Autoscale a CrateDB Cloud Cluster

How to scale your CrateDB Cloud cluster using the REST API.

CrateDB’s fully managed solution CrateDB Cloud comes with a REST API, which enables you to automate a lot of tasks. The tutorial demonstrates how to use the REST API to scale your CrateDB Cloud Cluster based on a threshold on the number of shards in a cluster.

Goal: The goal of the demo is to show how easy it is to scale out a CrateDB Cloud Cluster, using a simple Python script.

Introduction

One of the unique features of CrateDB is that it is a multi-model distributed SQL database with a shared-nothing architecture. This makes it the perfect database for those environments where scalability is needed.

CrateDB’s fully managed solution comes with a REST API which enables you to automate a lot of tasks. As CrateDB is world famous for its, highly scalable architecture, one of those tasks could be to scale a cluster out (horizontally) when needed, during a peak, for example, and back when possible to reduce costs.

CrateDB-Scaling-Process|690x389

This small blog shows how you can use the API to scale out and scale back a cluster when needed.

Prerequisites

  • A running CrateDB Cloud Cluster. If you don’t have one yet, please follow this link to get started. Note: To be able to scale a cluster you need to have a dedicated-tier cluster.

  • Admin UI access to the cluster.

  • An active API Key/Secret combination. See API-KEY for more info. Note: These keys are cloud-dependent. So, if you have multiple clusters across different cloud vendors, you need to generate different keys.

  • You will need both the organization ID and the cluster ID to be able to execute the correct API calls.

  • A Python runtime environment.

Manual Scaling

Let’s do things step by step first to give you a “feel” of what is possible.

Open a terminal and execute the following:

export APIKEY='USE YOUR API_KEY'
export APISECRET='USE YOUR API_SECRET'
export auth_data=${APIKEY}:${APISECRET}

Now that the authorization is defined, we can start calling APIs. For example:

curl -s -u  $auth_data \
https://console.cratedb.cloud/api/v2/organizations/ | jq

Output:

[
  {
    "dc": {
      "created": "2023-08-24T13:53:34.691000+00:00",
      "modified": "2023-10-05T09:30:54.109000+00:00"
    },
    "email": "wierd@nonofyourbusiness.org",
    "id": "xxxxx-8055-4c48-b332-d06d583c1999",
    "name": "My Organization",
    "notifications_enabled": true,
    "plan_type": 3,
    "project_count": 1,
    "role_fqn": "org_admin"
  }
]

Or if you want to know the current amount of nodes in your cluster. Here, you need to use the organization ID that was projected in the previous API call. For the next steps to work fluently, we can populate some variables.

orgid=$(curl -s -u $auth_data https://console.cratedb.cloud/api/v2/organizations/ | jq -r '.[0].id')
echo ${orgid}

If you have multiple clusters running you need to filter using that cluster_name to get the clusterid.

# Replace these values with your authentication data and the name of the cluster you want to select
cluster_name="autoscaling-demo"

# Use jq to filter the output based on the cluster name and extract its ID
clusterid=$(curl -s -u $auth_data https://console.cratedb.cloud/api/v2/clusters/ | jq -r '.[] | select(.name == '\"$cluster_name\"') | .id')
echo ${clusterid}

Get the number of nodes

numnodes=$(curl -s -u $auth_data https://console.cratedb.cloud/api/v2/clusters/${clusterid}/ | jq -r '.num_nodes')
echo ${numnodes}

Output

3

So, if you want to scale out, you can call the scale API. Note that the clusterid is needed.

curl -u $auth_data -X PUT "https://console.cratedb.cloud/api/v2/clusters/${clusterid}/scale/" -H "accept: application/json" -H "Content-Type: application/json" -d "{\"product_unit\":2}"

Extra info: product_unit is the number of nodes minus 1, as 1 is considered the minimum number. In case you want to increase a 3 node cluster you need to use 3 as the value for product_unit.

jobid=$(curl -u $auth_data -X PUT "https://console.cratedb.cloud/api/v2/clusters/${clusterid}/scale/" -H "accept: application/json" -H "Content-Type: application/json" -d "{\"product_unit\":$numnodes}" | jq -r '.last_async_operation.id')

The job can be used to query the operations API to get the info on the status.

