Resilience in Elastic Cloud Serverless

In Elastic Cloud Serverless, Elastic manages all infrastructure resilience automatically. Unlike ECH, ECE, ECK, or self-managed deployments, there are no nodes to configure, no replica counts to tune, and no availability zone settings to manage. Resilience is built into the platform at the infrastructure level.

Serverless uses a stateless storage architecture in which indexed data is written to cloud-provider object storage rather than stored as local replicas on cluster nodes. This provides high data durability:

• Object storage is replicated across multiple physical locations within a region by the underlying cloud provider, independently of compute node health.
• Data durability does not depend on Elasticsearch shard replication. A node failure has no impact on data that has already been written — the data exists independently of the compute layer.

When a client receives a successful write response from Elasticsearch, the data has been durably persisted to the relevant cloud service provider's resilient object storage — Amazon S3, Google Cloud Storage, or Azure Blob Storage. For detailed information on the durability and recoverability guarantees of these stores, refer to their respective documentation. From the perspective of Elasticsearch, the Recovery Point Objective (RPO) for events such as node failures or even an availability zone outage is zero - there is no data loss to recover.

To learn more about the stateless architecture underpinning Serverless, refer to Elastic's serverless architecture.

Serverless manages placement of nodes into availability zones within each region - no configuration is required. Replacement of a failed node, including to a different availability zone when necessary, is automated and requires no user intervention.

Elastic maintains a service level agreement (SLA) for Serverless project availability. Refer to the Elastic Cloud Serverless Service Level Agreement for details.

Serverless does not currently offer a self-service recovery workflow for data loss resulting from user-initiated operations, such as accidental index deletion or bulk document removal. If you experience this type of data loss, contact Elastic Support to discuss your options.

All resilience in Serverless is regional. Projects run in a single cloud region, and there is no built-in support for multi-region architectures:

• Cross-region replication is not available.
• There is no automatic failover to an alternative region if the region itself becomes unavailable.

If your requirements include multi-region resilience, you can deploy separate projects in different regions and manage data routing or synchronization at the application level. Refer to Serverless regions for the supported options.