Yeah, etcd was the main culprit, but latency was 150-300ms in my case. At 3 nodes, it was relatively stable (had an issue every week or so that lasted < 5 min), but at 4 the camel's back broke.