Well I don't count with a lot more 4.x releases - maybe I'm wrong-headed . For master branch and cdh6 we'd be looking at spark 2.x
Part of the success of a project is about version stability. Not a lot of corporate projects can afford keep upgrading to the latest versions - think about it, you're in production, with a few thousand lines code running spark 1.6 ... And for upgrading to 4.14 you need to review all of this spark code -and maybe recompile to scala 2.11 btw- . It doesn't make sense. Until now 4.x was pretty stable and in my opinion it should've never been migrated to spark 2 and java 8. Minor versions should keep certain stability in terms of dependencies. All these changes should've come with phoenix 5. But you're right it needs a sensible solution as 4.14.1 is already out and compiled with java8. On Fri, 30 Nov 2018, 23:28 Thomas D'Silva <[email protected] wrote: > Spark 1.6 is really old and doesn't support the newer Datasource v2 api > that we have been looking at integrating with. > As Alex points out you will might end up having to revert a lot more > commits in the future. > Seems like the queryserver and phoenix-spark modules on the cdh branch > would end up diverging a lot from the standard open source branch. > > > On Fri, Nov 30, 2018 at 2:23 PM Alex Araujo <[email protected]> wrote: > > > > Only a downgrade to spark 1.6 ( > > changes are only needed in a few IT, basically going back from Datasets > to > > Dataframes) and going back to Avatica 1.10 ( involving reverting > > PHOENIX-4755, PHOENIX-4750 and PHOENIX-4805 ). > > > > We're talking about the 4.x branches, right? Doesn't seem prudent to do > it > > there as down-streamers may already be relying on the newer versions. > > > > On Fri, Nov 30, 2018 at 4:18 PM Pedro Boado <[email protected]> > wrote: > > > > > Thinking about typical server installation in a corporate environment > I'd > > > keep everything compatible with the same JVM version. > > > > > > I've gone down the route for the cdh branch. Full JDK 7 compatibility > > > doesn't require changes in phoenix-core. Only a downgrade to spark 1.6 > ( > > > changes are only needed in a few IT, basically going back from Datasets > > to > > > Dataframes) and going back to Avatica 1.10 ( involving reverting > > > PHOENIX-4755, PHOENIX-4750 and PHOENIX-4805 ). > > > > > > > > > On Fri, 30 Nov 2018, 18:57 Thomas D'Silva <[email protected] > > > <[email protected]> wrote: > > > > > > > We could allow individual submodules like the queryserver, or > > > phoenix-spark > > > > to be built with their own compiler configuration (1.8+). > > > > This would allow these modules to use Java 1.8 features. I think this > > > would > > > > be a good compromise given that they depend on > > > > features that are provided by versions of spark and avatica that no > > > longer > > > > support Java 1.7. > > > > We can still ensure phoenix-core supports Java 1.7. You would have to > > > skip > > > > building modules that require Java 1.8, WDYT? > > > > > > > > On Thu, Nov 29, 2018 at 6:11 PM Jaanai Zhang <[email protected] > > > > > > wrote: > > > > > > > > > I'd vote for keep using java7 on 4.x branches. if upgrades to > java8, > > it > > > > > will impact users who want to upgrade the latest 4.x branches. they > > > must > > > > > consider using java8 in their running environments, maybe their > > > > libraries > > > > > do not support java8, then they have to give up to upgrade. So I > > think > > > > that > > > > > drops support java7 is not friendly for some users. > > > > > > > > > > > > > > > ---------------------------------------- > > > > > Jaanai Zhang > > > > > Best regards! > > > > > > > > > > > > > > > > > > > > Pedro Boado <[email protected]> 于2018年11月30日周五 上午6:13写道： > > > > > > > > > > > I'd vote for keep compiling 4.x branches in java7. It makes sense > > as > > > > it's > > > > > > just a new minor release. > > > > > > > > > > > > It's pretty easy reverting back to spark 1.6 and also avatica > > > > dependency > > > > > > could be reverted to the previous version. > > > > > > > > > > > > On 29 Nov 2018 21:41, "Thomas D'Silva" <[email protected]> > > > wrote: > > > > > > > > > > > > We have traditionally followed HBase's java support (see > > > > > > https://hbase.apache.org/book.html#basic.prerequisites). The > > > > > > phoenix-queryserver module has a dependency on Avatica which does > > not > > > > > > support Java 1.7. The phoenix-spark module depends on spark 2.3.2 > > > which > > > > > > also does not support Java 1.7. Do folks feel we should continue > to > > > > > provide > > > > > > support Java 1.7 on the 1.x branches? > > > > > > > > > > > > > > > > > > > > > On Fri, 30 Nov 2018, 18:57 Thomas D'Silva <[email protected] > wrote: > > > > > > > We could allow individual submodules like the queryserver, or > > > phoenix-spark > > > > to be built with their own compiler configuration (1.8+). > > > > This would allow these modules to use Java 1.8 features. I think this > > > would > > > > be a good compromise given that they depend on > > > > features that are provided by versions of spark and avatica that no > > > longer > > > > support Java 1.7. > > > > We can still ensure phoenix-core supports Java 1.7. You would have to > > > skip > > > > building modules that require Java 1.8, WDYT? > > > > > > > > On Thu, Nov 29, 2018 at 6:11 PM Jaanai Zhang <[email protected] > > > > > > wrote: > > > > > > > > > I'd vote for keep using java7 on 4.x branches. if upgrades to > java8, > > it > > > > > will impact users who want to upgrade the latest 4.x branches. they > > > must > > > > > consider using java8 in their running environments, maybe their > > > > libraries > > > > > do not support java8, then they have to give up to upgrade. So I > > think > > > > that > > > > > drops support java7 is not friendly for some users. > > > > > > > > > > > > > > > ---------------------------------------- > > > > > Jaanai Zhang > > > > > Best regards! > > > > > > > > > > > > > > > > > > > > Pedro Boado <[email protected]> 于2018年11月30日周五 上午6:13写道： > > > > > > > > > > > I'd vote for keep compiling 4.x branches in java7. It makes sense > > as > > > > it's > > > > > > just a new minor release. > > > > > > > > > > > > It's pretty easy reverting back to spark 1.6 and also avatica > > > > dependency > > > > > > could be reverted to the previous version. > > > > > > > > > > > > On 29 Nov 2018 21:41, "Thomas D'Silva" <[email protected]> > > > wrote: > > > > > > > > > > > > We have traditionally followed HBase's java support (see > > > > > > https://hbase.apache.org/book.html#basic.prerequisites). The > > > > > > phoenix-queryserver module has a dependency on Avatica which does > > not > > > > > > support Java 1.7. The phoenix-spark module depends on spark 2.3.2 > > > which > > > > > > also does not support Java 1.7. Do folks feel we should continue > to > > > > > provide > > > > > > support Java 1.7 on the 1.x branches? > > > > > > > > > > > > > > > > > > > > >
