They should be handled as
-
Performance: Not critical in control
-
Logging: For control plane
- Use sugared logger.
- Prefer log level=INFO in order to form an audit log. If a running setup suddenly degrades, it is likely to coincide with one of the control operations. Emit relevant state wrt the control operation. For data plane stick to metrics for hot path. For exceptional we don't care about performance but so many logs during some failure will stress the system unnecessarily. Use throttled logs or metrics.