> On March 9, 2016, 10:03 a.m., Jiang Yan Xu wrote: > > src/slave/containerizer/mesos/isolators/disk/xfs.cpp, lines 573-576 > > <https://reviews.apache.org/r/44342/diff/3/?file=1279808#file1279808line573> > > > > We do need to handle orphans, as you've commented in `cleanup()`. > > > > `orphans` here are known to the containerizer so `cleanup()` will be > > called by it to remove the quota. However if there's not an entry for it in > > `infos`, `cleanup()` ignores it. > > > > The concern is that if the project ID of the orphan is not unassigned > > from the sandbox and the ID is not tracked by the isolator, next time it's > > used by a new task, it'll share the same quota with the orphan.
OK it looks to me that we do need to recursively scan all the sandboxes (we can probably `glob` them) and determine whether they belong to an recovered container, known orphan and unknown orphan and then clean up the extraneous ones. - Jiang Yan ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/44342/#review122156 ----------------------------------------------------------- On March 7, 2016, 10:38 a.m., James Peach wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/44342/ > ----------------------------------------------------------- > > (Updated March 7, 2016, 10:38 a.m.) > > > Review request for mesos, Jie Yu and Jiang Yan Xu. > > > Bugs: MESOS-4828 > https://issues.apache.org/jira/browse/MESOS-4828 > > > Repository: mesos > > > Description > ------- > > Track sandbox directory usage by dynamically assigning XFS project > quotas. We track a range of XFS project IDs, assigning a project ID > and a project quota to each sandbox as it is created. When the task > reaches the quota, writes will fail with EDQUOT, and the task will have > an opportunity to handle that. > > Quotas are not applied to volume resources since the isolator interface > has no insight into the volume lifecycle. Thus it is not currently > possible to accurately assign and reclaim project IDs. > > If LOW is the lower bound of the project ID range and HIGH is the upper > bound, you can show the currently allocated project quotas using the > xfs_quota command: > > $ xfs_quota -x -c "report -a -n -L LOW -U HIGH" > > To show the project ID assigned to the file PATH, use the xfs_io command: > > $ xfs_io -r -c stat PATH > > > Diffs > ----- > > configure.ac a20382e8d425eb297492a6e6c2c75ea59be097c2 > docs/configuration.md 305ba2c801c2060db6dcb4ef83c1043aaa7d520c > docs/mesos-containerizer.md 15fb5bdbe74e059614b8948108f32cd04b623305 > src/Makefile.am a41e95ddeb838fdebf4ced953c4a29181916e261 > src/slave/containerizer/mesos/containerizer.cpp > af3ff5750649497d8852b4761c78d4cae5455a02 > src/slave/containerizer/mesos/isolators/disk/xfs.hpp PRE-CREATION > src/slave/containerizer/mesos/isolators/disk/xfs.cpp PRE-CREATION > src/slave/flags.hpp feb095da4521f678c96f4cc53bdfda262d350388 > src/slave/flags.cpp 6e3fd69c06eefd40bc0e5c222ea72f34144c5534 > > Diff: https://reviews.apache.org/r/44342/diff/ > > > Testing > ------- > > Manual testing on Fedora 23 w/ XFS. Make check on Fedora and OS X. > > > Thanks, > > James Peach > >
