What CloudWatch metrics are available for Lambda?
Available AWS CloudWatch Metrics for Lambda Functions
Amazon CloudWatch provides a range of built-in metrics for monitoring AWS Lambda performance and behavior. These metrics help you understand how your function is performing and identify potential issues such as errors, timeouts, and throttling.
- 1
Invocations – Number of times the function is invoked.
- 2
Duration – Execution time of the function (in milliseconds).
- 3
Errors – Number of failed invocations due to function errors (excluding throttles and system faults).
- 4
Throttles – Number of invocation requests that were throttled due to concurrency limits.
- 5
IteratorAge – Age of the last record for stream-based invocations (e.g., DynamoDB, Kinesis).
- 6
ConcurrentExecutions – Number of function instances executing at a given time.
- 7
UnreservedConcurrentExecutions – Available concurrency not reserved for other functions.
- 8
ProvisionedConcurrencyUtilization – Usage of provisioned concurrency vs. what was allocated.
- 9
ProvisionedConcurrencyInvocations – Number of invocations handled by provisioned concurrency.
- 1
Diagnose performance bottlenecks and cold starts using Duration and ConcurrentExecutions.
- 2
Track availability and error rates with Errors and Throttles.
- 3
Adjust scaling and concurrency settings based on usage trends.
- 4
Monitor stream event sources using IteratorAge to prevent lag.
Scenario Questions
- 1
You need to set up an alarm that notifies the team when a Lambda function starts failing. Which CloudWatch metric would you monitor and how would you configure the alarm?
- 2
If you notice that a Lambda function is taking longer than expected, which built‑in CloudWatch metric would you look at to verify its execution time?
- 3
When a function is being throttled, which metric tells you that, and what does a non‑zero value indicate?
- 1
Your recent deployment caused a spike in errors for a Lambda that processes S3 events. Walk me through how you would use CloudWatch metrics to pinpoint whether the issue is code errors, throttling, or downstream service latency.
- 2
You have a Lambda that processes a high volume of requests and you see the ConcurrentExecutions metric approaching the account limit. How would you decide between increasing the limit, adding provisioned concurrency, or refactoring the function?
- 3
Explain why the IteratorAge metric might be high for a Lambda triggered by a Kinesis stream, and what steps you would take to investigate.
- 1
Design a monitoring dashboard for a suite of Lambda functions that need to meet an SLA of <200 ms latency and <0.1 % error rate. Which metrics would you surface, how would you aggregate them, and what alert thresholds would you set?
- 2
Your organization wants to reduce CloudWatch costs while still retaining visibility into Lambda performance. Discuss the trade‑offs of using built‑in metrics versus publishing custom metrics, and how you would decide which to enable.
- 3
During a traffic surge, you observe high Throttles and increased Duration for a critical Lambda. How would you architect a solution to automatically mitigate these issues at scale?
- 1
We are migrating a legacy monolith to a microservice architecture using many Lambda functions. How would you establish a unified observability strategy across teams, considering metric naming, retention, cross‑region aggregation, and alert fatigue?
- 2
Explain how you would integrate Lambda CloudWatch metrics into a centralized SRE incident response system that correlates with other services, and what challenges you anticipate around metric granularity and latency.
- 3
If a compliance requirement mandates storing all Lambda invocation logs for 7 years, how would you design the metric and log pipeline to balance retention, cost, and query performance?
Follow-up Questions
- ❓How would you differentiate between a cold start and a genuine performance regression using these metrics?
- ❓What impact does enabling Lambda Insights have on the metric set and cost?
- ❓Can you describe a situation where you’d need to publish a custom metric alongside the built‑in ones?