Topic lessonTopic: Monitoring and recovering event processing
Read CloudWatch alarm states when data is missing
An alarm evaluates a condition on the metric you configure. Follow an image-processing delay example to distinguish observed breaches from an interval that lacked enough data.
Read Period, evaluation periods N, and datapoints-to-alarm M, then choose what to inspect for breaching, non-breaching, and missing observations.
This assumes you know that CloudWatch can aggregate a metric over a period.
Match the alarm state to the metric it evaluated
A metric alarm compares each evaluated period’s statistic with a threshold. In an M-out-of-N alarm, M is the required number of breaching points and N describes the number of periods evaluated.
- Metric → PeriodSelect the same metric
- Period → BreachesDistinguish breaches from missing points
- Breaches → StateReview the reason for the state
- Metric
- Time spent waiting for image processing
- Period
- One-minute values
- Breaches
- Two of the three evaluated points
- State
- Check the reason and underlying data
Evaluate a threshold that needs two of three points
| Observations | Interpretation | Next check |
|---|---|---|
| 45, 80, and 90 seconds | Two breaches | Meets the example threshold condition. |
| 45, 80, and 50 seconds | One breach | With all three points present, this falls short. |
| Every point in the evaluation range is missing | Missing data treatment is missing | INSUFFICIENT_DATA. Check when the metric is published. |
A default sliding evaluation can retrieve extra older points when data is missing. An empty slot among the last three plotted points does not determine the alarm state by itself. Read the evaluation reason and its data range.
Check when the metric is expected to publish
A metric that normally publishes continuously. If expected points stop arriving, check the resource, delivery of telemetry, Region, and dimensions. Choose missing-data behavior with its notification and automation effects in mind.
A metric that publishes only on a particular event. A metric designed to publish only on errors can have no points during normal periods. Confirm the reporting rules and whether a gap is expected.
Choose what each alarm state establishes
Choose an option to see the result and explanation.
Check, then explain in your own words
- Record the threshold, period, M, and N.
- Read when the state changed and its reported reason.
- Verify when the metric is expected to publish.
Check your understanding: The dashboard has a gap while the alarm is OK. What do you verify first?
Check the evaluation range, missing-data setting, and publication conditions.
Alarm state describes its configured evaluation. Use independent service evidence to assess whether the workload is healthy.
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.
- Alarm evaluationChecked:
Scope and conditionsPeriod, M out of N, and evaluation windows.
Used in this lessonMatch the alarm state to the metric it evaluated Evaluate a threshold that needs two of three points
Scope and conditionsMissing data choices and evaluation range.
Used in this lessonMatch the alarm state to the metric it evaluated Evaluate a threshold that needs two of three points Check when the metric is expected to publish