We also shut down all of the schedulers, upgrade to the new version, then bring them all back up. This isn’t a hard requirement, but it has made things a bit simpler for us in practice to always know we’re dealing with the same version of the schedulers on all hosts at once.
> On Sep 21, 2015, at 10:55 AM, Zameer Manji <[email protected]> wrote: > > I don't think the 0.7 -> 0.9 upgrade path has been tested. I think it is > advisable to upgrade from 0.7 to 0.8 then later upgrade from 0.8 to 0.9. I > think Jeff's upgrade steps are reasonable and shouldn't give you any > trouble. Note that upgrading the executors won't affect existing tasks. If > you want the current tasks on the system to adopt the latest executor you > will need to upgrade the executor and then roll your tasks so they can be > relaunched with the updated binary. This isn't necessary to do but keep > this in mind if you are expecting tasks to leverage the latest executor > functionality. > > On Sun, Sep 20, 2015 at 8:30 PM, Jeff Schroeder <[email protected]> > wrote: > >> When I did this, I updated the executors, then the observers, and then the >> schedulers, one at a time, in a rolling fashion. Seemed to work fine, but >> the clusters were lightly utilized. >> >> On Sun, Sep 20, 2015 at 9:40 PM, Mauricio Garavaglia < >> [email protected]> wrote: >> >>> Hi guys, >>> >>> I'm about to upgrade a cluster from 0.7 to 0.9 and was wondering what are >>> the things to consider to make it as seamless as possible? >>> >>> For example, update the executors first and then the schedulers; just >>> update the schedulers one at a time, etc. >>> Thanks! >>> >>> Mauricio >>> >> >> >> >> -- >> Jeff Schroeder >> >> Don't drink and derive, alcohol and analysis don't mix. >> http://www.digitalprognosis.com >> >> -- >> Zameer Manji >> >> <http://www.digitalprognosis.com>
