Split memberlist

[JBodkin-Amphora]

I'm running mimir on Azure Kubernetes using spot instances across 3 node pools, with each node pool in a different az availability zone. There are 2 distributors (running in different zones) and 3 ingesters (one in each zone). When Azure pre-empts the node, I'm seeing issues where two different memberlists/rings are getting created under the classic architecture, instead of joining to the existing ring which leads to data not being ingested.

Is there some configuration options I should set? I'm using the mimir helm chart and I haven't overridden any options for the memberlist except for the cluster label

cortex_ring_members{name="ingester", state="ACTIVE"}

In the above image, you can see that 3 ingesters were active and then splits into 2 and 1. I'm expecting a small difference in the graph for a few minutes rather than for it to run in a split memberlist for hours on end.