I have a 7 node cluster. I have an app 'A' having 3 replicas running in random 3 nodes. I have another app 'B' having 5 replicas, now I have to assign only 1 of the replicas of 'B' to the same node as 'A'. Other 4 replicas can run in different nodes.
Node 1 - A1
Node 2 - B2
Node 3 - B3
Node 4 - A2, B1
Node 5 - B4
Node 6 - A3
Node 7 - B5By those rules, you can achieve something like bellow:
Node 1: A1, B1
Node 2: A2, B2
Node 3: A3, B3
Node 4: B4
Node 5: B5
... ...
But if you add taint to Node:1,2 such that no B pods will be able to be scheduled on those nodes. And add taint to Node:4,5,6,7 such that no A will be able to be scheduled on those nodes.
Then you will end up having the following distribution:
Node 1 - A1
Node 2 - A2
Node 3 - A3, B1
Node 4 - B2
Node 5 - B3
Node 6 - B4
Node 7 - B5