Production incident Friday evening. With Datadog, Slack alert arrives in fifteen minutes — dashboard ready, APM traces. With home Prometheus, Grafana is empty: scrape target changed after last deploy, nobody updated prometheus.yml. Same goal — visibility — two opposing models.
Prometheus: data at yours, full control, ops load. SaaS: time-to-value, bill scaling with metrics.
Prometheus: metric sovereignty
Pull model, local TSDB, Alertmanager, Grafana ecosystem. Strengths: no metric vendor lock-in, controlled retention, rich labels, Kubernetes standard. Weaknesses: non-trivial HA, cardinality explosion, no native APM, you maintain upgrades.
Hosting: dedicated VPS, Kubernetes sidecar, or managed Prometheus at some European clouds. Also see Prometheus metrics for choosing right signals.
SaaS (Datadog, Grafana Cloud, New Relic)
Push agents, template dashboards, often bundled APM/logs. Strengths: fast integrations, log/trace correlation, vendor support. Weaknesses: unpredictable per-host or series cost, data at US third parties by default, query lock-in.
| Criterion | Prometheus | SaaS |
|---|---|---|
| Time-to-value | Weeks | Hours |
| Cost at scale | Fixed infra + ops | Variable, can explode |
| GDPR / residency | You choose | Contract + region |
| APM | External (Jaeger…) | Often included |
Data control: what goes to SaaS
Metrics can contain client names in labels, URLs, order IDs. Logs and traces: frequent personal data. Require EU region, DPA, configurable retention before production. Self-hosted Prometheus in EU: you remain responsible for not labelling PII.
The climax: install speed ≠ incident speed
Decide and move forward without blind spots
Count hosts and estimated series, then simulate twelve-month SaaS bill. Assess who updates targets after each deploy. Choose SaaS if under five services and no SRE; Prometheus if mature Kubernetes. Consider hybrid if sensitive infra local + SaaS aggregates. See guides and comparator.
Frequently asked questions
Is Prometheus really free?
No licence, but compute, storage, ops time, Alertmanager/Grafana HA — and hidden bill if poorly sized.
When is SaaS monitoring more rational?
Small team without SRE, need turnkey APM/integrations, short deadline, predictable budget accepting per-host/series cost.
Does Prometheus suit GDPR?
Yes if EU-hosted and labels without personal data. Metric labels can leak emails/URLs — cardinality discipline required.
Can you combine both?
Yes: local Prometheus + remote write to long-term SaaS, or SaaS alerts + Prometheus for sensitive infra metrics.
Useful monitoring still alerts after the third architecture change — not the one installed on go-live day then forgotten.
