Catalog
concept#Architecture#Software Engineering#Observability#Reliability

Feedback Control

A control-theory concept that uses feedback to regulate systems, ensuring stability, setpoint achievement, and disturbance rejection.

Feedback control is a conceptual pattern from control theory that measures and adjusts system behavior via feedback to achieve setpoints or reject disturbances.
Established
Medium

Classification

  • Medium
  • Technical
  • Architectural
  • Intermediate

Technical context

Monitoring and APM systems (e.g. Prometheus, Datadog)Orchestration and auto-scaling mechanismsConfiguration and feature-flag systems

Principles & goals

Measure before acting: decisions are based on observable metrics.Clearly defined setpoints and tolerances.Layered application: local control loops with higher-level coordination.
Build
Enterprise, Domain, Team

Use cases & scenarios

Compromises

  • Stable systems can be destabilized by incorrect rules.
  • Lack of transparency in automated interventions complicates root cause analysis.
  • Dependence on measurement infrastructure introduces new failure points.
  • Roll out incrementally and test first in non-critical areas.
  • Introduce comprehensive observability and explainable logs for decisions.
  • Configure conservative default parameters and safety limits.

I/O & resources

  • Real-time metrics and telemetry
  • Defined setpoints and SLOs
  • Mechanisms to execute control commands (actuators)
  • Control commands to actuators or configurations
  • Metrics for evaluating control behavior
  • Logs and audit trails of interventions

Description

Feedback control is a conceptual pattern from control theory that measures and adjusts system behavior via feedback to achieve setpoints or reject disturbances. It defines measurement, comparison and corrective-action loops and supports stability and robustness. It is applicable across software architecture, operations and product quality.

✔

  • Increased stability through automatic disturbance compensation.
  • Improved SLO adherence via continuous adjustment.
  • Reduced manual intervention and faster response times.

✖

  • Requires reliable, clean metrics; poor data causes miscontrol.
  • Designing controllers can be complex and technically demanding.
  • Overreaction and oscillation possible with unsuitable parameters.

  • Control error

    Difference between setpoint and actual value over time; core metric to evaluate control quality.

  • Control-loop latency

    Time between measurement and execution of corrective action; affects stability and responsiveness.

  • Oscillation/overshoot

    Amplitude and frequency of oscillations after a disturbance; indicator of over-control.

PID loop for latency control

A simple PID controller adjusts request rate to a service based on measured latency.

SLO-based autoscaler

An autoscaler uses SLO deviations as feedback to dynamically increase or decrease pods.

Queue length regulation

A controller reduces input to a queue when lengths reach a critical threshold.

1

Instrumentation: Define and implement relevant metrics and measurement points.

2

Design: Choose setpoints, tolerances and control strategy (e.g. PID, throttling).

3

Integration: Connect actuators and implement safe interfaces.

4

Validation: Test behavior under load and disturbances and fine-tune parameters.

⚠️ Technical debt & bottlenecks

  • Missing or insufficient telemetry hinders later adoption.
  • Hardcoded control parameters in legacy components.
  • No versioning of control strategies and missing tests.
Measurement latency can limit control qualityUnreliable metrics lead to miscontrolLimited actuator capacity prevents fast response
  • Auto-scaling solely based on user count without considering latency
  • Controller with too high gain causing persistent oscillations
  • Decisions made without audit logging, hiding root causes
  • Ignoring delays in measurement and actuator paths.
  • Non-robust parameters that only work in lab conditions.
  • Neglecting safety and compliance aspects of interventions.
Basics of control theory and system dynamicsObservability and monitoring skillsExperience with system architecture and operational automation
Availability and quality of telemetryRequirement for stable response timesScalability of actuators and control instances
  • • Reliable measurement infrastructure must be available.
  • • Control algorithms must not jeopardize system stability.
  • • Privacy and access rights for metric data must be respected.