On Tue, Jun 9, 2015 at 2:35 PM, Aiman Parvaiz <ai...@flipagram.com> wrote:
> Thanks Sean, in this scenario also I would end up running 2 versions of > Cassandra as I am planning to do a rolling upgrade and hence zero downtime. > Upgrading in place one node at a time would lead to running 2 versions, > please let me know if I am missing something here. > Running a cluster with nodes of two different versions during a rolling upgrade, for the duration of the upgrade, is (mostly) supported [1]. Modifying cluster topology (adding or removing or replacing nodes) during such an upgrade is not. It is a fair statement that with very large clusters and very slow upgradesstables, the ability of any operator to operate in this manner approaches not-possible. I don't know how people with truly huge clusters deal with this race. =Rob [1] some operations are not supported possible with some combinations of versions. for example one cannot repair in some cases.