[MESOS-2845] Command tasks lead to a mixing of revocable / non-revocable cpus and memory within the container.

• Type: Bug
• Status: Accepted
• Priority: Major
• Resolution: Unresolved
• Component/s: agent
• Labels: twitter
• Epic Link: Oversubscription
• Assignee: Unassigned
• Reporter: Benjamin Mahler
• Created: 10/Jun/15 19:58
• Updated: 09/Jul/16 10:48

Description

Due to the hack here, where we add a small set of resources into the command executor:

ExecutorInfo Slave::getExecutorInfo(
    const FrameworkID& frameworkId,
    const TaskInfo& task)
{
  if (task.has_command()) {
    ...

    // XXX: These are always non-revocable.

    // Add an allowance for the command executor. This does lead to a
    // small overcommit of resources.
    executor.mutable_resources()->MergeFrom(
        Resources::parse(
          "cpus:" + stringify(DEFAULT_EXECUTOR_CPUS) + ";" +
          "mem:" + stringify(DEFAULT_EXECUTOR_MEM.megabytes())).get());
  }
  ...
}

The obvious extension here would be to make these revocable, but would be great to remove this hack entirely.

Seems to originate in r/22251 from MESOS-1417.

FYI idownes jieyu

Issue Links

• depends upon: MESOS-1718 Command executor can overcommit the agent.
• relates to: MESOS-1417 Slave should not send terminal status update before containerizer update is finished