apache mesos 入门（4）——mesos中的资源

mesos根据角色（ROLE）在不同的framework之间分配资源，在启动master时可以指定系统中的角色和每种角色的权重，例如：

mesos-master --ip=127.0.0.1 --work_dir=/var/lib/mesos --roles=qarole,devrole --weights='qarole=1,devrole=3'

指定了两种角色parole和devrole，权重分别是1和3.

在设置FrameworkInfo时可以设置对应的角色

/**
 * Describes a framework.
 */
message FrameworkInfo {
  // Used to determine the Unix user that an executor or task should
  // be launched as. If the user field is set to an empty string Mesos
  // will automagically set it to the current user.
  required string user = 1;

  // Name of the framework that shows up in the Mesos Web UI.
  required string name = 2;

  // Note that 'id' is only available after a framework has
  // registered, however, it is included here in order to facilitate
  // scheduler failover (i.e., if it is set then the
  // MesosSchedulerDriver expects the scheduler is performing
  // failover).
  optional FrameworkID id = 3;

  // The amount of time (in seconds) that the master will wait for the
  // scheduler to failover before it tears down the framework by
  // killing all its tasks/executors. This should be non-zero if a
  // framework expects to reconnect after a failure and not lose its
  // tasks/executors.
  //
  // NOTE: To avoid accidental destruction of tasks, production
  // frameworks typically set this to a large value (e.g., 1 week).
  optional double failover_timeout = 4 [default = 0.0];

  // If set, framework pid, executor pids and status updates are
  // checkpointed to disk by the slaves. Checkpointing allows a
  // restarted slave to reconnect with old executors and recover
  // status updates, at the cost of disk I/O.
  optional bool checkpoint = 5 [default = false];

<span style="color:#ff0000;">  // Used to group frameworks for allocation decisions, depending on
  // the allocation policy being used.
  optional string role = 6 [default = "*"];</span>

  // Used to indicate the current host from which the scheduler is
  // registered in the Mesos Web UI. If set to an empty string Mesos
  // will automagically set it to the current hostname if one is
  // available.
  optional string hostname = 7;

  // This field should match the credential's principal the framework
  // uses for authentication. This field is used for framework API
  // rate limiting and dynamic reservations. It should be set even
  // if authentication is not enabled if these features are desired.
  optional string principal = 8;

  // This field allows a framework to advertise its web UI, so that
  // the Mesos web UI can link to it. It is expected to be a full URL,
  // for example http://my-scheduler.example.com:8080/.
  optional string webui_url = 9;

  message Capability {
    enum Type {
      // Receive offers with revocable resources. See 'Resource'
      // message for details.
      // TODO(vinod): This is currently a no-op.
      REVOCABLE_RESOURCES = 1;

      // Receive the TASK_KILLING TaskState when a task is being
      // killed by an executor. The executor will examine this
      // capability to determine whether it can send TASK_KILLING.
      TASK_KILLING_STATE = 2;
    }

    required Type type = 1;
  }

  // This field allows a framework to advertise its set of
  // capabilities (e.g., ability to receive offers for revocable
  // resources).
  repeated Capability capabilities = 10;

  // Labels are free-form key value pairs supplied by the framework
  // scheduler (e.g., to describe additional functionality offered by
  // the framework). These labels are not interpreted by Mesos itself.
  // Labels should not contain duplicate key-value pairs.
  optional Labels labels = 11;
}

slave启动时会检测当前节点的资源情况，包括cpu、内存、磁盘等，slave默认会预留1G（或者50%）的内存，5G（或者50%）的磁盘以保证自己及操作系统的正常运行，剩下的资源会上报master以供调度。当然也可以手动指定应该上报的资源量，例如，

mesos-slave --master=127.0.0.1:5050 --resources='cpus:4;mem:8192;ports:[1000-5000,31000-32000];'

slave还可以静态的设置各个角色的资源量，例如：

mesos-slave --master=127.0.0.1:5050 --resources='cpus(prod):1;mem(prod):4096;cpus(qa):2;mem(qa):1024;cpus:13;mem:11264' --attributes='operating_system:ubuntu;cpu_class:haswell;zone:us_east;rack:22'

其中prod角色：1cpu、4096内存

qa角色：2cpu、1024内存

默认角色：13cpu、11264内存

除了上报slave节点的资源情况，也可以标记当前节点的一些属性以供scheduler使用，比如是否ssd硬盘，操作系统类型等

mesos-slave --master=127.0.0.1:5050 --resources='cpus:4;mem:8192;ports:[1000-5000,31000-32000];' --attributes='operating_system:ubuntu;cpu_class:haswell;zone:us_east;rack:22'