Why not take it all at once and avoid the need for 2 deprecation cycles? With that change folks with existing Mesos clusters would be able to deploy Aurora without changing the attributes flag on every slave (which requires rebooting the entire cluster to take effect).
On Mon, Oct 6, 2014 at 5:24 PM, Zameer Manji <zma...@twopensource.com> wrote: > I currently don't plan on removing 'host' right now or in the future. If > there is a need to remove it I would prefer to tackle it later. > > On Mon, Oct 6, 2014 at 5:13 PM, Kevin Sweeney <kevi...@apache.org> wrote: > > > Will 'host' be deprecated as well? > > > > On Mon, Oct 6, 2014 at 4:30 PM, Zameer Manji <zma...@twopensource.com> > > wrote: > > > > > I have put up a review <https://reviews.apache.org/r/26391/> which > > removes > > > the special casing and handling of the ‘rack’ constraint. This is a > > > breaking change because Aurora will no longer inject a rack constraint > > of 1 > > > if a job omits a rack constraint value. However this means that Aurora > > will > > > be able to schedule jobs on to slaves which don’t have the ‘rack’ > > attribute > > > and slaves that want to accept tasks from Aurora don’t need to have the > > > ‘rack’ attribute set. > > > > > > Before the next release of Aurora (0.6.0) you should check if you are > > > relying on the injected value for scheduling decisions and if so > > explicitly > > > encode a rack limit of 1 in the job's aurora config. > > > > > > -- > > > Zameer Manji > > > > > > > > > -- > Zameer Manji >
