I think the crux of the issue is that nobody's done the work to find out. On Sat, Oct 28, 2017 at 8:24 PM, Andrew Purtell <[email protected]> wrote:
> Is there anything in 1.4* not in 1.2 that would warrant that? Otherwise I > agree, not requiring an intermediate upgrade step would be best. Requiring > a double upgrade would be super operator unfriendly. > > * - Should everything go reasonably well we will have a 1.4.0 release > before December. I'm going to do the first RC next week. > > > > On Oct 28, 2017, at 5:09 PM, Mike Drob <[email protected]> wrote: > > > > Ok, looks like there is some enough feelings that we don't need to worry > > about downgrades. > > > > What about the other part of Sean's question? Do we need to support > rolling > > upgrades to 2.0 from any 1.x, or is it fair game to require a specific > > minimum version? > > > > If we felt that it simplified things, I'd even be happy with a minimum > > 1.4->2.0 upgrade path, but 1.4 doesn't exist yet and I don't feel like > > that's something we can dictate to users. Maybe it's ok to set the > minimum > > line at 1.2? If we end up moving the stable pointer, that makes for a > > stronger argument for a newer minimum version. > > > > Mike > > > >> On Sat, Oct 28, 2017 at 4:46 PM, Josh Elser <[email protected]> > wrote: > >> > >> +1 -- well put, Andrew. > >> > >> > >>> On 10/28/17 1:17 PM, Andrew Purtell wrote: > >>> > >>> I would not like to see downgrades as a goal. This would be new. We've > >>> not done it before. Laudible goal, but we are clearly stretched > already. > >>> > >>> > >>>> On Oct 28, 2017, at 10:08 AM, Mike Drob <[email protected]> wrote: > >>>> > >>>> If downgrades are a later goal, does that mean somebody could go from > >>>> some > >>>> 1.x to 2.0 to 2.y then back to 1.x? > >>>> > >>>>> On Fri, Oct 27, 2017 at 10:42 PM, Sean Busbey <[email protected]> > wrote: > >>>>> > >>>>> I'd like to make downgrades a non-goal. I'd love us to support > >>>>> downgrades eventually, but that's a feature in its own right and I > >>>>> don't think we have time to get it done and still have a 2.0.0 GA in > a > >>>>> reasonable time frame. > >>>>> > >>>>> On Fri, Oct 27, 2017 at 10:40 PM, Sean Busbey <[email protected]> > >>>>>> wrote: > >>>>>> A recent JIRA about our hfile format[1] has got me thinking about > >>>>>> expectations for upgrading. The specifics of that JIRA aren't > terribly > >>>>>> important; it's the general issue I want to talk about. A > >>>>>> simplification of the mismatch in expectations between two groups is > >>>>>> that some folks place the bar for "we support rolling upgrade" at > >>>>>> rolling upgrade from 1.y.z* versions generally and others are > >>>>>> comfortable requiring an initial upgrade to some later 1.y.z version > >>>>>> first. > >>>>>> > >>>>>> Have we documented what our goals are for upgrades this major > release? > >>>>>> Do we know what we have to do to get there? I've seen a few one-off > >>>>>> JIRAs to fix particular problems, but not really a plan. > >>>>>> > >>>>>> We should discuss here a bit. > >>>>>> > >>>>>> When things have some consensus is anyone willing to take point on > >>>>>> writing up the results in a scope document of sorts? I have a few > good > >>>>>> examples to point you to, though they're all for features. > >>>>>> > >>>>>> > >>>>>> [1]: https://issues.apache.org/jira/browse/HBASE-19052 > >>>>>> > >>>>> > >>>>> >
