Datadog, New Relic, and Grafana: how observability becomes a surprise bill

Grafana is not one commercial product

Grafana OSS is a visualization application. The bill sits in Prometheus, Loki, Tempo, object storage, compute, backups, and the person operating them. Grafana Cloud is a managed service with separate meters for metrics, logs, traces, profiles, and solution-specific host or container hours.

Grafana Cloud Pro currently starts with a $19 platform fee. Its published log model separates processing, writing, and retention; the first 50 GB is included, then list rates begin at $0.05, $0.40, and $0.10 per GB respectively. Metrics are measured through active series and data points per minute, not log volume.^[7][8]

Datadog separates log processing from indexing. New Relic primarily prices stored ingest plus users or compute. A dashboard that looks similar across the three products can therefore have a completely different cost driver.

A small logging change can multiply a monthly invoice

New Relic publishes 100 GB of free monthly ingest and $0.40 per GB beyond that for Original Data. On the free edition, ingest and platform access stop after the 100 GB allowance until upgrade or the next month. On a paid edition, the same stream becomes usage-based spend.^[4][5]

Datadog publishes log processing from $0.10 per uncompressed GB on demand and standard 15-day indexing from $1.70 per million events on an annual plan. One hundred million compact log lines can therefore have a much larger indexing line than processing line. Event count and bytes both matter.^[1][2]

The first chart applies published self-service rates to a deliberately simple 30-day model. It excludes users, hosts, APM, support, tax, contracts, compression differences, and historical rehydration.

Cardinality is a multiplier, not a label-count statistic

Datadog defines a custom metric by the combination of metric name and tag values. Grafana Cloud bills metric series using active series and DPM. A label such as user ID, session ID, request ID, or raw URL can turn one useful metric into thousands or millions of series.^[3][8]

The chart is a worst-case combinatorial model for one metric with three environments, three regions, five statuses, and twenty endpoints. Adding tenant and user dimensions creates a different series for every observed combination. Real data is often sparse, but a launch is a bad time to discover that it is not sparse enough.

Grafana Adaptive Metrics and Datadog Metrics without Limits can reduce indexed cardinality. Those tools help after collection has been designed; they do not make arbitrary identifiers good metric labels.^[10][3]

Configure the cost boundary before the dashboards

A Datadog daily index quota is a real hard limit for logs stored in that index, but matching logs can still appear in Live Tail, go to archives, and generate log-derived metrics. It is not a universal account spend cap. Grafana Cloud billing alerts notify after a threshold is exceeded. New Relic ingest drop rules prevent selected data from being stored and billed.^[2][9][6]

Start at the collector: drop health-check noise, redact secrets, sample successful traces, cap attributes, and route high-value audit data separately. Then alert on the provider's own estimated-usage metrics from an independent channel.

• Measure bytes, events, active series, DPM, sampled spans, RUM sessions, and synthetic runs independently.
• Reject user IDs, request IDs, timestamps, and unbounded URLs as metric labels in review or CI.
• Set index quotas and retention by signal class; archive only when restore and query economics are understood.
• Test a temporary DEBUG deployment and verify that cost alarms fire before the planned monthly envelope is consumed.
• Keep a provider-independent raw evidence path for the few signals required to explain a production incident.