Datadog’s Logging without Limits* lets you dynamically decide what to include or exclude from your indexes for storage and query, at the same time many types of logs are meant to be used for telemetry to track trends, such as KPIs, over long periods of time. Log-based metrics are a cost-efficient way to summarize log data from the entire ingest stream. This means that even if you use exclusion filters to limit what you store for exploration, you can still visualize trends and anomalies over all of your log data at 10s granularity for 15 months.
With log-based metrics, you can generate a count metric of logs that match a query or a distribution metric of a numeric value contained in the logs, such as request duration.
You can also create metrics from an Analytics search by selecting the “Generate new metric” option from the Export menu.
Add a new log-based metric
Input a query to filter the log stream: The query syntax is the same as for the Log Explorer Search. Only logs ingested with a timestamp within the past 20 minutes are considered for aggregation.
Select the field you would like to track: Select * to generate a count of all logs matching your query or enter a log attribute (fo example, @network.bytes_written) to aggregate a numeric value and create its corresponding count, min, max, sum, and avg aggregated metrics. If the log attribute facet is a measure, the value of the metric is the value of the log attribute.
Add dimensions to group by: By default, metrics generated from logs do not have any tags unless explicitly added. Any attribute or tag dimension that exists in your logs (for example, @network.bytes_written, env) can be used to create metric tags. Metric tags names are equal to the originating attribute or tag name, without the @. Log-based metrics are considered custom metrics. Avoid grouping by unbounded or extremely high cardinality attributes like timestamps, user IDs, request IDs, or session IDs to avert impacting your billing. See the Log Security page for more information on using this feature for HIPAA-enabled customers.
Add percentile aggregations: For distribution metrics, you can optionally generate p50, p75, p90, p95, and p99 percentiles. Percentile metrics are also considered custom metrics, and billed accordingly.