Recognising overkill.
WHAT IT SOLVES
Scheduling across many hosts Recovery from host failure Rolling deployments at scale Autoscaling Uniform configuration across many services
WHAT IT COSTS
A cluster to operate and upgrade Concepts every engineer must learn Diagnosis requiring real expertise Infrastructure charges before any workload runs
WHEN IT IS NOT JUSTIFIED
One or two applications One host Infrequent deployment Steady, modest load A small team with no cluster experience
WHAT TO USE INSTEAD
Containers on one or two servers, with a proxy A platform service running containers for you Plain virtual machines, where containers add nothing
WHY THAT IS NOT A LESSER CHOICE
Most applications never reach the scale where orchestration pays.
WHAT ADOPTING IT PREMATURELY COSTS
Time spent on infrastructure rather than the product Outages caused by machinery nobody understands
WHAT THE HONEST TEST IS
Can someone on the team diagnose it at three in the morning?
WHAT TO DO IF THE ANSWER IS NO
Use something simpler, or buy the expertise.
WHAT TO RESIST
Adopting it because it is expected.
WHAT TO REVISIT
The decision, when the constraints actually change.
