Logging

Important operations in a service should always be logged. Logging information needs to be categorized to be of a real value. A common list of categories is debug, info, warning, error, and fatal. Logging information should be collected by a central log aggregation service and not be stored on an individual node of the cluster. Aggregated logs are easy to parse and filter for relevant information.