Skip to content

OpenVidu Elastic administration: DigitalOcean#

DigitalOcean

The deployment of OpenVidu Elastic on DigitalOcean is automated using Terraform CLI. The Master Node is a single Droplet, while Media Nodes are plain Droplets created and removed by a DigitalOcean Function that acts as the autoscaler.

Internally, the DigitalOcean Elastic deployment mirrors the On Premises Elastic deployment, allowing you to follow the same administration and configuration guidelines of the On Premises Elastic documentation. However, there are specific considerations unique to the DigitalOcean environment that are worth keeping in mind:

How Media Nodes are managed

  • Terraform deploys the autoscaler as a DigitalOcean Function (namespace <STACK_NAME>-autoscaler, function autoscaler/check) and a scheduled trigger named <STACK_NAME>-autoscale-cron that invokes it every four minutes.
  • Media Nodes created by the autoscaler are Droplets named <STACK_NAME>-media-<TIMESTAMP>-<RANDOM> and tagged <STACK_NAME>-media-node-tag. They are not managed by Terraform, and they are not part of any Droplet Autoscale Pool.
  • minNumberOfMediaNodes, maxNumberOfMediaNodes, initialNumberOfMediaNodes, scaleTargetCPU and mediaNodeInstanceType are baked into the function when it is deployed. They are changed by editing terraform.tfvars and running terraform apply, which redeploys the function: there is no autoscaling setting to edit in the DigitalOcean console.
  • If fixedNumberOfMediaNodes is greater than 0, no autoscaler function is deployed and Media Nodes are Terraform-managed Droplets named <STACK_NAME>-media-node-<N>.

Cluster shutdown and startup#

The Master Node is a Droplet that you power off and on, while Media Nodes are ephemeral: they are drained and re-created instead of being powered off. The process for starting and stopping these components differs:

To shut down the cluster, stop the autoscaler, then remove the Media Nodes, and finally power off the Master Node.

  1. From the directory containing your Terraform state, remove the autoscaler so that no new Media Nodes are created:

    terraform destroy -target='null_resource.deploy_autoscaler_function'

    This deletes only the scheduled trigger, the function and its namespace. The rest of the deployment is untouched.

    Info

    In a deployment with a fixed number of Media Nodes (fixedNumberOfMediaNodes greater than 0) there is no autoscaler function. Skip this step and power off the <STACK_NAME>-media-node-<N> Droplets the same way as the Master Node in step 4.

  2. Drain every Droplet tagged <STACK_NAME>-media-node-tag as described in Removing a Media Node gracefully. Each Media Node waits for its active Rooms to end and then deletes itself.

  3. After confirming that no Media Node is left, navigate to the DigitalOcean Droplet Web .
  4. Select the droplet called <STACK_NAME>-master-node. Click on it to go to the Master Node instance, there click on "Power" and then "Turn off" the droplet. Turn Off Master Node

To start the cluster, start the Master Node first and then let the autoscaler re-create the Media Nodes.

  1. Navigate to the DigitalOcean Droplet Web .

  2. Select the droplet named <STACK_NAME>-master-node, then go to "Power" and then "Turn on" the droplet.

    Turn on Master Node

  3. Wait until the instance is running.

  4. Redeploy the autoscaler from the directory containing your Terraform state:

    terraform apply

    Terraform re-creates the function and its scheduled trigger and invokes the function once immediately, so the cluster goes back to max(minNumberOfMediaNodes, initialNumberOfMediaNodes) Media Nodes without waiting for the first scheduled run.

    Info

    With a fixed number of Media Nodes, power on the <STACK_NAME>-media-node-<N> Droplets instead, following steps 2 and 3.

Removing a Media Node gracefully#

Media Nodes are removed through the <STACK_NAME>-draining tag. Every Media Node checks its own tags every two minutes and, as soon as the draining tag is present, it waits for its active Rooms to conclude and then deletes its own Droplet. This is exactly what the autoscaler does on a scale-in decision, and you can trigger it manually on any Media Node:

  1. Navigate to the DigitalOcean Droplet Web and click on the Media Node you want to remove. Media Nodes created by the autoscaler are named <STACK_NAME>-media-<TIMESTAMP>-<RANDOM>.
  2. Open the "Tags" section of the droplet and add the tag <STACK_NAME>-draining.
  3. In the same section, remove the tag <STACK_NAME>-media-node-tag so the autoscaler stops counting this droplet as an active Media Node.
  4. Within two minutes the Media Node starts its graceful shutdown. The droplet disappears once its active Rooms have finished.

