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

(Updated Dec. 2, 2015, 11:25 a.m.)

Review request for mesos, Bernd Mathiske, Greg Mann, Artem Harutyunyan, Jan 
Schlicht, and Till Toenshoff.


Fixed typo in comment.  Tested on a few more platforms.

Summary (updated)

Fix flaky MemoryPressureMesosTests

Bugs: MESOS-3586

Repository: mesos


The existing tests will check that "low" pressure events occur at least as 
often as "medium" pressure events (this is the documented behavior).  However, 
the order of events and the order in which we process said events is not 
guaranteed.  When we collect the pressure events via a counter, there may be 
some events that are in-flight, and thereby not accounted for in the counters.

This patch modifies MemoryPressureMesosTests to wait for memory pressure events 
to stop before checking the counts for correctness.
The tests now stop the memory-pressure-triggering task and then wait for all 
events to be processed before checking the counters.

Diffs (updated)


Diff: https://reviews.apache.org/r/40849/diff/

Testing (updated)

On Debian 8, Ubuntu 14, Centos 7 (Thanks Jan!):
`make check`
`sudo bin/mesos-tests.sh --gtest_filter="*MemoryPressureMesosTest*" 
--gtest_repeat=-1 --gtest_break_on_failure`

^ Ran the above for a couple minutes or ~100 times.  It was previously failing 
~1/5 times.

Note on Centos 6 (Thanks Greg!):
[ RUN      ] MemoryPressureMesosTest.CGROUPS_ROOT_Statistics
../../src/tests/mesos.cpp:849: Failure
Value of: _baseHierarchy.get()
  Actual: "/cgroup"
Expected: baseHierarchy
Which is: "/tmp/mesos_test_cgroup"
Multiple cgroups base hierarchies detected:
Mesos does not support multiple cgroups base hierarchies.
Please unmount the corresponding (or all) subsystems.
../../src/tests/mesos.cpp:932: Failure
(cgroups::destroy(hierarchy, cgroup)).failure(): Failed to remove cgroup 
'/tmp/mesos_test_cgroup/perf_event/mesos_test': Device or resource busy
[  FAILED  ] MemoryPressureMesosTest.CGROUPS_ROOT_Statistics (12 ms)

Note on Ubuntu 14:
There is some other flakiness (in Agent recovery).  This will be tracked in 


Joseph Wu

Reply via email to