Service Level for Yandex Compute Cloud
This document defines Service Level of Yandex Compute Cloud (“Service”) and is an integral part of Yandex.Cloud Customer Agreement (“Agreement”) and SLA available at: https://yandex.com/legal/cloud_sla. Capitalized terms used herein but not defined herein shall have the meanings set forth in Agreement and Linked Documents.
|
Service Level |
99.95% |
|
GPU Cluster |
99.5% |
|
Single Dedicated Host |
99,9% |
Compensation Amount:
Availability of a single Virtual Machine:
|
Service Uptime Percentage in Accrual Period |
Compensation, % of Service Fee for Service in Accrual Period |
|
< 99.95%-99.00% |
10.0% |
|
< 99.00%-95.00% |
15.0% |
|
less than 95.00% |
30.0% |
Definition of a single Virtual Machine Unavailability:
Loss of connectivity or a boot disk of a Virtual Machine.
Loss of connectivity refers to the unavailability of the data transmission medium and the inability to transmit and receive correctly generated (in accordance with the relevant international RFC standards) IP packets on all network interfaces of a single Virtual Machine connected to subnets of the Virtual Private Cloud Service due to a malfunction on the Yandex side.
Compensation for consumed Service in case of Unavailability of a single Virtual Machine is calculated as % depends on the cost of the Virtual Machine that was Unavailable.
In case of Unavailability of several Virtual Machines or all Virtual Machines in the Availability Zone, Compensation for consumed Service is calculated as % from the cost of all Virtual Machines for which Unavailability was recorded in the corresponding Availability Zone.
SLA does not apply to cases of loss of connectivity due to circumstances beyond Yandex’ control, including cases of DoS or DDoS attacks on Customer’s Virtual Machine or Customers actions that caused loss of connectivity, failure or disconnection of network interfaces of a single Virtual Machine.
Availability of a group of non-replicated Disk`s:
|
Basis for Compensation calculating in Accrual Period |
Compensation, % of Service Fee for Service in Accrual Period |
|
Every fact that a non-replicated Disk placement group is Unavailable |
10.0% |
Definition of a group of non-replicated Disk`s Unavailability:
The transition of two or more Disks from the same placement group to a faulty state in an 8-hour time interval (when after the transition of the first Disk to a faulty state within 8 hours, another Disk from the same group became faulty).
A failed Disk state means that the Disk goes to the "ERROR" state (the disk state is displayed in the Console or via the https://cloud.yandex.ru/docs/cli/quickstart).
Determination of unavailability of GPU Cluster:
- Lack of connectivity via high-speed bus between virtual machines that are members of the cluster for 5 minutes or more, subject to the "Running" status of such virtual machines, due to problems on the Yandex side.
The following are not covered by the Service Level Agreement:
- lack of connectivity via high-speed bus between virtual machines that are members of the cluster due to circumstances beyond Yandex's control; unavailability of the Service due to custom settings of firewall, routing and guest operating system rules of virtual machines that are cluster members; unavailability of the Service due to its integration by the Customer with an external application, or due to the Customer's installation of an application that resulted in inoperability of virtual machines that are cluster members.
Preventive Maintenance of the Service:
- Yandex shall be entitled to interrupt the operation of the Service for necessary preventive maintenance, including on business days. Such cases will not be considered as interruptions in providing access to the Service if Yandex notifies the Customer in advance by e-mail three (3) calendar days prior to the start of the preventive maintenance work causing interruptions in the Service.
Availability of a Single Dedicated Host
|
Service Uptime Percentage in Accrual Period |
Compensation, % of Service Fee for Single Dedicated Host for which Unavailability was recorded in Accrual Period |
|
< 99.9%-99.00% |
10.0% |
|
< 99.00%-95.00% |
15.0% |
|
less than 95.00% |
30.0% |
Determining The Unavailability of a Single Dedicated Host:
Loss of network connectivity (external and internal subnets) or the Dedicated Host as a whole as a result of an accident.
The Service Level Agreement does not apply to cases of loss of external connectivity that occurred due to circumstances beyond Yandex's control, including in cases of Dos or Ddos attacks on the Customer's Dedicated Host or Customer actions that caused the failure or disconnection of network interfaces on the Dedicated Host.
Loss of network connectivity refers to the unavailability of the data transmission medium and the inability to transmit and receive correctly generated (in accordance with the relevant international RFC standards) Ethernet frames and IP packets on all network interfaces of the Dedicated Host due to a malfunction on the Yandex side.
If several Dedicated Hosts in the group or all Dedicated Hosts in the Availability Zone are Unavailable, the Compensation for consumed Service is calculated as % from the cost of all Dedicated Hosts for which Unavailability has been fixed in the corresponding Availability Zone.
The Customer independently monitors data security on the Dedicated Host, controls access to this data, and organizes backups of information hosted on the Dedicated Host.
A planned server maintenance or upgrade event is not considered an unavailability event, accompanied by a notification to the Customer at least 14 business days before the event, with the Customer's option to postpone work data for another 14 days (the total time before the start of work is up to 28 days) and the issuance of a replacement server before the event date.
Definitions
Dedicated Host means a physical server designed to host Virtual Machines.
______________________
Date of placement: October 03, 2025
Effective date: October 17, 2025
Previous version of the document: https://yandex.com/legal/cloud_sla_compute/01062025/
Previous version of the document: https://yandex.com/legal/cloud_sla_compute/27092024/
Previous version of the document: https://yandex.com/legal/cloud_sla_compute/01112023/
Previous version of the document: https://yandex.com/legal/cloud_sla_compute/23042019/