Olaf:
Thanks for spotting on the CI issue.

Roman:
Although that's doable, I think that adds complexity and hidden knowledge
in our pipeline. I'm a +1 to roll it back until upstream fixed the problem.

Regarding to CI, another problem is if it's always failing, then no body
will look into it. That's why I'd like to separate out ARCH64. I'll try to
tweak the sleep time and see if it gets around with PPC64LE. Meanwhile, I
think we also need to add deployment matrix in CI as well, so that the
issue like HBase won't start on PPC64LE won't happened again. I'll spend
some effort on that.

What's your plan on the CI improvements? I'd love to know.

 Evans

2017-04-30 13:04 GMT+08:00 Olaf Flebbe <[email protected]>:

> Hi Roman,
>
> you remember we are symlinking jars? esp. zookeeper. It was feeling weird
> to me at that time, now it will be desaster area. (because of  netty
> dependency changes in zookeeper).
>
> If we use old client we have to remove the symlink functionality in hadoop
> build code as well.
>
> Right now i am hesitant to have different zookeeper versions in bigtop
> (client and server).
>
> olaf
>
> > Am 30.04.2017 um 04:33 schrieb Roman Shaposhnik <[email protected]>:
> >
> >> On Tue, Apr 25, 2017 at 11:07 PM, Olaf Flebbe <[email protected]> wrote:
> >> Hi,
> >>
> >> There may be issues with netty and zookeeper within hadoop depending on
> each
> >> other.
> >> https://issues.apache.org/jira/browse/HADOOP-12928
> >> https://issues.apache.org/jira/browse/HADOOP-13413
> >>
> >> We have a dependency on curator as well
> >> https://issues.apache.org/jira/browse/HADOOP-14187 ( only in hadoop 3
> alpha)
> >>
> >> I didn't see all the mess when I initialy reviewed b00ac0936
> >>
> >> I think we should roll back commit b00ac0936  to zookeeper-3.4.6   T I
> have
> >> to revoke my LGTM ;-(
> >
> > One option is to have Zookeeper 3.4.6 as a service, but relax Hadoop
> dependency
> > back to the default. Typically old clients work pretty well with new
> Zookeeper
> > servers.
> >
> > What do y'all think?
> >
> > Thanks,
> > Roman.
>

Reply via email to