Optimize underprovisioned Cloud SQL instances

The underprovisioned instance recommender helps you detect instances that have high CPU and/or memory utilization. It then provides recommendations for how to optimize the instance. This page describes how this recommender works and how to use it.

How it works

When there is high CPU and/or memory utilization detected, you see a recommendation to increase the size of your instance to reduce CPU and/or memory utilization at peak.

Pricing

The underprovisioned instance recommender is in the Standard Recommender pricing tier.

Before you begin

Before you can view recommendations and insights, do the following:

  • Ensure that you have the required roles so that you have permission to view and work with insights and recommendations.
    Tasks Roles
    View recommendations One of these roles: recommender.cloudsqlViewer or cloudsql.viewer.
    Apply recommendations One of these roles: recommender.cloudsqlAdmin, cloudsql.editor, or cloudsql.admin.
    For more information about roles, see understanding roles and granting IAM permissions.

  • Enable the Recommender API.

    Enable the API

List underprovisioned instance recommendations

Console

To list recommendations about underprovisioned instances, follow these steps:

  1. Go to the Cloud SQL Instances page.

    Go to Cloud SQL Instances

  2. Click View all on the resize Cloud SQL instance recommender banner.

Alternatively, follow these steps:

  1. Go to the Recommendation Hub.

    Go to the Recommendation Hub

    For more information, see Getting started with Recommendation Hub.

  2. In the Resize Cloud SQL instances to improve performance card, click View all.

  3. Click the instances with the Resize Cloud SQL instance recommendation.

gcloud

To list recommendations about underprovisioned instances, run the gcloud recommender recommendations list command as follows:

gcloud recommender recommendations list \
--project=PROJECT_ID \
--location=LOCATION \
--recommender=google.cloudsql.instance.UnderprovisionedRecommender

Replace the following:

  • PROJECT_ID: Your project ID.
  • LOCATION: A region, such as us-central1.

API

To list recommendations about underprovisioned instances by using the Recommendations API, call the recommendations.list method as follows:

GET https://recommender.googleapis.com/v1beta1/projects/PROJECT-ID/locations/LOCATION/recommenders/google.cloudsql.instance.UnderprovisionedRecommender/recommendations

Replace the following:

  • PROJECT_ID: Your project ID.
  • LOCATION: A region, such as us-central1.

If the recommender detects underprovisioned instances, it lists them in a table. Each row shows the instance name, a brief recommendation, the current machine type, the recommended machine type, the location, and the last refresh date.

View insights and detailed recommendations

Console

To view insights and detailed recommendations about an underprovisioned instance by using the Google Cloud console, click the recommendation button in the list of instances.

gcloud

To view insights and detailed recommendations about underprovisioned instances by using gcloud, run the gcloud recommender insights list command as follows:

gcloud recommender insights list \
--project=PROJECT_ID \
--location=LOCATION \
--insight-type=google.cloudsql.instance.INSIGHT_TYPE

Replace the following:

  • PROJECT_ID: Your project ID.
  • LOCATION: A region, such as us-central1.
  • INSIGHT_TYPE: The ID of the insight type is UnderprovisionedCpuUsageInsight, UnderprovisionedMemoryUsageInsight, or OomProbabilityInsight, which provide insights about CPU usage, memory, and OOM probability, respectively.

API

To view insights and detailed recommendations about underprovisioned instances by using the Recommendations API, call the insights.list method as follows:

GET https://recommender.googleapis.com/v1beta1/projects/PROJECT-ID/locations/LOCATION/insightTypes/google.cloudsql.instance.INSIGHT_TYPE/insights

Replace the following:

  • PROJECT_ID: Your project ID.
  • LOCATION: A region, such as us-central1.
  • INSIGHT_TYPE: The ID of the insight type as `UnderprovisionedCpuUsageInsight, UnderprovisionedMemoryUsageInsight, or OomProbabilityInsight, which provide insights about CPU usage, memory, and OOM probability, respectively.

The following table lists the insights and recommendations that the Cloud SQL underprovisioned instance recommender might generate to help you avoid bottlenecks from high CPU and memory usage and minimize the likelihood of out-of-memory events. The subtypes are visible in the gcloud and API results.

Insights Recommendations
Based on the current CPU utilization trends, the instance is flagged as having high CPU usage.
Subtype: HIGH_CPU_UTILIZATION 		Increase CPU size or reduce CPU utilization.
Subtype: CHANGE_INSTANCE_SIZE
Based on the current memory utilization trends, the instance is flagged as having high memory usage.
Subtype: HIGH_MEMORY_UTILIZATION. 		Increase memory size or reduce memory utilization.
Subtype: CHANGE_INSTANCE_SIZE
Based on the current usage trends, the instance is flagged as being at the risk of incurring future OOM events.
Subtype: HIGH_OOM_PROBABILITY. 		Increase memory size or reduce memory utilization.
Subtype: CHANGE_INSTANCE_SIZE

Apply recommendations

Evaluate the recommendations carefully and do any of the following:

  • To examine the recommendation, click View instance.

    A banner appears at the top of the page providing a recommendation.

  • To dismiss the recommendation so that it's de-emphasized and appears dimmed, click Dismiss.

  • To close the panel without applying or dismissing the recommendation, click Cancel.

Optimize the performance of your instance

To optimize your instance's performance, do the following:

  • Edit the instance, if required, to increase the vCPUs or memory. Alternatively, switch to a machine type with more vCPUs and more memory. You don't need to rightsize the instance exactly as recommended. Use your judgement and resize based on how you intend to provision the instance. Keep in mind that the bigger the size, the greater the operational latency.

You can also export these recommendations to BigQuery and then examine them using Looker Studio or Looker.

What's next