[MESOS-1558] Removing bind mount in the network isolator has race condition

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Target Version/s: 0.20.0
• Epic Link: Container Network Monitoring
• Sprint: Q2 Sprint 5
• Assignee: Jie Yu
• Reporter: Jie Yu
• Shepherd: Vinod Kone
• Created: 01/Jul/14 20:59
• Updated: 14/Jul/14 16:56
• Resolved: 14/Jul/14 16:56

Description

Say there are two containers: A and B. Here is the sequence of events:

1) A is launched and running
2) B is cloned
3) Before B executes child isolation scripts which remove all bind mounts under /var/run/netns, A exits.
4) A tries to remove it's corresponding bind mount under /var/run/netns
5) A gets a "device is busy" error