> On July 18, 2018, 1:51 p.m., James Peach wrote: > > src/slave/flags.cpp > > Lines 1338 (patched) > > <https://reviews.apache.org/r/67914/diff/2/?file=2059409#file2059409line1338> > > > > Rather than introducing a new flag, lets use > > `container_disk_watch_interval` or `disk_watch_interval` ... probably the > > former? > > Ilya Pronin wrote: > I think if we do this then we should rather use the latter because disk > GC is driven by `disk_watch_interval` and `gc_delay` (maybe use which one of > those 2 is the smallest?). `container_disk_watch_interval` will most likely > be set to a smaller value to promptly terminate containers that have breached > their quota (that's true in our cluster). Running sandboxes check at that > frequently would be wasteful. > > James Peach wrote: > OK, `disk_watch_interval` works for me too.
Done. - Ilya ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/67914/#review206203 ----------------------------------------------------------- On July 23, 2018, 4:21 p.m., Ilya Pronin wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/67914/ > ----------------------------------------------------------- > > (Updated July 23, 2018, 4:21 p.m.) > > > Review request for mesos and James Peach. > > > Bugs: MESOS-9007 > https://issues.apache.org/jira/browse/MESOS-9007 > > > Repository: mesos > > > Description > ------- > > Currently upon container destruction its project ID is unallocated by > the isolator and removed from the container work directory. However due > to API limitations we can't unset project IDs on symlinks that may exist > inside the directory. Because of that the project may still exist until > the container directory is garbage collected. If the project ID is > reused for a new container, any lingering symlinks that still have that > project ID will contribute to disk usage of the new container. Typically > symlinks don't take much space, but still this leads to inaccuracy in > disk space usage accounting. > > This patch postpones project ID reclaiming until sandbox GC time. The > isolator periodically checks if sandboxes of terminated containers still > exist and deallocates project IDs of the ones that were removed. Checks > are performed every "disk_watch_interval". This mechanism can be > improved in the future if we introduce a way for the isolators to learn > about disk GCs. > > Current number of available project IDs can be tracked with the new > "containerizer/mesos/disk/project_ids_free" and > "containerizer/mesos/disk/project_ids_total" metrics. > > > Diffs > ----- > > src/slave/containerizer/mesos/isolators/xfs/disk.hpp > 0891f7709aa4f98758a727856d58e6177d46adca > src/slave/containerizer/mesos/isolators/xfs/disk.cpp > 25f52a43b34b141bdaf7c448817423cf4264e22a > src/tests/containerizer/xfs_quota_tests.cpp > dc18a8a59d1eb7fae3592ef6ba8c046e4f46ee4a > > > Diff: https://reviews.apache.org/r/67914/diff/4/ > > > Testing > ------- > > Added `ROOT_XFS_QuotaTest.ProjectIDReclaiming` test that verifies that > project ID is reclaimed and reused after sandbox GC. Ran `sudo make check`. > > > Thanks, > > Ilya Pronin > >
