Cloud · Azure Data Engineering

Multi-region disaster recovery for Azure data lakes with paired regions

azureHardazureadls-gen2disaster-recoveryhigh-availability

Interview question

How do you architect cross-region disaster recovery for Azure Data Lake Storage Gen2 with defined RTO and RPO?

How to answer

The short answer

Enterprise analytics platforms supporting business operations must maintain continuity if an entire cloud region experiences a prolonged outage (such as power failure or severe physical infrastructure damage).

Azure provides regional resilience through Azure Paired Regions and geo-redundant storage architectures.

Storage redundancy: GRS versus RA-GRS

Azure Data Lake Storage Gen2 supports geo-redundant storage options:

  • Geo-Redundant Storage (GRS): Replicates 3 copies of data within the primary region using locally redundant storage (LRS), and asynchronously replicates a fourth, fifth, and sixth copy to a designated secondary paired region hundreds of miles away.
  • Read-Access Geo-Redundant Storage (RA-GRS): Provides secondary read endpoints (mystorage-secondary.dfs.core.windows.net), allowing reporting workloads to query secondary replicas during normal operations.

Customer-managed account failover

Historically, failover to secondary regions was initiated exclusively by Microsoft during catastrophic disasters.

With Customer-Managed Planned Account Failover:

  • If an operational outage occurs in East US, the data operations team triggers an account failover via Azure CLI, PowerShell, or Azure Portal.
  • The secondary storage in West US becomes the primary read-write account.
  • The storage account DNS record updates automatically: clients continue connecting to mystorage.dfs.core.windows.net without modifying connection strings in pipelines or Spark jobs.
  • The RPO depends on asynchronous replication lag (typically under 15 minutes), and the RTO is under 1 hour to complete DNS propagation.