The server only runs a single replica and the reverted Redis-adapter
stack means two overlapping pods cannot safely share state. During a
rolling update we want the old pod killed first, brief downtime, then
the new pod started — never two pods live at once.
replicas: 1 (unchanged)
maxUnavailable: 0 -> 1
maxSurge: 1 -> 0
Kirill Dubovitskiy
9b64c68b21