The namespaces/pid isolator can be used to isolate each container in a separate pid namespace with two main benefits:
Visibility: Processes running in the container (executor and descendants) are unable to see or signal processes outside the namespace.
Clean termination: Termination of the leading process in a pid namespace will result in the kernel terminating all other processes in the namespace.
You can turn on this isolator by specifying the --isolation agent flag (i.e., --isolation=namespaces/pid,...). Note that filesystem/linux isolator is required for turning on pid namespace isolator.
The Launcher will use (2) during destruction of a container in preference to the freezer cgroup, avoiding known kernel issues related to freezing cgroups under OOM conditions.
/proc will be mounted for containers so tools such as ps will work correctly.
To enable the PID Namespace isolator, append namespaces/pid to the --isolation flag when starting the agent. By default, each container will have its own PID namespace if this isolator is enabled.
Framework users can allow a container to share pid namespace with its parent by setting the ContainerInfo.linux_info.share_pid_namespace field to true. If the container is a top level container, it will share the pid namespace with the agent. If the container is a nested container, it will share the pid namespace with its parent container. The container will have its own pid namespace if the ContainerInfo.linux_info.share_pid_namespace field is set to false.
As a security measure, operators can disallow any container to share the agent’s PID namespace by setting the agent flag --disallow_sharing_agent_pid_namespace to true. If this agent flag is set as true and the framework requests to launch a top level container which shares its pid namespace with the agent, the container launch will be rejected.