# check status:
curl -s -u $auth_data https://console.cratedb.cloud/api/v2/clusters/${clusterid}/operations/ | jq --arg jobid "$jobid" '.operations[] | select(.id == $jobid)'

The status should change from IN_PROGRESS to SUCCEEDED:

{
  "dc": {
    "created": "2024-02-23T13:04:22.960000+00:00",
    "modified": "2024-02-23T13:06:52.157000+00:00"
  },
  "entity": "CLUSTER",
  "entity_id": "xxxxx-b44e-446f-adca-a092fd1df0aa",
  "feedback_data": {
    "message": "Successfully scaled cluster to 4 nodes."
  },
  "id": "xxxxx-3d33-4c4a-8c9c-93202931ed1c",
  "non_sensitive_data": {
    "current_num_nodes": 3,
    "current_product_unit": 2,
    "target_num_nodes": 4,
    "target_product_unit": 3
  },
  "status": "SUCCEEDED",
  "type": "SCALE"
}

Automatic Scaling

We created a Python script that can be used as a starting point to enable autoscaling (out and back). In this script, we check the number of shards in a cluster, and when this crosses a threshold, the cluster will be scaled out. When the threshold goes a certain number, the cluster will be scaled back.

Download the autoscaling script from GitHub and run it as indicated in the README.

You can control the autoscale behavior by defining the MAX_NUM_SHARDS and by tweaking the number of shards in the cluster. For example, by creating an extra table with x number of shards to see the autoscale in action. In the script, the MAX_NUM_SHARDS is set to 30, but this can be changed to a number of your liking. It is also good to understand that the script will trigger a scale-out when the number of shards is higher than 80% of the MAX_NUM_SHARDS. Assuming the default number of 30, this will mean that when the number of shards exceeds 24, the autoscale will kick in.

In this example, I created a 3-node CrateDB Cloud cluster and created this table:

CREATE TABLE ta (
    "keyword" TEXT INDEX using fulltext,
    "ts" TIMESTAMP,
    "day" TIMESTAMP GENERATED ALWAYS AS date_trunc('day', ts)
)
CLUSTERED INTO 24 SHARDS;

This will create 8 primary shards per node plus 8 replicas. This can be checked by looking at the number of shards. This can be done for example using the console by running this:

SELECT node ['name'], count(*)
FROM sys.shards
GROUP BY node ['name']
ORDER BY 1
LIMIT 100;

In this example, the amount of shards is 16 per node.

node['name']	    count(*)
data-hot-0	        16
data-hot-1	        16
data-hot-2	        16

Run the Python script, assuming you filled in the API_KEY API_SECRET organization_id and cluster_id.

To trigger the scale-out you can add a table. For example:

CREATE TABLE tb (
    "keyword" TEXT INDEX using fulltext,
    "ts" TIMESTAMP,
    "day" TIMESTAMP GENERATED ALWAYS AS date_trunc('day', ts)
)
CLUSTERED INTO 18 SHARDS;

As you can see in the output of the running Python script this change is triggering a scale-out:

Current avg number of shards: 16.8
Nothing to do!
Current avg number of shards: 28.0
Start scaling out from 3 to 4
Scaling in progress...................
Scaled up successfully!
Current avg number of shards: 21.0
Nothing to do!

The scaling of a cluster takes time depending on the amount of data that needs to be rebalanced. In this example, we don’t really have any data, so that will be pretty quick (+/- 3mins).

When you delete the tb table, you will see that the autoscale script will scale the cluster back to 3 nodes.

Current avg number of shards: 21.0
Nothing to do!
Current avg number of shards: 12.0
Start scaling down from 4 to 3
Scaling in progress...................
Scaled down successfully!
Current avg number of shards: 16.0

Conclusion

With this simple example, I showed how easy it is to scale out a CrateDB cluster either through manual or automated API calls.