Question

At 14:10 a rolling deploy of the orders service begins (8 of 24 pods on the new version). Within two minutes, error rate climbs to ~12% with a tight cluster of `NULL value in column "fulfillment_channel" violates not-null constraint` on INSERT. The migration that ran in the deploy's pre-step added the column `fulfillment_channel TEXT NOT NULL` to `orders`. The new pods write the column; the old pods (still 16 of them) do not. The DB CPU is normal, replication lag is 0, latency p99 unchanged. The deploy is configured `maxUnavailable=0, maxSurge=25%`, so old and new pods serve concurrently for the full rollout window. How do you triage and mitigate, and what's the durable fix?