Stream and Ingest Data with Azure Event Hubs

This skill covers @azure/event-hubs, the TypeScript SDK for Azure Event Hubs, Azure's high-throughput event-streaming service. An EventHubProducerClient batches and sends events (createBatch(), batch.tryAdd({ body }), sendBatch()), optionally pinned to a specific partitionId or hashed consistently by partitionKey for ordering. An EventHubConsumerClient, scoped to a consumer group like "$Default", subscribes with processEvents/processError handlers; for production use it pairs with a BlobCheckpointStore (backed by Azure Blob Storage) so processing position survives restarts, calling context.updateCheckpoint() after each processed batch.

Use this skill when ingesting or emitting real-time event streams at scale - IoT telemetry, application events, or log/metric pipelines - from a TypeScript service. It is specifically for Event Hubs' publish/subscribe streaming model with partitions and checkpoints; it is not a general message-queue client (no per-message dead-lettering or FIFO queue semantics of, say, Service Bus) and consumers should not checkpoint after a processing failure, since that would silently drop the unprocessed events - the SDK relies on re-delivery from the last checkpoint for at-least-once processing.

Sends take structured events (body, optional properties, contentType, correlationId) grouped into a batch. Consumption yields ReceivedEventData objects per partition, exposing properties, sequenceNumber, enqueuedTimeUtc, and offset, delivered through a PartitionContext that also exposes updateCheckpoint. startPosition controls where a subscription begins reading per partition: "@earliest", "@latest" (new events only), a specific numeric offset, or an enqueuedOn timestamp. getEventHubProperties() returns the hub's partition IDs, and getPartitionProperties(id) returns lastEnqueuedSequenceNumber for lag monitoring against what a consumer has actually processed.

npm install @azure/event-hubs @azure/identity
npm install @azure/eventhubs-checkpointstore-blob @azure/storage-blob

Authenticates via DefaultAzureCredential from @azure/identity. Checkpointing integrates with @azure/storage-blob through BlobCheckpointStore, storing consumer progress per partition per consumer group in a blob container. maxBatchSize/maxWaitTimeInSeconds options on subscribe() control how events are grouped into processEvents calls. The processError handler distinguishes transient MessagingErrors, which the SDK automatically retries, from fatal errors that need explicit handling - a distinction worth checking before deciding whether an error needs to bubble up to alerting or can be logged and left to the SDK's own retry logic.

TypeScript developers building real-time ingestion or streaming pipelines on Azure - IoT telemetry collectors, event-driven microservices, or analytics pipelines needing exactly-once-style processing guarantees. The skill's best practices push toward always checkpointing in production, batching sends, using partition keys to keep related events ordered, running separate consumer groups per downstream pipeline, closing producer/consumer clients when done, and actively monitoring consumer lag against lastEnqueuedSequenceNumber.