Yes, 

If you rely just on fat jar. For dev it is great tool, but for production may 
not be so much.



On Jul 19, 2015, at 8:46 PM, Ken Sipe <[email protected]> wrote:

> the point of a fat jar is to not have jars “outside”… it would make sense 
> that would cause challenges.   We have a fat jar now because 1) that is the 
> way it was original dev and 2) the ability to distribute all things the 
> needed by the executor in a runtime jar.    Fat jars work great when they are 
> all inclusive, all stand alone.   if in a shared class path space then bad 
> things are sure to happen:)
> 
> ken
>> On Jul 19, 2015, at 10:08 PM, Yuliya Feldman <[email protected]> 
>> wrote:
>> 
>> I am wondering about usage of fat jar in principal - fat jar unless it has 
>> the same versions of jars inside fat one and outside can create issues. slf 
>> jars , mesos jars
>> 
>> What is your experience? Ours so far was quite negative
>> 
>> 
>> 
>> On Jul 19, 2015, at 6:46 PM, Ken Sipe <[email protected]> wrote:
>> 
>>> 
>>>> On Jul 17, 2015, at 5:51 PM, Adam Bordelon <[email protected]> wrote:
>>>> 
>>>> Ken, Swapnil is looking at turning the executor into a YARN NM AuxService
>>>> anyway, so it might end up without a main too. Then we wouldn't need the
>>>> application plugin at all?
>>>> 
>>> actually I don’t think we need the application plugin right now.   I think 
>>> the capsule uses the application property for the main for configuration.  
>>> We can go without it I believe.  When we make the executor a NM AuxService, 
>>> we may need to look at new packaging.   I believe the capsule is a 
>>> “runnable” fat jar and we will need just a standard fat jar at that point.  
>>>  I’ll look into it.
>>> 
>>> ken
>>> 
>>>> On Fri, Jul 17, 2015 at 3:31 PM, Ken Sipe <[email protected] 
>>>> <mailto:[email protected]>> wrote:
>>>> 
>>>>> Hey Patrick,
>>>>> 
>>>>> I was the one that refactored myriad in to a multi-project structure.   I
>>>>> took what was originally there and tried to maintain the semantics into 
>>>>> the
>>>>> new structure (meaning it was my intention to not change anything other
>>>>> than the structure).  At first glance this appears to be a miss on my
>>>>> part.   There is no main class in the scheduler.  The scheduler is a YARN
>>>>> resource manager.
>>>>> 
>>>>> Looking it over in more detail, I’m not sure we even need the application
>>>>> plugin for the executor.  It is required for the building of the capsule,
>>>>> but that could be declared in the capsule instead of using the
>>>>> configuration of the application.  It is worth more investigation.
>>>>> 
>>>>> As for long, there are no other projects other than the executor that need
>>>>> the application plugin.  I have refactored this and have submitted it as
>>>>> PR-120:  https://github.com/mesos/myriad/pull/120 <
>>>>> https://github.com/mesos/myriad/pull/120 
>>>>> <https://github.com/mesos/myriad/pull/120>>
>>>>> 
>>>>> Thanks for helping to identify this.
>>>>> 
>>>>> Ken
>>>>> 
>>>>>> On Jul 17, 2015, at 3:50 PM, Patrick Wong <[email protected]> wrote:
>>>>>> 
>>>>>> Hello Myriad Team,
>>>>>> 
>>>>>> I noticed that the application plugin is applied to all subprojects. Does
>>>>>> this mean that you are supposed to be able to run "gradle installDist" to
>>>>>> get a working local installation for each subproject, and "gradle run" to
>>>>>> run them?
>>>>>> 
>>>>>> --
>>>>>> Patrick Wong
>>>>>> 510.386.7205
>>>>>> 
>>>>>> mapr.com
>>> 
> 

Reply via email to