kube-controller log gatherer doesn't limit the number of bytes read. The log can be pretty huge in some long living clusters (tens of MB).
The Insights operator should read the full log every time but rather last few hundreds of KB max.