RED, USE, Golden: When Each Method Fits

Three observability methodologies. RED for services, USE for resources, Golden for general. The decision rule.

RED for services

RED, USE, and Golden are three observability frameworks that capture different perspectives on system health. Each is useful; each fits different system types. The mature observability practice uses all three at appropriate levels; understanding when each applies is the discipline.

What RED provides:

• Rate, Errors, Duration.: RED captures three signals per service or endpoint. Rate is requests per second; Errors is the percentage failing; Duration is the latency distribution. Together they describe the service's behavior.
• Per-endpoint.: RED is naturally per-endpoint. Each route or RPC method has its own RED metrics. The granularity matches the service's structure.
• The natural fit for HTTP and RPC services.: Request-response services produce RED metrics naturally. Most observability platforms support RED out of the box.
• Most user-facing services use RED.: The framework matches the service shape; nearly all user-facing services are HTTP or RPC. RED is the default observability for these.
• Enable by default.: New services should have RED metrics from launch. The team's standard service templates include RED instrumentation; the metrics appear without per-service work.

RED is the framework for service-shaped systems. The metrics capture what users care about: how fast, how often, how reliable.

USE for resources

USE captures resource-level signals. Where RED watches services, USE watches the resources serving them. The two complement each other; using both produces complete coverage.

• Utilisation, Saturation, Errors.: USE captures three signals per resource. Utilization is how much is used; Saturation is how overloaded; Errors are resource-level failures. Together they describe the resource's behavior.
• Per-resource.: Each resource (CPU, memory, disk, queue, connection pool) has its own USE metrics. The granularity matches the resource boundary.
• Catches infrastructure pressure.: Issues at the resource level often precede issues at the service level. USE catches them earlier; the team responds before service degradation.
• Use alongside RED.: USE and RED complement. RED shows service symptoms; USE shows resource causes. The combination supports diagnosis: when RED degrades, USE often shows why.
• RED watches services; USE watches resources serving them.: The two perspectives are layered. Service watching catches user-visible issues; resource watching catches infrastructure issues. Both are necessary.

USE is the framework for resource-shaped systems. The metrics capture infrastructure pressure that causes service issues.

Golden for everything

Golden Signals are the universal four metrics. They work for any system; they are the right starting point when in doubt about which framework to use.

• Latency, Traffic, Errors, Saturation.: Four signals: how slow (latency), how busy (traffic), how broken (errors), how loaded (saturation). The four cover most observability needs.
• The universal four.: The signals work for services and resources both. Latency applies to either; traffic applies to either; errors and saturation similarly. The universality is the value.
• Works on any system.: Whatever the system shape, Golden Signals apply. The framework is the safe default; specialized frameworks (RED, USE) are refinements when the system shape is clear.
• When in doubt, start with Golden.: The team faces a new system or unfamiliar component; start with Golden. The four signals get the team to good observability quickly; specialization can follow.
• Refine into RED or USE.: Once the system shape is clear, the team can refine. Service-shaped systems get RED layered on top; resource-shaped systems get USE; complex systems get both.

RED, USE, and Golden are layered frameworks. Nova AI Ops integrates with observability data, supports all three perspectives, and helps teams produce the right signals for the systems they operate.