Skip to content

How to implement a streaming at scale solution in Azure

License

Notifications You must be signed in to change notification settings

chetanmsft/streaming-at-scale

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

page_type languages products statusNotificationTargets description
sample
azurecli
csharp
json
sql
scala
azure
azure-container-instances
azure-cosmos-db
azure-databricks
azure-data-explorer
azure-event-hubs
azure-functions
azure-kubernetes-service
azure-sql-database
azure-stream-analytics
azure-storage
azure-time-series-insights
How to setup an end-to-end solution to implement a streaming at scale scenario using a choice of different Azure technologies.

Streaming at Scale

License

The samples shows how to setup an end-to-end solution to implement a streaming at scale scenario using a choice of different Azure technologies. There are many possible way to implement such solution in Azure, following Kappa or Lambda architectures, a variation of them, or even custom ones. Each architectural solution can also be implemented with different technologies, each one with its own pros and cons.

More info on Streaming architectures can also be found here:

Here's also a list of scenarios where a Streaming solution fits nicely

A good document the describes the Stream Technologies available on Azure is:

Choosing a stream processing technology in Azure

The goal of this repository is to showcase a variety of common architectural solutions and implementations, describe the pros and the cons and provide you with a sample script to deploy the whole solution with 100% automation.

Ingestion Processing Serving

Running the samples

Please note that the scripts have been tested on Ubuntu 18 LTS, so make sure to use that environment to run the scripts. You can run it using Docker, WSL or a VM:

Just do a git clone of the repo and you'll be good to go.

Each sample may have additional requirements: they will be listed in the sample's README.

Streamed Data

Streamed data simulates an IoT device sending the following JSON data:

{
    "eventId": "b81d241f-5187-40b0-ab2a-940faf9757c0",
    "complexData": {
        "moreData0": 57.739726013343247,
        "moreData1": 52.230732688620829,
        "moreData2": 57.497518587807189,
        "moreData3": 81.32211656749469,
        "moreData4": 54.412361539409427,
        "moreData5": 75.36416309399911,
        "moreData6": 71.53407865773488,
        "moreData7": 45.34076957651598,
        "moreData8": 51.3068118685458,
        "moreData9": 44.44672606436184,
        [...]
    },
    "value": 49.02278128887753,
    "deviceId": "contoso-device-id-000154",
    "deviceSequenceNumber": 0,
    "type": "CO2",
    "createdAt": "2019-05-16T17:16:40.000003Z"
}

Duplicate event handling

Event delivery guarantees are a critical aspect of streaming solutions. Azure Event Hubs provides an at-least-once event delivery guarantees. In addition, the upstream components that compose a real-world deployment will typically send events to Event Hubs with at-least-once guarantees (i.e. for reliability purposes, they should be configured to retry if they do not get an acknowledgement of message reception by the Event Hub endpoint, though the message might actually have been ingested). And finally, the stream processing system typically only has at-least-once guarantees when delivering data into the serving layer. Duplicate messages are therefore unavoidable and are better dealt with explicitly.

Depending on the type of application, it might be acceptable to store and serve duplicate messages, or it might desirable to deduplicate messages. The serving layer might even have strong uniqueness guarantees (e.g. unique key in Azure SQL Database). To demonstrate effective message duplicate handling strategies, the various solution templates demonstrate, where possible, effective message duplicate handling strategies for the given combination of stream processing and serving technologies. In most solutions, the event simulator is configured to randomly duplicate a small fraction of the messages (0.1% on average).

Integration tests

End-to-end integration tests are configured to run. You can check the latest closed pulled requests ("View Details") to navigate to the integration test run in Azure DevOps. The integration test suite deploys each solution and runs verification jobs in Azure Databricks that pull the data from the serving layer of the given solution and verifies the solution event processing rate and duplicate handling guarantees.

Available solutions

At present time the available solutions are

Implement a stream processing architecture using:

  • Kafka on Azure Kubernetes Service (AKS) (Ingest / Immutable Log)
  • Azure Databricks (Stream Process)
  • Cosmos DB (Serve)

Implement stream processing architecture using:

  • Event Hubs (Ingest)
  • Event Hubs Capture (Store)
  • Azure Storage (Azure Data Lake Storage Gen2)
  • Azure Databricks (Stream Process)
  • Delta Lake (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Azure Databricks (Stream Process)
  • Azure SQL (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Azure Databricks (Stream Process)
  • Cosmos DB (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log) with Kafka endpoint
  • Azure Databricks (Stream Process)
  • Cosmos DB (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Azure Databricks (Stream Process)
  • Delta Lake (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Azure Functions (Stream Process)
  • Azure SQL (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Azure Functions (Stream Process)
  • Cosmos DB (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Stream Analytics (Stream Process)
  • Cosmos DB (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Stream Analytics (Stream Process)
  • Azure SQL (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Stream Analytics (Stream Process)
  • Event Hubs (Serve)

Implement a stream processing architecture using:

  • HDInsight Kafka (Ingest / Immutable Log)
  • Flink on HDInsight or Azure Kubernetes Service (Stream Process)
  • HDInsight Kafka (Serve)

Implement a stream processing architecture using:

  • Event Hubs Kafka (Ingest / Immutable Log)
  • Flink on HDInsight or Azure Kubernetes Service (Stream Process)
  • Event Hubs Kafka (Serve)

Implement a stream processing architecture using:

  • Event Hubs Kafka (Ingest / Immutable Log)
  • Azure Functions (Stream Process)
  • Cosmos DB (Serve)

Implement a stream processing architecture using:

  • HDInsight Kafka (Ingest / Immutable Log)
  • Azure Databricks (Stream Process)
  • Azure SQL Data Warehouse (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Azure Data Explorer (Stream Process / Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Microsoft Data Accelerator on HDInsight and Service Fabric (Stream Process)
  • Cosmos DB (Serve)

Implement a stream processing architecture using:

  • Event Hubs (Ingest / Immutable Log)
  • Time Series Insights (Stream Process / Serve / Store to Parquet)
  • Azure Storage (Serve for data analytics)

Implement a stream processing architecture using:

  • IoT Hub (Ingest)
  • Azure Functions (Stream Process)
  • Azure SQL (Serve)

Implement a stream processing architecture using:

  • Azure Storage (Azure Data Lake Storage Gen2) (Ingest / Immutable Log)
  • Azure Databricks (Stream Process)
  • Delta Lake (Serve)

Implement a stream processing architecture using:

  • IoT Hub (Ingest)
  • Azure Digital Twins (Model Management / Stream Process / Routing)
  • Time Series Insights (Serve / Store to Parquet)
  • Azure Storage (Serve for data analytics)

Note

Performance and Services change quickly in the cloud, so please keep in mind that all values used in the samples were tested at them moment of writing. If you find any discrepancies with what you observe when running the scripts, please create an issue and report it and/or create a PR to update the documentation and the sample. Thanks!

About

How to implement a streaming at scale solution in Azure

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C# 57.8%
  • Shell 25.1%
  • Scala 6.2%
  • Java 5.8%
  • HCL 3.6%
  • TSQL 1.1%
  • Other 0.4%