A stakeholder asks for "five nines". How do you reason about availability targets and their cost?
What they are really testing: Senior signal: translates nines into real downtime and cost, knows each nine is roughly an order of magnitude harder, that dependencies multiply, and pushes back on over-targeting.
A real interview question
A stakeholder asks for "five nines". How do you reason about availability targets and their cost?
What most people say
drag me
“I would set up redundancy and failover to hit five nines.”
It jumps to implementation without reckoning with what five nines costs (about 5 minutes downtime a year, near-impossible with human involvement) or whether it is justified. A senior translates the target, prices it, and challenges whether it is needed.
The follow-ups they ask next
How much downtime does each nine allow per year?
Roughly: 99% ~3.65 days, 99.9% ~8.8 hours, 99.99% ~53 minutes, 99.999% ~5 minutes. Translating to concrete downtime is the first step in any availability conversation.
Why can adding a dependency lower your availability?
Availability multiplies across dependencies in a chain: two 99.9% services in series give ~99.8%. Each added hard dependency can only lower end-to-end availability unless you add redundancy/fallbacks around it.
What the interviewer is listening for
- Translates nines to concrete downtime
- Knows each nine is ~10x cost
- Knows dependencies multiply + pushes back on over-targeting
What sinks the answer
- Jumps to "add redundancy" with no cost reckoning
- Cannot translate nines to downtime
- Accepts five nines uncritically
If you genuinely do not know
Say this instead of freezing. Reasoning out loud from what you do know beats silence every single time, and a good interviewer is listening for exactly that.
“Make it concrete: [99.9% ~8.8h/yr, 99.99% ~53min, 99.999% ~5min (no human in the loop)]. Each nine is [~10x more cost: multi-region, automated sub-minute failover, no SPOFs]. Note [availability multiplies across dependencies, so a chain is lower than any one]. Then [anchor the SLO to real business impact and push back on round numbers, often 99.9% is plenty].”
Keep going with reliability
Foundation
What is a Region versus an Availability Zone, and why does the difference matter?
Foundation
What is the difference between vertical scaling and horizontal scaling, and when would you reach for each?
Foundation
What does "highly available" actually mean, and how do you achieve it?
Foundation
Why are stateless services easier to scale and make reliable than stateful ones?
Junior
How should a service retry a failed call to a dependency, and what is a retry storm?
Junior
What is idempotency, and why does it matter for reliable distributed systems?
Knowing the answer is not the same as recalling it under pressure
Sign in to send the questions you fumble to spaced recall, so they come back right before you would forget them, and learn the concepts behind them with hands-on labs.
Start free