Thanks for the feedback, Alan!

I think it's fine to have a separate bucket for data corruption. However,
many principles here probably apply to that as well.

I like your other thoughts as well and am open to other suggestions. I aim
at establishing a process that is to be followed rather than no process at
all as today. Thus, other comments/suggestions are also greatly welcome.

Thanks,
Xuefu

On Tue, Nov 24, 2015 at 9:57 AM, Alan Gates <[email protected]> wrote:

> Xuefu, thanks, I think this is a good start.  I've inlined a few responses.
>
> Xuefu Zhang <[email protected]>
> November 24, 2015 at 7:26
> Hi all,
>
> In the last contributor meetup, the topic around data correctness or data
> corruption is rather concerning. Not only is the number of such issues that
> have been reported recently, but also the way that Hive community is
> handling these issues. The latter is the the topic of this discussion. I
> think everyone agrees that the current practice is problematic and that
> Hive community should treat data correctness more seriously. Therefore, I'd
> like to find a "standard" procedural that we should follow. Here are my
> initial thought:
>
> 1. JIRA should be correctly labeled and the title should reflect data
> correctness.
>
> I think we should have data corruption (which is now possible with ACID)
> as a separate category that is treated in the same way (ie we would have
> separate labels for correctness and corruption issues).
>
> 2. JIRA should bear adequate description about the issue, including
> affected version, JIRA that incurred the issue, any workaround, etc.
> 3. Once confirmed, advisory message should be sent to user@ and @dev
> regarding the problem.
> 4. Once the JIRA is closed, a message should be sent again to the lists
> advising the availability of the fix.
>
> I don't think this is necessary.  The people who are concerned about the
> JIRA can follow it.  That way they will know when it's committed and when
> it's released and in what version, without polluting the mailing list with
> it.  The email to the user and dev lists when we discover the issue should
> include instruction on how to follow the JIRA.
>
> A couple of other thoughts:
> 5) We should make it easy for users to discover known correctness and
> corruption issues in released versions of Hive.  This can be as simple as a
> wiki page with a link to a JIRA search that will return all these issues
> for each released Hive version.
> 6) We need to think about how we quickly produce maintenance releases to
> address these issues.  Finding a correctness issue and then not releasing a
> fix for 6 months will not be good.  I think we should set a goal of
> releasing a maintenance release with at least the correctness or corruption
> issue repaired within 6 weeks of discovering the issue.
>
> Alan.
>
>
> I know these may not be all clear or actionable, but I hope we can have
> concrete steps to follow at the end of this discussion.
>
> Please share your thoughts.
>
> Thanks,
> Xuefu
>
>

Reply via email to