там ещё товарищ о юзкейсе пишет: “I have a use-case for pod affinity. I would like to run HDFS in HA mode on kubernetes. There are 3 types of nodes: namenodes, journalnodes and datanodes. For performance, it makes sense to run journalnodes on the same nodes as namenodes. From my understanding, each pod is assigned an IP address, but in the case of HDFS, the namenode and journalnodes need to have their own IP addresses, so pod affinity would be a great candidate for this.”