Warning

Do not power off a Media Node to remove it. The autoscaler counts Droplets by tag, so a powered-off Media Node still counts towards minNumberOfMediaNodes and maxNumberOfMediaNodes while reporting no CPU metrics, and it never runs its graceful shutdown script.

If you do not need the graceful behavior, destroy the droplet directly ("Destroy" in the console). Active Rooms on it are interrupted, and the autoscaler brings the number of Media Nodes back to minNumberOfMediaNodes on its next run.

Info

With a fixed number of Media Nodes (fixedNumberOfMediaNodes greater than 0) the tag watcher is not installed, so tagging has no effect. Instead, SSH into the Media Node and run /usr/local/bin/graceful_shutdown.sh: it waits for the active Rooms to end and then deletes the droplet. Run terraform apply afterwards to re-create it.

Change the instance size#

It is possible to change the instance size of both the Master Node and the Media Nodes. The Master Node is resized from the DigitalOcean console, while the Media Node size is a Terraform variable because Media Nodes are created by the autoscaler. The following section details the procedures:

Warning

This procedure requires downtime, as it involves stopping the Master Node.

  1. Shutdown the cluster.

    Info

    You can stop only the Master Node droplet to change its droplet size, but it is recommended to stop the whole cluster to avoid any issues.

  2. Go to the DigitalOcean Droplet Web and locate the resource with the name <STACK_NAME>-master-node and click on it.

  3. Click on "Upsize" and select the Droplet size you desire and click on "Resize"

    Change droplet size master

  4. Start the cluster.

  1. Go to the terraform.tfvars file and set mediaNodeInstanceType to the Droplet size you want.

  2. Open a terminal and run the following command:

    terraform apply

  3. Confirm the change that Terraform proposes. The Media Node size is baked into the autoscaler function, so Terraform redeploys it with the new value.

  4. Running Media Nodes keep their current size: only Media Nodes created after the apply use the new one. To roll out the change immediately, drain the running Media Nodes as described in Removing a Media Node gracefully; the autoscaler replaces them with Droplets of the new size on its next run.

Info

With a fixed number of Media Nodes, terraform apply applies the new size to the <STACK_NAME>-media-node-<N> Droplets that Terraform manages, which interrupts the Rooms running on them. To avoid that, first SSH into each Media Node and run /usr/local/bin/graceful_shutdown.sh (it waits for the active Rooms to end and then deletes the droplet), and then run terraform apply to re-create them with the new size.

Media Nodes Autoscaling Configuration#

You can modify the autoscaling configuration of the Media Nodes via terraform.tfvars file and terraform apply:

  1. Go to the terraform.tfvars file and change the config related to autoscaling, such as:

    • scaleTargetCPU
    • minNumberOfMediaNodes
    • maxNumberOfMediaNodes
    • initialNumberOfMediaNodes

  2. Open a terminal and write the following command once you've changed the value/s. 

    terraform apply

  3. Confirm the change that Terraform proposes (it will redeploy the autoscaler function with the new values), and the changes will take effect. The function is invoked once right after being redeployed, so the new limits apply without waiting for the next scheduled run. Running Media Nodes are not affected. Terraform output autoscale change

How the autoscaler uses these values

  • On a run where no Media Node exists, the cluster is brought straight to max(minNumberOfMediaNodes, initialNumberOfMediaNodes) Media Nodes. The same logic re-creates missing Media Nodes: whenever the number of Media Nodes drops below minNumberOfMediaNodes, the autoscaler creates the ones needed.
  • On every other run, the average CPU usage of the last four minutes across all Media Nodes is compared against scaleTargetCPU: above it, one Media Node is added (never exceeding maxNumberOfMediaNodes); below it, one Media Node is drained (never going below minNumberOfMediaNodes).
  • Setting minNumberOfMediaNodes equal to maxNumberOfMediaNodes keeps an exact number of Media Nodes while still using the autoscaler, so failed or drained nodes are replaced automatically.
  • If initialNumberOfMediaNodes is greater than maxNumberOfMediaNodes, the extra Media Nodes are drained again on the following runs.

