Many teams start with Prometheus and Grafana, then add a logging tool, then a tracing tool, then struggle to connect them. The Grafana LGTM stack brings all of these together under one design philosophy.
This guide explains each component, how they work together and whether to self-host or use Grafana Cloud.
Many teams start with Prometheus and Grafana, then add a logging tool, then a tracing tool, then struggle to connect them.
What is the Grafana LGTM stack?
LGTM is an acronym for four open-source projects from Grafana Labs:
| Component | Signal | What it does |
|---|---|---|
| Loki | Logs | Log aggregation that indexes labels rather than full text |
| Grafana | Visualisation | Dashboards, exploration, alerting across many data sources |
| Tempo | Traces | High-scale distributed tracing backed by object storage |
| Mimir | Metrics | Horizontally scalable, long-term Prometheus-compatible metrics storage |
Many teams also add Pyroscope for continuous profiling and Grafana Alloy, Grafana Labs' OpenTelemetry Collector distribution, for collection.
How does Loki keep log costs low?
Loki indexes only a small set of labels (such as service, environment and namespace) and stores the log content compressed in object storage. Queries filter by labels first, then scan the matching logs. This makes ingestion and storage cheaper than full-text indexing systems, at the cost of designing labels carefully. Too many unique label values hurt performance, the same cardinality problem seen in metrics.
Why use Mimir instead of plain Prometheus?
Prometheus is excellent for scraping and short-term storage on a single server. At scale, teams need long-term retention, high availability and a global view across many clusters. Mimir provides that while remaining compatible with PromQL and Prometheus remote write, so existing dashboards and alerts keep working.
How does Tempo work?
Tempo stores traces in object storage without heavy indexing, making it cost-effective to keep large volumes. Traces are found via trace IDs from logs and metrics (exemplars), or with TraceQL queries. The result is easy navigation: from a metric spike, to an exemplar trace, to the logs for that exact request.
Correlating metrics, traces and logs in Grafana
Illustration in progress
How do the components work together?
The real power is correlation inside Grafana:
- A dashboard shows a latency spike in Mimir metrics.
- An exemplar links to a slow trace in Tempo.
- The trace's span links to its logs in Loki.
- A profile in Pyroscope shows which function consumed the CPU.
This flow is what turns monitoring into observability.
Should you self-host LGTM or use Grafana Cloud?
| Factor | Self-hosted LGTM | Grafana Cloud |
|---|---|---|
| Licence cost | Open source (AGPLv3) | Usage-based subscription |
| Operational effort | High: scaling, upgrades, storage | Managed by Grafana Labs |
| Features | Core projects | Core plus ML, Sift, Assistant, incident, synthetics |
| Control | Full | High, with managed infrastructure |
| Best for | Teams with strong platform engineering capacity | Teams that want to focus on using, not running, observability |
Self-hosting looks free but requires engineers to run a distributed system at scale. For most organisations, the managed option costs less in total once engineering time is counted.
How Crozaint approaches the LGTM stack
Crozaint delivers unified observability on Grafana Cloud using Mimir, Loki and Tempo, with OpenTelemetry instrumentation. In our 8-week engagement we deploy the platform, instrument services, design SLOs and dashboards, engineer high-signal alerting and train your team to run it independently.
We also enable Grafana Cloud's AI features: Grafana ML for anomaly detection, Sift for incident investigation and Grafana Assistant for natural-language queries. AI assists; your engineers stay in control.
Common mistakes to avoid
- Using high-cardinality labels in Loki
- Self-hosting without budgeting engineering time
- Deploying the components without linking exemplars and trace IDs
- Migrating dashboards one-to-one instead of redesigning around SLOs
- Running separate Grafana instances per team with no shared standards
Conclusion
The LGTM stack gives you metrics, logs and traces that work together, built on open standards. Choose managed or self-hosted based on your team's capacity, not just licence cost.
Considering Grafana Cloud? Book a 30-minute discovery call with Crozaint.