[
https://issues.apache.org/jira/browse/HIVE-9454?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14291459#comment-14291459
]
Julian Hyde commented on HIVE-9454:
-----------------------------------
It's unfair to say I've been "using hive as a test bed". We have a test suite
for Calcite that we're quite happy with. It serves its purpose -- it keeps
Calcite from regressing.
You want to release Hive, so you need a release of Calcite. I am trying to do
integration -- identify the incompatibilities between Hive and the new version
of Calcite -- before we make that release. If that doesn't get done, you might
well end up with a release of Calcite that is not suitable to make a release of
Hive.
I have been telegraphing for some time on the Calcite dev list that a Calcite
release is imminent. I made a release candidate on Thursday and started a vote.
Given that Hive uses Calcite, at least one Hive committer ought to be actively
participating on the Calcite dev list, participating in that vote, and doing
integration testing.
The only integration testing that happened occurred because I pushed a new
snapshot. That shit was going to hit the fan at some point between now and Hive
15. It's actually a good thing that it happened now.
I'm coming round to the idea that it is a bad idea for Hive to develop on a
snapshot of Calcite. But what are the alternatives? Given that it takes at
least 6 days for an incubator project such as Calcite to make a release,
development could very easily become stalled by a small bug or missing feature
in Calcite. The ideal would be a Hive-specific snapshot of Calcite, controlled
by Hive developers, but (a) Apache nexus doesn't seem to allow multiple
versions of snapshots, (b) the Apache release process doesn't allow releases on
snapshots, and (c) this would require proactive efforts by Hive committers to
integrate with Calcite ahead of a Calcite release. Whatever we decide, it needs
more proactive involvement from the Hive side.
There is an urgent need for a decision on the Calcite 1.0 release vote. We have
sufficient votes for a release, and I could close the vote in just over an
hour, but I won't. There is a non-binding -1 from [~jpullokkaran] due to
incompatibilities but we haven't figured out whether the cause is on the Hive
side or the Calcite side. I'd like to close the vote as soon as possible, but I
need Hive developers to either log bugs or let the vote pass. "We haven't had
time to do integration testing" (I'm paraphrasing a little) is not a valid
reason for a -1.
> Test failures due to new Calcite version
> ----------------------------------------
>
> Key: HIVE-9454
> URL: https://issues.apache.org/jira/browse/HIVE-9454
> Project: Hive
> Issue Type: Bug
> Reporter: Brock Noland
> Attachments: HIVE-9454.1.patch
>
>
> A bunch of failures have started appearing in patches which seen unrelated. I
> am thinking we've picked up a new version of Calcite. E.g.:
> http://ec2-174-129-184-35.compute-1.amazonaws.com/jenkins/job/PreCommit-HIVE-TRUNK-Build/2488/testReport/org.apache.hadoop.hive.cli/TestCliDriver/testCliDriver_auto_join12/
> {noformat}
> Running: diff -a
> /home/hiveptest/54.147.202.89-hiveptest-1/apache-svn-trunk-source/itests/qtest/../../itests/qtest/target/qfile-results/clientpositive/auto_join12.q.out
>
> /home/hiveptest/54.147.202.89-hiveptest-1/apache-svn-trunk-source/itests/qtest/../../ql/src/test/results/clientpositive/auto_join12.q.out
> 32c32
> < $hdt$_0:$hdt$_0:$hdt$_0:$hdt$_0:src
> ---
> > $hdt$_0:$hdt$_0:$hdt$_0:$hdt$_0:$hdt$_0:src
> 35c35
> < $hdt$_0:$hdt$_0:$hdt$_1:$hdt$_1:$hdt$_1:src
> ---
> > $hdt$_0:$hdt$_0:$hdt$_1:$hdt$_1:$hdt$_1:$hdt$_1:src
> 39c39
> < $hdt$_0:$hdt$_0:$hdt$_0:$hdt$_0:src
> ---
> > $hdt$_0:$hdt$_0:$hdt$_0:$hdt$_0:$hdt$_0:src
> 54c54
> < $hdt$_0:$hdt$_0:$hdt$_1:$hdt$_1:$hdt$_1:src
> ---
> > $hdt$_0:$hdt$_0:$hdt$_1:$hdt$_1:$hdt$_1:$hdt$_1:src
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
