20k+ nodes and 200k+ allocs. To be fair, Kubernetes cannot support this large of a cluster.

Most of my issues with it aren't related to the scale though. I wasn't involved in the operations of the cluster (though I did hear many "fun" stories from that team), I was just a user of Nomad trying to run a few thousand stateful allocs. Without custom resources and custom controllers, managing stateful services was a pain in the ass. Critical bugs would also often take years to get fixed. I had lots of fun getting paged in the middle of the night because 2 allocs would suddenly decide they now have the same index (https://github.com/hashicorp/nomad/issues/10727)