> On Oct. 6, 2016, 6:09 p.m., Jiang Yan Xu wrote: > > src/tests/persistent_volume_tests.cpp, lines 1062-1066 > > <https://reviews.apache.org/r/52288/diff/1/?file=1510563#file1510563line1062> > > > > It's important that the task uses only a portion of cpus/mem so the > > rest can be offered to framework2. Perhaps emphasize this in the comment as > > well? > > > > The benefit of not using the the shared volume in the task is actually > > not obvious to me: we kill the task before destroying the volume anyways. > > If it's for simplicity that we don't use the volume, we should comment on > > the reason being so.
Yes, I did not add the shared volume for this task as it is not needed for verifying rescind of offers. Added a comment regarding that. > On Oct. 6, 2016, 6:09 p.m., Jiang Yan Xu wrote: > > src/tests/persistent_volume_tests.cpp, lines 1068-1071 > > <https://reviews.apache.org/r/52288/diff/1/?file=1510563#file1510563line1068> > > > > Don't reuse the variable `offers` (even though it's reset to pending, > > this detail is very implicit). Based on the pattern in the tests, the variable `offers` and `offer` are reused. I agree it might make it cleaner to have different variables used but I think we should do it as a larger refactor (if we do it at all). For now, I have a `offers1` and `offers2` which represent offers from the 2 frameworks respectively. > On Oct. 6, 2016, 6:09 p.m., Jiang Yan Xu wrote: > > src/tests/persistent_volume_tests.cpp, lines 1102-1103 > > <https://reviews.apache.org/r/52288/diff/1/?file=1510563#file1510563line1102> > > > > Since `task` doesn't use all resources, we should expect `famework2` to > > receive the offer immediately right? Updated based on the minor modification in flow. - Anindya ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/52288/#review151635 ----------------------------------------------------------- On Sept. 26, 2016, 11:58 p.m., Anindya Sinha wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/52288/ > ----------------------------------------------------------- > > (Updated Sept. 26, 2016, 11:58 p.m.) > > > Review request for mesos and Jiang Yan Xu. > > > Bugs: MESOS-6257 > https://issues.apache.org/jira/browse/MESOS-6257 > > > Repository: mesos > > > Description > ------- > > When a framework issues a DESTROY of a shared volume, and that volume > is not in use by a running or a pending task, we rescind the offers > from frameworks to which the shared volume is present in pending offers > so that the volume is not assigned to any task in a future ACCEPT call. > At that time, we need to recover the resources as well for proper > accounting of such resources by the allocator. > > > Diffs > ----- > > src/master/master.cpp 66a672f6d16233e96b29e330a9e6c474546fa851 > src/tests/persistent_volume_tests.cpp > 2ab44d11d159162dfcac9d0791b651ed059b8164 > > Diff: https://reviews.apache.org/r/52288/diff/ > > > Testing > ------- > > All tests passed. > > > Thanks, > > Anindya Sinha > >
