Contents of this repo have been described in detail in this blog post.

This tool enables containerized load testing of Azure Data Explorer (ADX) clusters by executing KQL queries from Docker containers. The KQL query is kept outside and is exposed to the tool as an online Python script (configured by QUERY_SCRIPT_URL). The script must contain a function get_query() which returns the KQL query. Having a function allows us to randomize aspects of the query. Here is an example Python script.

For reporting query performance, ADX's built-in management command .show queries can be used. In order to measure E2E query latency, an optional Application Insights instrumentation can also be provided via APPINSIGHTS_INSTRUMENTATIONKEY.

The tool requires Azure AD for Authentication. Please follow this document on how to provision an AAD application and assign it relevant permissions on the ADX cluster.

CLUSTER_QUERY_URL=https://<ADX_CLUSTER>.<REGION>.kusto.windows.net
CLIENT_ID=<AAD_CLIENT_ID>
CLIENT_SECRET=<AAD_SECRET>
TENANT_ID=<AAD_TENANT>
DATABASE_NAME=adx_db
QUERY_SCRIPT_URL=https://.../query.py
TEST_ID=my_stressful_test
QUERY_CONSISTENCY=weakconsistency
APPINSIGHTS_INSTRUMENTATIONKEY=<APPINSIGHTS_INSTRUMENTATIONKEY>
QUERIES_TOTAL=100

• If TEST_ID is not provided, a guid will be generated.
• Default QUERY_CONSISTENCY value is weakconsistency. This document describes Query consistency in detail.
• Application Insights instrumentation will be ignored if APPINSIGHTS_INSTRUMENTATIONKEY is not provided.
• The tool will run indefinitely if QUERIES_TOTAL is not provided.

Create a .env file with above configuration, then run;

docker run -it --rm --env-file .env syedhassaanahmed/azure-kusto-load-test

Generating concurrent load requires a Kubernetes (k8s) cluster. Here are some of the options to create a cluster;

• Use the in-built Kubernetes in Docker Desktop for Windows.
• Install Minikube.
• Create an AKS cluster in Azure.

Once the k8s cluster is up and running, modify the above environment variables in deployment.yaml file and run the following;

kubectl apply -f deployment.yaml

Logs from a successfully running deployment can be viewed by;

kubectl logs -l app=adx-load-test

To stop the load tests;

kubectl delete deployment adx-load-test

For a given test run my_stressful_test;

E2E duration of all completed queries.

customMetrics
| where name == "query_time" and customDimensions.test_id == "my_stressful_test"
| summarize percentiles(value, 5, 50, 95) by bin(timestamp, 1m)
| render timechart

Duration of all completed queries as measured by the ADX query engine.

.show queries 
| where Database == "<DATABASE_NAME>" 
    and State == "Completed"
    and Text endswith "TEST_ID=my_stressful_test"
| extend Duration = Duration / time(1s)
| summarize percentiles(Duration, 5, 50, 95) by bin(StartedOn, 1m)
| render timechart

Number of queries/second issued during a test run.

.show queries 
| where Database == "<DATABASE_NAME>" and Text endswith "TEST_ID=my_stressful_test"
| summarize TotalQueriesSec=count() by bin(StartedOn, 1s)  
| render timechart

Correlation between average query duration and disk misses

.show queries
| where Database == "<DATABASE_NAME>"
    and State == "Completed"
    and Text endswith "TEST_ID=my_stressful_test"
| summarize DiskMisses = avg(toint(CacheStatistics.Disk.Misses)), 
    Duration = avg(Duration / 1s) 
    by bin(StartedOn, 1m)
| render timechart

Find Test IDs for tests executed in last 7 days

.show queries 
| where Database == "<DATABASE_NAME>" and StartedOn > ago(7d)
| parse Text with * "TEST_ID=" TestId
| where isnotempty(TestId)
| distinct TestId