Overview with affiliate links. We receive a commission if you click through via us. Read how
Compare

High availability in the cloud: what it means for architecture

High availability is about absorbing failures with redundancy, zones and automatic failover; costs and complexity increase with ambition.

Door de redactie van ClouwinoGepubliceerd Laatst bijgewerkt

Short answer

High availability means a service keeps working when a component fails. You achieve it with redundancy, spreading across zones, health checks and automatic failover. Choose the level based on what downtime costs you; every extra layer adds complexity and cost.

Who is it for: Developers, DevOps engineers and technical buyers who design or assess the availability of a cloud application.

What does high availability mean?

Availability is the proportion of time that a service is usable. High availability (HA) is a design goal: no single component may be a point of failure that brings the whole service down. A single point of failure is a component whose failure stops the service.

HA is something different from disaster recovery. HA absorbs failures of components, without noticeable interruption. Disaster recovery restores after a major disruption, such as the loss of an entire location, with a certain recovery time and data loss time.

Which building blocks do you use?

Redundancy is running components in multiples. A load balancer distributes traffic across multiple servers and skips faulty ones based on health checks. An availability zone is a separate location within a region with its own power and network; by spreading components across zones, you can survive the loss of a zone.

The hardest layer is data. A stateless application (without local state) is easy to multiply. A database requires replication and a choice between synchronous, with certainty about the last write but more latency, and asynchronous, faster but with a risk of losing the most recent changes.

How do you determine the level?

Translate availability into business terms: what does an hour of downtime cost, and how much data can you afford to lose? Two concepts help. RTO (recovery time objective) is the maximum recovery time you accept. RPO (recovery point objective) is the maximum amount of data loss, expressed in time.

Read providers’ availability promises critically. Ask what a commitment relates to, how it is measured, what the compensation is and which components are excluded. A service only has HA if your own design makes use of it; one server in one zone remains vulnerable.

How do you test it?

Practise failures. Deliberately switch off a component and see whether the service keeps running and alarms go off. Also test database failover and the recovery of a zone. A design that has not been tested is an assumption.

Concepts around availability

Concepts you encounter in design and contracts.
TermWhat does it meanCheck question
RedundancyRunning components in duplicateWhich components are still a point of failure?
Availability zoneSeparate location within a regionIs everything in one zone or spread out?
FailoverSwitching over to a standbyAutomatic or manual, and how fast?
RTOMaximum recovery timeHas the time been tested and is it achievable?
RPOMaximum acceptable data lossDoes the replication method fit this value?

Who is it not ideal for?

Frequently asked questions

What is the difference between HA and fault tolerance?

Fault tolerance tries to make failure completely invisible, often with expensive duplicate execution. HA accepts a short interruption during failover.

Do I need multiple regions?

Only if losing an entire region is unacceptable. For many applications, multiple zones within a region are enough.

What is split brain?

A situation in which two components cannot see each other and both act as leader, leading to conflicting data. Good clusters prevent this with a quorum.

What about maintenance?

HA makes maintenance without interruption possible: update one component at a time while the rest keeps running.

Ask us
Compare Get in touch

Are you sure?