Summary
- Yandex restored access to 97.16% of affected cloud customer data.
- The remaining 2.84% is unrecoverable because of equipment damage and absent backups.
- The company is changing backup defaults and distributing infrastructure across additional locations.
Russian technology company Yandex has confirmed that 2.84% of customer data affected by October drone attacks on its data centres is permanently unrecoverable, despite restoring access to the remaining 97.16%.
The company attributes the losses to critical equipment damage and the absence of backups for the affected customer data.
The recovery figures were disclosed on 10 November, with subsequent reporting examining the operational consequences and the changes being made to Yandex Cloud’s backup arrangements.
Three facilities were damaged during drone strikes between 8 and 11 October, affecting data centres at Sasovo in the Ryazan region, Kaluga, and Vladimir.
The attacks disrupted Yandex’s infrastructure and cloud services, with operations at the Sasovo and Vladimir locations reportedly halted and equipment at Kaluga also affected.
Yandex has said that specialised supercomputers at two of the affected sites were not damaged and were subsequently relocated.
The company has not disclosed the number of customers affected by permanent data loss or the total volume of information that could not be recovered.
The percentages refer to customer data rather than the proportion of customers affected. They should not be interpreted as evidence that 2.84% of Yandex Cloud customers lost all their information.
The incident illustrates the difference between restoring service availability and recovering customer data.
A cloud service may resume operation after damaged infrastructure has been replaced or workloads have been moved elsewhere, but that does not guarantee restoration of individual records or storage volumes.
Data recovery depends on whether usable copies exist and whether they remain accessible after the underlying incident.
In this case, Yandex says the unrecoverable information was associated with equipment damage and missing backups.
Customers who used the platform’s backup tools experienced no data loss, according to the company.
Yandex has consequently changed its approach to backup configuration.
Previously, customers had to enable the relevant protection manually. Backups are now activated by default when creating new virtual machines, although users may choose to disable them.
The change may improve protection for new workloads, but it does not establish that existing virtual machines have automatically acquired the same safeguards.
Nor does enabling a backup feature alone demonstrate that all potential failure scenarios have been addressed.
The effectiveness of recovery depends on where backups are stored, how frequently they are created, the retention arrangements, and whether the recovery process has been tested.
Physical separation is particularly important when the triggering event affects an entire building or several facilities.
Copies held within infrastructure exposed to the same incident can fail together, even when individual servers or storage devices are duplicated.
A geographically distributed design can reduce that risk, but only if its operational dependencies are properly understood.
Network connectivity, management systems, access controls, and data consistency all affect the ability to recover services from a separate location.
Yandex says it is distributing its cloud computing resources across additional independent facilities in different Russian regions.
The company has not provided enough detail to establish the exact resilience configuration or recovery objectives associated with those changes.
Physical damage and recovery design
The attacks demonstrate how physical infrastructure incidents can create permanent consequences when data protection depends on equipment that is no longer usable.
Cloud availability architectures are often designed to tolerate failures of individual components, such as drives, servers, or network equipment.
More extensive incidents can affect shared electricity systems, cooling infrastructure, entire availability zones, or multiple sites.
The recovery strategy must therefore account for the scale of disruption against which protection is intended.
Replication can maintain service continuity by preserving multiple copies of active data, but it may also propagate accidental deletion or corruption.
Backups provide a different form of protection through independently recoverable copies and defined recovery points.
The two approaches serve related but distinct purposes.
Responsibilities also vary between cloud providers and their customers. Some services include automatic replication, while independent backups must be selected and configured separately.
The terms and technical configuration of each product determine where those responsibilities lie.
Yandex’s reported losses show the consequences of that distinction under severe physical disruption.
The company’s change to default backup settings may reduce the number of workloads operating without configured protection, while its geographic redistribution programme addresses exposure to concentrated physical failures.
Neither measure reverses information already lost.
The incident provides a concrete example of why resilience must be assessed through recoverable data, usable alternative infrastructure, and tested restoration procedures rather than service availability claims alone.
