서버리스 관측 가능성의 과제

Serverless functions, like AWS Lambda, offer incredible scalability and cost efficiency. However, their unique characteristics introduce distinct challenges for observability compared to traditional long-running applications.

Understanding these challenges is key to building effective monitoring and troubleshooting strategies for your serverless applications.

One of the biggest challenges is the ephemeral nature of serverless functions. They only exist for the duration of an invocation and then disappear.

• No Persistent Host: There's no long-lived server to install monitoring agents on.
• Short-Lived Context: Application state and local logs are gone after execution.
• Data Must Be Externalized: Observability data (logs, metrics, traces) must be immediately pushed to external services.

Serverless applications are often highly distributed and event-driven. A single user request might trigger a chain of multiple functions, queues, and databases.

Tracing the full journey of a request, especially across asynchronous boundaries (like messages in a queue), becomes a complex task. You need to link together disparate pieces of information.

A 'cold start' occurs when a serverless function is invoked after a period of inactivity. The platform needs to initialize the execution environment, which adds latency to the invocation.

• Increased Latency: Cold starts can significantly impact user experience.
• Difficult to Predict: Their occurrence depends on traffic patterns and platform management.
• Requires Specific Monitoring: You need to distinguish cold start durations from regular execution times.

Serverless computing is typically priced per invocation and execution duration. This model makes cost efficiency paramount, and observability plays a crucial role.

By monitoring invocation counts, function durations, and memory usage, you can identify inefficient functions, optimize resource allocation, and prevent unexpected cloud bills.

Logs are the foundation of serverless observability. Most serverless platforms automatically capture stdout/stderr to a managed logging service (e.g., AWS CloudWatch Logs, Azure Monitor Logs).

• Structured Logging: Always output logs in a structured format (like JSON) to make them machine-readable and easy to query.
• Contextual Information: Include request IDs, function names, and other relevant metadata in every log entry.
• Centralization: Forward logs from the platform's native service to a centralized logging system (like ELK Stack or Splunk) for advanced analysis.

Serverless platforms usually provide essential metrics out-of-the-box. These are vital for understanding function health and performance without manual instrumentation.

• Invocations: Total number of times a function was called.
• Errors: Number of invocations that resulted in an error.
• Duration: Time taken for the function to execute (distinguish between average, p99).
• Throttles: When the function execution was limited by concurrency limits.
• Memory Usage: How much memory the function actually consumed compared to its configured limit.

Distributed tracing is critical for understanding complex serverless workflows. It links individual function invocations into a single, end-to-end request journey.

Tools like AWS X-Ray or OpenTelemetry SDKs (covered in a later course) help propagate context and trace IDs across function boundaries, even for asynchronous calls. This allows you to visualize the entire flow and pinpoint performance bottlenecks.

For example, a trace ID might be passed in an event payload or HTTP header:

To master serverless observability, integrate these practices into your development workflow:

• Structured Logging: Always use JSON for your logs.
• Context Propagation: Implement mechanisms to pass trace IDs and other context across all services.
• Granular Metrics: Beyond default metrics, add custom metrics for key business logic.
• Proactive Alerting: Set up alerts for critical metrics like errors, throttles, and high durations.
• Cost Awareness: Regularly review observability data to optimize resource allocation and manage costs.

Which of the following are significant challenges when observing serverless functions?

In this lesson, we explored the unique challenges of observing serverless functions, including their ephemeral nature, distributed architecture, and the impact of cold starts.

We also covered key strategies for effective serverless observability, focusing on structured logging, essential metrics, and the importance of distributed tracing to gain end-to-end visibility in these dynamic environments.