Making Your Cluster Elastic

Making Your Cluster Elastic

Creating Receivers

Suppose you want a cluster to scale out by one node each time an event occurs, you can create a receiver for this task:

$ senlin receiver-create --type webhook --cluster mycluster \
    --action CLUSTER_SCALE_OUT so_receiver_1
+------------+---------------------------------------------------+
| Property   | Value                                             |
+------------+---------------------------------------------------+
| action     | CLUSTER_SCALE_OUT                                 |
| actor      | {                                                 |
|            |   "trust_id": "b2b8fd71c3d54f67ac14e5851c0117b8"  |
|            | }                                                 |
| channel    | {                                                 |
|            |   "alarm_url": "<WEBHOOK_URL>"                    |
|            | }                                                 |
| cluster_id | 30d7ef94-114f-4163-9120-412b78ba38bb              |
| created_at | 2017-02-08T02:08:13Z                              |
| domain_id  | -                                                 |
| id         | 5722a2b0-1f5f-4a82-9c08-27da9982d46f              |
| location   | -                                                 |
| name       | so_receiver_1                                     |
| params     | {}                                                |
| project_id | 36d551c0594b4cc99d1bbff8bf202ec3                  |
| type       | webhook                                           |
| updated_at | -                                                 |
| user_id    | 9563fa29642a4efdb1033bf8aab07daa                  |
+------------+---------------------------------------------------+

The command above creates a receiver named so_receiver_1 which can be used to initiate a CLUSTER_SCALE_OUT action on the cluster my_cluster. From the output of this command, you will find an alarm_url value from the channel property. This will be the URL for you to trigger the scaling operation.

Note

You are expected to treat the alarm_url value as a secret. Any person or software which knows this value will be able to trigger the scaling operation on your cluster. This may not be what you wanted.

The default type of receiver would be “webhook”. You may choose to create a “message” type of receiver if you have the zaqar messaging service installed. For more details, please refer to Receiver.

Triggering Scaling

Once you have received a channel from the created receiver, you can use it to trigger the associated action on the specified cluster. The simplest way to do this is to use the cURL command as shown below:

$ curl -X POST <WEBHOOK_URL>

Once the above request is received by the senlin-api, your cluster will be scaled out by one node. In other words, a new node is created into the cluster.

Creating Scaling Policies

Senlin provides some builtin policy types to control how a cluster will be scaled when a relevant request is received. A scaling request can be a simple CLUSTER_SCALE_OUT or CLUSTER_SCALE_IN action which can accept an optional count argument; it can be a more complex CLUSTER_RESIZE action which can accept more arguments for fine-tuning the scaling behavior.

In the absence of such arguments (which is not uncommon if you are using a 3rd party monitoring software which doesn’t have the intelligence to decide each and every argument), you can always use scaling policies for this purpose.

Below is a sample YAML file (examples/policies/scaling_policy.yaml) used for creating a scaling policy object:

type: senlin.policy.scaling
version: 1.0
properties:
  event: CLUSTER_SCALE_IN
  adjustment:
    type: CHANGE_IN_CAPACITY
    number: 2
    min_step: 1
    best_effort: True
    cooldown: 120

To create a policy object, you can use the following command:

$ openstack cluster policy create \
  --spec-file examples/policies/scaling_policy.yaml \
  policy1
+------------+--------------------------------------+
| Field      | Value                                |
+------------+--------------------------------------+
| created_at | 2016-12-08T02:41:30.000000           |
| data       | {}                                   |
| domain     |                                      |
| id         | 3ca962c5-68ce-4293-9087-c73964546223 |
| name       | policy1                              |
| project_id | 36d551c0594b4cc99d1bbff8bf202ec3     |
| spec       | {                                    |
|            |   "version": 1.0,                    |
|            |   "type": "senlin.policy.scaling",   |
|            |   "properties": {                    |
|            |     "adjustment": {                  |
|            |       "min_step": 1,                 |
|            |       "cooldown": 120,               |
|            |       "best_effort": true,           |
|            |       "number": 1,                   |
|            |       "type": "CHANGE_IN_CAPACITY"   |
|            |     },                               |
|            |     "event": "CLUSTER_SCALE_IN"      |
|            |   }                                  |
|            | }                                    |
| type       | senlin.policy.scaling-1.0            |
| updated_at | None                                 |
| user_id    | 9563fa29642a4efdb1033bf8aab07daa     |
+------------+--------------------------------------+

The next step to enforce this policy on your cluster is to attach the policy to it, as shown below:

$ openstack cluster policy attach --policy policy1 mycluster
Request accepted by action: 89626141-0999-4e76-9795-a86c4cfd531f

$ openstack cluster policy binding list mycluster
+-----------+-------------+---------------------------+------------+
| policy_id | policy_name | policy_type               | is_enabled |
+-----------+-------------+---------------------------+------------+
| 3ca962c5  | policy1     | senlin.policy.scaling-1.0 | True       |
+-----------+-------------+---------------------------+------------+

In future, when your cluster is about to be scaled in (no matter the request comes from a user or a software or via a receiver), the scaling policy attached will help determine 1) how many nodes to be removed, 2) whether the scaling operation should be done on a best effort basis, 3) for how long the cluster will not respond to further scaling requests, etc.

For more information on using scaling policy, you can refer to Scaling Policy.

Creative Commons Attribution 3.0 License

Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.