New three node cluster in AKS constantly electing a new master

EventStoreDB
1

I’ve created a new AKS cluster and installed eventstore via its helm chart with the following values:
clusterSize: 3

persistence:

enabled: true

size: 128Gi

``

The TCP connections between the pods seem to be dropping consistently, causing a lot of churn in the master election process (so much so that the pods are eating between 0.5 and 2 cores at idle doing only master election). There are no other workloads running in the cluster, and nothing is attempting to use eventstore, it’s completely idle. The admin UI is available, and one can see the master cycling there as well. The eventstore nodes do not seem to progress past PreReplica. The AKS node size is D4s_v3, if that’s relevant.

Here’s a link to the logs from one of the evenstore nodes

Thanks for any help

2

I’ve also attempted building a new deployment running 5.0.2 instead of the default 4.1.1-hotfix1, and encounter the same behavior.

3

This appears not to be an issue on clusters with accelerated networking enabled, and also AKS clusters based on VMSS seem to have a regression over the old VMAS method, and do not properly enable accelerated networking