Silence. The kind of silence that isn't empty—it's full of failed heartbeats, disconnected clients, and the distant sound of dashboards turning red.
But the cluster was live. Four thousand active sessions. Three replicas of the order-processing database. If he ran the patch in , nodes would update one by one—seamless, safe, standard.
“I know what it means.”
He pulled up the change request dashboard. His eyes skimmed over the numbers: active transactions, replication lag, customer SLAs. If he did this now, the order system would vanish for at least forty-five minutes. The on-call manager would scream. The VP of Engineering would ask why he hadn’t scheduled a maintenance window.
But opatchauto had just vetoed that.
This operation will patch all nodes in non-rolling mode. Database services on all instances will be interrupted until patching completes on all nodes. Proceed? [y/n]
“Why we ran opatchauto-72030 in non-rolling mode—and why I’d do it again.”
The entire cluster would go down. All nodes. At once. Patch applied to the Oracle home while the databases were offline. Then a full restart. It was the nuclear option—sledgehammer surgery.
The first node went dark. Then the second. Then the third.
It was the kind of line that made Leo’s coffee taste like sand. He stared at the screen, the green-on-black terminal casting sharp shadows under his eyes. Two days with no sleep, and now this—.
Leo typed:
opatchauto-72030: Prerequisite check failed — rolling mode not supported for this patch on RAC with custom service affinities.
He closed the terminal, finished his cold coffee, and wrote the post-mortem email subject line before sleep took him:
