I crafted an OpenStack HOT template to automatically deploy a single-datacenter, 3-node Cassandra cluster (1 seed node only) from scratch. However, I'm unable to deterministically create it without bootstrap token collisions, since all cassandra nodes are initialized at the same time: sometime all 3 nodes are shown with nodetool status, sometimes one is ignored. I may have an idea on how to circumvent this by using a shared volume between virtual machines to "sequentialize" the initialization processes, but that would be the last resort.
Any ideas? Should I try to manually allocate the initial tokens (How? dis/advantages?)? I'm using the default configuration with auto_bootstrap=false for all nodes. Thank you