Which companies set up Prometheus, Grafana and Loki observability as a managed service?
17 Nov 2026 · 6 min read

Short answer: There are two kinds. Hosted platforms such as Grafana Cloud, or Amazon Managed Service for Prometheus with Amazon Managed Grafana, run the backends for you, but your team still instruments the services and writes the dashboards and alerts. Engineering partners such as iDefender set up the whole stack in your own cloud account and keep it running.
Choose hosted if you have engineers who will own instrumentation and alert design. Choose a partner if you want the outcome (useful dashboards and alerts people trust) delivered and maintained.
What each piece does
| Component | Job | Notes |
|---|---|---|
| Prometheus | Collects and stores metrics, evaluates alert rules | Short retention by default; add Thanos, Grafana Mimir or VictoriaMetrics for long-term storage |
| Alertmanager | Groups, deduplicates, silences and routes alerts | Where most alert-fatigue fixes live |
| Grafana | Dashboards and exploration across metrics, logs and traces | Keep dashboards as code (JSON or Terraform), not hand-edited |
| Loki | Stores logs, indexing only labels, not full text | Much cheaper to run than full-text indexing; queried with LogQL |
| Grafana Alloy | Agent that ships logs (and metrics, traces) | Replaces Promtail, which Grafana has deprecated |
Hosted platform or partner: what you actually get
| Hosted platform | Engineering partner (in your account) | |
|---|---|---|
| Who runs the backends | The vendor | The partner, in your cloud |
| Instrumentation of your services | Your team | The partner, with your team |
| Dashboards, SLOs, alert rules | Your team | The partner, as code in your repo |
| Data location | Vendor's region | Your account and region |
| Cost model | Usage-based (series, log volume) | Your infrastructure plus the engagement |
| Lock-in | Low (open formats) but billing scales with volume | Low; everything is open source and in your repo |
Step by step, either way
- Instrument first. Expose metrics from every service (Prometheus client libraries or OpenTelemetry), emit structured JSON logs, and add a request ID that appears in both.
- Install on Kubernetes with the community charts.
kube-prometheus-stackbrings Prometheus, Alertmanager, Grafana and the standard Kubernetes dashboards; Loki has its own Helm chart. - Decide retention. Prometheus for days to weeks; long-term metrics in object storage via Thanos or Mimir; Loki logs in object storage (S3 or equivalent) with a retention period agreed per environment.
- Keep labels under control. High-cardinality labels (user IDs, request IDs) in metrics or Loki labels are the most common cause of cost blow-ups and slow queries. Put them in the log line, not the label.
- Write alerts on symptoms, not causes. Alert on error rate and latency against an SLO; keep CPU and disk as dashboard panels or low-urgency tickets.
- Everything as code. Dashboards, alert rules and routing live in Git and deploy through the pipeline, so a lost Grafana pod loses nothing.
How iDefender does this
We instrument what you already run, build a baseline from your own traffic, agree SLOs with your on-call team and then tune alerting until every page is one a human should act on. A typical baseline takes 4 to 6 weeks. Our core stack is Prometheus, Grafana, ELK and Alertmanager; deliverables are instrumented services, dashboards provisioned as code, SLOs with error budgets and tuned alert routing. See AIOps services or book a call.
Frequently asked questions
Is Loki a replacement for Elasticsearch?
For many teams, yes for logs. Loki indexes only labels, so it's cheaper to store and run, but searches across large time ranges are slower than a full-text index. Teams that need heavy full-text search or analytics on log content often keep Elasticsearch or OpenSearch.
Can Prometheus keep metrics for a year?
Not on its own in a practical way. Add Thanos, Grafana Mimir or VictoriaMetrics to keep long-term metrics cheaply in object storage.
Is Grafana Cloud cheaper than running it ourselves?
At small volumes, usually yes, because you pay nothing for the people running it. At high metric series or log volumes, self-hosting in your own account often costs less, but only if someone owns it.
Do we need OpenTelemetry?
Not to start. Prometheus client libraries and structured logs are enough. OpenTelemetry is worth adopting when you add distributed tracing or want one instrumentation layer for several backends.

