Fluentd starts as the source for collecting logs from all the containers. It uses the same underlying sources that the command kubectl logs uses. Within the cluster, every container that is operating is generating logs that are handled in some fashion by the container runtime, the most common of which is Docker, which maintains log files for every container on each of the hosts.
The Minikube add-on that sets up Fluentd configures it with a ConfigMap, which references where to load these log files, and includes additional rules to annotate the log data with information from Kubernetes. As Fluentd runs, it keeps track of these log files, reading in the data as it is updated from each container, parsing the log file ...
