Run Self-Managed Kusto Kafka Connector Serverless in Azure Container Instances

Posted by nielsb on Monday, September 6, 2021

A couple of weeks ago, I posted how I set up Kafka to run serverless in Azure by deploying Confluent Cloud.

If you have followed my blog lately, you have probably seen that I am interested in Azure Data Explorer and that I have a couple of conference talks coming. One being:

  • How to do Real-Time Analytics Using Apache Kafka and Azure Data Explorer: We are looking at how to stream events from Apache Kafka to Azure Data Explorer (ADX) and perform user-facing analytics in near real-time.

The question is how to connect Kafka with ADX? You normally connect Kafka with another system using a Kafka Connect connector, and fortunately a connector exists for connecting Kafka with ADX: the Kafka Connect Kusto Sink Connector.

However, since I am running managed Kafka (Confluent Cloud, remember), I need a managed connector to run it in Confluent Cloud’s Kafka Connect. In the previous paragraph, I mentioned I was fortunate as we had a Kafka connector for ADX. Unfortunately, it is not a managed connector, so I cannot run it in Confluent Cloud - bummer!

So, this post looks at the various options we have if we want to use the Kafka Connect Kusto Sink Connector connecting Confluent Cloud in Azure with Azure Data Explorer. However, if you are not interested in neither Kafka nor ADX, the post may still be of use for you. The reason being it also covers running Docker images in Azure Container Instances (ACI).

NOTE: Even though this post comes out of me wanting to use the Kafka Connect Kusto Sink Connector, the post is NOT about the usage or the configuration of the connector. That is covered in a future post.


Usually, the credits “roll” at the end of the movie post, but I feel I should start with the credits as this post would not have happened if it wasn’t for this blog post:

As I said, Robin Moffat is a Kafka Guru, and if you are into Kafka, then you MUST read his blog.


This section is here if you want to follow along. I am listing what you need if you’re going to deploy and run a container in ACI; not all required components for Kafka and ADX:

  • An Azure account. If you don’t have an Azure subscription, sign up for a free account.
  • Azure CLI. The Azure CLI is a cross-platform command-line tool to connect to Azure and execute administrative commands on Azure resources. You find downloads and install instructions here.
  • Docker Desktop: We will build a Docker image, so we need Docker Desktop.
  • Something to build the image from. The image I build for this post includes the Kusto Sink Connector, which I download from here.

The version (2.0) of the Kusto Sink Connector I downloaded is not the latest you find on the release page, but I could not get the 2.1 version to work.

In addition to the above, I have Confluent Cloud deployed as per my post mentioned in the beginning. As I want to connect Kafka to Azure Data Explorer, I also have Azure Data Explorer installed.


Before we get down into it, let us get an understanding of some of the components in this post:

  • Azure Data Explorer
  • Confluent Cloud
  • Kafka Connect
  • Kafka Connect Kusto Sink Connector

Azure Data Explorer

Azure Data Explorer is a fully-managed big data analytics cloud platform and data-exploration service that ingests structured, semi-structured (like JSON) and unstructured data. The service then stores this data and answers analytic ad hoc queries on it with seconds of latency. It is a full-text indexing and retrieval database, including time series analysis capabilities, machine learning, regular expression evaluation, and text parsing.

It is ideal for analyzing large volumes of diverse data from any data source, such as websites, applications, IoT devices, and more. Essentially it is a distributed database running on a cluster of compute nodes in Azure.

Confluent Cloud

In my post setting up Confluent Cloud I said it is a resilient, scalable streaming data service based on Apache Kafka, delivered as a fully managed service. It is Confluent Platform, running as a managed service in the cloud, and you can run it on Azure, AWS, and Google Cloud.

As it is Confluent Platform, you get so much more than just Kafka. You get built-in stream processing through ksqlDB, schema registry for data integrity, managed Kafka Connect connectors for data sources/sinks, and more.

Kafka Connect

Kafka Connect is a tool for streaming data between Apache Kafka and other systems scalable and reliable. The way you move data between systems and Kafka is using connectors, and there are two flavors of connectors:

  • Source connectors that understand how to interact with the source system send records into Kafka
  • Sink connectors that propagate records from Kafka topics to other systems.

Kafka Connect is a JVM process, and it operates separately from the Kafka Broker. The process loads connectors, and the connectors know how to interact with Kafka and the source/sink systems. The connectors are written in Java and packaged into .jar files.

Kafka Connect Kusto Sink Connector

The Kusto Sink Connector is a Kafka Connect connector. It is - as the name implies - a sink connector, dequeuing events from Kafka topics and ingesting them into Azure Data Explorer. The ingestion is, for now, queued ingestion leveraging the Azure Data Explorer Java SDK, i.e. batch mode.

Since the Kusto connector is not a managed connector, we need to decide where and how to run it.


Robin covered the various options in his post a lot better than I ever could, so I won’t repeat that. Furthermore, seeing that I say in the title of this post Azure Container Instances (ACI), it is probably safe to assume that’s the option we’ll go with. As a picture says more than a thousand words, our solution looks something like so:

Figure 1: Kafka - ADX Architecture

We see in Figure 1 how we have:

  • Kafka in Azure (Confluent Cloud).
  • an Azure Container Instances running the Kusto Sink Connector.
  • data ingested from the connector into Azure Data Explorer.

