[MESOS-2016] docker_name_prefix is too generic

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.23.0
• Component/s: containerization, docker
• Target Version/s: 0.23.0
• Epic Link: Recover docker in slave container
• Sprint: Mesosphere Q4 Sprint 2 - 11/14, Mesosphere Q4 Sprint 3 - 12/7, Mesosphere Q1 Sprint 1 - 1/23, Mesosphere Q1 Sprint 2 - 2/6, Mesosphere Q1 Sprint 3 - 2/20, Mesosphere Q1 Sprint 4 - 3/6, Mesosphere Q1 Sprint 5 - 3/20, Mesosphere Sprint 10
• Assignee: Timothy Chen
• Reporter: Jay Buffington
• Shepherd: Benjamin Hindman
• Created: 30/Oct/14 16:44
• Updated: 11/Oct/15 10:33
• Resolved: 25/May/15 17:46

Description

From docker.hpp and docker.cpp:

    // Prefix used to name Docker containers in order to distinguish those
    // created by Mesos from those created manually.
    extern std::string DOCKER_NAME_PREFIX;

    // TODO(benh): At some point to run multiple slaves we'll need to make
    // the Docker container name creation include the slave ID.
    string DOCKER_NAME_PREFIX = "mesos-";

This name is too generic. A common pattern in docker land is to run everything in a container and use volume mounts to share sockets do RPC between containers. CoreOS has popularized this technique.

Inevitably, what people do is start a container named "mesos-slave" which runs the docker containerizer recovery code which removes all containers that start with "mesos-" And then ask "huh, why did my mesos-slave docker container die? I don't see any error messages..."

Ideally, we should do what Ben suggested and add the slave id to the name prefix.

Issue Links

• breaks: MESOS-3699 After upgrade version to 0.23.0, then If a task that has been executing from version 0.22.X is killed, the docker container has been orphaned.