Grafana

Grafana Labs surpassed $400M ARR with 7,000+ enterprise customers as of September 2025. 70% of Fortune 50 companies use Grafana. Key customers include Anthropic, NVIDIA, Salesforce, and Microsoft. Grafana Labs was named a Leader in the 2025 Gartner Magic Quadrant for Observability Platforms (furthest in Completeness of Vision) and ranked #13 on the Forbes Cloud 100. Open-source Grafana has 60K+ GitHub stars with active ongoing development.

LGTM stands for Loki (log aggregation), Grafana (visualization), Tempo (distributed tracing), and Mimir (long-term metrics storage). Together they form a complete open-source observability platform maintained by Grafana Labs. Grafana Alloy (formerly Grafana Agent) collects telemetry from applications and infrastructure and routes it to the appropriate LGTM component. The LGTM stack is the open-source alternative to managed observability platforms like Datadog, designed around OpenTelemetry-native, vendor-neutral instrumentation.

Grafana Cloud is the fully managed SaaS offering — Grafana Labs operates Loki, Tempo, Mimir, and Grafana infrastructure. You send telemetry to Grafana Cloud endpoints; no servers to manage. The free tier includes 50GB logs/month, 10K Prometheus metrics series, and 50GB traces. Self-hosted Grafana requires operating the full stack on your own infrastructure — more control, data sovereignty, but operational overhead for upgrades, storage management, and availability. We recommend Grafana Cloud for teams without dedicated SRE capacity and self-hosted for regulated environments with strict data residency.

High cardinality happens when metric labels have unbounded values — user IDs, request UUIDs, or full URL paths as label values. Prevention: review instrumentation code before deployment, reject labels with cardinality > 1,000 unique values, use recording rules to pre-aggregate high-cardinality series, and set Prometheus memory limits with alerts at 80% utilization. Detection: Grafana's built-in TSDB Status endpoint shows top labels by series count. Remediation: drop high-cardinality labels in Prometheus relabeling, or aggregate them via recording rules before they explode storage.

Grafana open-source is free (MIT license) and self-hosted. Grafana Cloud Free tier: 50GB logs/month, 10K metrics series, 50GB traces/month, 3 users — free forever. Grafana Cloud Pro starts at $29/user/month with higher limits. Grafana Enterprise adds SSO, audit logging, and enterprise data source plugins with negotiated pricing. Self-managed operational cost is infrastructure (Kubernetes cluster for the LGTM stack) plus engineering time. Share your observability requirements and we'll provide deployment and implementation cost estimates.

Three approaches: Grafana's provisioning API (YAML config files mounted to Grafana at startup — dashboards defined in JSON and auto-loaded), Terraform Grafana provider (dashboards defined in Terraform, deployed via CI/CD), and Jsonnet with Grafonnet (programmatic dashboard generation for teams with many similar dashboards). We use Terraform Grafana provider for most engagements — dashboards are version-controlled alongside other infrastructure, reviewed in PRs, and deployed automatically. Dashboard JSON is exported from Grafana UI and committed to Git.

Grafana Loki indexes only log labels (not the full log line), making storage dramatically cheaper than Elasticsearch at high log volume. Loki is designed for Kubernetes workloads — auto-labeling from pod metadata, structured LogQL queries, and native Grafana integration. Elasticsearch indexes every word in every log line, enabling full-text search that Loki's label-based approach can't match. Choose Loki for cost-effective Kubernetes log aggregation and correlation with Prometheus metrics. Choose Elasticsearch/OpenSearch when full-text search, complex log parsing, or existing Elasticsearch investment are requirements.

Applications emit trace spans via OpenTelemetry SDKs — every service call records start time, duration, error status, and attributes. Grafana Alloy collects traces and forwards to Tempo. Grafana Tempo stores traces indexed by trace ID; TraceQL queries search by service name, duration, error status, or custom attributes. In Grafana, exemplars on Prometheus metrics link data points to specific traces — you see a P99 latency spike, click the exemplar, and see the exact trace with all spans that caused it.

We define SLIs as Prometheus recording rules (success_rate = success_requests / total_requests), SLO targets as Grafana thresholds, and error budgets as remaining margin (SLO target - current success rate). Grafana panels visualize rolling 28-day SLO compliance, hourly success rate, and multi-window burn rate alerts. Alertmanager fires when burn rate predicts SLO breach — 1-hour window catching fast burns, 6-hour window catching slow burns. We follow the Google SRE Book patterns and provide dashboard templates adapted to your specific SLI definitions.

We offer Grafana managed support covering dashboard maintenance and optimization (query performance tuning, panel redesign as systems evolve), Prometheus cardinality reviews and recording rule additions, Loki label schema evolution as new services are added, Alertmanager routing updates for team structure changes, Grafana and LGTM component upgrades, and on-call runbook updates. We also provide Grafana training for engineering teams new to PromQL, LogQL, and dashboard design.