Oh, and yes - the data being ingested is published to Kafka from the publisher we see in the upper left-hand corner.

In this post, the term Azure Container Instances has been mentioned a couple of times. What is that, then?

Azure Container Instances

ACI gives you an easy way to run containers in the Azure cloud, eliminating the need to manage virtual machines (VMs) or using more complex container orchestration services, like Kubernetes. For me, just testing this out, ACI is “good enough”.

There are a couple of ways you can deploy and run a container in ACI:

  • Utilise the integration between Docker and Azure and execute docker run. You do this mostly in test scenarios.
  • Deploy the container image to ACI and run it in ACI.

Even though I said I am using ACI for testing at the beginning of this section, I will not use docker run but instead do a “proper” deployment to ACI.

Create Kusto Sink Connector Image

We start with creating the Dockerfile for the image we want to deploy to ACI:

Figure 2: Dockerfile

As we need to run the connector in Kafka Connect, we base the image on the cp-server-connect-base image (the FROM statement). This image contains the bare minimum for Kafka Connect.

In the Pre-Reqs section, I mentioned how I downloaded the Kusto connector. I downloaded it as a zip file and unzipped it to the same directory the Dockerfile is in. In line 3, we see how I copy the connector’s .jar file to /usr/share/java in the base image. That path is a “well known” path to load connectors from.

The ...OVERRIDE_POLICY=All statement on line 5 allows this connector to override consumer and producer properties to not impact all connectors running in that worker.

Lines 7 - 10 in the docker file are core config security settings that need to be set at the Kafka connect worker level and need to be “baked” into the Docker image. These settings are related to authentication and authorization against the Confluent Cloud Kafka.

Build the Image

Having created the Dockerfile, we can build the image:

Figure 3: Docker Build

In Figure 3, we see us executing the docker build command and the outcome. When the build has finished, we can check that all looks OK by running docker images:

Figure 4: Kusto Docker Image

Success, we have an image, as we see in Figure 4. Well, at least partial success; we don’t know if it works yet. Let us find out.

Run Locally

To find out if it works, we can run the container image locally using docker-compose. We create a docker-compose.yml file containing the bare minimum for running the connector:

Figure 5: Docker Compose File

We see in Figure 5 the compose file I use to test that my container works. As mentioned before, the file contains the required properties to get this connector up and running. What you see outlined in red are properties naming topics needed to store Kafka offsets, configs and statuses. After we have “spun up” the container, we can check for these topics in our Kafka installation. Let us execute `docker-compose up -d':

Figure 6: Run Docker Compose

In Figure 6, we see that it looks like everything has worked OK and that we have created a connector instance kusto-conn-1. We can confirm that it has worked by checking the topics mentioned above or execute a REST call against the Kafka Connect API to GET the available connectors:

Figure 7: GET Connectors

From what we see in Figure 7 it looks like we are OK. We see, outlined in blue, the GET call on port 8083, and in the result below, we see the connector outlined in red.

Azure Container Instances

When we have confirmed that our image works, it is time to deploy it to ACI. When we run a container in ACI, the container is stored in Azure Container Registry (ACR).

We’ll create a new container registry in a second, but before we do that, let us log in to Azure and set the subscription to use. To log in, we run az login. The command may take a second or two, and a dialog in your browser may ask you for login credentials. When login is done, you will see a JSON blob with information about the subscriptions you have access to. Choose the one you want to use:

# az login
# az login above returns a JSON blob with subscriptions.
# set the subscription you want to use
az account set --subscription 00000000-0000-0000-0000-000000000000 

Code Snippet 1: Set Subscription

After we have logged in, we execute the code in Code Snippet 1 to set the subscription we want to use. And no, the subscription Id in Code Snippet 1 is not mine.

We are almost at the stage where we can create the ACR, but we need one more thing before creating the ACR. That one more thing is a resource group. I will use an existing resource group for this post, so I will not create a new one. If you need to create a resource group, you do:

az group create --name name-of-rg --location azure-location  

Code Snippet 2: Create Resource Group

To see what locations are available for the --location parameter in Code Snippet 2, you can execute: az account list-locations.

Right, we now have a resource group. Let us press on.

Create Azure Container Registry

ACR is a private Docker registry service, similar to Dockerhub. As with Dockerhub, you push container images to your container registry.

To create a container registry, we use az acr create ...:

az acr create --resource-group rg-kafka \
              --name nielsblog1 \
              --sku Basic \
              --admin-enabled true \

Code Snippet 3: Create Container Registry

In Code Snippet 3, we see how we create a container registry. The first two parameters define the resource group to create it in: rg-kafka, and the name of the registry: nielsblog1. You may ask what the last two parameters are:

  • --sku: this parameter defines the service tier: Basic, Standard, or Premium. Read more about service tiers here.
  • --admin-enabled: an admin user account is included when creating a container registry. The account is disabled by default. For testing purposes, you may want to have it enabled, so that is why I have included it in the creation. Read more about the admin account here.

