On Tue, Mar 24, 2015 at 11:02 PM, James Taylor <jamestay...@apache.org> wrote:
> The master branch already includes PHOENIX-1642, so we just keep it > there. No need to revert anything or cherry-pick anything. Every > commit being done to 4.x-HBase-1.x is being done for master (that's > why it's just wasted overhead until it's needed). > Like the pom.xml version, and https://issues.apache.org/jira/browse/PHOENIX-1766 have to be reverted if we fork the 4.x-HBase-1.0 branch from master, no? > > Your plan sounds fine, except this step isn't necessary (and no revert > of anything currently in master is necessary): > - After we fork 4.x-HBase-1.0, we cherry-pick PHOENIX-1642. > > Thanks, > James > > On Tue, Mar 24, 2015 at 10:53 PM, Enis Söztutar <e...@apache.org> wrote: > > On Tue, Mar 24, 2015 at 5:09 PM, James Taylor <jamestay...@apache.org> > > wrote: > > > >> I'm fine with a 4.4 release for HBase 1.0, but it depends on demand - > >> do our users need this? > > > > > > I think so. HBase-1.0 usage is picking up, and we already saw users > asking > > for it. Though as usual, everything depends on whether there is enough > > bandwidth to do the actual work (in terms of release, testing, porting, > > etc). > > > > > >> I think doing that out of master will work and > >> we can create a branch for the release like we do with our other > >> releases. > >> > >> When sub tasks of PHOENIX-1501 are ready, I think we'd want to put > >> those in master and prior to that we'll need to create a > >> 4.x-HBase-1.0. So we'll save the overhead of maintaining duplicate > >> branches until that point. > >> > >> Make sense? > >> > > > > Forking 4.4 release for HBase-1.0 from master seems strange. We have to > > revert back the version, and make sure that it is really identical to the > > 4.x-HBase-0.98 branch except for PHOENIX-1642. However, if you think that > > an extra branch is really a lot overhead, maybe we can do this: > > - Delete 4.x-HBase-1.x now. > > - Keep 4.x-HBase-0.98 and master branches. > > - Fork 4.x-HBase-1.0 branch when whichever of these happens earlier: > > -- 4.4 is about to be released > > -- PHOENIX-1501, or PHOENIX-1681 or PHOENIX-1763 needs to be > committed. > > - After we fork 4.x-HBase-1.0, we cherry-pick PHOENIX-1642. > > - When PHOENIX-1501/PHOENIX-1681 and PHOENIX-1763 is ready and > HBase-1.1.0 > > is released we can fork 1.1 branch. > > > > Will that work? I am up to doing the work as long as we have a plan. > > > > Enis > > > > > >> > >> On Tue, Mar 24, 2015 at 4:50 PM, Enis Söztutar <e...@apache.org> wrote: > >> >> > >> >> We've been putting stuff on feature branches that need more time. > When > >> >> PHOENIX-1681 or other sub tasks of PHOENIX-1501 are ready (after > >> >> HBASE-12972 is in), we'll need a branch specific to HBase 1.1. Until > >> >> then, I think it's just unneeded overhead. > >> > > >> > > >> > That is why the branch for 1.1 is not created yet. The current branch > >> > 4.x-HBase-1.x supports ONLY HBase-1.0 release, not 1.1 release. I had > >> named > >> > the branch 1.x hoping that it will support both, but it seem that we > >> cannot > >> > do this. Should we rename the branch to 4.x-HBase-1.0 so that it is > >> > explicit? I am assuming that we are still interested in making a > Phoenix > >> > release (4.4) which will support HBase-1.0.x series. If that is not > the > >> case > >> > and we only want to support 1.1, we can make the decision at the > expense > >> of > >> > users who want to run Phoenix with HBase 1.0 releases. > >> > > >> > HBASE-12972 will not be committed to HBase-1.0.x releases, only to > 1.1.x > >> > releases, which (with the other changes like HBASE-11544) requires yet > >> > another branch or release vehicle for HBase-1.1 support, different > than > >> 1.0 > >> > support. > >> > > >> >> > >> >> > >> >> >> > >> >> >> > > >> >> >> > On Tuesday, March 24, 2015, Enis Söztutar <enis....@gmail.com > >> >> >> > <javascript:_e(%7B%7D,'cvml','enis....