From: Mateus Costa, operations, Pinecrest Travel
"We added a GPU node last night and since then nothing new schedules. Bookings is running on its old Pods but can't scale. Find out what we did wrong and fix it without undoing the GPU reservation."
Done when: the evidence and the taint are saved, the bookings Pods run, the GPU node keeps its reservation, and the summary names cause and fix.
A scheduling incident in four steps: keep the scheduler's evidence, find what changed on the nodes, fix the one mistake without undoing the intended change, and write a summary someone can check.
The scheduler's evidence and the node it points to.