K8s 1.26->1.33. 4 hours clicking “upgrade” in Oracle Cloud.
Turns out the nodepool is static, not dynamic. So I had to manually “add” node to each AZ, separately specify network for each! Here I fondly recalled the GKE, which automaitcally upgrades its nodes, hidden during maintenance.
And there I manually clicked “upgrade” in web-console, three times with pauses. Created nodepool on the second try, drained old nodes one by one (downtime of evicted pods 14 minutes! what the hell with the network there?) Continue clicking button three times… Created one more nodepool, drain old nodes… Two more clicks and decided to stop here - too much time spent already, and holy balls, v1.34 will be alive till the end of October 2026.