@gmail.com');>> wrote: > >> >> >> > > >> >> >> >> You mean get rid of 4.x-HBase-1.x branch? It is already created > >> and > >> >> >> >> has > >> >> >> >> PHOENIX-1642. It builds with HBase-1.0, but not 1.1. > >> >> >> >> > >> >> >> >> Enis > >> >> >> >> > >> >> >> >> On Sun, Mar 22, 2015 at 12:33 PM, James Taylor > >> >> >> >> <jamestay...@apache.org> > >> >> >> >> wrote: > >> >> >> >> > >> >> >> >>> I think we can stick with just 4.x-HBase-0.98 and master > branch > >> for > >> >> >> >>> now until we need to work simultaneously on a Phoenix release > >> that > >> >> >> >>> supports both HBase 1.0 and HBase 1.1. Seems like the earliest > >> >> >> >>> would > >> >> >> >>> be closer to an HBase 1.1 release. Any idea when that might > be? > >> >> >> >>> Otherwise, the overhead of keeping master in sync with > >> >> >> >>> 4.x-HBase-1.x > >> >> >> >>> is wasted effort (as they'll be exactly the same until then). > >> >> >> >>> > >> >> >> >>> Thoughts? > >> >> >> >>> > >> >> >> >>> On Fri, Mar 20, 2015 at 4:53 PM, James Taylor > >> >> >> >>> <jamestay...@apache.org> > >> >> >> >>> wrote: > >> >> >> >>> > Is this fixed yet? If not, would it be possible for you to > set > >> >> >> >>> > the > >> >> >> >>> > pom > >> >> >> >>> > to HBase-1.0.1 instead so that master will build? Just don't > >> want > >> >> >> >>> > to > >> >> >> >>> > leave it in a broken state. > >> >> >> >>> > Thanks, > >> >> >> >>> > James > >> >> >> >>> > > >> >> >> >>> > On Thu, Mar 19, 2015 at 7:31 PM, Enis Söztutar < > >> e...@apache.org> > >> >> >> >>> wrote: > >> >> >> >>> >> About the 4.x-HBase-1.x branch, it seems that I have spoken > >> too > >> >> >> >>> >> soon. > >> >> >> >>> >> Current branch head does not compile with latest > >> >> >> >>> >> HBase-1.1.0-SNAPSHOT: > >> >> >> >>> >> > >> >> >> >>> >> It seems the RegionScanner changes are the problem. Let me > >> look > >> >> >> >>> >> into > >> >> >> >>> how we > >> >> >> >>> >> can resolve those for future compatibility. > >> >> >> >>> >> > >> >> >> >>> >> Enis > >> >> >> >>> >> > >> >> >> >>> >> On Thu, Mar 19, 2015 at 2:15 PM, Enis Söztutar < > >> e...@apache.org> > >> >> >> >>> wrote: > >> >> >> >>> >> > >> >> >> >>> >>> Hi, > >> >> >> >>> >>> > >> >> >> >>> >>> As per private PMC threads and the dev discussions [1], I > >> have > >> >> >> >>> created two > >> >> >> >>> >>> new branches for 4.x development for supporting both > >> HBase-0.98 > >> >> >> >>> >>> and > >> >> >> >>> >>> HBase-1.0 versions. The goal is to have 4.4.0 and 4.5.0, > etc > >> >> >> >>> releases which > >> >> >> >>> >>> support both of the HBase versions and possibly > HBase-1.1.0+ > >> as > >> >> >> >>> >>> well. > >> >> >> >>> >>> > >> >> >> >>> >>> See [1] for why the branches are needed (this seems like > the > >> >> >> >>> >>> least > >> >> >> >>> bad > >> >> >> >>> >>> approach). Here are the changes I did for this: > >> >> >> >>> >>> > >> >> >> >>> >>> BRANCH CHANGES: > >> >> >> >>> >>> - Committed PHOENIX-1642 to master > >> >> >> >>> >>> - Created branch-4.x-HBase-0.98. Pushed to git repo > >> >> >> >>> >>> - Created branch-4.x-HBase-1.x. Pushed to git repo > >> >> >> >>> >>> - Changed versions to be 4.4.0-HBase-0.98-SNAPSHOT and > >> >> >> >>> >>> 4.4.0-HBase-1.x-SNAPSHOT respectively in above branches > >> >> >> >>> >>> - Cherry-picked PHOENIX-1642 to branch-4.x-HBase-1.x > >> >> >> >>> >>> - Deleted branch named "4.0". (there is no rename of > branches > >> >> >> >>> >>> in > >> >> >> >>> >>> git) > >> >> >> >>> >>> > >> >> >> >>> >>> I have named the branch 4.x-HBase-1.x instead of suffix > >> >> >> >>> >>> HBase-1.0 > >> >> >> >>> >>> in > >> >> >> >>> hopes > >> >> >> >>> >>> that further HBase-1.1, 1.2 can be supported in this > branch > >> and > >> >> >> >>> >>> we > >> >> >> >>> can get > >> >> >> >>> >>> away without branching again for 1.1. See especially > >> >> >> >>> >>> HBASE-12972. > >> >> >> >>> >>> We > >> >> >> >>> can > >> >> >> >>> >>> change this later on if it is not the case. > >> >> >> >>> >>> > >> >> >> >>> >>> > >> >> >> >>> >>> JENKINS CHANGES: > >> >> >> >>> >>> - Disabled Phoenix-4.0 job (Lets