Read metrics from anywhere your stack lives.
Observer evaluates the same signals your engineers debug with, then publishes the verdict. If your service emits a number, Observer can read it.
Most status pages are a second system. Someone watches a dashboard, decides something is wrong, and updates the page by hand. Observer closes that gap: it reads the metric your team already trusts, applies a threshold you define, and publishes the result. The dashboard and the status page finally agree, because they read the same number.
Five ways in. One way out.
Prometheus, OpenTelemetry, CloudWatch, and two layers of probe. Each one arrives differently. Each one leaves as the same numeric verdict, the only thing a status page ever sees.
- Prometheus
- PromQL against a remote-read endpoint. The same expressions your engineers already debug with.
- OpenTelemetry
- OTLP push from any collector or SDK. Application metrics arrive without a scrape target to maintain.
- AWS CloudWatch
- Pulled with IAM role assumption. RDS, ELB, Lambda, anything AWS already emits on your behalf.
- HTTP probes
- Status code, response time, JSON-path extraction, body match, mutual TLS. The endpoint becomes the metric.
- Network probes
- TCP connect, DNS resolution, TLS certificate expiry, ICMP reachability. Signal from below the application.
Reading the number is the easy part.
Turning a reading into a status the page can trust takes a few deliberate decisions and one rule: nothing flips until it holds.
- 01
Define the threshold
Say what working means as a comparison: p99 under 300ms, error rate under one percent, certificate expiry over 30 days.
- 02
Set a dwell time
Choose how long a breach must hold before it counts. A single noisy sample never flips the page on its own.
- 03
Bind it to a service
Attach the metric to a service, and its verdict surfaces on every status page that service belongs to.
- 04
Let the agent run
Observer evaluates the check every 10 to 30 seconds, inside your network, and pushes only the verdict outward.
- 05
Watch the transitions
A confirmed state change updates the page and, if you want, drafts an incident for on-call to approve.
The same shape, whatever the source.
A few of the signals teams point at Observer, and where each one ends up.
API p99 latency from Prometheus flips the page to Degraded once the latency holds, not on the first spike.
RDS CPU from CloudWatch feeds an SLO burn alert while there is still budget left to spend.
A custom metric pushed over OTLP drives a service health verdict on a single customer's scoped page.
A TLS certificate expiry probe warns 30 days out, long before anyone is paged about it at 2am.