Spot Instances

Overview

Spot Instances are unused virtual machine resources offered by cloud providers at significantly reduced prices. They are ideal for cost-effective computing but are temporary by nature and can be abruptly terminated by the cloud provider. Spot Instances are available on AWS, Google Cloud Platform (GCP), and Microsoft Azure.

Information on this page

You can view a list of Spot Instances in the project and perform various actions on them depending on the Spot status.

There are three types of states on the emma platform: ON (running spot), OFF (stopped by the user) and PREEMPTED (interrupted by the provider; this Spot Instance is trying to restart).

Spot Instance states

Powered on

• A regular state of the Spot Instance, up and running
• Displayed in the interfaces as: ON
• Billed both for compute and storage

Powered off

• A regular state of the Spot Instance. The main characteristic of this status is that the user powers off the spot by himself
• Displayed in the interfaces as: OFF
• Billed for storage only

Preempted

• This status means that the Spot Instance was interrupted by the provider. The main characteristic of this status is that a spot will try to restart automatically
• Displayed in the interfaces as: PREEMPTED
• Billed for storage only

Actions you can perform on this page

Actions you can perform on a Spot Instance via buttons:

• +Add spot VM: create a new Spot Instance.

Actions you can perform on a Spot Instance via the ellipsis menu:

• Start: power the Spot Instance on. This action is allowed only for instances with the status OFF.
• Shutdown: power the Spot Instance off. This action is only allowed for instances with the ON and PREEMPTED statuses.
• Reboot: restarts your Spot Instance.
• Change price (applicable only for AWS and Azure): adjust the price of a Spot Instance in the OFF or PREEMPTED statuses to improve its chances of being powered on.
• Rename: change the name of the Spot Instance. Applicable in all statuses except FAILED.
• Delete: delete your Spot Instance permanently. All associated data will be removed upon confirmation of the request.

About Spot Instances

What are Spot Instances used for?

Spot Instances are best suited for workloads that:

• can handle interruptions or restarts;
• do not require continuous execution;
• include batch processing;
• use distributed computing and require large-scale computations.

Duration of Spot Instance existence

Spot Instances run while resources are available and the price remains acceptable. Maximum duration is unlimited. Spot Instances are interrupted by the provider when it needs the resources for higher-priority tasks.

The providers notify the spot user just before the spot termination:

The notification is sent to the Spot Instance.

Key recommendations for using Spot Instances

• Use Spot Instances to reduce costs for fault-tolerant workloads.
• Ensure data is regularly saved to durable storage.
• Use monitoring tools to detect and handle interruptions in real time.

What happens when the Spot Instance is interrupted by the provider

The Spot Instance can be interrupted by the cloud provider at any time when the capacity is needed elsewhere. When a Spot VM is interrupted by the provider:

1. The interrupted Spot Instance enters a "stopped" state instead of being completely terminated and changes its status to PREEMPTED.
2. The data stored on persistent storage (e.g., attached disks or network-based storage) remains intact and unaffected by the interruption.
3. Once the interruption conditions are resolved and capacity becomes available, the automated recovery attempts to restart the Spot Instance. The result of restart attempts is subject to resource availability and the original configuration of the Spot Instance.