如何创建涉及多指标的AWS CloudWatch告警：API网关与Lambda场景

Let’s break down exactly how to configure these CloudWatch Alarms using Metric Math—this is a common setup I’ve implemented for production APIs, so I’ll walk you through each step clearly.

1. API Gateway Alarm: 5XX Error Rate ≥10% (5-Minute Window) #

To trigger an alert when 10% or more of requests result in 5XX errors over 5 minutes, you’ll need to calculate the error rate using two base metrics combined with a math expression:

• First, add your API Gateway’s 5XXError metric, set the statistic to Sum, and choose a 5-minute period.
• Next, add the Count metric (total requests) for the same API/stage, also set to Sum with a 5-minute period.
• Create a Metric Math expression like this:

  ERROR_RATE = m1 / m2
  

  Where m1 is the reference ID for your 5XXError Sum metric, and m2 is the ID for your total Count Sum metric.
• Set your alarm condition to trigger when ERROR_RATE is greater than or equal to 0.1 (10%).
• Finish by configuring your notification target (e.g., an SNS topic for email/Slack alerts) and other standard alarm settings.

2. API Gateway Alarm: 6% of Requests Exceed 4-Second Latency #

For this alert, we need to track both the percentage of slow requests and confirm the maximum latency threshold is hit. Here’s how:

• Add a Latency metric for your API/stage, set the statistic to Count, and add a filter: Latency > 4000 (since latency is tracked in milliseconds). Use a 5-minute period. This counts how many requests took longer than 4 seconds.
• Add the total Count metric again (same as step 1) to get overall request volume.
• Create a Metric Math expression for the slow request rate:

  SLOW_REQUEST_RATE = m1 / m2
  

  Where m1 is the count of requests with latency >4000, and m2 is total requests.
• Optionally, add a second Latency metric with statistic Maximum to confirm the peak latency exceeds 4000ms.
• Set your alarm condition to trigger when both SLOW_REQUEST_RATE ≥ 0.06 (6%) and MAX_LATENCY > 4000 are true. This ensures you’re only alerted when a meaningful portion of requests are slow, not just a single outlier.

3. Quick Lambda Alarm Configurations (Since You Mentioned It) #

While your main ask is API Gateway, here are a couple of critical Lambda alarms you might want to set up alongside:

• Error Rate Alarm: Calculate Errors Sum / Invocations Sum, trigger when the rate exceeds your threshold (e.g., 5%).
• Throttle Alarm: Track Throttles Sum / Invocations Sum—throttles are a sign you need to adjust concurrency limits.
• Duration Alarm: Monitor the P95 or Maximum of the Duration metric, and alert if it’s approaching your function’s timeout (e.g., 80% of the timeout value).

内容的提问来源于stack exchange，提问作者Darshan Ambhaikar