Skip to content
AAAhtisham Ashraf
000
Insights

Cloud

Availability is mostly the absence of surprises

High availability is not produced by clever architecture. It is produced by removing the conditions under which a system can be surprised.

6 min read

Infrastructure that grows alongside a business tends to accumulate environments faster than it accumulates structure. Each one was added for a good reason. Collectively they become a set of systems that share a network and very little else — different configuration conventions, different access rules, different assumptions about who is responsible for what.

Availability problems in estates like this are rarely dramatic. They are the accumulated cost of every place where two environments disagree.

Consistency beats optimality

It is tempting to tune each environment for its workload. In practice, an estate where every environment is configured the same way is easier to keep available than one where each is configured optimally. Consistency is what makes an incident diagnosable: if the answer to "how is this configured" is the same everywhere, the search space during an outage collapses.

Reliable technology should work so consistently that the business rarely has to think about it.

The unglamorous list

  • Environment boundaries enforced at the network layer, not by naming convention.
  • Least privilege applied to infrastructure and data services as a default rather than an exception.
  • Storage and database services configured against public exposure, verified rather than assumed.
  • Monitoring that maps to the availability target, so the alert and the objective describe the same thing.
  • A rollback path that has been used recently enough to be trusted.

None of this is interesting to design. All of it is what four nines is actually made of.

Cost and capability are not opposites

Reducing infrastructure cost is often framed as a trade against capability or headcount. Usually the real trade is against complexity. Environments that exist because nobody was certain they could be removed, services provisioned for a peak that never arrived, duplicated tooling from two eras of the same platform — removing those lowers cost and raises availability at the same time, because there is less to be surprised by.

Related topics

  • Cloud Architecture
  • Infrastructure
  • Operations