On Mon, Jun 12, 2017 at 08:47PM, Olaf Flebbe wrote:
> Hi Cos,
> 
> The deployment CI job of bigtop-trunk use the repo and packages of the last
> bigtop-release. So any change in package (i.e. fixes) will not be reflected
> in the provisioner jobs. I am really tired in arguing against this.

Ah, I am not crazy then! Thanks for confirming my sanity, Olaf!

So, the next question is how we can combine the night package build with
Evans' deployment job. Yes, I know - the provisioning matrix server a
different purpose to polish our Puppet code. But, and that's a big but -
while polishing Puppet we might find out that it is a package that is
problematic (BIGTOP-2165 is a case in point). In wish case we need to test the
latest packages. 

I wonder if we can collect all packages together and build a repo on a nightly
basis? Or perhaps we can directly use Jenkins' artifacts? I did the former for
another project a while back and it is a-ok if you have just a handful of
packages. In our case we are talking about Gigabytes of stuff being copied
between the slaves, and I don't think it would be tht efficient. May be we can
use a shared location to store the binary artifacts?

Thoughts?
  Cos

> > Am 12.06.2017 um 20:40 schrieb Konstantin Boudnik <[email protected]>:
> > 
> > So, I got BIGTOP-2165 in and it has addressed the ignite service startup 
> > issue
> > in my local testing, but I still see failing build in our CI [1]
> > 
> > Evans, looks like I need your help to sort it our. Here's how I deploy the
> > cluster:
> > - first I provision hdfs/yarn/mapreduce components from our 1.2 repo
> > - then I change the repo URL to my local one, where I have a freshly built
> >  ignite packages, and run 'puppet apply' manually
> > - the service comes up just fine and I see the process running. I also can
> >  stop it using 'service ignite-hadoop stop'
> > 
> > I was looking into the deployment matrix's config, but had a bit of a hard
> > time deciphering where the packages are coming from. Is there a chance of a
> > race-condition where the deployment might be using the packages from the 
> > last
> > build or something or a sort? I guess the easiest answer to this would be to
> > wait until the next provisioner cycle and see if the situation with CentOS'
> > Ignite deployment has improved.
> > 
> > This bug was driving me nuts for the past two years, and now that I thought 
> > I
> > have fixed it there are two different outcomes in different environments. So
> > please accept my apologies if I am barking at the wrong tree ;)
> > 
> > Thanks for your help!
> > 
> > [1] 
> > https://ci.bigtop.apache.org/view/Provisioner/job/Bigtop-trunk-deployments/BUILD_SLAVE=docker-slave-02,COMPONENTS=ignite_hadoop,OS=centos6/54/console
> > 
> > Cos
> > 
> > On Thu, Jun 08, 2017 at 10:03AM, Konstantin Boudnik wrote:
> >> Cool! I am trying to get some Ignite's dev@ help with BIGTOP-2108 -
> >> hopefully, it won't be long before it is fixed. Thanks!
> >> --
> >>  Take care,
> >> Konstantin (Cos) Boudnik
> >> 2CAC 8312 4870 D885 8616  6115 220F 6980 1F27 E622
> >> 
> >> Disclaimer: Opinions expressed in this email are those of the author,
> >> and do not necessarily represent the views of any company the author
> >> might be affiliated with at the moment of writing.
> >> 
> >> 
> >> On Thu, Jun 8, 2017 at 8:55 AM, Evans Ye <[email protected]> wrote:
> >>> OK. I agreed with you.
> >>> Considered that most of us are making contributions in free time.
> >>> Let's make the release less stressed but more solid.
> >>> 
> >>> I plan to make 1.2.1 release at the end of the June.
> >>> That should make plenty of time for blockers to be removed.
> >>> 
> >>> Finally, I'd like to thank you all for taking actions to our deployment
> >>> issues.
> >>> 
> >>> 
> >>> 2017-06-08 1:11 GMT+08:00 Konstantin Boudnik <[email protected]>:
> >>> 
> >>>> On Thu, Jun 08, 2017 at 12:57AM, Evans Ye wrote:
> >>>>> Cos,
> >>>>> Thanks a lot!
> >>>>> 
> >>>>> Olaf,
> >>>>> My original thought was to remove those puppet code which no one has
> >>>>> interest to fix it.
> >>>>> But you're right. Some of them are broken at package level...
> >>>>> 
> >>>>> I've dig a bit in and found some of them are failing due to Java not
> >>>>> presented before daemons starting up.
> >>>>> I've uploaded a patch in BIGTOP-2775, which should fix the problem.
> >>>>> But we still got many others failing for various reasons.
> >>>>> 
> >>>>> Roman,
> >>>>> Good idea. We can track them via JIRAs.
> >>>>> 
> >>>>> All,
> >>>>> I still hope that 1.2.1 can be released before DataWorks Summit.
> >>>>> However, at this point I'd like to get everybody's thoughts that whether
> >>>> we
> >>>>> should release it and mark these stuffs as known issues, or we get them
> >>>>> fixed and then release a fully functional release.
> >>>>> Which one you're in favor?
> >>>> 
> >>>> Eitherway is good for me. Making the release before the mid of next week 
> >>>> is
> >>>> only possible if we have everything in place right now and only need to 
> >>>> do
> >>>> the
> >>>> release itself and vote on it. Otherwise, the timeframe is too tight...
> >>>> 
> >>>> Cos
> >>>> 
> >>>>> 2017-06-07 6:15 GMT+08:00 Roman Shaposhnik <[email protected]>:
> >>>>> 
> >>>>>> On Sun, Jun 4, 2017 at 9:32 PM, Evans Ye <[email protected]> wrote:
> >>>>>>> Hi Bigtopers,
> >>>>>>> 
> >>>>>>> I've shared this on my Apache Big Data talk, but haven't got a
> >>>> chance to
> >>>>>>> share with you yet.
> >>>>>>> This is a Jenkins matrix job to show the healthy status of our
> >>>> component
> >>>>>>> deployment code.
> >>>>>>> Each run deploys hdfs + Y on X, where X is the OS and Y is the
> >>>> component
> >>>>>>> shown on the matrix:
> >>>>>>> 
> >>>>>>> https://ci.bigtop.apache.org/view/Provisioner/job/Bigtop-
> >>>>>> trunk-deployments/45/
> >>>>>> 
> >>>>>> This is super awesome! Thanks for spending time on putting it together!
> >>>>>> 
> >>>>>>> Generally speaking, the status of our deployment code needs a lot of
> >>>>>> effort
> >>>>>>> to polish.
> >>>>>>> So, for 1.2.1 release, I think we should make some changes:
> >>>>>>> 
> >>>>>>> 1). Remove unmaintained failing components
> >>>>>>> 2). Get maintained components fixed
> >>>>>>> 
> >>>>>>> 1) might be easier to achieve, while 2) takes time.
> >>>>>>> In order to maintain a good code quality, I suggest proceed with 1).
> >>>>>>> 
> >>>>>>> Any different thoughts?
> >>>>>> 
> >>>>>> Shall we at least start with filing JIRAs for everything that's
> >>>>>> failing -- we can discuss
> >>>>>> the particulars on the JIRAs then.
> >>>>>> 
> >>>>>> I can try filing some tomorrow if nobody beats me to it (please do ;-))
> >>>>>> 
> >>>>>> Thanks,
> >>>>>> Roman.
> >>>>>> 
> >>>> 
> 


Reply via email to