Eight to ten seconds.
That was the site blip we expected while moving load balancers and core routers onto separate power circuits.
The maintenance would force failovers across the network. Firewalls, routers, and load balancers all had to move. We did not expect a prolonged outage, but pretending customers would see nothing was not useful.
So I gave the team the number.
The reason for the work was larger than the interruption. Too much critical equipment shared the same power risk. A single failure could take out components that were supposed to protect one another.
Redundancy only counts when the redundant pieces do not fail together.
We had already learned that lesson from a firewall failover that did not happen automatically. This maintenance addressed the physical version of the same mistake. Separate boxes were not enough if power could remove both at once.
That sounds obvious. Data centers are full of expensive ways to forget it. Two devices can have separate names, separate cables, and separate configurations while still depending on one breaker, one rack, or one badly placed power strip.
The maintenance was physical risk reduction. Move the equipment. Use distinct circuits. Trigger the failovers on purpose. Watch what happens.
The short blip was the price of proving the new arrangement before a real power event did it for us.
Customers would not care why the site disappeared for eight seconds. They should not have to. Our responsibility was to make the planned interruption small and the unplanned one less likely.
I promised an update when the work was complete. Until then, eight to ten seconds was both the plan and the test.
If it lasted longer, we had learned something.