June 2023

Serverless computing is a cloud service model that allows developers to run code without provisioning or managing servers. Serverless applications are composed of functions that are triggered by events and run on demand. Serverless computing offers many benefits, such as scalability, performance, cost-efficiency, and agility.

However, serverless computing also introduces new challenges for observability and monitoring. Observability is the ability to measure and understand the internal state of a system based on the external outputs. Monitoring is the process of collecting, analyzing, and alerting on the metrics and logs that indicate the health and performance of a system.

Observability and monitoring are essential for serverless applications because they help developers troubleshoot issues, optimize performance, ensure reliability, and improve user experience. However, serverless applications are more complex and dynamic than traditional applications, making them harder to observe and monitor.

Some of the challenges of serverless observability and monitoring are:

• Lack of visibility: Serverless functions are ephemeral and stateless, meaning they are created and destroyed on demand, and do not store any data or context. This makes it difficult to track the execution flow and dependencies of serverless functions across multiple services and platforms.
• High cardinality: Serverless functions can have many variations based on input parameters, environment variables, configuration settings, and runtime versions. This creates a high cardinality of metrics and logs that need to be collected and analyzed.
• Distributed tracing: Serverless functions can be triggered by various sources, such as HTTP requests, messages, events, timers, or other functions. This creates a distributed tracing problem, where developers need to correlate the traces of serverless functions across different sources and services.
• Cold starts: Serverless functions can experience cold starts, which are delays in the execution time caused by the initialization of the function code and dependencies. Cold starts can affect the performance and availability of serverless applications, especially for latency-sensitive scenarios.
• Cost optimization: Serverless functions are billed based on the number of invocations and the execution time. Therefore, developers need to monitor the usage and cost of serverless functions to optimize their resource allocation and avoid overspending.

AWS and Azure are two of the leading cloud providers that offer serverless computing services. AWS Lambda is the serverless platform of AWS, while Azure Functions is the serverless platform of Azure. Both platforms provide observability and monitoring features for serverless applications, but they also have some differences and limitations.

In this article, we will compare AWS Lambda and Azure Functions in terms of their observability and monitoring capabilities, including their native features and third-party software reviews and recommendations.

Native Features

Both AWS Lambda and Azure Functions provide native features for observability and monitoring serverless applications. These features include:

• Metrics: Both platforms collect and display metrics such as invocations, errors, duration, memory usage, concurrency, and throughput for serverless functions. These metrics can be viewed on dashboards or queried using APIs or CLI tools. Metrics can also be used to create alarms or alerts based on predefined thresholds or anomalies.
• Logs: Both platforms capture and store logs for serverless functions. These logs include information such as start and end time, request ID, status code, error messages, custom print statements, etc. Logs can be viewed on consoles or queried using APIs or CLI tools. Logs can also be streamed or exported to external services for further analysis or retention.
• Tracing: Both platforms support distributed tracing for serverless functions. Distributed tracing allows developers to track the execution flow and latency
  of serverless functions across different sources and services. Tracing can help identify bottlenecks errors, failures or performance issues in serverless applications.

Both platforms use open standards such as OpenTelemetry or W3C Trace Context for tracing. However, there are also some differences between AWS Lambda and Azure Functions in terms of their native features for observability and monitoring.

Some of these differences are:

• Metrics granularity: AWS Lambda provides metrics at a 1-minute granularity by default while Azure Functions provides metrics at a 5-minute granularity by default
  However, both platforms allow users to change the granularity to a lower or higher level depending on their needs
• Metrics aggregation: AWS Lambda aggregates metrics by function name function version or alias (if specified), region (if specified) or globally (across all regions). Azure Functions aggregates metrics by the function name (or function app name), region (if specified) or globally (across all regions).
• Logs format: AWS Lambda logs are formatted as plain text with a timestamp prefix. Azure Functions logs are formatted as JSON objects with various fields such as timestamp, level, message, category, functionName, invocationId, etc.
• Logs retention: AWS Lambda logs are stored in Amazon CloudWatch Logs service for 90 days by default (or longer if specified by users). Azure Functions logs are stored in Azure Monitor service for 30 days by default (or longer if specified by users)
• Tracing integration: AWS Lambda integrates with AWS X-Ray service for tracing. AWS X-Ray provides a web console and an API for viewing traces and analyzing the performance of serverless applications on AWS. Azure Functions integrates with Azure Application Insights service for tracing. Azure Application Insights provides a web console and an API for viewing traces and analyzing the performance of serverless applications on Azure.