As we know in a Redis Cluster, data is divided into shards, with each shard being managed by a master node and one or more replica nodes. The problem I want to tackle is the case where master and replica should never be taken down together, as I will be losing the keys stored there and decrease the overall availability of the system. I have already set podAntiAffinity so to prefer to schedule pods on different nodes, so now I want to make sure that nodes that hold a "pair" of master-slave shard are never taken down altogether, there is the PodDistributionBudget option and setting it to 1 for maxUnavailable, or some preStopHook magic where we block if one of the paired nodes is not available, but they do not feel natural to me, so I wonder if there is like a concept for failure-domains or upgrade-domains that I am missing on?
- Here's a decent guide on getting this set up: dragonflydb.io/guides/redis-kubernetes scroll down to the "persistence and data durability" part for some pointers.Tijmen– Tijmen2024-08-06 13:42:36 +00:00Commented Aug 6, 2024 at 13:42
Add a comment |