wtf( )unctionsystem design, drawn
← all problemsCloud PracticeMedium

The console fix nobody wrote down

During an incident an engineer widened a security group by hand. The infrastructure code hasn't run in three weeks, so the next apply — for something unrelated — will either silently revert that fix or fail on unexpected state.

Nobody can say which of forty accounts still match their code. The repository describes a desired state and nothing continuously checks reality against it.

Close the loop between what the code says and what exists.
Components — tap one, then tap a slot on the diagram
!Nobody can say which of forty accounts still match their code. The next apply may revert an incident fix.

Boundaries, outermost first: Cloud accounts: Console changes (out of band; FAILED: unrecorded), Actual state Outside every boundary: Infra code (desired state), Alerting, an empty slot for the runs on a timer Connections: Infra code controls runs on a timer — desired (step 1) Actual state controls runs on a timer — observed (step 2) Console changes publishes to Actual state — drift enters here runs on a timer controls Alerting — differences (step 3)

Infra codedesired state
Console changesout of bandunrecorded
Actual state
Alerting