- You have a system whereby every instance has a 75%-79% uptime.
- Each instance's 21%-25% daily scheduled maintenance window can (generally) be arbitrarily-scheduled by the ops team.
How do you near guarantee that there's always one instance available?
If you said:
"manage a redundant cluster of at least 2 instances with non-overlapping maintenance windows"
Then you were correct!
BONUS QUESTIONS:
Q: What's the max number of ice cream machines found in a McDonald's location?
Q: What's the cluster size of the McDonald's ice cream monitoring system?
Q: How much money does McDonald's have?
Q: Why the fuck?
If you said: "1", "at least 3", "metric fuck-tons", and "I don't know", respectively, then you were correct!
The answer to #4 is probably that the goal isn't to always have ice cream. The goal is to run a profitable business.
Doubling the cost of the ice cream infrastructure to avoid a 25% downtime may not be a profitable venture -- especially if the downtime is scheduled to occur while the store is closed.
Also, you can use data to schedule the maintenance when the probability of someone ordering ice cream is low. That 21-25% downtime results in significantly fewer people wanting ice cream not being able to get it.
If you said:
Then you were correct!BONUS QUESTIONS:
Q: What's the max number of ice cream machines found in a McDonald's location?
Q: What's the cluster size of the McDonald's ice cream monitoring system?
Q: How much money does McDonald's have?
Q: Why the fuck?
If you said: "1", "at least 3", "metric fuck-tons", and "I don't know", respectively, then you were correct!