Edit

Share via


Use Azure Advisor recommendations to optimize your Azure Data Explorer cluster

Azure Advisor analyzes Azure Data Explorer cluster configurations and usage telemetry, and offers personalized, actionable recommendations to help you optimize your cluster.

Access Azure Advisor recommendations

There are two ways to access the Azure Advisor recommendations:

View Azure Advisor recommendations for your Azure Data Explorer cluster

  1. In the Azure portal, go to your Azure Data Explorer cluster page.

  2. In the left menu, under Monitoring, select Advisor recommendations. A list of recommendations opens for that cluster.

    Screenshot of Azure Advisor recommendations for an Azure Data Explorer cluster.

View Azure Advisor recommendations for all clusters in your subscription

  1. In the Azure portal, go to the Advisor resource.

  2. In Overview, select one or more subscriptions to get recommendations.

  3. Select Azure Data Explorer Clusters and Azure Data Explorer Databases in the second dropdown.

    Screenshot of Azure Advisor resource page.

Use Azure Advisor recommendations

Azure Advisor offers different recommendation types. Use the relevant type to optimize your cluster.

  1. In Advisor, under Recommendations, select Cost to view cost recommendations.

    Screenshot of the Azure Advisor interface showing the selection of recommendation type.

  2. Select a recommendation from the list.

    Screenshot of the Azure Advisor interface showing a list of recommendations.

  3. The window shows a list of clusters relevant to the recommendation. Recommendation details vary for each cluster and include the recommended action.

    Screenshot showing a list of clusters with relevant recommendations in Azure Advisor.

Recommendation types

Cost, performance, reliability, and service excellence recommendations are available.

Important

Your actual yearly savings may vary. The yearly savings presented are based on 'pay-as-you-go' prices. These potential savings don't take into account Azure Reserved Virtual Machine Instance (RIs) billing discounts.

Cost recommendations

The Cost recommendations are for clusters that can be changed to reduce cost without compromising performance. Cost recommendations include:

Unused running Azure Data Explorer cluster

A cluster is considered unused and running if it is in the running state and has neither ingested data nor run queries in the past five days. In some cases, clusters may be automatically stopped. In the following cases, the cluster won't automatically stop and a recommendation will be shown:

The recommendation is to stop the cluster to reduce cost while preserving the data. If the data isn't needed, consider deleting the cluster to increase your savings.

Unused stopped Azure Data Explorer cluster

A cluster is considered unused and stopped if it has been stopped for at least 60 days.

The recommendation is to delete the cluster to reduce cost.

Caution

Stopped clusters may still contain data. Before deleting the cluster, verify that the data is no longer needed. Once the cluster is deleted, the data will no longer be accessible.

Change Data Explorer clusters to a more cost effective and better performing SKU

The recommendation Change Data Explorer clusters to a more cost effective and better performing SKU is for a cluster operating under a nonoptimal SKU. This updated SKU should reduce your costs and improve overall performance. We have calculated the required instance count that meets the cache requirements of your cluster, while ensuring that performance won't be negatively impacted.

As part of the recommendation, we recommend enabling Optimized Autoscale if not yet enabled. Optimized Autoscale will perform a more in-depth analysis of the cluster's performance, and if needed, will further scale-in the cluster. This results in more cost reductions. The Optimized Autoscale recommendations include minimum and maximum instance count recommendations. The Max value is set to the recommended SKU instance count. If the cluster has plans to organically grow, it's recommended to manually increase this Max number. If Optimized Autoscale is already configured on your cluster, in some cases the recommendation may suggest increasing the Max instance count.

The SKU recommendation takes into account the current zones definitions of a cluster and if the cluster supports zones will only recommend target SKUs that have a minimum of two zones. Adding more compute availability zones doesn't incur extra costs.

The advisor SKU recommendation is updated every few hours. The recommendation checks for capacity availability of the selected SKU in the region. However, it's important to note that capacity availability is dynamic and changes over time.

Note

The advisor SKU recommendation doesn't support clusters with Virtual Network or managed private endpoint configurations.

Reduce cache for Azure Data Explorer tables

The reduce Azure Data Explorer table cache period for cluster cost optimization recommendation is given for a cluster that can reduce its table's cache policy. This recommendation is based on the query look-back period during the last 30 days. To see where savings are possible, you can view the most relevant five tables per database for potential cache savings. This recommendation is only offered if the cluster can scale-in or scale-down after a cache policy change. Advisor checks if the cluster is "bounded by data," meaning the cluster has low CPU and low ingestion utilization, but because of high data capacity the cluster can't scale-in or scale-down.

Enable Optimized autoscale

The recommendation enable Optimized autoscale is given when enabling Optimized autoscale would have reduced the instance count on a cluster. This recommendation is based on usage patterns, cache utilization, ingestion utilization, and CPU. To make sure you don't exceed your planned budget, add a maximum instance count when you enable Optimized autoscale.

Performance recommendations

The Performance recommendations improve the performance of your Azure Data Explorer clusters. Performance recommendations include the following:

Update cache policy for Azure Data Explorer tables

The review Azure Data Explorer table cache-period policy for better performance recommendation is given for a cluster that requires a different look-back period time filter, or a larger cache policy. This recommendation is based on the query look-back period of the last 30 days. Most queries run in the last 30 days accessed data not in the cache, which can increase the query run-time. You can view the top five tables per database that accessed out-of-cache data, ordered by querying percentage.

You may also get a performance recommendation to reduce the cache policy. This can happen if the cluster is data-bound. A cluster is data-bound if the data to be cached according to the caching policy is larger that the total size of the cluster's cache. Reducing the cache policy for data-bound clusters reduce the number of cache misses and potentially improves performance.

Operational Excellence recommendations

The Operational Excellence or "best practice" recommendations are recommendations whose implementation doesn't improve cost or performance immediately but can benefit the cluster in the future. This includes reducing the table cache policy to match usage patterns.

Reduce table cache policy to match usage patterns

This recommendation focuses on updating the cache policy based on actual usage during the last month to reduce the hot cache for a table. Unlike the previous cost recommendation, this particular recommendation is applicable to clusters where the number of instances is determined by CPU and ingestion load rather than the amount of data stored in the hot cache. In such cases, changing the cache policy alone is insufficient to reduce the number of instances, further optimizations such as changing the SKU, reducing CPU load, and enabling autoscale are recommended to efficiently scale in. This recommendation can be useful for tables where the actual query lookback based on usage patterns is lower than the configured cache policy. However, reducing the cache policy won’t directly lead to cost savings. The number of cluster instances is determined by CPU and ingestion load, irrespective of the amount of data stored in the hot cache. Therefore, removing data from the hot cache won't directly cause the cluster to scale in.

Reliability recommendations

The Reliability recommendations help you ensure and improve the continuity of your business-critical applications.

Reliability recommendations include the following:

Cluster uses subnet without delegation

This recommendation is for a virtual network cluster that uses a subnet without delegation for 'Microsoft.Kusto/clusters'. When you delegate a subnet to a cluster, you allow that service to establish basic network configuration rules for the subnet, which helps the cluster operate its instances in a stable manner.

Cluster uses subnet with invalid IP configuration

The recommendation is given to a virtual network cluster where the subnet is also used by other services. The recommendation is to remove all other services from the subnet and only use it for your cluster.