INFRASTRUCTURE LEARNINGINFRASTRUCTURE LEARNING
インフラ大全インフラ大全IT INFRASTRUCTURE / LEARNING
Infrastructure learning guide
Find a lesson or lab/

Esc to close · Search terms stay on your device

LEARNING日本語

Topic lessonTopic: Monitoring and recovering event processing

Read CloudWatch metrics with matching periods and statistics

Compare graphs only after checking the metric identity, aggregation period, and statistic. An image-processing example shows how the same measurements can yield different values.

What you will learn

Identify a time series, distinguish Period from Statistic, and choose a denominator when comparing intervals with different request volumes.

Prerequisite

A metric records observations over time. This lesson uses fictional custom-metric data.

Select the same time series and aggregation before comparing

A CloudWatch metric is identified by its namespace, name, and dimensions. Period sets the time interval used for aggregation. Statistic determines how the observations in that interval are summarized.

Compare measurements under the same conditions
Compare measurements under the same conditions1: Observation — Choose the metric identity — Time series. 2: Time series — Set the period and statistic — Aggregation. 3: Aggregation — Check the units before comparing — Comparison123ObservationTime seriesAggregationComparisonCompare measurements under the same conditions1: Observation — Choose the metric identity — Time series. 2: Time series — Set the period and statistic — Aggregation. 3: Aggregation — Check the units before comparing — Comparison123ObservationTime seriesAggregationComparison
  1. Observation → Time seriesChoose the metric identity
  2. Time series → AggregationSet the period and statistic
  3. Aggregation → ComparisonCheck the units before comparing
Observation
Investigate failed image conversions
Time series
Choose the environment and job
Aggregation
Use the same one-minute Sum
Comparison
Check units and denominator

Compare the average and maximum of five observations

Five individual measurements in a single minute
ObservationStatisticWhat it shows
200, 200, 200, 200, and 1,200 msAverage = 400 msArithmetic mean of five observations.
The same five durationsMaximum = 1,200 msSlowest observed invocation in the interval.
The application publishes a value of 1 only on a failureSum = total of the reported failure valuesInterpret an empty interval using the publication conditions.

The 400 ms average and 1,200 ms maximum summarize the same requests. Use the maximum to find evidence of a slow request, then check timestamps and request IDs in the logs to identify it.

Check whether different aggregations are comparable

Align the units. Compare counts over equal intervals, or convert to a rate if that answers the question. Raw counts from one-minute and five-minute periods can differ even when traffic is unchanged.

Weight aggregate means by sample count. If interval A had 10 requests averaging 100 ms and interval B had 90 averaging 300 ms, the combined mean is (10×100 + 90×300) ÷ 100 = 280 ms. Each individual request contributes equally.

Spot mistakes in metric comparisons

Choose an option to see the result and explanation.

Two metrics share a name but use different environment dimensions. What do you check first?
Read the answer and reason

Answer: The environment and full metric identity · Identify which environment each time series describes.

Which statistic shows the slowest of five measured invocations?
Read the answer and reason

Answer: Maximum · Maximum is the largest observation in the period.

How do you combine the averages of intervals with 10 and 90 requests?
Read the answer and reason

Answer: Weight the means by 10 and 90 · Weight each interval’s mean by its number of requests.

Check, then explain in your own words

  1. Record the namespace, metric name, and dimensions.
  2. Choose the same period and statistic.
  3. Include the total request count when comparing failure counts.
Check your understanding: The Sum of failures doubled since yesterday. What else do you compare?

Compare the observed workload, interval length, and total number of requests.

If these differ, the count alone cannot explain the change in failure rate.

Sources and conditions checked

The official references used in this lesson are collected here with their scope, conditions and corresponding sections. Independent expert review and live AWS comparison have not been performed.