AWS Reliability Pillar

What is Reliability

Reliability is the ability to recover from disruption, dynamically acquire resources and mitigate disruptions.

Common Disruptions

Hardware Failure
Failure of any hardware component in the system, including in hosts, storage, network, or elsewhere.

Deployment Failure
Failure caused directly as a result of a software, hardware, network, or configuration deployment. This includes both automated and manual changes. The rest of the buckets specifically do not meet this definition.

Load Induced
Load related failures can be triggered by a change in behavior, either of a specific caller or in the aggregate, or by the service reaching a tipping point. Load failures can occur in the network.

Data Induced
An input or entry is accepted by the system that it can’t process (“poison pill”)

Credential Expiration
Failure caused by the expiration of a certificate or credential.

Failure of a dependent service results in failure of the monitored service. A 99.99% available application with hard dependencies on 2 external systems available at 99.99%, can only have a 99.97% uptime (.9999 * .9999 * .9999 = .9997). Converted, that is allowed downtime increase from 52m 37s to 2h 37m 47s in a year.

Power supply or environmental condition failure has an impact on hardware availability.

Identifier Exhaustion
Exceeding available capacity, a throttling limit was hit, an ID ran out, or a resource that is vended t customers is no longer available.