> On Aug. 22, 2016, 3:09 a.m., Jie Yu wrote: > > src/slave/containerizer/mesos/isolators/cgroups/subsystems/memory.cpp, > > lines 166-169 > > <https://reviews.apache.org/r/49851/diff/21/?file=1480170#file1480170line166> > > > > Realized an issue. What if the operator changes the flags to add memory > > cgroup support later. During recovery, we don't recovery the container. > > That means the subsequent update will fail later. > > > > I am wondering if we should add some logic in the cgroups isolator to > > remember the subsystems that we recovered (also in launch path), and only > > call `update`, `wait`, `usage`, `status` on those subsystems? > > > > Can you follow up on that? > > Qian Zhang wrote: > Jie, I remembered that we discussed this issue before, currently we don't > support restarting agent with a different set of isolators. So I think this > is a general issue for all isolators rather than specific to cgroups > isolator, maybe we need a dedicated project to handle it for all isolators? > > Jie Yu wrote: > Yeah, but that means we cannot safely add more cgroups systems without > reboot the box. That's really unfortunate. > > haosdent huang wrote: > Let me file a ticket and submit follow up patches to fix this. > > haosdent huang wrote: > When restart Mesos Agent with different cgroups subystems, current > unified cgroups isolator behaviour are same with the old one. They ignore it > in recover and continue to call `usage` and `status` on them. Because we > ignore failure in `usage` and `status`, restart Mesos Agent with different > cgoups subsystems is allowed. > > I create a ticket here https://issues.apache.org/jira/browse/MESOS-6063 > to track this.
what about 'update', 'update' failures won't be ingored. I think this is a pretty high priority issue. Could you please help solve it asap? I think this will be a blocker if we want to turn on other subsystem support in a backwards compatible way. - Jie ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/49851/#review146267 ----------------------------------------------------------- On Aug. 21, 2016, 9:29 a.m., haosdent huang wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/49851/ > ----------------------------------------------------------- > > (Updated Aug. 21, 2016, 9:29 a.m.) > > > Review request for mesos, Gilbert Song, Jie Yu, Qian Zhang, and Zhengju Sha. > > > Bugs: MESOS-5045 > https://issues.apache.org/jira/browse/MESOS-5045 > > > Repository: mesos > > > Description > ------- > > Implemented `MemorySubsystem`. > > > Diffs > ----- > > src/CMakeLists.txt ff51705202569b85922d877ef95750d7943372d8 > src/Makefile.am 61c941f42bdccdf1745f72f875cb5c5a9901dc76 > src/slave/containerizer/mesos/isolators/cgroups/cgroups.hpp > 9b2d33ec3b023058d00c6671464cc9cd092f653b > src/slave/containerizer/mesos/isolators/cgroups/cgroups.cpp > 0c724523a8b46f74a52214ae2b223f41f8dc2d58 > src/slave/containerizer/mesos/isolators/cgroups/constants.hpp > 4a7b556bb74cfda0a74e1b9b5d4bb226b84751c1 > src/slave/containerizer/mesos/isolators/cgroups/subsystem.hpp > 06f4400966ba467623556901b38d12e69fbbbd04 > src/slave/containerizer/mesos/isolators/cgroups/subsystem.cpp > df701d2fa7ce7622e6a32904ba3a99fd36b02df6 > src/slave/containerizer/mesos/isolators/cgroups/subsystems/memory.hpp > PRE-CREATION > src/slave/containerizer/mesos/isolators/cgroups/subsystems/memory.cpp > PRE-CREATION > > Diff: https://reviews.apache.org/r/49851/diff/ > > > Testing > ------- > > > Thanks, > > haosdent huang > >
