Horizontally Scaling to More than Number of Agent Nodes

[lhr0909]

I am wondering if we can scale the ES cluster to have more nodes than the # of agent nodes. I am currently using the latest mesos-es-scheduler container mode with latest image and with the following config on a 3-master HA 3-agent Mesos v1.0.1 cluster:

{
  "id": "/elasticsearch-scheduler",
  "cmd": null,
  "cpus": 0.2,
  "mem": 512,
  "disk": 0,
  "instances": 1,
  "container": {
    "type": "DOCKER",
    "volumes": [],
    "docker": {
      "image": "mesos/elasticsearch-scheduler",
      "network": "HOST",
      "portMappings": null,
      "privileged": false,
      "parameters": [],
      "forcePullImage": false
    }
  },
  "env": {
    "JAVA_OPTS": "-Xms128m -Xmx256m"
  },
  "portDefinitions": [
    {
      "port": 10000,
      "protocol": "tcp",
      "labels": {}
    }
  ],
  "args": [
    "--zookeeperMesosUrl",
    "zk://10.0.0.2:2181,10.0.0.243:2181,10.0.0.149:2181/mesos",
    "--frameworkName",
    "elasticsearch-scheduler",
    "--elasticsearchClusterName",
    "es-mesos",
    "--elasticsearchCpu",
    "1",
    "--elasticsearchDisk",
    "10240.0",
    "--elasticsearchNodes",
    "5",
    "--elasticsearchRam",
    "1024.0"
  ]
}

With this config I could only spawn 3 ES nodes for the cluster, and I am wondering if we could schedule more than that on 3 agents. Resource-wise there is definitely enough (CPU, disk and RAM) for 2 more, so I am just wondering why we couldn't make it to work.

also I am wondering if --elasticsearchDisk and --elasticsearchRam options are for the overall cluster or per-node. Thanks!

[krishnabigdata]

Hi,

As per the elasticsearch configuration and persistence volume requirement for elasticsearch you should run only 1 ES node per Mesos slave agent and the mesos slave offer as well configured to allocate resource per host basis only. It is the recommended approach. If you need more detail, please let me know.