Hi Sergio, - What is the release cadence in Hive? Is there a scheduled time to do major releases, like every 3 months?
There have been discussions in the past about release cadence, and people seem to agree that more frequent releases would be good. However, in practice, the release cadence has varied over time. It depends on someone volunteering to work on the release. - Do we have some kind of quality check before doing a release? or is it just by voting that packages and jars are correct? Many of the releases in last few years were driven by committers who are also employed at Hortonworks. In case of those releases, those apache releases were also being used as input for the next Hortonworks Data Platform releases, and were going through system/integration tests at hortonworks. Unfortunately, with the tests we have in Apache, we don't have good way to verify if the release actually works at scale, with kerberos, HDFS encryption, LDAP etc. Hive 1.1.0 was driven by Brock Noland who was at cloudera, maybe that one went through system/integration tests at Cloudera. When people Vote on the release, we have to rely on what is available in apache, which is the unit test results and contents of the tar.gz. - Is there a document that mentions what kind of changes must be added and not to minor and major versions? When can we break api, features, etc? when a change must be on Hive3 or on Hive2.2? I am not aware of a formal document. The general practice seems to have been to try and keep backward compatibility in a major line. There have been some backward incompatible changes where the behavior was considered a bug. The backward incompatible changes in Decimal datatype in first few releases would be a good example of that. I think we should work on a more formal document around this, like the one in hbase - https://hbase.apache.org/book.html#hbase.versioning Thanks, Thejas On Tue, Oct 4, 2016 at 10:36 AM, Sergio Pena <[email protected]> wrote: > Hi All, > > Anyone knows about Hive release cadence? > > I would like to understand this topic, and see if there are some > improvements we can do for the process, and create some documentation too. > Also, I sometimes wonder what things we can break between Hive 2.x major > versions, when we should mark features as deprecated and when remove them > completely from the version. > > I look forward hearing from you. > > - Sergio > > On Wed, Sep 28, 2016 at 4:34 PM, Sergio Pena <[email protected]> > wrote: > > > Hi All, > > > > I was looking in the Hive Wiki regarding this topic, but I don't find > > anything about how often we do releases in Hive. > > So, I wanted to send this email to ask a few questions regarding > releases: > > > > - What is the release cadence in Hive? Is there a scheduled time to do > > major releases, like every 3 months? > > What about minor version releases? > > > > - Do we have some kind of quality check before doing a release? > > or is it just by voting that packages and jars are correct? > > > > - Is there a document that mentions what kind of changes must be added > and > > not to minor and major versions? > > When can we break api, features, etc? when a change must be on Hive3 or > > on Hive2.2? > > > > Btw, I'd like to include a testing phase on every release to validate its > > quality before the final release. > > There is some work we're doing on PTest to allow that in ths jira: > > https://issues.apache.org/jira/browse/HIVE-14744 > > > > - Sergio > > >
