With CRIs different than Docker gaining more and more popularity, we need to improve the user experience of kubeadm by not having to supply --cri-socket on every command if something different than Docker is used.
Most of the other CRIs have well known default paths for Unix domain sockets, thus detecting the default installations of CRIs on the local machine should not be too difficult.
Currently kubeadm will fail if --cri-socket is not specified and Docker is not installed on the system. But the bigger problem here is that, if --cri-socket is forgotten and Docker is installed, the wrong CRI could be used and the operation to succeed (like with kubeadm init).
/area UX
/area ecosystem
/kind feature
With CRIs different than Docker gaining more and more popularity, we need to improve the user experience of kubeadm by not having to supply
--cri-socketon every command if something different than Docker is used.Most of the other CRIs have well known default paths for Unix domain sockets, thus detecting the default installations of CRIs on the local machine should not be too difficult.
Currently kubeadm will fail if
--cri-socketis not specified and Docker is not installed on the system. But the bigger problem here is that, if--cri-socketis forgotten and Docker is installed, the wrong CRI could be used and the operation to succeed (like with kubeadm init)./area UX
/area ecosystem
/kind feature