[MESOS-9076] Mesos agent will be wrongly treated as unknown orphaned container if `--cgroups_root` has a leading slash

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: agent, containerization
• Assignee: Unassigned
• Reporter: Qian Zhang
• Created: 16/Jul/18 06:59
• Updated: 16/Jul/18 07:06

Description

When agent is started with the following flags:

• --cgroups_root=<a value with a leading slash>, e.g., /mesos.
• --agent_subsystems=<some cgroups subsystems>, e.g., memory.
• --isolation=<some cgroups subsystems which contain the ones specified in --agent_subsystems>, e.g., cgroups/cpu,cgroups/mem.

we will see the agent will be treated as an unknown orphaned container:

I0716 14:55:59.969400 3892 containerizer.cpp:718] Recovering Mesos containers
I0716 14:55:59.970304 3888 linux_launcher.cpp:299] Recovering Linux launcher
I0716 14:55:59.973687 3894 containerizer.cpp:1025] Recovering isolators
W0716 14:55:59.978680 3891 cgroups.cpp:352] Couldn't find the cgroup '/mesos/slave' in hierarchy '/sys/fs/cgroup/cpu,cpuacct' for container slave
I0716 14:55:59.981603 3892 memory.cpp:478] Started listening for OOM events for container slave
I0716 14:55:59.983178 3892 memory.cpp:590] Started listening on 'low' memory pressure events for container slave
I0716 14:55:59.985761 3892 memory.cpp:590] Started listening on 'medium' memory pressure events for container slave
I0716 14:55:59.987036 3892 memory.cpp:590] Started listening on 'critical' memory pressure events for container slave
I0716 14:55:59.988668 3893 cgroups.cpp:320] Cleaning up unknown orphaned container slave