We plan to upgrade our cluster, which is currently running from 2.0.9 to 2.2.6. According to the documentation, and some blogs people update cassandra inplace, i.e. Remove the node from the ring update and add it again. We are skeptical of this approach, because everything can go wrong (this is a high transaction base with a lot of QPS).
So, we planned to add a new data cluster to the cluster, which should update the version of cassandra (2.2). Thus, the installation should have two data centers, one old (2.0.9) and another new (2.2.6)
This data center is just a backup. When the data center becomes stable, we will change the client connection to this data center, and if it works well, we will go with this data center and close the old data center, otherwise we can return to the old data center and debug what went wrong So.
Is this process enough or should we upgrade?
Is it possible to use two versions of cassandra (2.0 and 2.2) in the data center.
Is there a drop in this approach?
source share