That was our problem! We had a disagreement of scheme! Three nodes in our six-node cluster were in a different scheme: once the node has completed the boot trap and joins the cluster, the option JVM_OPTS=”$JVM_OPTS -Dcassandra.replace_address=” must be removed from the cassandra file env.sh, otherwise the node will not be able to start during a reboot. This is a lack of function cassandra.replace_address. Many operators usually worry about replacing a dead node and therefore forget to update the file cassandra-env.sh after the order is complete. This is the reason why cassandra-7356 has been increased and a new option has been added; cassandra.replace_address_first_boot. This option works after Cassandra starts for the first time and the replacement node is inserted into the cluster. Then, the option is ignored on all subsequent reboots. However, this solution is not a miracle weapon, as it requires the appropriate node to be available for a consistent range moment during a boot time. This may not always be possible, and in such cases, there are two options: Metadata#getKeyspaces returns a client representation of the database schema: I guess it`s more of a function of the number of nodes; We have 2 nodes in deployment-prep, 6 in dev (2 instances x 3 hosts) and 18 in production (3 instances * 6 nodes). And FWIW, we had some problems with the elaboration of schemas in the development (not to this extent, but more than the preparation of the availability). I didn`t try to restart Reaper after the first error, but I can`t leave schema conflicts in the cluster, and it looks like migrating Reaper creates this lag, which is why I`m reporting it…