keep it around for a > couple > >> of > >> >> >> >>> >>> days > >> >> >> >>> just > >> >> >> >>> >>> in case) > >> >> >> >>> >>> - Created new jobs for these two branches: > >> >> >> >>> >>> > >> >> >> >>> >>> > >> https://builds.apache.org/view/All/job/Phoenix-4.x-HBase-0.98/ > >> >> >> >>> >>> https://builds.apache.org/job/Phoenix-4.x-HBase-1.x/ > >> >> >> >>> >>> > >> >> >> >>> >>> The build should be similar to the previous 4.0 branch > >> builds. > >> >> >> >>> >>> > >> >> >> >>> >>> > >> >> >> >>> >>> JIRA CHANGES: > >> >> >> >>> >>> - Renamed release version 4.4 in jira to 4.4.0 > >> >> >> >>> >>> > >> >> >> >>> >>> > >> >> >> >>> >>> Further changes coming shortly unless objection: > >> >> >> >>> >>> - Delete jenkins job > >> >> >> >>> >>> https://builds.apache.org/view/All/job/Phoenix%202.0/ > (does > >> >> >> >>> >>> not > >> >> >> >>> seem to > >> >> >> >>> >>> be used for more than 1 year) > >> >> >> >>> >>> - Delete jenkins job > >> >> >> >>> https://builds.apache.org/view/All/job/Phoenix-2.0/ > >> >> >> >>> >>> - Delete jenkins job > >> >> >> >>> https://builds.apache.org/view/All/job/Phoenix-4.0/ > >> >> >> >>> >>> > >> >> >> >>> >>> > >> >> >> >>> >>> How does this affect development and releases? > >> >> >> >>> >>> - Current master is version 5.0.0-SNAPSHOT. It builds > with > >> >> >> >>> >>> HBase-1.0.1-SNAPSHOT (from apache snapshots repo). > >> >> >> >>> >>> - branch-4.x-HBase-0.98 is very similar to old 4.0 > branch. > >> It > >> >> >> >>> builds with > >> >> >> >>> >>> HBase-0.98.9-hadoop2 > >> >> >> >>> >>> - branch-4.x-HBase-1.x is forked from > branch-4.x-HBase-0.98 > >> >> >> >>> >>> and > >> >> >> >>> builds > >> >> >> >>> >>> with HBase-1.0.1-SNAPSHOT. > >> >> >> >>> >>> - There should be two release artifacts (or releases > >> >> >> >>> simultaneously) for > >> >> >> >>> >>> 4.4 release. One will have version 4.4.0-HBase-0.98 and > the > >> >> >> >>> >>> other > >> >> >> >>> >>> 4.4.0-HBase-1.x. We can make it so that the RM creates > both > >> >> >> >>> >>> releases > >> >> >> >>> at the > >> >> >> >>> >>> same time, and the VOTE applies to both releases. > >> >> >> >>> >>> - All changes MUST be committed to both branches for > future > >> >> >> >>> >>> 4.x > >> >> >> >>> releases > >> >> >> >>> >>> unless it is HBase version specific. There is no way to > >> >> >> >>> >>> auto-enforce > >> >> >> >>> it, so > >> >> >> >>> >>> all committers should take this into account. The patches > >> might > >> >> >> >>> differ > >> >> >> >>> >>> sligtly. Before the release RM may do some manual checks > to > >> >> >> >>> >>> ensure > >> >> >> >>> that > >> >> >> >>> >>> every patch is commmitted to both branches. > >> >> >> >>> >>> - Old 4.0 is deleted from git repository. Please > re-check or > >> >> >> >>> >>> rename > >> >> >> >>> your > >> >> >> >>> >>> local branches. Please do not push anything there (as it > will > >> >> >> >>> re-create the > >> >> >> >>> >>> branch). > >> >> >> >>> >>> - There is only one jira version 4.4.0, which should > apply > >> >> >> >>> >>> equally > >> >> >> >>> to > >> >> >> >>> >>> both release versions. If needed we can differentiate > these > >> in > >> >> >> >>> >>> jira > >> >> >> >>> as > >> >> >> >>> >>> well. Let me know. > >> >> >> >>> >>> - Before the 4.4.0 release, RM should fork both 4.x > branches > >> >> >> >>> >>> and > >> >> >> >>> name > >> >> >> >>> >>> them 4.4-HBase-XXX. At that time, we will have 1 master > >> branch, > >> >> >> >>> >>> 2 > >> >> >> >>> >>> of > >> >> >> >>> 4.x > >> >> >> >>> >>> branches and 2 of 4.4 branches. > >> >> >> >>> >>> > >> >> >> >>> >>> Let me know if you have further concerns. Let's see how > well > >> >> >> >>> >>> this > >> >> >> >>> process > >> >> >> >>> >>> works. > >> >> >> >>> >>> > >> >> >> >>> >>> Thanks, > >> >> >> >>> >>> Enis > >> >> >> >>> >>> > >> >> >> >>> >>> Ref: > >> >> >> >>> >>> [1] http://search-hadoop.com/m/lz2la1GgkPx > >> >> >> >>> >>> > >> >> >> >>> >>> > >> >> >> >>> > >> >> >> >> > >> >> >> >> > >> >> > > >> >> > > >> > > >> > > >> >
