What | Removed | Added |
---|---|---|
Status | NEW | IN_PROGRESS |
CC | sgrunert@suse.com |
(In reply to Richard Brown from comment #6) > (In reply to Sascha Grunert from comment #5) > > (In reply to Marco Vedovati from comment #4) > > > (In reply to Richard Brown from comment #3) > > > > If you do that you won���t ever be able to install podman on a machine with a > > > > custom chi config, such as one using cri-o/kubernetes > > > > > > > > That seems silly, given one of the use cases for podman is troubleshooting > > > > containers on systems with cri-o/kubernetes > > > > > > Sascha WDYT? > > > > I think we could change the libpod.conf to comment out > > `cni_default_network`. Then podman should select the config with the highest > > priority in `cni_plugin_dir`. Then we could make podman-cni-config > > mandatory, because user would still be able to put higher prio configs in. > > Good theory, in practice this will not work > > Most upstream CNI config files, which of course are generated at container > runtime, are not prioritised > > So CNI tries to load all of them > > So this plan would make Podman toxic to any system > > This is why podman implemented the default CNI network parameter, but it has > not yet been adopted by CRI-O or any other CNI consumer... I might miss a point here, but right now we have the 'issue' that libpod.conf does specify: cni_default_network = "podman". This does not exist in a default installation of podman, right? If there is nothing we can do here because of clashing use cases I would recommend to close this issue and we think about a proper upstream solution to provide stronger defaults.