> On June 4, 2014, 7:12 p.m., Vinod Kone wrote:
> > src/master/http.cpp, lines 543-561
> > <https://reviews.apache.org/r/22223/diff/2/?file=603297#file603297line543>
> >
> >     How about pushing this into slave model?
> 
> Yifan Gu wrote:
>     Yes, putting the "orphan_tasks" in their parent slaves seems more 
> structural. 
>     So do you mean moving this piece of code into model(const Slave& slave) 
> {}?
>     
>     But I don't know how to get the "master.frameworks" in that function 
> without changing the interface of the function... 
>     Did I misunderstand what you said?

I see. Then what you have is good for now.


> On June 4, 2014, 7:12 p.m., Vinod Kone wrote:
> > src/master/http.cpp, line 563
> > <https://reviews.apache.org/r/22223/diff/2/?file=603297#file603297line563>
> >
> >     How about just calling them "unknown_frameworks" or maybe 
> > "disconnected_frameworks"? but disconnected is an overloaded term because a 
> > framework might be disconnected even when the master doesn't failover.
> 
> Yifan Gu wrote:
>     Yeah, I see. unknown_frameworks seems better IMO. Because I found that 
> the newly elected master actually "forgets" the framework. Even the framework 
> tried to re-register with the master, the master will reply a 
> FrameworkRegisteredMessage.
>     
>     I don't know if the registerar will be modified later to help the master 
> memorize the frameworks. 
>     But for now, unknown_frameworks is great. Thanks! Vinod!

yup. registrar in the future will store framework infos and hence master will 
remember them through failovers.


- Vinod


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/22223/#review44738
-----------------------------------------------------------


On June 3, 2014, 11:58 p.m., Yifan Gu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/22223/
> -----------------------------------------------------------
> 
> (Updated June 3, 2014, 11:58 p.m.)
> 
> 
> Review request for mesos, Adam B, Niklas Nielsen, Ross Allen, and Vinod Kone.
> 
> 
> Repository: mesos-git
> 
> 
> Description
> -------
> 
> This is a patch to add *orphan_tasks* and *not_yet_reregistered_frameworks" 
> fields in the state.json. So that users can get more information about what's 
> going on when master failover happens.
> 
> For the WebUI part, I think that will have Ross to have a look at this later.
> 
> Thanks!
> 
> 
> Diffs
> -----
> 
>   src/master/http.cpp b565dc6 
>   src/tests/master_tests.cpp 34df121 
> 
> Diff: https://reviews.apache.org/r/22223/diff/
> 
> 
> Testing
> -------
> 
> Added MasterTest.OrphanTasks to test:
> 1, Before the framework re-registers with the master:
>   Tasks that belongs to that frameworks will appear in the ["orphan_tasks"] 
> field.
>   That frameworks will appear in the ["not_yet_reregistered_frameworks"] 
> field.
> 2, After the framework re-registers with the master:
>   There will be no more orphan tasks and not-yet-reregistered frameworks.
> 
> 
> Thanks,
> 
> Yifan Gu
> 
>

Reply via email to