If the Kubernetes pod is dead or cannot start, in this case, we cannot access to that pod in order to read its status logs.
However, we can see those logs by reading the master node. You can add a toleration to the agent podspec to get it to deploy on the masters. That said, the master are pretty noisy, so you may want to consider the potential impact on your logging costs.
I found an example for you->
tolerations: - key: "node-role.kubernetes.io/master" operator: "Exists" effect: "NoSchedule"
The daemonset respects the default taints and tolerations set by Kubernetes. If you want to pick up master node logs, you'll need to add a toleration to our default daemonset podspec.