to acquire the necessary
resources to start your compute instance until its wait time ends. If
Compute Engine acquires the resources within this time and you have
sufficient quota for those resources, then the compute instance state changes to
PROVISIONING. Otherwise, you encounter an error and Compute Engine
deletes the compute instance. You can optionally delete the compute instance
before the wait time ends if you no longer need it.
PROVISIONING state
After you create, restart, or resume a compute instance, Compute Engine
starts allocating resources for the compute instance and sets its state to
PROVISIONING. When Compute Engine finishes allocating resources, the
compute instance transitions to the STAGING state.
STAGING state
In the STAGING state, Compute Engine prepares the compute instance for
first boot due to one of the following reasons:
Compute Engine is still creating and configuring the compute instance.
You, or a scheduled operation, restart or resume the compute instance.
In this state, the compute instance isn't running yet. When it does start
running or starts booting up, the compute instance enters the RUNNING state.
RUNNING state
In the RUNNING state, Compute Engine is booting up the
compute instance or the compute instance is running. From the RUNNING state, a
compute instance can transition to PENDING_STOP, STOPPING, SUSPENDING, or
REPAIRING depending on the action taken, system events, or compute instance
configuration.
PENDING_STOP state
In the PENDING_STOP state, the compute instance starts to gracefully shut
down. This shutdown process happens only if you've enabled
graceful shutdown. The
compute instance enters this state if you stop or delete the compute instance,
or Compute Engine is doing so automatically for a scheduled stop or
deletion. The compute instance state changes to STOPPING when one of the
following happens:
You manually end the graceful shutdown.
The graceful shutdown period times out. If any tasks are still running, then Compute Engine forcefully stops them.
STOPPING state
In the STOPPING state, the compute instance is shutting down its guest OS.
This process happens in the following scenarios:
You, or a scheduled operation, stop or delete the compute instance.
A host error occurs.
Unless you configure the compute instance to skip the guest OS shutdown, the shutdown time depends on the compute instance type. However, if the compute instance stops due to a host error, the shutdown time might differ from the expected length. After the guest OS shuts down, the ongoing stop or delete operation ends as follows:
Stop operation: the compute instance state changes to
TERMINATED.Delete operation: Compute Engine permanently deletes the compute instance and all its attached resources.
TERMINATED state
When a compute instance enters the TERMINATED state, Compute Engine
completes the stop operation. The compute instance remains in the TERMINATED
state until you, or a scheduled operation, restart or delete it. If you request
to restart the compute instance, but Compute Engine can't allocate your
requested resources, then the restart request fails and the compute instance
remains in the TERMINATED state. Otherwise, the restart request succeeds and
the compute instance state changes to PROVISIONING.
REPAIRING state
In the REPAIRING state, Compute Engine starts a
repair operation on the compute instance.
Compute Engine repairs a compute instance if it encounters an internal
error or the compute instance's host server is unavailable due to maintenance.
While a compute instance is in repair, the following happens:
You can't use the compute instance.
The service level agreement (SLA) doesn't cover the compute instance.
If Compute Engine successfully repairs the compute instance, then it
returns the compute instance state to its original state before the repair
operation began. This state can be STAGING, RUNNING, SUSPENDING, or
STOPPING. If your compute instance is configured to automatically restart
(automaticRestart) after the repair operation completes, then you can
optionally stop the compute instance during the repair process. This action
prevents the compute instance from automatically restarting after the repair
completes, leaving the compute instance in the TERMINATED state.
SUSPENDING state
In the SUSPENDING state, Compute Engine starts the suspend operation
for the compute instance. You can only suspend
specific compute instances.
When the suspend operation completes, the compute instance state changes to
SUSPENDED.
SUSPENDED state
In the SUSPENDED state, Compute Engine completes the suspend
operation. You can resume or delete the compute instance. If you request to
resume the compute instance, but Compute Engine can't allocate your
requested resources, then the resume request fails and the compute instance
remains in the SUSPENDED state. Otherwise, the resume request succeeds and the
compute instance state changes to PROVISIONING.
A compute instance can remain in the SUSPENDED state for up to 60 days. After
that time, Compute Engine changes the compute instance state to
TERMINATED.
Compute instance pricing
During a compute instance lifecycle, you incur charges for the compute instance as follows:
For CPU usage, you incur charges when the compute instance is in the following states:
RUNNINGPENDING_STOP
For memory usage, you incur charges when the compute instance is in the following states:
RUNNINGPENDING_STOPSUSPENDINGSUSPENDED
For attached resources like disks or external IP addresses, you incur charges until the resources exist, regardless of the compute instance state.
For more information, see the pricing for compute instances.
What's next
Learn how to view the state of one or more compute instances by doing the following:
Learn more about suspending, stopping, or resetting compute instances.
Learn more about host events.
Learn how to delete a compute instance.