Tim asked: We haven't generally had clear guarantees about what OS versions we're > going to support. I ran into issues with > https://issues.apache.org/jira/browse/IMPALA-8508 where we couldn't run > python 3 on centos 6.4, for example. > Would it be possible to explicitly state what OS versions we're not going > to support in Impala 4.0?
+1 We tried to set clear expectations for Impala 3.0 when it started, see IMPALA-7273, IMPALA-7274, and IMPALA-6826 (for adding Ubuntu 18.04) I agree with Tim's proposal that the project should clearly state which operating system versions it supports. I propose that the following OS platforms should be retired for Impala 4.0: - CentOS 6: according to https://wiki.centos.org/About/Product maintenance updates will stop at the end of this November, and it also has an ancient kernel version. CentOS 7 has been available for almost 6 years now, so it is quite mature. Additionally, dropping CentOS 6 would also let us drop the Python 2.6 compatibility requirement for our Python scripts. - Ubuntu 14.04: https://ubuntu.com/about/release-cycle shows that it has been in Extended Security Maintenance (the last phase of the lifecycle) for almost a year now. Ubuntu 16.04 and 18.04 have been available for years, and 20.04 is expected later this spring. I wonder what the community's opinion is about SLES 12 (mentioned in IMPALA-7273). We could also express our intention to support CentOS 8 and Ubuntu 20.04 (when it becomes available), but these are just additions, not breaking changes. On Thu, Feb 13, 2020 at 1:58 AM Tim Armstrong <tarmstr...@cloudera.com> wrote: > Sounds good to me - that approach has been successful before - i.e. > switching the default in the major version, deprecating the old one, and > then removing it some number of releases later. Although in some cases we > have been slow to remove the code, e.g. the old decimal behaviour is still > accessible. > > On Wed, Feb 12, 2020 at 4:37 PM Vihang Karajgaonkar <vih...@cloudera.com> > wrote: > > > Yes, I think changing to local catalog can be done in 4.0 branch. Based > on > > the feedback received here I think we can start removing the code for 3.x > > catalog implementation after we fill in some of the gaps in local > catalog. > > We can selectively decide which ones are (eg. external data sources, HDFS > > caching) important and need to be supported in local catalog mode. Having > > just one mode of catalog will simplify implementation of newer features > > like catalog HA, fine-grained partition level metadata a lot. > > > > On Wed, Feb 12, 2020 at 3:46 PM Tim Armstrong <tarmstr...@cloudera.com> > > wrote: > > > > > Do we plan to switch the default catalog implementation to the local > > > catalog as well? > > > > > > On Wed, Jan 29, 2020 at 3:41 PM Tim Armstrong <tarmstr...@cloudera.com > > > > > wrote: > > > > > > > We haven't generally had clear guarantees about what OS versions > we're > > > > going to support. I ran into issues with > > > > https://issues.apache.org/jira/browse/IMPALA-8508 where we couldn't > > run > > > > python 3 on centos 6.4, for example. > > > > > > > > Would it be possible to explicitly state what OS versions we're not > > going > > > > to support in Impala 4.0? > > > > > > > > On Mon, Jan 27, 2020 at 12:05 PM Joe McDonnell < > > > joemcdonn...@cloudera.com> > > > > wrote: > > > > > > > >> Given the positive feedback, I will begin moving forward on the > > proposed > > > >> plan. I will start a new thread to discuss the Impala 3.4 release, > and > > > we > > > >> can start collecting/discussing the breaking changes that we want > for > > > >> Impala 4.0. > > > >> > > > >> Feedback about the plan is still welcome on this thread. > > Alternatively, > > > >> any > > > >> concerns can be raised in a new thread here on dev@. > > > >> > > > >> Thanks, > > > >> Joe > > > >> > > > >> On Wed, Jan 22, 2020 at 9:06 AM Laszlo Gaal < > laszlo.g...@cloudera.com > > > > > > >> wrote: > > > >> > > > >> > +1 > > > >> > > > > >> > I'd also add that bumping the major version of Impala opens the > > window > > > >> > for introducing breaking changes. > > > >> > > > > >> > I don't intend to hijack this mail thread for that purpose, > > > >> > but I'd like to suggest compiling such a list in the context of > the > > > >> version > > > >> > bump. > > > >> > > > > >> > Thanks, > > > >> > > > > >> > - LaszloG > > > >> > > > > >> > On Tue, Jan 21, 2020 at 6:48 PM Anurag Mantripragada < > > > >> anu...