Keeping Cattle or a Pets in IT affecting Toil Budget and Error Budget

We all know a colleague who administers a cluster, yet there’s that one server — sometimes the entire cluster — they quietly won’t let anyone replace. They log in to check on it, tweak a config by hand, and restart a service just to be happy when it’s happy. This is called a pet in IT, and the job done to it is babysitting.

The cost of care is measured in hours. How many hours do admins spend fixing problematic instances—patching them, tracking down old configuration changes? This is time spent on toil (toil budget). Undocumented or manual modifications make the instance harder to duplicate, turning it into a snowflake in the glass.

Ideally, a DevOps engineer keeps their toil budget low, pushing for automation to give themselves some headroom in the error budget.

The opposite —cattle —are the instances you can easily replace; tear them down and spin them back up from a declarative config. Version-controlled, surely it is. Managing instances in code, handling lifecycles, and resources between tenants. Committing code and an ops posture like a pro values versioning and policies within bounds.