Storage redundancy: LRS vs ZRS vs GRS vs RA-GRS
Verdict: Stay in-region but survive a zone loss: ZRS. Single datacentre only: LRS. Copy to a paired region: GRS. Read the secondary before failover: RA-GRS. Data-residency rules rule out every geo option.
| Criterion | LRS | ZRS | GRS | RA-GRS |
|---|---|---|---|---|
| Scope of copies | One datacentre | Three or more zones in the primary region | Primary region plus a paired secondary region | Primary plus paired secondary, readable |
| Survives zone loss | No | Yes | Yes (primary is still local) | Yes |
| Stays in region | Yes | Yes | No | No |
| Secondary read access | None | None | Only after account failover | Anytime, via the -secondary endpoint |
| Choose when | Lowest cost, non-critical | In-region data residency plus zone resilience | Cross-region disaster recovery | Test or read the secondary while primary is up |
Rules
- ZRS copies data synchronously across three or more availability zones in the primary region, giving zone resilience without leaving the region - the choice under data-residency rules.
- Only read-access variants (RA-GRS, RA-GZRS) expose the secondary region for reads; the endpoint appends -secondary to the account name so apps can validate before an outage.
- SSD (premium) Azure Files shares support only LRS and ZRS; geo options apply to HDD SMB shares.
- For in-region backup vaults that must survive a zone loss, ZRS keeps every copy in the primary region while tolerating one zone failure.
Traps
- GRS and RA-GRS replicate to a distant paired region, so both violate an in-region data-residency requirement even though they add resilience.
- Plain GRS keeps the secondary unreadable unless an account failover happens; only the RA- variants allow reads while the primary is up.
- ZRS managed disks reach twelve nines of yearly durability versus eleven nines for LRS - the figures are commonly swapped.