Durability is the ability to preserve data when failures occur. It asks how much data a system can lose, and under which failure scenarios. Availability asks whether the system is up; the two concerns are related, but meeting one does not guarantee the other. section 2.2.2
Set durability according to the data’s value and recovery path, not by assuming every record needs the strongest guarantee. Losing a derived search index may mean rebuilding it from the source database; losing orders or customer records can have commercial, regulatory, or legal consequences. Caches, short-lived queues, and temporary sessions may also tolerate lower durability when their loss is recoverable. section 2.2.2
Table 2.2 Durability tiers
|
Tier
|
Durability
|
Architecture implications
|
|
Minimal
|
Any server failure may lose recent writes, up to the
backup point.
|
Single-node database instance, no replication. Only
suitable for ephemeral or low-value data.
|
|
Moderate
|
Survives single-node crashes. Risk remains for recent
writes if the primary fails before replication completes.
|
Multiple instances with async replication of writes, daily
backups.
|
|
High
|
Strong guarantee: committed writes survive single-node
failure.
|
Multiple instances with synchronous write replication in
one region/datacenter, point-in-time recovery. Slightly higher write latency.
|
|
Very high
|
Survives regional disasters and ensures global durability.
|
Multi-region replication (sync intra-region, async
inter-region), point-in-time recovery. Higher complexity and latency.
|
Use the durability tiers to compare the failure protection each architecture provides, from a single node with backups to multi-region replication. Stronger protection can increase latency and complexity. section 2.2.2
Make the acceptable loss explicit for each failure scenario. Vendor loss probabilities may be unavailable, so compare concrete guarantees instead. Consider durability alongside availability when discussing architectural choices, and use tiered requirements to establish shared expectations before solution design. section 2.2.2 section 2.5







