Log-based metrics

What they are #

A log-based metric counts, sums or averages values found in your logs on a fixed schedule. It gives you a metric from data you are already sending, with no agent to deploy.

Metric kinds #

• Count — how many records matched. The default, and the one to reach for first.
• Sum — the total of a numeric field across matching records.
• Average — the mean of a numeric field across matching records.

Configuration #

• Name — how the metric appears in the metrics explorer.
• Type — count, sum or average.
• Filter — which logs contribute.
• Field — the numeric attribute to aggregate, for sum and average.
• Group by — the dimensions to break the metric down by.
• Schedule — how often the metric is computed.

{
  "name": "checkout.failed_orders",
  "type": "count",
  "filter": "service:checkout status:error @payment.declined:true",
  "groupBy": ["env", "payment.provider"],
  "schedule": "*/5 * * * *"
}

Counting semantics #

A bucket that receives several records sums their counts; every other kind is averaged. This keeps a five-minute bucket correct whether it received one record or ten thousand.

Grouping dimensions #

Group by any attribute you already send. Grouping by an attribute with a very large number of distinct values produces a metric with a very large number of series, so choose dimensions that stay bounded.

Using them #

The metric appears in the metrics explorer alongside everything else, and can be charted, put on a dashboard and used in monitors.