Ensure proper metrics registration and lifecycle management to maintain observability system stability. Handle metric registration/unregistration correctly during configuration changes and avoid common pitfalls that can cause panics or data loss. Key practices: 1. **Use AlreadyRegisteredError pattern for shared metrics**: When multiple components share the same metrics, handle registration conflicts gracefully by retrieving existing metrics from the registry instead of creating new ones. 2. **Manage registration during config reloads**: Properly unregister metrics before re-registering to avoid panics, but be aware this causes counter resets. Consider keeping metrics registered and reusing them when possible. 3. **Register all required collectors**: When using custom registries, ensure all necessary collectors (GoCollector, ProcessCollector) are registered, not just the default ones. Example of proper AlreadyRegisteredError handling: ```go func newDiscovererMetrics(reg prometheus.Registerer) *zookeeperMetrics { m := &zookeeperMetrics{ failureCounter: prometheus.NewCounter(prometheus.CounterOpts{ Name: "zookeeper_failures_total", Help: "The total number of ZooKeeper failures.", }), } // Handle shared metrics between components if err := reg.Register(m.failureCounter); err != nil { if are, ok := err.(prometheus.AlreadyRegisteredError); ok { // Use the existing metric instead of the new one m.failureCounter = are.ExistingCollector.(prometheus.Counter) } else { // Handle other registration errors return nil } } return m } ``` This prevents registration panics while ensuring metrics remain functional across component lifecycles and configuration changes.