----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/38569/#review99831 -----------------------------------------------------------
src/slave/containerizer/isolators/filesystem/linux.cpp (lines 73 - 74) <https://reviews.apache.org/r/38569/#comment156842> After then sentense `using the bind backend)` add `because cleanup operations within the work_dir can be propagted to all container namespaces.`? src/slave/containerizer/isolators/filesystem/linux.cpp (line 84) <https://reviews.apache.org/r/38569/#comment156818> Is `workDirMount` better? src/slave/containerizer/isolators/filesystem/linux.cpp (lines 97 - 98) <https://reviews.apache.org/r/38569/#comment156811> So this seems to work but oh my godness the the way `mount` command interacts with the syscall is very implicit. I can confirm this: ``` # /home/jyx/tmp/ is created by mount command and /home/jyx/tmp/target is mounted by syscall. [root tmp]# mount --make-shared target mount: can't find /home/jyx/tmp/target in /etc/fstab or /etc/mtab [root tmp]# mount --make-rshared . # OK. ``` So the rule seemed to be **as long as the arguments provided to the `mount` command themselves are mounted by the the command, we are OK**. (I guess because whatever is recurively done by the command is using syscall directly so it's fine) So I suspect that the `mount --make-rslave /` command inside the container **should work** because there is always a root `/` mount. I think we should seek consistency: use fs::mount() as long as it doesn't break `mount` commands. Or stick with `mount` command whenever possible? - Jiang Yan Xu On Sept. 21, 2015, 2:23 p.m., Jie Yu wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/38569/ > ----------------------------------------------------------- > > (Updated Sept. 21, 2015, 2:23 p.m.) > > > Review request for mesos, Jojy Varghese, Timothy Chen, and Jiang Yan Xu. > > > Bugs: MESOS-3483 > https://issues.apache.org/jira/browse/MESOS-3483 > > > Repository: mesos > > > Description > ------- > > Made slave's work_dir a shared mount in LinuxFilesystemIsolator. > > See ticket for motivation. > > > Diffs > ----- > > src/slave/containerizer/isolators/filesystem/linux.cpp > f1e6f7519bdeeff7790fff63e7a9cb3075001758 > src/tests/environment.cpp e40cde23daba25b0b61b567ee73682c67e7acbdc > src/tests/utils.cpp 498c9aa1442c4e937a74364073b2267dde6ffcb2 > > Diff: https://reviews.apache.org/r/38569/diff/ > > > Testing > ------- > > sudo make check > > > Thanks, > > Jie Yu > >
