----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/34135/#review92480 -----------------------------------------------------------
src/slave/containerizer/isolators/filesystem/posix.cpp (lines 183 - 185) <https://reviews.apache.org/r/34135/#comment146681> I just realized an issue here (not your fault). Since we don't checkpoint 'resources' in this isolator, when slave restarts and recovers, the 'resources' in Info will be empty, but there are symlinks exists in the sandbox. We'll end up with trying to create already exist symlinks (and fail). I think this is easy to resolve because we can just ignore EEXIST. Also, it's likely that we fail to remove some symlinks because they do not exist (e.g., a task finishes but the slave crashes after containerizer->update is called but before the status update is checkpointed). - Jie Yu On July 12, 2015, 4:46 a.m., Ian Downes wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/34135/ > ----------------------------------------------------------- > > (Updated July 12, 2015, 4:46 a.m.) > > > Review request for mesos, Chi Zhang, Paul Brett, Timothy Chen, and Vinod Kone. > > > Repository: mesos > > > Description > ------- > > Moved code from Mesos Containerizer to filesystem isolators > - filesystem/posix (symlinks, doesn't support container rootfs) > - > The filesystem/posix isolator will be automatically included if no > filesystem/ isolator is specified. > > > Diffs > ----- > > src/Makefile.am e5b5d36f0ac160e5a3a9fdc50b31c060a413ce2c > src/slave/containerizer/isolators/filesystem/posix.hpp PRE-CREATION > src/slave/containerizer/isolators/filesystem/posix.cpp PRE-CREATION > src/slave/containerizer/mesos/containerizer.cpp > 47d146125dfd4ea909e7ec9d94f41cfa11d035e5 > > Diff: https://reviews.apache.org/r/34135/diff/ > > > Testing > ------- > > existing persistent volumes tests. > > > Thanks, > > Ian Downes > >
