6.6 Reliability measures

What this heading covers

The four families describe different mechanisms and they need different data. Faultlessness is about the absence of failure in operation or test. Availability is about the proportion of time the service is usable. Fault tolerance is about continuing to operate when a component fails. Recoverability is about restoring service and data after a failure. A programme that reports availability alone has said nothing about how the product behaves when something breaks, and a faultlessness count without an operating window cannot be compared. Reliability measures are the ones most often built from operational data, which means the observability of the product is part of the measurement method.

Record position: characteristic 5 of 9, carrying 4 of the 40 measure families in clause 6.

What the heading asks of a team

The common confusion

Reliability is often reported as uptime. Availability is one of four families, and a product can be available, fragile, slow to recover and full of failures at the same time.

The measure families in this characteristic

• 6.6.2 Faultlessness measures
• 6.6.3 Availability measures
• 6.6.4 Fault tolerance measures
• 6.6.5 Recoverability measures

What to put in place before publication

1. Write the failure definition and the severity scale the reliability measures will use.
2. Define the service window and the measurement period for availability, including planned maintenance.
3. Plan the fault injection or the failure observation that will evidence fault tolerance.
4. Write the recovery time objective and the recovery point objective, and measure against them.
5. Connect the measures to the operational telemetry, and record the coverage of that telemetry.

Where teams usually fall short

• Availability computed from an uptime monitor that only checks one endpoint.
• Fault tolerance claimed from architecture diagrams, with no injected failure on record.

What happens next

Read faultlessness and availability together, then fault tolerance and recoverability. The first pair describes normal operation; the second describes a bad day.