Scheduler fails to register as framework if logstash role isn't defined

[mwl]

On minimal configuration, with only file input enabled, the scheduler fails to register the framework

[sadovnikov]

Marathon task JSON:

{
  "id": "/logstash-scheduler",
  "cpus": 0.2,
  "mem": 256.0,
  "instances": 1,
  "container": {
    "type": "DOCKER",
    "docker": {
      "image": "mesos/logstash-scheduler:latest",
      "network": "BRIDGE"
    }
  },
  "env": {
    "ZK_URL": "zk://172.17.0.3:2181/mesos",
    "LOGSTASH_ELASTICSEARCH_URL": "http://172.17.0.5:33000",
    "LOGSTASH_HEAP_SIZE": "256",
    "LOGSTASH_EXECUTOR_IMAGE": "mesos/logstash-executor",
    "LOGSTASH_EXECUTOR_VERSION": "latest",
    "EXECUTOR_HEAP_SIZE": "128",
    "ENABLE_SYSLOG": "false",
    "ENABLE_FILE": "true",
    "EXECUTOR_FILE_PATH": "/Users/vik/projects/cs/devbox/sources/mesos-elk-demo/build/elk.log, /Users/vik/projects/cs/devbox/sources/mesos-elk-demo/build/*/*.log"
  }
}

stderr output of the task

I0121 09:02:52.177243    34 sched.cpp:1024] Got error 'Role 'logstash' is not present in the master's --roles'
I0121 09:02:52.177374    34 sched.cpp:1805] Asked to abort the driver
I0121 09:02:52.179143    34 sched.cpp:1070] Aborting framework '

and the process does not register as Mesos framework

[mwl]

A quick fix might be to change the mesos role, by adding MESOS_ROLE=* to env object.

[sadovnikov]

The suggested configuration change help. The executor registers now with Mesos Master and framework starts. However I'd expect Scheduler to check if role logstash is defined and then, if present, use it

[mwl]

I don't think that's the right approach. We'll use the user defined role (in --mesos-role=S)

[mwl]

@sadovnikov I'm pretty confident it'll work on a default cluster now (with no logstash role) as long as syslog and collectd input plugins are disabled.