Constraining Applications to Portworx nodes

Whenever possible, Portworx should be deployed on all nodes within a Mesos/Mesosphere/DCOS cluster. Portworx clusters can scale up to 1000 nodes. Portworx clusters can also include nodes that contribute no storage, but operates as “head-only” nodes to facilitate auto-scaling groups.

However, when Portworx cannot be installed on all nodes, Mesos ‘constraints’ should be used to ensure that services depending on Portworx will only get scheduled on nodes that are part of a Portworx cluster.

Install DCOS CLI or Apache Mesos

For Mesosphere, follow the instructions for installing Mesosphere DCOS and the DCOS CLI. Use the DCOS CLI command dcos node to identify which nodes in the Mesos cluster are the Agent nodes.

If not using Mesosphere, then follow the instructions appropriate for your OS distribution and environment to install both Apache Mesos and Marathon.

Add constraints on agent nodes

If the size of your Mesos cluster is larger than the maximum number of nodes supported for a Portworx release, or if it is not possible to install Portworx on all nodes in the Mesos cluster, then you will need to use Mesos “constraints”, in order to restrict/constrain jobs that use Portworx volumes to only run on Mesos agent nodes where Portworx is running. (Please check the Portworx release notes for maximum Portworx cluster size). Otherwise, Portworx, Inc. recommends simply deploying Portworx on all nodes in a Mesos cluster, thereby avoiding the need to use “constraints”.

The following steps are required only for configuring and using “constraints”.

For each Mesosphere Agent node or Mesos agent that is participating in the Portworx cluster, specify Mesos attributes that allow for affinity of tasks to nodes that are part of the Portworx cluster.

If using Mesosphere/DCOS:

echo MESOS_ATTRIBUTES=pxfabric:pxclust1 >> /var/lib/dcos/mesos-slave-common
rm -f /var/lib/mesos/slave/meta/slaves/latest
systemctl restart dcos-mesos-slave.service
systemctl status dcos-mesos-slave.service -l

If using Apache Mesos:

mkdir -p /etc/default/mesos-slave/attributes
echo pxclust1 > /etc/default/mesos-slave/attributes/pxfabric
rm -f /var/lib/mesos/slave/meta/slaves/latest
systemctl restart mesos-slave
systemctl status mesos-slave -l

Please note that restarting a dcos/mesos agent to change the MESOS_ATTRIBUTES will cause the node to register under a new UUID.

Deploy Portworx with ‘constraints’

When deploying the Portworx framework you would then add the following constraint to node.placement_constraint

pxfabric:LIKE:pxclust1

This should be used in addition to the “hostname:UNIQUE” constraint.

Consume Portworx with constraints

The following example shows how to use constraints for running mysql only on nodes where Portworx is installed.

{
    "id": "mysql",
    "cpus": 0.5,
    "mem": 256,
    "instances": 1,
    "container": {
        "type": "DOCKER",
        "docker": {
            "image": "mysql:5.6.27",
            "parameters": [
                    {
                       "key": "volume-driver",
                       "value": "pxd"
                    },
                    {
                       "key": "volume",
                       "value": "mysql_vol:/var/lib/mysql"
                    }],
            "network": "BRIDGE",
              "portMappings": [
                {
                  "containerPort": 3306,
                  "hostPort": 32000,
                  "protocol": "tcp"
                }
                ]
        }
    },
    "constraints": [
            [
              "pxfabric",
              "LIKE",
              "pxclust1"
            ]],
    "env": {
        "MYSQL_ROOT_PASSWORD": "password"
    },
      "minimumHealthCapacity" :0,
      "maximumOverCapacity" : 0.0
}

Download example