Fix tez build, which was failing because of some docker daemon issue on the
build slave. :)

2015-09-25 19:06 GMT+02:00 Evans Ye <[email protected]>:

> With nodeps setting added now our CI is almost back to normal.
> There're still 5 components failing but the other's are good.
> I need to update our build slaves so that mahout can resume to normal.
> I don't know why the others still failing.
> Will look into them when back from budapest ;)
>
> 2015-09-22 2:50 GMT+08:00 Konstantin Boudnik <[email protected]>:
>
>> I think for now it would be reasonable to just turn off the dependency
>> graph
>> building. To do that we need to add
>>     -Dbuildnodeps=true
>> to the gradle command line.
>>
>> Cos
>>
>> On Tue, Sep 22, 2015 at 01:40AM, Evans Ye wrote:
>> > Yup, and I think we've discussed that before in BIGTOP-1906.
>> >
>> > Right now I don't have any clue how to build that server up, but that's
>> a
>> > good way to go since our users would somehow need to set there internal
>> > jars repository for hadoop apps developer to download dependencies.
>> >
>> > At this point, I'm thinking that maybe Cos can give me a hint of code
>> > snippet how to turn the dependency build off. If there's no such
>> feature I
>> > can try to add it so that we can resume the CI status.
>> >
>> >
>> > 2015-09-21 20:36 GMT+08:00 Jay Vyas <[email protected]>:
>> >
>> > > Thanks Evans.  I like your idea of separate Bigtop artifacts stored
>> > > somewhere...
>> > >
>> > > If Bigtop can publish its own jars to its own nexus (or else to
>> upstream
>> > > apaches maven server) that is ideal I guess!
>> > >
>> > > > On Sep 21, 2015, at 4:19 AM, Evans Ye <[email protected]> wrote:
>> > > >
>> > > > Hi all !
>> > > >
>> > > > I spent some time looking into this and found that the massive job
>> > > failing
>> > > > is telling us that the build dependency feature is now working!
>> Which is
>> > > > because of every build that has dependency setting start from
>> building
>> > > > upstream packages locally. That fills up the disk size on every
>> build
>> > > > slaves.
>> > > > As I mentioned earlier, our CI build env is a sub-optimal design
>> because
>> > > of
>> > > > limited disk size on instances. I have no choice to split components
>> > > > equally on 4 slaves manually to ease the disk consumption.
>> > > >
>> > > > Maybe I should start trying nexus server to put built jars there?
>> > > > Or I can specify an option to tell gradle to fetch jars directly
>> from
>> > > > public maven instead of building upstream components locally?
>> > > >
>> > > >
>> > > > Thanks,
>> > > > Evans
>> > >
>>
>
>

Reply via email to