Design business continuity solutions

Key terms

• RPO: maximum tolerable data loss, measured in time. RTO: maximum tolerable downtime. SLA: Microsoft’s financially backed availability commitment for a service.
• High availability (HA) keeps the service running through a local fault (zone, host). Disaster recovery (DR) restores it after a region loss. Backup protects against deletion, corruption and ransomware, and is not a substitute for either HA or DR.
• Composite SLA: services in series multiply (99.95% x 99.99% = about 99.94%). Redundant paths in parallel improve it: 1 - (1 - A)^n. Adding a dependency in the request path always lowers the composite.

Azure paired regions and zones

• Most regions have 3 availability zones: physically separate datacentres with independent power, cooling and networking. Zone-redundant services replicate across zones automatically. Zonal resources are pinned to one zone and you build redundancy yourself.
• Paired regions (for example East US and West US) get sequential platform updates and prioritised recovery, and are the replication target for GRS and for many default DR choices. Some newer regions have no pair.
• Availability sets protect against host and rack failures inside one datacentre: fault domains (shared power and network) and update domains (planned maintenance, up to 20). Use zones in preference when the region supports them. A VM can be in a set or a zone, not both.

Backup and disaster recovery

Site Recovery vs Backup

You usually need both. A frequent trap: Site Recovery faithfully replicates corruption and deletion, so it can’t replace backup.

Azure Site Recovery

• Scenarios: Azure VM to another region, VMware or Hyper-V or physical to Azure (using the configuration server or Hyper-V site), and Azure to on-premises.
• Recovery Services vault holds the replication configuration. Recovery plans orchestrate failover order for multi-tier apps (groups, scripts, manual steps, Automation runbooks).
• Test failover runs in an isolated network with no impact to production. Planned failover (no data loss), unplanned failover (accepts data loss), failback reverses replication.
• Crash-consistent recovery points every 5 minutes; app-consistent points at a configured interval (for example hourly) for databases. Retention up to 15 days.
• For Azure VM to Azure DR, the target region, resource group, VNet and cache storage account are chosen at enablement. The mobility service extension is installed automatically. Zone-to-zone DR within a region is also supported.
• Pair with Traffic Manager or Front Door to redirect users, and keep IP addressing decisions (same or different IP) in the plan. Use Azure Automation or a recovery plan to reattach load balancers and public IPs.
• Azure Migrate replication shares technology but is for migration, not DR.

Azure Backup: compute

• Vault types: Recovery Services vault (VMs, MARS, MABS, SQL/HANA in VM, Azure Files) and Backup vault (blobs, managed disks, PostgreSQL, AKS, Kubernetes). A vault stores only items in its own region for restore and supports Cross Region Restore with GRS.
• Vault redundancy: LRS, ZRS or GRS (default GRS). Choose before the first backup.
• Instant restore snapshots are kept locally 1 to 5 days for fast recovery. Policy retention: daily, weekly, monthly, yearly, up to 99 years.
• Security: soft delete (14 days, extendable), multi-user authorization (Resource Guard), immutable vaults, MFA for critical operations, and encryption with CMK. These counter ransomware.
• Central management: Backup Center / Business Continuity Center, Azure Policy to auto-enable backup, Backup Explorer reports.

Backup and recovery: databases

• Restores to a new database (not in place). Dropped databases can be restored within retention.
• Use LTR for compliance retention of years; PITR covers recent operator error.

Backup and recovery: unstructured data

Object replication copies blobs asynchronously to another account or region and can be used as a lightweight DR or locality copy. Immutable storage protects against tampering.

Hybrid and meeting recovery objectives

Work backwards from the numbers:

Strategies: backup and restore, pilot light, warm standby and active-active, ordered by rising cost and falling recovery time. Test failovers regularly and document runbooks. For on-premises workloads, ASR to Azure replaces a secondary datacentre.

High availability

Compute

• Proximity placement groups trade availability for latency, so avoid them when HA is the requirement.
• Capacity reservations guarantee VM capacity in a zone for DR.
• Mix: place a stateless web tier across zones, and keep state in zone-redundant data stores.
• Azure Load Balancer Standard is zone-redundant by default for new frontends. Basic is retired.

Relational data

• Failover group vs geo-replication: pick failover groups when the application needs one stable connection string and automatic or group failover; geo-replication is the lower-level primitive for single databases.
• Failover group failover can lose recent data (async). Business Critical zone-redundant keeps synchronous replicas across zones.
• Hyperscale supports geo-replicas and named replicas (named replicas are not DR targets).

Semi-structured and unstructured data

• Cosmos DB: add regions for reads and automatic failover (priority list), or multi-region writes for active-active. Availability zones in a region for zone redundancy. SLA 99.99% single-region, 99.999% multi-region reads (and writes with multi-write).
• Storage accounts: ZRS for zone failures, GZRS/RA-GZRS for zone and region. Customer-managed account failover promotes the secondary to primary (RPO from last sync time, with data loss possible, account becomes LRS afterwards until reconfigured). Microsoft-managed failover only in major disasters.
• Azure Files: ZRS and GZRS options on standard shares, ZRS on premium. Azure File Sync across servers.
• Blob object replication for active DR copies, and Front Door or Traffic Manager to route readers.
• Event Hubs and Service Bus: zone redundancy by default in supported regions, geo-replication (and geo-disaster recovery pairing, metadata only) for namespaces.
• Key Vault replicates automatically to the paired region (read-only during failover). Azure Cache for Redis / Managed Redis use zone redundancy and geo-replication on the Enterprise or Premium tiers.

Traffic redirection for DR