Don't take my word for it, though I think I remember reading somewhere that your topology would keep running even if Nimbus is down (of course tasks on failed nodes will not be reallocated). So as far as your client-facing components are concerned you will probably not experience any downtime on a Nimbus upgrade.
I don't run mission-critical topologies so I don't know much about doing an upgrade with minimal downtime. On Sat, Jan 17, 2015 at 12:39 PM, R. Tyler Croy <[email protected]> wrote: > I dug through more of the documentation on the new-fangled Storm project > website and I didn't find docs speaking to upgrades of clusters. > > We currently have a number of topologies built against 0.9.2-incubating > and a > number of clusters running 0.9.2-incubating and I'd like to start charting > a > course for upgrading to 0.9.3 across the board for a number of reasons. > > Is it possible to do a piece-meal supervisor/nimbus upgrade? In effect: > > * Bring new 0.9.3 supervisors online > * Let 0.9.2 supervisors die > * ?? nimbus ?? (I'm really not sure how to zero-downtime upgrade nimbus > nodes) > > Or are you folks finding yourselves doing big-bang upgrades where you're > either > replacing or upgrading the whole cluster in place? > > I'm certain this isn't going to be a terrifically easy process but there's > some > Kafka fixes in 0.9.3 that will probably make it worth my time :) > > > Cheers > -R Tyler Croy >
