Posted On: May 23, 2019
Metric filters in CloudWatch Logs allow you to turn log data into numerical CloudWatch metrics that you can graph or set an alarm on. With metric filters, you can, for example, create CloudWatch metrics that count the number of errors in your logs, or that represent a latency value from your web servers. You can then visualize those metrics in CloudWatch Dashboards or use them to define alarm thresholds in CloudWatch Alarms.
Starting today, CloudWatch Logs is introducing support for percentiles on metric filters. Percentiles are particularly useful when applied to metrics that exhibit large variances. They help users understand the distribution of metric values, and can be critical to understand outliers or unusual metric behaviors. Users can visualize and alarm on p90, p95, p99, p99.9 or any other percentile.
Percentiles are produced automatically. All users have to do is select what percentile they wish to use when visualizing a metric or defining an alarm threshold. As with all other CloudWatch metrics, percentile metrics can be accessed using the CloudWatch console, SDK or API.
For more information about CloudWatch Logs metric filters, please read the documentation.