Which is most important?

It depends on user impact. Active user-facing errors are usually the highest priority, but latency and saturation can be just as urgent when they prevent users from completing work.

Runframe

MONITORING & OBSERVABILITY

The Four Golden Signals

The four key metrics that represent the health of a system: Latency, Traffic, Errors, and Saturation.

Four Golden Signals

The four key metrics that represent the health of a system: Latency, Traffic, Errors, and Saturation.

"The Vital Signs"

Just as a doctor checks heart rate and blood pressure, an SRE checks the Four Golden Signals.

1. Latency

The time it takes to service a request.

Tip: Distinguish between latency for successful requests and latency for failed requests; mixing them can hide what users actually experience.

2. Traffic

A measure of how much demand is being placed on your system.

Web: Requests per second (RPS).
Audio: Concurrent streams.

3. Errors

The rate of requests that fail.

Explicit: HTTP 500s.
Implicit: HTTP 200s with "Success: False" body (content errors).

4. Saturation

How "full" your service is.

CPU usage, Memory, Disk I/O.
Many systems degrade before they reach 100% utilization, so watch constrained resources and set utilization targets before latency spikes.

ExThe Slow Disk

“A service was slow, but CPU and Memory were low. No errors were firing.”

Impact

Latency increased from 100ms to 2s.

Resolution

The team checked "Saturation" and found Disk I/O was at 100%. A logging process was spamming the disk. They throttled the logger, and latency recovered.