Tip

Every autoscaler run returns its full log in the activation result. You can review its decisions in the DigitalOcean console under "Functions", in the <STACK_NAME>-autoscaler namespace.

Change Fixed Number of Media Nodes#

You can change the fixed number of Media Nodes in case you put a number of fixed Media Nodes by following these steps:

  1. Go to the terraform.tfvars file and set fixedNumberOfMediaNodes to the number of Media Nodes you want.
  2. Open a terminal and write the following command once you've changed the value. 
    terraform apply
  3. Confirm the change that Terraform proposes. Terraform creates or destroys <STACK_NAME>-media-node-<N> Droplets until their number matches the new value.

Warning

Lowering the value destroys Droplets without draining them, so the Rooms running on them are interrupted. To avoid this, SSH into the highest-numbered Media Nodes (they are the ones Terraform removes first) and run the /usr/local/bin/graceful_shutdown.sh script, which waits for the active Rooms to end and then deletes the droplet. Then lower fixedNumberOfMediaNodes and run terraform apply.

Activate Scale In when Fixed Number of Media Nodes#

You can activate or deactivate the scale in when you decide you need autoscale option activated or not.

  1. Go to the terraform.tfvars file and change the config related to autoscaling, such as:

    • fixedNumberOfMediaNodes need to be set to 0.
    • scaleTargetCPU if you don't want the default.
    • minNumberOfMediaNodes if you don't want the default.
    • maxNumberOfMediaNodes if you don't want the default.
    • initialNumberOfMediaNodes if you don't want the default.

  2. Open a terminal and write the following command once you've changed the value/s. 

    terraform apply

  3. Confirm the change that Terraform proposes (it will destroy the fixed media nodes and deploy the autoscaler function), and the changes will take effect. The autoscaler is invoked right after being deployed and creates max(minNumberOfMediaNodes, initialNumberOfMediaNodes) Media Nodes. Terraform output autoscale change

Warning

The fixed Media Node Droplets are destroyed without being drained, so their active Rooms are interrupted. Drain them first with the /usr/local/bin/graceful_shutdown.sh script if you need a graceful transition.

  1. Go to the terraform.tfvars file and change the config related to autoscaling, such as:

    • fixedNumberOfMediaNodes need to be set to the value of your desire.

  2. Open a terminal and write the following command once you've changed the value/s. 

    terraform apply

  3. Confirm the change that Terraform proposes. Terraform removes the autoscaler function (trigger, function and namespace), deletes every Droplet tagged <STACK_NAME>-media-node-tag or <STACK_NAME>-draining, and creates the <STACK_NAME>-media-node-<N> Droplets. When the apply finishes, check in the Droplets console that the expected number of Media Nodes is running. Terraform output autoscale change

Warning

The Media Nodes are deleted without being drained, so their active Rooms are interrupted. Drain them first as described in Removing a Media Node gracefully if you need a graceful transition.

Administration and configuration#

Regarding the administration of your deployment, you can follow the instructions in section On Premises Elastic Administration.

Regarding the configuration of your deployment, you can follow the instructions in section Changing Configuration. Additionally, the How to Guides offer multiple resources to assist with specific configuration changes.

In addition to these, a DigitalOcean deployment provides the capability to manage global configurations by downloading secrets.env file of the bucket and changing it, then upload it again. Here are the detailed steps:

  1. Navigate to the DigitalOcean Spaces Object Storage and click on the bucket that you are using for the deployment.

  2. Download the secrets.env file that is in the bucket.

    Secrets.env download

  3. Open it and edit the values of the credential of your choice.

  4. Upload the edited secrets.env to the bucket, select private file and replace it. Secrets.env upload Secrets.env replace
  5. Restart the Master Node by shutting it down and then starting it again. Changes will be applied automatically in all the nodes of your OpenVidu Elastic deployment.

Backup and Restore#

Review the Backup and restore OpenVidu deployments guide for recommended backup workflows.