----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/66001/#review199445 -----------------------------------------------------------
I've taken a first pass, but before we get into the details of the style and implementation, I'd like to address the overall direction. We need to narrow the scope of this change to make it easier to land. to that end, we should: - remove the addition of nesting support - remove code that deals with quota timers - remove all the configuration options - Move the tests into separate patches For configuration, we can use `container_disk_watch_interval` for the polling interval, and (for now) hard-code a 10MB red zone (i.e. make the hard limit 10MB larger than the soft limit). To match the behaviour of the `disk/du` isolator we just kill containers that violate the soft limit for any period of time. To enable the killing behaviour make an explicit flag `xfs_kill_containers`. Later, we can consider whether this could be per-container. - James Peach On March 16, 2018, 10:43 a.m., Harold Dost wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/66001/ > ----------------------------------------------------------- > > (Updated March 16, 2018, 10:43 a.m.) > > > Review request for mesos and James Peach. > > > Bugs: MESOS-6575 > https://issues.apache.org/jira/browse/MESOS-6575 > > > Repository: mesos > > > Description > ------- > > New Flags for disk/xfs isolator > - This patch adds a number of flags to handle switching the limit in the > `disk/xfs` isolator to allow apps to go over their limit and for mesos > to kill them if they have gone over their limit. > > New Flags: > - xfs_disk_hard_limit_offset_pct - Use the `disk` as the soft limit and > set the hard limit to be some percentage above the soft limit. > Allowing > for containers to surpass a desired allocation and making them > killable. > - xfs_disk_hard_limit_offset - Use the `disk` as the soft limit and set > the hard limit to some number of bytes specified above the > applications to be a soft limit instead of a hard limit. > - xfs_kill_after_grace_period - This will kill tasks if they breach the > grace period configured using `xfs_quota -x -c "timer -p <time>"` > - xfs_kill_check_interval - The frequency with which a container will be > checked for soft limit violations. > > Functionality > - Add head room to the hard limit as a percentage or specified amount > for each container. This is specified at a flag level and not > customizable on a per container basis. > - Provide the ability for an application to be killed after the > configured grace period for projects is violated. > - Add an interval between which the watcher will check for violations. > > > Diffs > ----- > > src/slave/containerizer/mesos/isolators/xfs/disk.hpp > 07e68a777aefba4dd35066f2eb207bba7f199d83 > src/slave/containerizer/mesos/isolators/xfs/disk.cpp > 8d9f8f846866f9de377c59cb7fb311041283ba70 > src/slave/containerizer/mesos/isolators/xfs/utils.hpp > e034133629a9c1cf58b776f8da2a93421332cee0 > src/slave/containerizer/mesos/isolators/xfs/utils.cpp > 2708524add1ff693b616d4fb241c4a0a3070520b > src/slave/flags.hpp 0c67bf214ceb93ae7ff088bec2648fa26ddac59e > src/slave/flags.cpp 962b07c1d701f4ab819b14730fbc116b981433bb > src/tests/containerizer/xfs_quota_tests.cpp > 64c3e1c3f0bc435897626cb0a13bc19c7cb1a4fe > > > Diff: https://reviews.apache.org/r/66001/diff/7/ > > > Testing > ------- > > > Thanks, > > Harold Dost > >
