> On June 5, 2014, 10:45 p.m., Ben Mahler wrote:
> > src/master/master.hpp, lines 317-327
> > <https://reviews.apache.org/r/22151/diff/3/?file=603279#file603279line317>
> >
> >     Here's a proposal to try to avoid mixing the meaning the failure 
> > conditions (invalid task information vs. authorization denied):
> >     
> >     // Returns the validation result for the task.
> >     // Returns a failure if validation fails.
> >     Future<Option<Invalid> > validateTask(...);
> >     
> >     This exposes the validation result as the type of the future. Since we 
> > probably don't want to bother creating an 'Invalid' type for now, we can 
> > re-use the 'Error' type as we did in a few other places:
> >     
> >     // Returns the validation result for the task.
> >     // Returns a failure if validation fails.
> >     Future<Option<Error> > validateTask(...);
> >     
> >     This looks familiar to the discussion in: 
> > https://reviews.apache.org/r/22162/

done


> On June 5, 2014, 10:45 p.m., Ben Mahler wrote:
> > src/master/master.hpp, lines 329-331
> > <https://reviews.apache.org/r/22151/diff/3/?file=603279#file603279line329>
> >
> >     Should we document the failure case for launchTask?

done


> On June 5, 2014, 10:45 p.m., Ben Mahler wrote:
> > src/master/master.cpp, line 2145
> > <https://reviews.apache.org/r/22151/diff/3/?file=603280#file603280line2145>
> >
> >     Looks like we would want this check even if the framework is null?

pulled it up.


> On June 5, 2014, 10:45 p.m., Ben Mahler wrote:
> > src/master/master.cpp, line 2151
> > <https://reviews.apache.org/r/22151/diff/3/?file=603280#file603280line2151>
> >
> >     What about post-incrementing here?
> >     
> >     const TaskInfo& task = tasks[index++];
> >     
> >     Or:
> >     
> >     const TaskInfo& task = tasks[index];
> >     index++;

can't do because we need to increment index even if the 'future' failed.


> On June 5, 2014, 10:45 p.m., Ben Mahler wrote:
> > src/master/master.cpp, lines 2153-2154
> > <https://reviews.apache.org/r/22151/diff/3/?file=603280#file603280line2153>
> >
> >     Could we add a TODO related to how we can achieve a synchronous 
> > launchTask->_launchTask instead of an asynchronous launchTask... 
> > ->_launchTasks?
> >     
> >     Might simplify things a bit and we wouldn't need to worry about this 
> > check here.

done


> On June 5, 2014, 10:45 p.m., Ben Mahler wrote:
> > src/master/master.cpp, line 2146
> > <https://reviews.apache.org/r/22151/diff/3/?file=603280#file603280line2146>
> >
> >     size_t?

done


> On June 5, 2014, 10:45 p.m., Ben Mahler wrote:
> > src/master/master.cpp, line 1476
> > <https://reviews.apache.org/r/22151/diff/3/?file=603280#file603280line1476>
> >
> >     The fact that this code has non-local knowledge (that the task is 
> > placed inside pendingTasks before any further validation takes place) is 
> > really tricky and seems error prone, it seems safer to just keep the 'ids' 
> > here for now?
> >     
> >     FWIW it seems the executor ids were left in the ResourceUsageChecker as 
> > well.

Unfortunately we cannot get away from not depending on non-local knowledge. 
Example: If launchTasks1(), launchTasks2(), _launchTasks1() happen in this 
order, local 'ids' is of no use to detect a duplicate id. I left executor ids 
asis in ResourceUsageChecker as an optimization (to use a hashmap instead of 
having to loop through pendingTasks to find the executor id). I will remove the 
optmization. I want to get to a place where the validators don't hold local 
state (the last remnant is 'usedResources' in ResourceUsageChecker).


> On June 5, 2014, 10:45 p.m., Ben Mahler wrote:
> > src/master/master.hpp, line 341
> > <https://reviews.apache.org/r/22151/diff/3/?file=603279#file603279line341>
> >
> >     Thoughts on using shared_ptr for the TaskInfoVisitors? Might help avoid 
> > a memory leak and is a quick way to simplify things a bit.

done


> On June 5, 2014, 10:45 p.m., Ben Mahler wrote:
> > src/common/protobuf_utils.hpp, line 51
> > <https://reviews.apache.org/r/22151/diff/3/?file=603277#file603277line51>
> >
> >     // ... for updates created by the master?

don't want a util function to know about its callers.


- Vinod


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/22151/#review44833
-----------------------------------------------------------


On June 3, 2014, 11 p.m., Vinod Kone wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/22151/
> -----------------------------------------------------------
> 
> (Updated June 3, 2014, 11 p.m.)
> 
> 
> Review request for mesos, Benjamin Hindman and Ben Mahler.
> 
> 
> Bugs: MESOS-1114
>     https://issues.apache.org/jira/browse/MESOS-1114
> 
> 
> Repository: mesos-git
> 
> 
> Description
> -------
> 
> See summary.
> 
> 
> Diffs
> -----
> 
>   src/Makefile.am ffde59be8683dd40cc5bc7cb88cd88c5bc91cf96 
>   src/common/protobuf_utils.hpp 0f653414bc1bb2b632ec8cd9c8bd7202a53d42e1 
>   src/master/hierarchical_allocator_process.hpp 
> 0c5e2e050cad96fafaf136232bd255b0ae3038cd 
>   src/master/master.hpp d4ef4bec7168179f2168e88d3727e50b0e2e68a1 
>   src/master/master.cpp 766a0e36a6e7a615e7b2974d9fee70bcef446719 
>   src/tests/master_authorization_tests.cpp PRE-CREATION 
> 
> Diff: https://reviews.apache.org/r/22151/diff/
> 
> 
> Testing
> -------
> 
> make check.
> 
> Will write new tests and update this review or will create a new one.
> 
> 
> Thanks,
> 
> Vinod Kone
> 
>

Reply via email to