When you execute the code in Code Snippet 3, it will take a little while. When it finishes, you will see a JSON blob with some information about the created registry.

Having created the registry, we can now log in to it: az acr login --name regname. We are almost ready to push our image to the registry, but there are two things we need to do before that.

Login Server

When we push an image to the registry, we need an address to push to; the login server. To retrieve the login server, we do:

az acr show --name registryname --query loginServer

Code Snippet 4: Get Login Server

Most likely, when you execute the code in Code Snippet 4, you get back, but it is good practice to explicitly retrieve the login server.


The second thing we need to do is get the credentials for the admin user we enabled in Code Snippet 3. We use the credentials later when we deploy our container:

Figure 8: ACR Credentials

In Figure 8, we see outlined in yellow:ish the code to retrieve the credentials. Outlined in red, we see the two passwords. These passwords are created when the admin account is enabled, and they can also be re-generated. Finally, outlined in blue is the user name to use for the admin account.

Push Image to ACR

We are getting there. Now, the time has come to push the image we built in Figure 3 to the ACR. We do it in a two-step process:

Tag the Image

Tag the image with the login server string:

Figure 9: Tag the Image

We see in Figure 9 outlined in:

  • Blue: the docker tag statement we use.
  • Yellow: the name of the image we want to tag.
  • Red: the “tagged” new name of the image.

Push the Image

Having tagged the image with the login server, we can push it to ACR:

Figure 10: Docker Push to ACR

When you run the code outlined in blue in Figure 10, you see how the image layers are pushed to the registry.

Looking at Figure 10 everything looks OK, but - look at the statement outlined in red. What is this “repository” thing?

It turns out that when we push an image to the ACR, we push it not directly into the ACR. Instead, a repository is created, and we push it into that repository. A repository is a collection of container images or other artefacts in a registry with the same name but different tags.

That explains why we, when looking for images in the ACR we do something like so:

Figure 11: ACR Repository with Image(s)

In Figure 11, we see how we execute az acr repository list ... when looking for images (outlined in yellow) and how the result comes back as an array (outlined in red).

Deploy the Container

It is now time to deploy and run the image in Azure Container Instances. To create the container, we use the az container create ... command:

az container create --resource-group rg-kafka `
>> --name nielsblog1 `
>> --image `
>> --restart-policy OnFailure `
>> --ip-address Public `
>> --ports 8083 `
>> --registry-login-server `
>> --registry-username nielsblog1 `
>> --registry-password some-super-secret-password

Code Snippet 5: Create Container - I

The code in Code Snippet 5 is part of the code we need to run to create the container. The parameters we see are straightforward. The only thing worth mentioning is the --ip-address Public and --ports 8083. We need to indicate that we need a public IP address and that port 8083 should be open.

In the previous paragraph, I mentioned that the code in Code Snippet 5 is only part of what we need to run. So what else do we need? Remember what we did when we tested the container locally, how we had a .yml file (Figure 5), with properties required to run Kafka Connect? We need the same here!

The question is, how do we supply those properties? The answer is that az container create has an --environment-variables parameter. This parameter is a list of environment variables for the container, where the list contains space-separated values in ‘key=value’ format, something like so:

Figure 12: Create Container - II

In Figure 12, we see the entire command, including the required properties for Kafka.

NOTE: Notice how the --environment-variables do not have an quotes around them. That is because I am running from PowerShell. If you run the command from command prompt you need the environment variables be enclosed in single quotes.

The text outlined in red at the bottom of Figure 12 shows that the screenshot is taken while the command executes. While the command is running, you can execute the following to see the state it is in:

 az container show --resource-group rg-kafka `
 >>--name nielsblog1 --query instanceView.state

Code Snippet 6: View Container State

Eventually, the creation finishes and results in a JSON blog with some information about the container. Most of the information in the blob is available from az container ... commands:

# get log information from the container
az container logs --resource-group rg-kafka --name nielsblog1

# get ip address information
az container show --resource-group  rg-kafka `
>>--name nielsblog1 --query ipAddress

Code Snippet 7: Get Logs & IP Address

The code in Code Snippet 7 retrieves log information from the running container and the public IP address. Looking at the logs, everything looks OK, so let us use the IP address we retrieved in Code Snippet 7 and do what we did in Figure 7 (but now against the container in Azure):

Figure 13: Get Connectors - II

Yay, from what we see in Figure 13 all is OK. We see the Kusto Sink Connector, outlined in red, being part of the returned result from the GET call. And in the GET call outlined in blue, we see we use the Azure IP address (highlighted in yellow). Well, you don’t necessarily know it is the Azure IP, but trust me - it is. Yay, again!


Wow, that was a lot! In this post, we saw how to:

  • Build a docker image for the Kusto Kafka Sink Connector.
  • Test it locally.
  • Create an Azure Container Registry.
  • Push the image to the registry.
  • Deploy the image to, and run it in Azure Container Instances.

Look out for a post covering how to configure and use the Kusto Sink Connector.

~ Finally

If you have comments, questions etc., please comment on this post or ping me.

comments powered by Disqus