Incident Management
Platform
AI
Dedicated compute
Dedicated compute keeps concurrent queries stable and complex analysis fast.
Learn moreDedicated compute
Dedicated compute keeps concurrent queries stable and complex analysis fast.
Learn morePrometheus + Grafana coexistence guide
A migration framework for teams that already collect with Prometheus and investigate in Grafana, covering Remote Write, long-term storage, labels, alert parity, and cross-signal workflows.
Guance publishes this guide and is one of the options discussed. Conclusions are limited to public official documentation reviewed on 17 August 2026; no cross-product performance benchmark or like-for-like price test was run.
Scope note: This guide does not assume a managed platform is always better and does not compare cost or performance without identical workloads.
Replay a Kubernetes incident to test whether metrics, logs, traces, and events form one evidence chain.
Short answer
Keep existing exporters, PromQL, rules, dashboards, and alerts while sending a controlled metric subset through Remote Write. Use the original path as the baseline, then test long-term storage, label governance, and whether responders can move from a metric alert to logs, traces, Kubernetes objects, releases, and user impact.
Evaluation criteria
Inventory Prometheus instances, exporters, ServiceMonitors, recording rules, and alerting rules
Measure active series, cardinality, scrape intervals, retention, query peaks, and long-term storage requirements
Record Grafana data sources, dashboards, variables, plugins, permissions, and notification dependencies
Test Remote Write queues, retries, filtering, authentication, TLS, and cross-region network cost
Replay the same alert and compare the path to logs, traces, pods, releases, and user impact
Decision matrix
Scroll horizontally to view the full table on a small screen.
Prometheus documents local storage separately from remote integrations, while Grafana defines data sources as the entry point for querying, visualisation, and alerting. Capture those dependencies before proposing change.
Remote Write reads samples from the WAL into queues and sends them to a receiver. A PoC must observe backlog, retry, throughput, filtering, and time consistency—not merely confirm that a chart appeared.
From a latency, error-rate, or resource alert, verify that responders can reach the relevant service trace, logs, pod events, deployment changes, and user experience. If the workflow is not shorter, a new interface alone is not a reason to migrate.
Migration path
FAQ
Not necessarily. Existing exporters and Prometheus can keep collecting while selected metrics are sent to DataKit through Remote Write. Change collection only when operating ownership or data governance requires it.
You can retain valuable dashboards and team habits. The PoC should first test whether metric evidence connects more naturally to logs, traces, Kubernetes, RUM, and events.
Monitor pending samples, retries, network throughput, filtering, cardinality, timestamps, query parity, and alert parity while preserving the original Prometheus path.
No. Prometheus currently labels 2.0 experimental. Verify sender, receiver, and version compatibility rather than treating it as a default requirement.
Next step
Bring instance counts, active series, rules, dashboards, retention, key alerts, and one past incident. We will help define evidence and rollback thresholds.
