----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/57190/#review167297 -----------------------------------------------------------
src/tests/role_tests.cpp Lines 812 (patched) <https://reviews.apache.org/r/57190/#comment239673> This might be better placed in `persistent_volume_tests.cpp`. src/tests/role_tests.cpp Lines 832 (patched) <https://reviews.apache.org/r/57190/#comment239680> "Helper function that starts a framework in a role and creates a persistent volume with the given id." src/tests/role_tests.cpp Lines 833 (patched) <https://reviews.apache.org/r/57190/#comment239681> No comma after "volume". src/tests/role_tests.cpp Lines 876 (patched) <https://reviews.apache.org/r/57190/#comment239683> "Create a task that uses the volume and checks that the volume does not contain any files leaked from any role." src/tests/role_tests.cpp Lines 879 (patched) <https://reviews.apache.org/r/57190/#comment239685> Seems like this would be a bit more readable if we split it into two or more separate statements. We could pull `offers.get()[0]` out into a separate variable, and maybe the filtered resources w/o the disk as well? src/tests/role_tests.cpp Lines 887 (patched) <https://reviews.apache.org/r/57190/#comment239696> We can't depend on `ls` and `grep` being available on Windows. We could write the test with `#ifdef`s, although that seems unfortunate. Not sure there's a better alternative, given the current agent volume creation logic... src/tests/role_tests.cpp Lines 907 (patched) <https://reviews.apache.org/r/57190/#comment239415> Why `ASSERT_EQ` rather than `EXPECT_EQ`? src/tests/role_tests.cpp Lines 918 (patched) <https://reviews.apache.org/r/57190/#comment239414> "in that case" src/tests/role_tests.cpp Lines 923 (patched) <https://reviews.apache.org/r/57190/#comment239688> Would be a bit more readable to move this function declaration up above the `runTask` calls. src/tests/role_tests.cpp Lines 937 (patched) <https://reviews.apache.org/r/57190/#comment239690> Do we really need the `v1` prefix to these variable names? Seems unnecessary. src/tests/role_tests.cpp Lines 957 (patched) <https://reviews.apache.org/r/57190/#comment239692> Capturing `CONTENT_TYPE` seems unnecessary. src/tests/role_tests.cpp Lines 967 (patched) <https://reviews.apache.org/r/57190/#comment239693> "Destroy both volumes. Even though the role `a/b/c` is a prefix of the role `a/b/c/d`, destroying the former role's volume `d` should not interfere with the latter's volume `id`." - Neil Conway On March 2, 2017, 5 p.m., Benjamin Bannier wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/57190/ > ----------------------------------------------------------- > > (Updated March 2, 2017, 5 p.m.) > > > Review request for mesos, Benjamin Mahler, Michael Park, and Neil Conway. > > > Bugs: MESOS-7047 > https://issues.apache.org/jira/browse/MESOS-7047 > > > Repository: mesos > > > Description > ------- > > This commit adjusts the way persistent volumes are stored on the > agent. Instead of interpreting the role of the volume as a literal > path, we replace `/` with ` ` when creating the path. This prevents > that subdirectories are created for volumes with hierarchical roles. > Directly interpreting the role as a path is undesirable as it can lead > to volumes overlapping (e.g., a volume with role `a/b/c/d` and id `id` > would be visible as `id` in a volume with role `a/b/c` and id `d`). > > > Diffs > ----- > > src/slave/paths.cpp 38ad1993aa36a627ec97a7865488677495ee4c5a > src/tests/role_tests.cpp 77f3d46a544a51ba71476e2f0735bb32758dd9e1 > > > Diff: https://reviews.apache.org/r/57190/diff/1/ > > > Testing > ------- > > make check > > > Thanks, > > Benjamin Bannier > >
