mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Kevin Klues <>
Subject Re: Review Request 50763: Updated Linux 'MountInfoTable' entries to be sorted as expected.
Date Fri, 05 Aug 2016 16:36:24 GMT

This is an automatically generated e-mail. To reply, visit:

(Updated Aug. 5, 2016, 4:36 p.m.)

Review request for mesos, Benjamin Mahler and Jie Yu.


CI was breaking because I assumed that the parent ID of the root mount point was always 1
(which is apparently not true). The updated code avoids this assumption.

Bugs: MESOS-5969

Repository: mesos


Many places in the codebase assume that the mountinfo table is sorted
according to the order: 'parent mount point < child mount point'.

However, in some cases this may not be true if (for example), a parent
mount point (say '/') is remounted to add some extra flags to it.
When this happens, the remounted file system will appear in the
mountinfo table at the point where it was remounted.

We actually encountered this problem in the wild for the case of '/'
being remounted after '/run' was mounted -- causing problems in the
'NvidiaVolume' which assumes the 'parent < child' ordering.

This commit fixes this problem by building the list of MountInfoTable
entries in sorted order when 'read()' is called. An optional flag can
be used to disable sorting produce the the original ordering.

Diffs (updated)

  src/linux/fs.hpp ec3b5b8cd6926b1f69ad499de1c13b989766a84e 
  src/linux/fs.cpp f57db80ad0d7235d47910e05d663c77e233f8228 
  src/tests/containerizer/fs_tests.cpp 4cfafad28daac6bed849992a254660117d7ff30b 



GTEST_FILTER="" make -j check
sudo src/mesos-tests

Appeared to have one unrelated flaky test fail: `ResourceOffersTest.ResourcesGetReofferedAfterTaskInfoError`
Rerunning the tests a second time passed.


Kevin Klues

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message