> On Sept. 22, 2017, 2:17 a.m., Jie Yu wrote:
> > src/resource_provider/registrar.cpp
> > Lines 134 (patched)
> > <https://reviews.apache.org/r/61528/diff/9/?file=1832372#file1832372line134>
> >
> >     Instead of wrapping a lot of `#ifndef` in the agent code, i'd suggest 
> > we use the memory based storage on windows and add a TODO in the code.
> >     
> >     We can document that windows agent does not support resource providers. 
> > What do you think?

That's a good idea. It prevents us from having to deal with any of this outside 
of the actual agent resource provider registrar. Also see my comment below.


> On Sept. 22, 2017, 2:17 a.m., Jie Yu wrote:
> > src/resource_provider/registrar.cpp
> > Lines 156 (patched)
> > <https://reviews.apache.org/r/61528/diff/9/?file=1832372#file1832372line156>
> >
> >     See my above comments. You can probably just wrap `storage` below.

I ended up introducing an `Owned<Storage> storage` so the AgentRegistrarProcess 
can use `storage` polymorphically. The Windows workaround is hidden behind a 
storage factory.


> On Sept. 22, 2017, 2:17 a.m., Jie Yu wrote:
> > src/resource_provider/registrar.cpp
> > Lines 201 (patched)
> > <https://reviews.apache.org/r/61528/diff/9/?file=1832372#file1832372line201>
> >
> >     what if fetch failed? Should we fail the recovery?

I believe by propagating the original `Future` in `recover` instead of managing 
our own `Promise` we will allow propagation of the failure to the caller.


- Benjamin


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/61528/#review185933
-----------------------------------------------------------


On Sept. 22, 2017, 5:03 p.m., Benjamin Bannier wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/61528/
> -----------------------------------------------------------
> 
> (Updated Sept. 22, 2017, 5:03 p.m.)
> 
> 
> Review request for mesos, Jie Yu and Jan Schlicht.
> 
> 
> Bugs: MESOS-7555
>     https://issues.apache.org/jira/browse/MESOS-7555
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> This patch adds a registry and registrar interface for resource
> provider managers. The registrar interface is modelled after the
> master registrar and supports similar operations. Currently a single,
> LevelDB-backed registrar is implemented which we plan to use for
> resource provider managers in agents.
> 
> Current the registry allows to add and remove resource provider IDs.
> 
> 
> Diffs
> -----
> 
>   src/CMakeLists.txt 1a828c4351ded36f51ccbbe67147da2f50b9cdb1 
>   src/Makefile.am 93ed2bf55447e3e470d9bea8a0b61ce78aad1900 
>   src/resource_provider/registrar.hpp PRE-CREATION 
>   src/resource_provider/registrar.cpp PRE-CREATION 
>   src/resource_provider/registry.hpp PRE-CREATION 
>   src/resource_provider/registry.proto PRE-CREATION 
>   src/slave/paths.hpp d021e6b85a2e5823ea088d65faf9cd85cfb57e28 
>   src/slave/paths.cpp e8724bf3e382211b5882728b86575de75981307f 
>   src/tests/resource_provider_manager_tests.cpp 
> 3bc56b51526e9dd188423f7349e74246c3295c77 
> 
> 
> Diff: https://reviews.apache.org/r/61528/diff/10/
> 
> 
> Testing
> -------
> 
> Tested on a number of platforms on internal CI.
> 
> 
> Thanks,
> 
> Benjamin Bannier
> 
>

Reply via email to