[MESOS-4711] Race condition in libevent poll implementation causes crash

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.28.0
• Fix Version/s: 0.24.2, 0.25.1, 0.26.1, 0.27.2, 0.28.0
• Component/s: libprocess
• Labels: mesosphere

Environment

CentOS 6.7 running in VirtualBox

• Sprint: Mesosphere Sprint 29
• Assignee: Alexander Rojas
• Reporter: Alexander Rojas
• Shepherd: Joris Van Remoortere
• Created: 18/Feb/16 19:29
• Updated: 26/Nov/18 12:22
• Resolved: 26/Nov/18 12:22

Description

The issue first arose in MESOS-3271, but can be reproduced every time by using the mentioned environment and running:

sudo ./bin/mesos-tests.sh --gtest_filter="MemoryPressureMesosTest.CGROUPS_ROOT_SlaveRecovery" --gtest_repeat=1000

The problem can be traced back to libevent_poll.cpp. If the event is triggered and the the future associated with the event is discarded, the situation arises in which pollCallback() starts executing just early enough to finish before pollDiscard() executes. If that happens, pollCallback() deletes the poll object and pollDiscard() is left with a dangling pointer which crashes when it executes the line event_active(ev, EV_READ, 0);.

Issue Links

• blocks: MESOS-3271 SlaveRecoveryTest/0.NonCheckpointingFramework is flaky.