@cloudera.com> > > > >> > wrote: > > > >> > > > > >> > > This makes sense. > > > >> > > +1 > > > >> > > > > > >> > > On Tue, Jan 21, 2020 at 9:03 AM Andrew Sherman < > > > asher...@cloudera.com > > > >> > > > > >> > > wrote: > > > >> > > > > > >> > > > +1 > > > >> > > > > > > >> > > > > > > >> > > > On Tue, Jan 21, 2020 at 8:28 AM Sahil Takiar < > > > >> takiar.sa...@gmail.com> > > > >> > > > wrote: > > > >> > > > > > > >> > > > > +1 makes sense to me. > > > >> > > > > > > > >> > > > > On Mon, Jan 20, 2020 at 4:55 PM Tim Armstrong < > > > >> > tarmstr...@cloudera.com > > > >> > > > > > > >> > > > > wrote: > > > >> > > > > > > > >> > > > > > I think this proposal make sense - we've done well in > > enabling > > > >> > > parallel > > > >> > > > > > development for different Hive versions so far, but it is > a > > > >> burden. > > > >> > > > E.g. > > > >> > > > > we > > > >> > > > > > still don't have precommit tests for Hive 3+ (I like that > > > name) > > > >> > and I > > > >> > > > > don't > > > >> > > > > > know that we want to go about making the suite of > precommit > > > >> tests > > > >> > > even > > > >> > > > > > larger. > > > >> > > > > > > > > >> > > > > > On Fri, Jan 17, 2020 at 4:29 PM Joe McDonnell < > > > >> > > > joemcdonn...@cloudera.com > > > >> > > > > > > > > >> > > > > > wrote: > > > >> > > > > > > > > >> > > > > > > I wanted to start a conversation around moving to > develop > > > >> against > > > >> > > > Hive > > > >> > > > > 3+ > > > >> > > > > > > by default. (I describe this as Hive 3+ because it is > > close > > > to > > > >> > Hive > > > >> > > > > > master, > > > >> > > > > > > which is well beyond any released Hive 3.) There has > been > > > >> > > > considerable > > > >> > > > > > > development effort towards implementing features > > integrating > > > >> > Impala > > > >> > > > > with > > > >> > > > > > > Hive 3+ and Hive ACID. This is currently developed under > > the > > > >> > > > > > > USE_CDP_HIVE=true configuration while regular > development > > > has > > > >> > > > continued > > > >> > > > > > > with Hive 2. The Hive 3+ development is now stable > enough > > to > > > >> be > > > >> > > used > > > >> > > > > for > > > >> > > > > > > regular development. It would be nice to reduce our test > > and > > > >> > > > > > compatibility > > > >> > > > > > > matrix and have a unified development environment. > > > >> > > > > > > > > > >> > > > > > > Changing the major version of Hive is a breaking change, > > so > > > it > > > >> > > would > > > >> > > > > > > require an Impala 4.x code line. I have a specific > > proposal, > > > >> but > > > >> > > this > > > >> > > > > is > > > >> > > > > > > mainly a frame for getting the discussion going. > > > >> > > > > > > > > > >> > > > > > > I propose that we release Impala 3.4.0 and then update > > > master > > > >> to > > > >> > > 4.0 > > > >> > > > > and > > > >> > > > > > > allow breaking changes until the Impala 4.0 release. The > > > main > > > >> > > > breaking > > > >> > > > > > > change would be to set USE_CDP_HIVE=true, enabling Hive > 3+ > > > >> > > > development > > > >> > > > > by > > > >> > > > > > > default. The Hive 2 configuration would be removed over > > > time. > > > >> > Other > > > >> > > > > > > breaking changes can be proposed and voted on. > > > >> > > > > > > > > > >> > > > > > > If there are developers interested in maintaining a 3.x > > > >> branch, > > > >> > we > > > >> > > > can > > > >> > > > > > > create this branch and add appropriate support to any > > > >> > > infrastructure > > > >> > > > > > (e.g. > > > >> > > > > > > bin/push_to_asf.py) to allow that. > > > >> > > > > > > > > > >> > > > > > > Thoughts? > > > >> > > > > > > > > > >> > > > > > > Thanks, > > > >> > > > > > > > > > >> > > > > > > Joe McDonnell > > > >> > > > > > > > > > >> > > > > > > > > >> > > > > > > > >> > > > > > > > >> > > > > -- > > > >> > > > > Sahil Takiar > > > >> > > > > Software Engineer > > > >> > > > > takiar.sa...@gmail.com | (510) 673-0309 > > > >> > > > > > > > >> > > > > > > >> > > > > > >> > > > > >> > > > > > > > > > >