Azure Functions with Java — Serverless Spring

Intermediate·2 min read·

MicrosoftStartupsEvent-driven architectures

Fast recall

Quick revision

Core answer

Azure Functions run Java code triggered by events (HTTP, timer, Service Bus, blobs). You annotate methods with @FunctionName and trigger annotations. Unlike Spring Boot, there's no embedded server — the Functions host manages the lifecycle and scales to zero.

Interview Answer2 min

Azure Functions runs Java code triggered by events — HTTP requests, queue messages, timer schedules, blob uploads — with per-invocation billing and no infrastructure to manage.

Function structure

A Java Azure Function is a class with a method annotated with @FunctionName and one or more binding annotations — @HttpTrigger for HTTP, @ServiceBusQueueTrigger for Service Bus, @TimerTrigger for scheduled execution
Input and output bindings are declared as method parameters or return types: @BlobInput reads a blob when the function fires; @CosmosDBOutput writes results to Cosmos DB without SDK calls
The Azure Functions runtime handles the trigger subscription and passes the event to your method; no Spring ApplicationContext is involved unless you use Spring Cloud Function

Cold starts — the Java challenge

Java functions have cold starts of 1-5 seconds after periods of inactivity — the JVM initializes and loads classes before the first invocation
Premium Plan provides pre-warmed instances that eliminate cold starts at the cost of always-on minimum compute; Consumption Plan is zero-cost when idle but has cold starts
Using Spring Cloud Function with Azure Functions adds ApplicationContext startup time to the cold start; keep Spring configuration minimal for latency-sensitive functions

Triggers and output bindings

@ServiceBusQueueTrigger processes messages from a Service Bus queue; the function completes successfully = message committed; throws exception = message abandoned (retry)
@TimerTrigger(schedule = "0 0 * * * *") runs every hour using a CRON expression — distributed timer that runs on one instance, not all instances
Output binding: return an object and annotate the method with @CosmosDBOutput to write the result to Cosmos DB without a database SDK call

Azure Functions is appropriate for: event-driven processing of Service Bus messages, S3/blob-triggered processing, scheduled batch jobs, lightweight HTTP endpoints with bursty traffic. Avoid for: long-running processes (limit: 10 min on Consumption, 60 min on Premium).

Interview-ready guide

Interview answer

Azure Functions runs Java code triggered by events — HTTP requests, queue messages, timer schedules, blob uploads — with per-invocation billing and no infrastructure to manage.

Function structure

A Java Azure Function is a class with a method annotated with @FunctionName and one or more binding annotations — @HttpTrigger for HTTP, @ServiceBusQueueTrigger for Service Bus, @TimerTrigger for scheduled execution
Input and output bindings are declared as method parameters or return types: @BlobInput reads a blob when the function fires; @CosmosDBOutput writes results to Cosmos DB without SDK calls
The Azure Functions runtime handles the trigger subscription and passes the event to your method; no Spring ApplicationContext is involved unless you use Spring Cloud Function

Cold starts — the Java challenge

Java functions have cold starts of 1-5 seconds after periods of inactivity — the JVM initializes and loads classes before the first invocation
Premium Plan provides pre-warmed instances that eliminate cold starts at the cost of always-on minimum compute; Consumption Plan is zero-cost when idle but has cold starts
Using Spring Cloud Function with Azure Functions adds ApplicationContext startup time to the cold start; keep Spring configuration minimal for latency-sensitive functions

Triggers and output bindings

@ServiceBusQueueTrigger processes messages from a Service Bus queue; the function completes successfully = message committed; throws exception = message abandoned (retry)
@TimerTrigger(schedule = "0 0 * * * *") runs every hour using a CRON expression — distributed timer that runs on one instance, not all instances
Output binding: return an object and annotate the method with @CosmosDBOutput to write the result to Cosmos DB without a database SDK call

03 · Deep dive