MaxSurge and MinAvailable calculated incorrectly about agones HOT 4 CLOSED

Need to dig in deeper, but feels like there is some overlap with #3438 here.

from agones.

It is possible that it is a contributing factor, but the issue here stands alone, and the situation is its own issue. Where in #3438 the Rollout gets stuck, here we are removing way too many Ready GameServers, to the point of having none left if your buffers are too low. I would naively assume the Rollout will still get stuck were this fixed, it would be get stuck slower.

from agones.

I don't disagree - more I'm thinking if we dig in and solve one, should probably solve the other at the same time. I expect they'll cross paths in implementation.

from agones.

Below(my) case is similar to issue #3438, and I believe that the feature in PR #3762 could also be beneficial here.

• If an update is made during live service in a game, new game server sets (GSS) will continuously receive allocations.
• In such cases, the new GSS will eventually match the fleet.spec value, maintaining a buffer amount of ready replicas.
• However, there can be scenarios where the old GSS remains, and the update gets stuck.
• According to this section in the code, if the number of minAvailable is greater than the ready replicas, the update halts.
• Typically, the number of ready replicas of the new GSS will be close to the buffer. If the buffer is 20% and the max unavailable value is also 20%, then the min available will be 80% of the fleet, meaning ready replicas cannot exceed this number.
• Using the new feature would calculate based on the ready replicas, thus eliminating such cases.

Additionally, if the ready count for the old GSS drops to zero, setting the replicas to zero could also be beneficial. This would facilitate a quicker transition to new servers, particularly in cases where the update has already been completed or there is a shortage of game servers.

from agones.