[MESOS-7050] IOSwitchboard FDs leaked when containerizer launch fails -- leads to deadlock

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.2.0
• Labels: debugging mesosphere
• Target Version/s: 1.2.0
• Epic Link: Container Attach/Exec Improvements
• Sprint: Mesosphere Sprint 51, Mesosphere Sprint 52
• Story Points: 2
• Assignee: Kevin Klues
• Reporter: Kevin Klues
• Shepherd: Jie Yu
• Created: 02/Feb/17 00:00
• Updated: 01/Mar/17 18:17
• Resolved: 01/Mar/17 18:15

Description

If the containizer launch path fails before actually
launching the container, the FDs allocated to the container by the
IOSwitchboard isolator are leaked. This leads to deadlock in
the destroy path because the IOSwitchboard does not shutdown until the
FDs it allocates to the container have been closed. Since the
switchboard doesn't shutdown, the future returned by its 'cleanup()'
function is never satisfied.

We need a general purpose method for closing the IOSwitchboard FDs when failing in the launch path.

Issue Links

• blocks: MESOS-6280 Task group executor should support command health checks.